snapshot of https://github.com/seaweedfs/seaweedfs.git (master, 2026-09-16; blobs > 131072B omitted)

seaweed-volume/Cargo.toml

@@ -0,0 +1,157 @@
+[package]
+name = "weed-volume"
+version = "0.1.0"
+edition = "2024"
+# The edition needs 1.85; the dependency tree needs more. Verified with
+# `cargo +1.91.1 check --all-targets` (1.90 fails on the AWS SDK).
+rust-version = "1.91.1"
+description = "SeaweedFS Volume Server — Rust implementation"
+
+[lib]
+name = "seaweed_volume"
+
+[[bin]]
+name = "weed-volume"
+path = "src/main.rs"
+
+[features]
+# Default: 5-byte offsets (8TB max volume size), matching production Go builds (-tags 5BytesOffset).
+# Disable with --no-default-features for 4-byte offsets (32GB max volume size).
+default = ["5bytes"]
+5bytes = []
+# Unstable redb cursor bulk-load for full_rebuild. Off in production.
+# Pulls redb's experimental_cursor (and therefore experimental-api-5).
+redb-experimental-cursor = ["redb/experimental_cursor"]
+
+[lints.clippy]
+# Protobuf message literals keep `..Default::default()` on purpose: it is
+# what lets a proto gain a field without touching every constructor.
+needless_update = "allow"
+
+[dependencies]
+# Async runtime
+tokio = { version = "1", features = ["full"] }
+tokio-stream = { version = "0.1", features = ["net"] }
+tokio-io-timeout = "1"
+
+# gRPC + protobuf
+tonic = { version = "0.14", features = ["tls-aws-lc"] }
+tonic-prost = "0.14"
+tonic-reflection = "0.14"
+prost = "0.14"
+prost-types = "0.14"
+
+# HTTP server
+axum = { version = "0.8", features = ["multipart"] }
+http-body = "1"
+hyper = { version = "1", features = ["full"] }
+hyper-util = { version = "0.1", features = ["tokio", "service", "server-auto", "http1", "http2"] }
+tower = { version = "0.5", features = ["util"] }
+tower-http = { version = "0.6", features = ["cors", "trace"] }
+
+# CLI
+clap = { version = "4", features = ["derive"] }
+
+# Metrics
+prometheus = { version = "0.13", default-features = false, features = ["process"] }
+
+# JWT
+jsonwebtoken = { version = "10", features = ["rust_crypto"] }
+
+# TLS
+rustls = "0.23"
+tokio-rustls = "0.26"
+rustls-pemfile = "2"
+
+# LevelDB (via RocksDB for better Rust support)
+# Using rusty-leveldb for pure Rust LevelDB
+rusty-leveldb = "3"
+
+# Disk-backed needle map (alternative to in-memory HashMap)
+redb = "4"
+
+# Reed-Solomon erasure coding
+reed-solomon-erasure = "6"
+
+# Logging
+tracing = "0.1"
+tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+# Config
+toml = "0.8"
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
+serde_urlencoded = "0.7"
+
+# CRC32 — using Castagnoli polynomial (CRC32-C), matching Go's crc32.Castagnoli
+crc32c = "0.6"
+crc32fast = "1"
+
+# xxhash64 - must match Go's cespare/xxhash for the heartbeat volume digest
+xxhash-rust = { version = "0.8", features = ["xxh64"] }
+
+# Memory-mapped files
+memmap2 = "0.9"
+
+# UUID
+uuid = { version = "1", features = ["v4"] }
+
+# HTTP client (for proxying, remote fetch)
+reqwest = { version = "0.12", default-features = false, features = ["rustls-tls", "stream", "multipart", "json"] }
+
+# Content hashing
+md-5 = "0.10"
+base64 = "0.22"
+
+# Compression
+flate2 = "1"
+
+# Image processing
+image = { version = "0.25", default-features = false, features = ["png", "jpeg", "gif", "webp"] }
+kamadak-exif = "0.5"
+
+# Multipart form-data parsing
+multer = "3"
+
+# MIME type guessing from file extensions
+mime_guess = "2"
+
+# Misc
+bytes = "1"
+rand = "0.10"
+chrono = "0.4"
+hex = "0.4"
+parking_lot = "0.12"
+dashmap = "6"
+thiserror = "1"
+anyhow = "1"
+async-trait = "0.1"
+futures = "0.3"
+async-stream = "0.3"
+x509-parser = "0.16"
+
+# Disk space checking
+sysinfo = "0.31"
+libc = "0.2"
+
+# AWS S3 SDK (for remote storage backends)
+aws-config = { version = "1", features = ["behavior-version-latest"] }
+aws-sdk-s3 = { version = "1.125.0", default-features = false, features = ["sigv4a", "http-1x", "default-https-client", "rt-tokio"] }
+aws-credential-types = "1"
+aws-types = "1"
+
+# pprof is Unix-only (requires libc/nix APIs not available on Windows)
+[target.'cfg(unix)'.dependencies]
+pprof = { version = "0.15", features = ["prost-codec"] }
+
+[dev-dependencies]
+tempfile = "3"
+
+[build-dependencies]
+tonic-prost-build = "0.14"
+# Ships protoc with the build so neither CI nor a developer needs a system
+# install, and so the version is pinned rather than whatever the platform's
+# package manager happens to carry.
+protoc-bin-vendored = "3"
+
+[patch.crates-io]
+reed-solomon-erasure = { path = "vendor/reed-solomon-erasure" }

seaweed-volume/DEV_PLAN.md

@@ -0,0 +1,105 @@
+# Rust Volume Server — Dev Plan
+
+## Current Status (2026-03-07)
+
+**HTTP tests**: 53/53 pass (100%)
+**gRPC tests**: 56/56 pass (100%) — includes TestVolumeMoveHandlesInFlightWrites with Rust multi-volume cluster
+**Rust integration tests**: 8/8 pass
+**S3 remote storage tests**: 3/3 pass
+**Total**: 117/117 (100%) + 8 Rust + 3 S3 tests
+**Rust unit tests**: 137 lib + 7 integration = 144
+
+## Completed Features
+
+All phases from the original plan are complete:
+
+- **Phase 1** — HTTP Core: CORS, OPTIONS, unsupported methods, static assets, path routing,
+  cookie validation, conditional headers, range requests, dedup 204, content-encoding,
+  readDeleted, chunk manifests, multipart validation, MD5 check, file size limit,
+  upload/download throttling, image resize/crop, download disposition
+- **Phase 2** — JWT/Security: signing keys from security.toml, token source precedence
+  (query > header > cookie), file_id claims, leeway=0
+- **Phase 3** — gRPC: maintenance mode, error message parity, ping routing, batch delete,
+  VolumeServerStatus (with real disk stats, data_center, rack), ReadVolumeFileStatus
+  (with timestamps)
+- **Phase 4** — Streaming gRPC: VolumeIncrementalCopy, CopyFile, ReceiveFile, ReadAllNeedles,
+  VolumeTailSender, VolumeCopy, VolumeTailReceiver, VacuumVolumeCheck
+- **Phase 5** — EC Shards: mount/unmount, delete, read, blob delete, rebuild, shards-to-volume,
+  copy, info
+- **Phase 6** — Advanced gRPC: ScrubVolume, ScrubEcVolume, Query, FetchAndWriteNeedle,
+  VolumeTierMoveDat (error paths)
+- **Phase 7** — Remote Storage: S3-compatible backend via aws-sdk-s3,
+  FetchAndWriteNeedle reads from S3/MinIO/SeaweedFS S3 and writes locally.
+  Supports all S3-compatible providers (AWS, Wasabi, Backblaze, Aliyun, etc.)
+- **Master Heartbeat** — Bidirectional streaming SendHeartbeat RPC, volume/EC registration,
+  leader changes, shutdown deregistration. Tested end-to-end with Go master.
+- **Production Sprint 1** — Quick wins:
+  - VolumeMarkReadonly master notification (triggers immediate heartbeat)
+  - Compaction throttling (`maybe_throttle_compaction()`)
+  - File size limit enforcement on upload
+  - `ts` query param for custom timestamps (upload + delete)
+  - TTL expiration check (was already implemented)
+  - Health check heartbeat status (returns 503 if disconnected from master)
+  - preStopSeconds graceful drain before shutdown
+  - S3 response passthrough headers (content-encoding, expires, content-language, content-disposition)
+  - .vif persistence for readonly state across restarts
+  - Webp image support for resize
+- **Production Sprint 2** — Compatibility:
+  - MIME type extraction from Content-Type header
+  - Stats endpoints (/stats/counter, /stats/memory, /stats/disk)
+  - JSON pretty print (?pretty=y) and JSONP (?callback=fn)
+  - Request ID generation (UUID if x-amz-request-id missing)
+  - Advanced Prometheus metrics (INFLIGHT_REQUESTS, VOLUME_FILE_COUNT gauges)
+- **Production Sprint 3** — Streaming & Multi-node:
+  - Streaming reads for large files (>1MB) via http_body::Body trait with spawn_blocking
+  - Meta-only needle reads (NeedleStreamInfo) to avoid loading full body for streaming
+  - Multi-volume Rust cluster support (RustMultiVolumeCluster test framework)
+  - TestVolumeMoveHandlesInFlightWrites now uses Rust volume servers
+  - CI skip list cleaned up (all tests pass with Rust)
+
+- **Production Sprint 4** — Advanced Features:
+  - BatchDelete EC shard support (ecx index lookup + ecj journal deletion)
+  - JPEG EXIF orientation auto-fix on upload (kamadak-exif + image crate)
+  - Async batched write processing (mpsc queue, up to 128 entries per batch)
+  - VolumeTierMoveDatToRemote/FromRemote (S3 multipart upload/download)
+  - S3TierRegistry for managing remote storage backends
+  - VolumeInfo (.vif) persistence for remote file references
+- **Production Sprint 5** — Upload Compatibility:
+  - TTL query parameter extraction during upload (`ttl=3m`)
+  - Auto-compression for compressible file types (text/*, .js, .css, .json, .svg, etc.)
+  - Seaweed-* custom metadata headers stored as needle pairs (JSON, max 64KB)
+  - Filename extraction from URL path stored in needle name field
+  - Upload response includes filename
+
+- **Production Sprint 6** — Storage & Networking:
+  - Redb disk-backed needle maps (pure Rust, no C deps) via `NeedleMap` enum
+  - Binary search for `VolumeIncrementalCopy` with `since_ns > 0`
+  - Proxy/redirect read modes for non-local volumes (master lookup, HTTP proxy, 301 redirect)
+
+## Remaining Work (Production Readiness)
+
+No major remaining items. All phases and production sprints are complete.
+
+## Test Commands
+
+```bash
+# Build
+cd seaweed-volume && cargo build --release
+
+# Run all Go integration tests with Rust volume server
+VOLUME_SERVER_IMPL=rust go test -v -count=1 -timeout 1200s ./test/volume_server/grpc/... ./test/volume_server/http/...
+
+# Run S3 remote storage tests
+VOLUME_SERVER_IMPL=rust go test -v -count=1 -timeout 180s -run "TestFetchAndWriteNeedle(FromS3|S3NotFound)" ./test/volume_server/grpc/...
+
+# Run specific test
+VOLUME_SERVER_IMPL=rust go test -v -count=1 -timeout 60s -run "TestName" ./test/volume_server/http/...
+
+# Run Rust unit tests
+cd seaweed-volume && cargo test
+
+# Test heartbeat with Go master
+weed master -port=9333 &
+seaweed-volume --port 8080 --master localhost:9333 --dir /tmp/vol1 --max 7
+curl http://localhost:9333/dir/status  # should show Rust volume server registered
+```

seaweed-volume/MISSING_FEATURES.md

@@ -0,0 +1,288 @@
+# Rust Volume Server — Missing Features Audit
+
+Comprehensive line-by-line comparison of Go vs Rust volume server.
+Generated 2026-03-07 from 4 parallel audits covering HTTP, gRPC, storage, and infrastructure.
+
+## Executive Summary
+
+| Area | Total Features | Implemented | Partial | Missing |
+|------|---------------|-------------|---------|---------|
+| gRPC RPCs | 48 | 43 (90%) | 2 (4%) | 3 (6%) |
+| HTTP Handlers | 31 | 12 (39%) | 10 (32%) | 9 (29%) |
+| Storage Layer | 22 | 6 (27%) | 7 (32%) | 9 (41%) |
+| Infrastructure | 14 | 5 (36%) | 4 (29%) | 5 (36%) |
+
+---
+
+## Priority 1 — Critical for Production
+
+### P1.1 Streaming / Meta-Only Reads
+- **Go**: `ReadNeedleMeta()`, `ReadNeedleData()`, `ReadPagedData()` — reads only metadata or pages of large files
+- **Go**: `streamWriteResponseContent()` streams needle data in chunks
+- **Go**: `AttemptMetaOnly` / `MustMetaOnly` flags in `ReadOption`
+- **Rust**: Reads entire needle into memory always
+- **Impact**: OOM on large files; 8MB file = 8MB heap per request
+- **Files**: `weed/storage/needle/needle_read.go`, `weed/server/volume_server_handlers_read.go`
+- **Effort**: Medium
+
+### P1.2 Download Proxy/Redirect Fallback (ReadMode)
+- **Go**: `ReadMode` config: "local" | "proxy" | "redirect"
+- **Go**: `tryProxyToReplica()` probes replicas, `proxyReqToTargetServer()` streams response
+- **Rust**: Always returns 404 for non-local volumes
+- **Impact**: Clients must handle volume placement themselves; breaks transparent replication
+- **Files**: `weed/server/volume_server_handlers_read.go:138-250`
+- **Effort**: Medium
+
+### P1.3 TLS/HTTPS Support
+- **Go**: `LoadServerTLS()`, `LoadClientTLS()`, cert/key loading from security.toml
+- **Go**: Applied to both HTTP and gRPC servers
+- **Rust**: No TLS at all — plain TCP only
+- **Impact**: Cannot deploy in secure clusters
+- **Files**: `weed/security/tls.go`, `weed/command/volume.go`
+- **Effort**: Medium (rustls + tokio-rustls already in Cargo.toml)
+
+### P1.4 VolumeMarkReadonly/Writable Master Notification
+- **Go**: `notifyMasterVolumeReadonly()` updates master with readonly state
+- **Rust**: Only sets local in-memory flag
+- **Impact**: Master keeps directing writes to readonly volume
+- **Files**: `weed/server/volume_grpc_admin.go`
+- **Effort**: Low
+
+### P1.5 Compaction/Maintenance Throttling
+- **Go**: `WriteThrottler` with `MaybeSlowdown()` for MB/s rate limiting
+- **Rust**: Flags parsed but no throttle implementation
+- **Impact**: Compaction/copy operations can saturate disk IO
+- **Files**: `weed/util/throttler.go`
+- **Effort**: Low
+
+### P1.6 File Size Limit Enforcement
+- **Go**: `fileSizeLimitBytes` checked on upload, returns 400
+- **Rust**: No enforcement — accepts any size
+- **Impact**: Can write files larger than volume size limit
+- **Files**: `weed/server/volume_server_handlers_write.go`
+- **Effort**: Low
+
+---
+
+## Priority 2 — Important for Compatibility
+
+### P2.1 `ts` Query Param (Custom Timestamps)
+- **Go**: Upload and delete accept `ts` query param for custom Last-Modified time
+- **Rust**: Always uses current time
+- **Impact**: Replication timestamp fidelity; sync from external sources
+- **Files**: `weed/server/volume_server_handlers_write.go`, `volume_server_handlers_admin.go`
+- **Effort**: Low
+
+### P2.2 Multipart Form Upload Parsing
+- **Go**: `needle.CreateNeedleFromRequest()` parses multipart forms, extracts MIME type, custom headers/pairs
+- **Rust**: Reads raw body bytes only — no multipart form parsing for metadata
+- **Impact**: MIME type not stored; custom needle pairs not supported
+- **Files**: `weed/storage/needle/needle.go:CreateNeedleFromRequest`
+- **Effort**: Medium
+
+### P2.3 JPEG Orientation Auto-Fix
+- **Go**: `images.FixJpgOrientation()` on upload when enabled
+- **Rust**: Not implemented (flag exists but unused)
+- **Impact**: Mobile uploads may display rotated
+- **Files**: `weed/images/orientation.go`
+- **Effort**: Low (exif crate)
+
+### P2.4 TTL Expiration Enforcement
+- **Go**: Checks `HasTtl()` + `AppendAtNs` against current time on read path
+- **Rust**: TTL struct exists but no expiration checking
+- **Impact**: Expired needles still served
+- **Files**: `weed/storage/needle/volume_ttl.go`, `weed/storage/volume_read.go`
+- **Effort**: Low
+
+### P2.5 Health Check — Master Heartbeat Status
+- **Go**: Returns 503 if not heartbeating (can't reach master)
+- **Rust**: Only checks `is_stopping` flag
+- **Impact**: Load balancers won't detect disconnected volume servers
+- **Files**: `weed/server/volume_server.go`
+- **Effort**: Low
+
+### P2.6 Stats Endpoints
+- **Go**: `/stats/counter`, `/stats/memory`, `/stats/disk` (whitelist-guarded)
+- **Rust**: Not implemented
+- **Impact**: No operational visibility
+- **Files**: `weed/server/volume_server.go`
+- **Effort**: Low
+
+### P2.7 Webp Image Support
+- **Go**: `.webp` included in resize-eligible extensions
+- **Rust**: Only `.png`, `.jpg`, `.jpeg`, `.gif`
+- **Impact**: Webp images can't be resized on read
+- **Files**: `weed/server/volume_server_handlers_read.go`
+- **Effort**: Low (add webp feature to image crate)
+
+### P2.8 preStopSeconds Graceful Drain
+- **Go**: Stops heartbeat, waits N seconds, then shuts down servers
+- **Rust**: Immediate shutdown on signal
+- **Impact**: In-flight requests dropped; Kubernetes readiness race
+- **Files**: `weed/command/volume.go`
+- **Effort**: Low
+
+### P2.9 S3 Response Passthrough Headers
+- **Go**: `response-content-encoding`, `response-expires`, `response-content-language` query params
+- **Rust**: Only handles `response-content-type`, `response-cache-control`, `dl`
+- **Impact**: S3-compatible GET requests missing some override headers
+- **Files**: `weed/server/volume_server_handlers_read.go`
+- **Effort**: Low
+
+---
+
+## Priority 3 — Storage Layer Gaps
+
+### P3.1 LevelDB Needle Maps
+- **Go**: 5 needle map variants: memory, LevelDB, LevelDB-medium, LevelDB-large, sorted-file
+- **Rust**: Memory-only needle map
+- **Impact**: Large volumes (millions of needles) require too much RAM
+- **Files**: `weed/storage/needle_map_leveldb.go`
+- **Effort**: High (need LevelDB binding or alternative)
+
+### P3.2 Async Request Processing
+- **Go**: `asyncRequestsChan` with 128-entry queue, worker goroutine for batched writes
+- **Rust**: All writes synchronous
+- **Impact**: Write throughput limited by fsync latency
+- **Files**: `weed/storage/needle/async_request.go`
+- **Effort**: Medium
+
+### P3.3 Volume Scrubbing (Data Integrity)
+- **Go**: `ScrubIndex()`, `scrubVolumeData()` — full data + index verification
+- **Rust**: Stub only in gRPC (returns OK without actual scrubbing)
+- **Impact**: No way to verify data integrity
+- **Files**: `weed/storage/volume_checking.go`, `weed/storage/idx/check.go`
+- **Effort**: Medium
+
+### P3.4 Volume Backup / Sync
+- **Go**: Streaming backup, binary search for last modification, index generation scanner
+- **Rust**: Not implemented
+- **Impact**: No backup/restore capability
+- **Files**: `weed/storage/volume_backup.go`
+- **Effort**: Medium
+
+### P3.5 Volume Info (.vif) Persistence
+- **Go**: `.vif` files store tier/remote metadata, readonly state persists across restarts
+- **Rust**: No `.vif` support; readonly is in-memory only
+- **Impact**: Readonly state lost on restart; no tier metadata
+- **Files**: `weed/storage/volume_info/volume_info.go`
+- **Effort**: Low
+
+### P3.6 Disk Location Features
+- **Go**: Directory UUID tracking, disk space monitoring, min-free-space enforcement, tag-based grouping
+- **Rust**: Basic directory only
+- **Impact**: No disk-full protection
+- **Files**: `weed/storage/disk_location.go`
+- **Effort**: Medium
+
+### P3.7 Compact Map (Memory-Efficient Needle Map)
+- **Go**: `CompactMap` with overflow handling for memory optimization
+- **Rust**: Uses standard HashMap
+- **Impact**: Higher memory usage for index
+- **Files**: `weed/storage/needle_map/compact_map.go`
+- **Effort**: Medium
+
+---
+
+## Priority 4 — Nice to Have
+
+### P4.1 gRPC: VolumeTierMoveDatToRemote / FromRemote
+- **Go**: Full streaming implementation for tiering volumes to/from S3
+- **Rust**: Stub returning error
+- **Files**: `weed/server/volume_grpc_tier_upload.go`, `volume_grpc_tier_download.go`
+- **Effort**: High
+
+### P4.2 gRPC: Query (S3 Select)
+- **Go**: JSON/CSV query over needle data (S3 Select compatible)
+- **Rust**: Stub returning error
+- **Files**: `weed/server/volume_grpc_query.go`
+- **Effort**: High
+
+### P4.3 FetchAndWriteNeedle — Already Implemented
+- **Note**: The gRPC audit incorrectly flagged this as missing. It was implemented in a prior session with full S3 remote storage support.
+
+### P4.4 JSON Pretty Print + JSONP
+- **Go**: `?pretty` query param for indented JSON; `?callback=fn` for JSONP
+- **Rust**: Neither supported
+- **Effort**: Low
+
+### P4.5 Request ID Generation
+- **Go**: Generates UUID if `x-amz-request-id` header missing, propagates to gRPC context
+- **Rust**: Only echoes existing header
+- **Effort**: Low
+
+### P4.6 UI Status Page
+- **Go**: Full HTML template with volumes, disks, stats, uptime
+- **Rust**: Stub HTML
+- **Effort**: Medium
+
+### P4.7 Advanced Prometheus Metrics
+- **Go**: InFlightRequestsGauge, ConcurrentUploadLimit/DownloadLimit gauges, metrics push gateway
+- **Rust**: Basic request counter and histogram only
+- **Effort**: Low
+
+### P4.8 Profiling (pprof)
+- **Go**: CPU/memory profiling, /debug/pprof endpoints
+- **Rust**: Flags parsed but not wired
+- **Effort**: Medium (tokio-console or pprof-rs)
+
+### P4.9 EC Distribution / Rebalancing
+- **Go**: 17 files for EC operations including placement strategies, recovery, scrubbing
+- **Rust**: 6 files with basic encoder/decoder
+- **Effort**: High
+
+### P4.10 Cookie Mismatch Status Code
+- **Go**: Returns 406 Not Acceptable
+- **Rust**: Returns 400 Bad Request
+- **Effort**: Trivial
+
+---
+
+## Implementation Order Recommendation
+
+### Sprint 1 — Quick Wins (Low effort, high impact) ✅ DONE
+1. ✅ P1.4 VolumeMarkReadonly master notification — triggers immediate heartbeat
+2. ✅ P1.5 Compaction throttling — `maybe_throttle_compaction()` method added
+3. ✅ P1.6 File size limit enforcement — checks `file_size_limit_bytes` on upload
+4. ✅ P2.1 `ts` query param — custom timestamps for upload and delete
+5. ✅ P2.4 TTL expiration check — was already implemented
+6. ✅ P2.5 Health check heartbeat status — returns 503 if not heartbeating
+7. ✅ P2.8 preStopSeconds — graceful drain delay before shutdown
+8. ✅ P2.9 S3 passthrough headers — content-encoding, expires, content-language, content-disposition
+9. ✅ P3.5 .vif persistence — readonly state persists across restarts
+10. ✅ P2.7 Webp support — added to image resize-eligible extensions
+11. ~~P4.10 Cookie 406~~ — Go actually uses 404 for HTTP cookie mismatch (406 is gRPC batch delete only)
+
+### Sprint 2 — Core Read Path (Medium effort) — Partially Done
+1. P1.1 Streaming / meta-only reads — TODO (medium effort, no test coverage yet)
+2. ✅ P1.2 ReadMode proxy/redirect — was already implemented and tested
+3. ✅ P2.2 Multipart form parsing — MIME type extraction from Content-Type header
+4. P2.3 JPEG orientation fix — TODO (low effort, needs exif crate)
+5. ✅ P2.6 Stats endpoints — /stats/counter, /stats/memory, /stats/disk
+6. ✅ P2.7 Webp support — done in Sprint 1
+7. ✅ P4.4 JSON pretty print + JSONP — ?pretty=y and ?callback=fn
+8. ✅ P4.5 Request ID generation — generates UUID if x-amz-request-id missing
+9. ✅ P4.7 Advanced Prometheus metrics — INFLIGHT_REQUESTS gauge, VOLUME_FILE_COUNT gauge
+
+### Sprint 3 — Infrastructure (Medium effort) — Partially Done
+1. ✅ P1.3 TLS/HTTPS — rustls + tokio-rustls for HTTP, tonic ServerTlsConfig for gRPC
+2. P3.2 Async request processing — TODO (medium effort)
+3. ✅ P3.3 Volume scrubbing — CRC checksum verification of all needles
+4. ✅ P3.6 Disk location features — MinFreeSpace enforcement, background disk monitor
+
+### Sprint 4 — Storage Advanced (High effort) — Deferred
+No integration test coverage for these items. All existing tests pass.
+1. P3.1 LevelDB needle maps — needed only for volumes with millions of needles
+2. P3.4 Volume backup/sync — streaming backup, binary search
+3. P3.7 Compact map — memory optimization for needle index
+4. P4.1 VolumeTierMoveDat — full S3 tiering (currently error stub)
+5. P4.9 EC distribution — advanced EC placement/rebalancing
+
+### Sprint 5 — Polish — Deferred
+No integration test coverage for these items.
+1. P4.2 Query (S3 Select) — JSON/CSV query over needle data
+2. ✅ P4.4 JSON pretty/JSONP — done in Sprint 2
+3. ✅ P4.5 Request ID generation — done in Sprint 2
+4. P4.6 UI status page — HTML template with volume/disk/stats info
+5. ✅ P4.7 Advanced metrics — done in Sprint 2
+6. P4.8 Profiling — pprof-rs or tokio-console

seaweed-volume/PARITY_PLAN.md

@@ -0,0 +1,230 @@
+# Rust Volume Server Parity Plan
+
+Generated: 2026-03-16
+
+## Goal
+
+Make `seaweed-volume` a drop-in replacement for the Go volume server by:
+
+- comparing every Go volume-server code path against the Rust implementation,
+- recording file-level ownership and verification status,
+- closing verified behavior gaps one logic change per commit,
+- extending tests so regressions are caught by Go parity suites and Rust unit/integration tests.
+
+## Ground Truth
+
+Primary Go sources:
+
+- `weed/server/volume_server.go`
+- `weed/server/volume_server_handlers*.go`
+- `weed/server/volume_grpc_*.go`
+- `weed/server/constants/volume.go`
+- `weed/storage/store*.go`
+- `weed/storage/disk_location*.go`
+- `weed/storage/volume*.go`
+- `weed/storage/needle/*.go`
+- `weed/storage/idx/*.go`
+- `weed/storage/needle_map*.go`
+- `weed/storage/needle_map/*.go`
+- `weed/storage/super_block/*.go`
+- `weed/storage/erasure_coding/*.go`
+
+Supporting Go dependencies that affect drop-in behavior:
+
+- `weed/command/volume.go`
+- `weed/security/*.go`
+- `weed/images/*.go`
+- `weed/stats/*.go`
+
+Primary Rust sources:
+
+- `seaweed-volume/src/main.rs`
+- `seaweed-volume/src/config.rs`
+- `seaweed-volume/src/security.rs`
+- `seaweed-volume/src/images.rs`
+- `seaweed-volume/src/server/*.rs`
+- `seaweed-volume/src/storage/*.rs`
+- `seaweed-volume/src/storage/needle/*.rs`
+- `seaweed-volume/src/storage/idx/*.rs`
+- `seaweed-volume/src/storage/erasure_coding/*.rs`
+- `seaweed-volume/src/remote_storage/*.rs`
+
+## Audit Method
+
+For each Go file:
+
+1. Map it to the Rust file or files that should own the same behavior.
+2. Compare exported entry points, helper functions, state transitions, wire fields, and persistence side effects.
+3. Mark each file `implemented`, `partial`, `missing`, or `needs verification`.
+4. Link each behavior to an existing test or add a missing test.
+5. Only treat a gap as closed after code review plus local verification.
+
+## Acceptance Criteria
+
+The Rust server is a drop-in replacement only when all of these hold:
+
+- HTTP routes, status codes, headers, and body semantics match Go.
+- gRPC RPCs match Go request validation, response fields, streaming behavior, and maintenance/read-only semantics.
+- Master heartbeat and topology metadata match Go closely enough that the Go master treats Rust and Go volume servers the same.
+- On-disk volume behavior matches Go for normal volumes, EC shards, tiering metadata, and readonly persistence.
+- Startup flags and operational endpoints that affect production deployment behave equivalently or are explicitly documented as unsupported.
+- Existing Go integration suites pass with `VOLUME_SERVER_IMPL=rust`.
+
+## File Matrix
+
+### HTTP server surface
+
+| Go file | Rust counterpart | Status | Comparison focus |
+| --- | --- | --- | --- |
+| `weed/server/volume_server.go` | `seaweed-volume/src/main.rs`, `seaweed-volume/src/server/volume_server.rs`, `seaweed-volume/src/server/heartbeat.rs` | partial | startup wiring, routers, heartbeat, shutdown, metrics/debug listeners |
+| `weed/server/volume_server_handlers.go` | `seaweed-volume/src/server/volume_server.rs`, `seaweed-volume/src/server/handlers.rs` | needs verification | method dispatch, OPTIONS behavior, public/admin split |
+| `weed/server/volume_server_handlers_admin.go` | `seaweed-volume/src/server/handlers.rs` | implemented | `/status`, `/healthz`, stats, server headers |
+| `weed/server/volume_server_handlers_helper.go` | `seaweed-volume/src/server/handlers.rs` | needs verification | JSON encoding, request parsing, helper parity |
+| `weed/server/volume_server_handlers_read.go` | `seaweed-volume/src/server/handlers.rs` | needs verification | JWT, conditional reads, range reads, proxy/redirect, chunk manifests, image transforms |
+| `weed/server/volume_server_handlers_ui.go` | `seaweed-volume/src/server/handlers.rs`, embedded assets | partial | UI payload and HTML parity |
+| `weed/server/volume_server_handlers_write.go` | `seaweed-volume/src/server/handlers.rs`, `seaweed-volume/src/images.rs` | needs verification | multipart parsing, metadata, compression, ts, delete semantics |
+| `weed/server/constants/volume.go` | `seaweed-volume/src/server/heartbeat.rs`, config defaults | needs verification | heartbeat timing, constants parity |
+
+### gRPC server surface
+
+| Go file | Rust counterpart | Status | Comparison focus |
+| --- | --- | --- | --- |
+| `weed/server/volume_grpc_admin.go` | `seaweed-volume/src/server/grpc_server.rs` | needs verification | readonly/writable, allocate/delete/configure/mount/unmount |
+| `weed/server/volume_grpc_batch_delete.go` | `seaweed-volume/src/server/grpc_server.rs` | implemented | batch delete, EC delete path |
+| `weed/server/volume_grpc_client_to_master.go` | `seaweed-volume/src/server/heartbeat.rs` | partial | heartbeat fields, leader changes, metrics settings from master |
+| `weed/server/volume_grpc_copy.go` | `seaweed-volume/src/server/grpc_server.rs` | needs verification | full copy streams |
+| `weed/server/volume_grpc_copy_incremental.go` | `seaweed-volume/src/server/grpc_server.rs` | needs verification | incremental copy binary search, timestamps |
+| `weed/server/volume_grpc_erasure_coding.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/storage/erasure_coding/*.rs` | needs verification | shard read/write/delete/mount/unmount/rebuild |
+| `weed/server/volume_grpc_query.go` | `seaweed-volume/src/server/grpc_server.rs` | needs verification | query validation and error parity |
+| `weed/server/volume_grpc_read_all.go` | `seaweed-volume/src/server/grpc_server.rs` | needs verification | read-all ordering and tail semantics |
+| `weed/server/volume_grpc_read_write.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/storage/*.rs` | needs verification | blob/meta/page reads, write blob semantics |
+| `weed/server/volume_grpc_remote.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/remote_storage/*.rs` | needs verification | remote fetch/write and tier metadata |
+| `weed/server/volume_grpc_scrub.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/storage/*.rs` | needs verification | scrub result semantics |
+| `weed/server/volume_grpc_state.go` | `seaweed-volume/src/server/grpc_server.rs` | implemented | GetState/SetState/Status |
+| `weed/server/volume_grpc_tail.go` | `seaweed-volume/src/server/grpc_server.rs` | needs verification | tail streaming and idle timeout |
+| `weed/server/volume_grpc_tier_download.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/remote_storage/*.rs` | needs verification | tier download stream/error paths |
+| `weed/server/volume_grpc_tier_upload.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/remote_storage/*.rs` | needs verification | tier upload stream/error paths |
+| `weed/server/volume_grpc_vacuum.go` | `seaweed-volume/src/server/grpc_server.rs`, `seaweed-volume/src/storage/*.rs` | needs verification | compact/commit/cleanup progress and readonly transitions |
+
+### Storage and persistence surface
+
+| Go file group | Rust counterpart | Status | Comparison focus |
+| --- | --- | --- | --- |
+| `weed/storage/store.go`, `store_state.go` | `seaweed-volume/src/storage/store.rs`, `seaweed-volume/src/server/heartbeat.rs` | partial | topology metadata, disk tags, server id, state persistence |
+| `weed/storage/store_vacuum.go` | `seaweed-volume/src/storage/store.rs`, `seaweed-volume/src/storage/volume.rs` | needs verification | vacuum sequencing |
+| `weed/storage/store_ec.go`, `store_ec_delete.go`, `store_ec_scrub.go` | `seaweed-volume/src/storage/store.rs`, `seaweed-volume/src/storage/erasure_coding/*.rs` | needs verification | EC lifecycle and scrub behavior |
+| `weed/storage/disk_location.go`, `disk_location_ec.go` | `seaweed-volume/src/storage/disk_location.rs`, `seaweed-volume/src/storage/store.rs` | partial | directory UUIDs, tags, load rules, disk space checks |
+| `weed/storage/volume.go`, `volume_loading.go` | `seaweed-volume/src/storage/volume.rs` | needs verification | load/reload/readonly/remote metadata |
+| `weed/storage/volume_super_block.go` | `seaweed-volume/src/storage/super_block.rs`, `seaweed-volume/src/storage/volume.rs` | implemented | super block parity |
+| `weed/storage/volume_read.go`, `volume_read_all.go` | `seaweed-volume/src/storage/volume.rs`, `seaweed-volume/src/server/handlers.rs` | needs verification | full/meta/page reads, TTL, streaming |
+| `weed/storage/volume_write.go` | `seaweed-volume/src/storage/volume.rs`, `seaweed-volume/src/server/write_queue.rs` | needs verification | dedup, sync/async writes, metadata flags |
+| `weed/storage/volume_vacuum.go` | `seaweed-volume/src/storage/volume.rs` | needs verification | compact and commit parity |
+| `weed/storage/volume_backup.go` | `seaweed-volume/src/storage/volume.rs`, `seaweed-volume/src/server/grpc_server.rs` | needs verification | backup/search logic |
+| `weed/storage/volume_checking.go` | `seaweed-volume/src/storage/volume.rs`, `seaweed-volume/src/storage/idx/mod.rs`, `seaweed-volume/src/server/grpc_server.rs` | needs verification | scrub and integrity checks |
+| `weed/storage/volume_info.go`, `volume_info/volume_info.go`, `volume_tier.go` | `seaweed-volume/src/storage/volume.rs`, `seaweed-volume/src/remote_storage/*.rs` | needs verification | `.vif` format and tiered file metadata |
+| `weed/storage/needle/*.go` | `seaweed-volume/src/storage/needle/*.rs` | needs verification | needle parsing, CRC, TTL, multipart metadata |
+| `weed/storage/idx/*.go` | `seaweed-volume/src/storage/idx/*.rs` | needs verification | index walking and binary search |
+| `weed/storage/needle_map*.go`, `needle_map/*.go` | `seaweed-volume/src/storage/needle_map.rs` | needs verification | map kind parity, persistence, memory behavior |
+| `weed/storage/super_block/*.go` | `seaweed-volume/src/storage/super_block.rs` | implemented | replica placement and TTL metadata |
+| `weed/storage/erasure_coding/*.go` | `seaweed-volume/src/storage/erasure_coding/*.rs` | needs verification | EC shard placement, encode/decode, journal deletes |
+
+### Supporting runtime surface
+
+| Go file | Rust counterpart | Status | Comparison focus |
+| --- | --- | --- | --- |
+| `weed/command/volume.go` | `seaweed-volume/src/config.rs`, `seaweed-volume/src/main.rs` | partial | flags, metrics/debug listeners, startup behavior |
+| `weed/security/*.go` | `seaweed-volume/src/security.rs`, `seaweed-volume/src/main.rs` | implemented | JWT and TLS loading |
+| `weed/images/*.go` | `seaweed-volume/src/images.rs`, `seaweed-volume/src/server/handlers.rs` | implemented | JPEG orientation and transforms |
+| `weed/stats/*.go` | `seaweed-volume/src/metrics.rs`, `seaweed-volume/src/server/handlers.rs` | partial | metrics endpoints, push-gateway integration |
+
+## Verified Gaps As Of 2026-03-08
+
+The startup/runtime gaps that were verified in the initial audit are now closed:
+
+1. Heartbeat metadata parity
+   Closed by `8ade1c51d` and retained in current HEAD.
+
+2. Dedicated metrics/debug listener parity
+   Closed by `fbe0e5829`.
+
+3. Master-provided metrics push settings
+   Closed by `fbe0e5829`.
+
+4. Slow-read tuning parity
+   Closed by `66e3900dc`.
+
+There are no remaining verified gaps from the initial startup/runtime audit. The broader line-by-line comparison batches below are still required to either confirm parity or surface new gaps.
+
+## Execution Status As Of 2026-03-16
+
+The file-by-file comparison and verification work executed in this round was:
+
+1. Startup and harness alignment
+   Compared `weed/command/volume.go`, `test/volume_server/framework/cluster*.go`, `seaweed-volume/src/config.rs`, and `seaweed-volume/src/main.rs` to ensure the Rust server is invoked with Go-compatible flags and is rebuilt from the current source during parity runs.
+
+2. HTTP admin surface
+   Compared `weed/server/volume_server_handlers_admin.go` against `seaweed-volume/src/server/handlers.rs` with emphasis on `/status` payload shape, disk-status fields, and volume ordering.
+
+3. gRPC admin surface
+   Compared `weed/server/volume_grpc_admin.go` against `seaweed-volume/src/server/grpc_server.rs` with emphasis on `Ping`, `VolumeConfigure`, readonly/writable flows, and error wrapping.
+
+4. Storage/index layout
+   Compared Go index-entry defaults in `weed/storage/types` and `weed/storage/idx/*.go` against the Rust default feature set in `seaweed-volume/Cargo.toml` and the Rust index reader/writer paths to confirm default binaries use the same offset width.
+
+5. End-to-end parity verification
+   Re-ran the Go HTTP and gRPC integration suites with `VOLUME_SERVER_IMPL=rust` after each fix to confirm wire-level compatibility.
+
+### Verified mismatches closed in this round
+
+- Rust parity runs could reuse a stale `weed-volume` binary across test invocations, hiding source and feature changes from the Go harness.
+- Rust defaulted to 5-byte index offsets, while the default Go `go build` path uses 4-byte offsets unless built with `-tags 5BytesOffset`.
+- Rust `/status` omitted Go fields in both `Volumes` and `DiskStatuses`, and did not sort volumes by `Id`.
+- Rust `Ping` treated an empty target as a self-ping and only performed a raw gRPC connect for filer targets; Go returns `remote_time_ns=0` for the empty request and performs a real filer `Ping` RPC.
+- Rust `VolumeNeedleStatus` dropped stored TTL metadata and reported `data_size` instead of Go’s `Size` field.
+- Rust multipart uploads ignored form fields such as `ts`, `ttl`, and `cm`, and also ignored part-level `Content-Encoding` and `Content-MD5`.
+- Rust only treated `dl=true` and `dl=1` as truthy, while Go accepts the full `strconv.ParseBool` set such as `dl=t` and `dl=True`.
+
+### Verification commands
+
+- `VOLUME_SERVER_IMPL=rust go test -count=1 -timeout 1200s ./test/volume_server/http/...`
+- `VOLUME_SERVER_IMPL=rust go test -count=1 -timeout 1200s ./test/volume_server/grpc/...`
+
+## Execution Plan
+
+### Batch 1: startup and heartbeat
+
+- Compare `weed/command/volume.go`, `weed/server/volume_server.go`, `weed/server/volume_grpc_client_to_master.go`, `weed/storage/store.go`, and `weed/storage/disk_location.go`.
+- Close metadata and startup parity gaps that affect master registration and deployment compatibility.
+- Add Rust unit tests for heartbeat payloads and config wiring.
+
+### Batch 2: HTTP read path
+
+- Compare `volume_server_handlers_read.go`, `volume_server_handlers_helper.go`, and related storage read functions line by line.
+- Verify JWT, path parsing, proxy/redirect, ranges, streaming, chunk manifests, image transforms, and response-header overrides.
+- Extend `test/volume_server/http/...` and Rust handler tests where parity is not covered.
+
+### Batch 3: HTTP write/delete path
+
+- Compare `volume_server_handlers_write.go` and write-related storage functions.
+- Verify multipart behavior, metadata, md5, compression, unchanged writes, delete edge cases, and timestamp handling.
+
+### Batch 4: gRPC admin and lifecycle
+
+- Compare `volume_grpc_admin.go`, `volume_grpc_state.go`, and `volume_grpc_vacuum.go`.
+- Verify readonly/writable flows, maintenance mode, status payloads, mount/unmount/delete/configure, and vacuum transitions.
+
+### Batch 5: gRPC data movement
+
+- Compare `volume_grpc_read_write.go`, `copy*.go`, `read_all.go`, `tail.go`, `remote.go`, and `query.go`.
+- Verify stream framing, binary search, idle timeout, and remote-storage semantics.
+
+### Batch 6: storage internals
+
+- Compare all `weed/storage` volume, needle, idx, needle map, and EC files line by line.
+- Focus on persistence rules, readonly semantics, TTL, recovery/scrub, backup, and memory/disk map behavior.
+
+## Commit Strategy
+
+- One commit for the audit/plan document if the document itself changes.
+- One commit per logic fix.
+- Every logic commit must include the smallest test addition that proves the new parity claim.

seaweed-volume/README.md

@@ -0,0 +1,143 @@
+# SeaweedFS Volume Server (Rust)
+
+A drop-in replacement for the [SeaweedFS](https://github.com/seaweedfs/seaweedfs) Go volume server, rewritten in Rust. It uses binary-compatible storage formats (`.dat`, `.idx`, `.vif`) and speaks the same HTTP and gRPC protocols, so it works with an unmodified Go master server.
+
+## Building
+
+Requires Rust 1.91.1+ (2024 edition), matching `rust-version` in `Cargo.toml`.
+The patch release matters: 1.91.0 does not build. The edition itself only needs
+1.85; the higher floor comes from the dependency tree — chiefly the AWS SDK — so
+it moves with those crates. CI builds on the latest stable.
+
+```bash
+cd seaweed-volume
+cargo build --release
+```
+
+The binary is produced at `target/release/seaweed-volume`.
+
+## Running
+
+Start a Go master server first, then point the Rust volume server at it:
+
+```bash
+# Minimal
+seaweed-volume --port 8080 --master localhost:9333 --dir /data/vol1 --max 7
+
+# Multiple data directories
+seaweed-volume --port 8080 --master localhost:9333 \
+  --dir /mnt/ssd1,/mnt/ssd2 --max 100,100 --disk ssd
+
+# With datacenter/rack topology
+seaweed-volume --port 8080 --master localhost:9333 --dir /data/vol1 --max 7 \
+  --dataCenter dc1 --rack rack1
+
+# With JWT authentication
+seaweed-volume --port 8080 --master localhost:9333 --dir /data/vol1 --max 7 \
+  --securityFile /etc/seaweedfs/security.toml
+
+# With TLS (configured in security.toml via [https.volume] and [grpc.volume] sections)
+seaweed-volume --port 8080 --master localhost:9333 --dir /data/vol1 --max 7 \
+  --securityFile /etc/seaweedfs/security.toml
+```
+
+### Common flags
+
+| Flag | Default | Description |
+|------|---------|-------------|
+| `--port` | `8080` | HTTP listen port |
+| `--port.grpc` | `port+10000` | gRPC listen port |
+| `--master` | `localhost:9333` | Comma-separated master server addresses |
+| `--dir` | `/tmp` | Comma-separated data directories |
+| `--max` | `8` | Max volumes per directory (comma-separated) |
+| `--ip` | auto-detect | Server IP / identifier |
+| `--ip.bind` | same as `--ip` | Bind address |
+| `--dataCenter` | | Datacenter name |
+| `--rack` | | Rack name |
+| `--disk` | | Disk type tag: `hdd`, `ssd`, or custom |
+| `--index` | `memory` | Needle map type: `memory`, `leveldb`, `leveldbMedium`, `leveldbLarge` |
+| `--readMode` | `proxy` | Non-local read mode: `local`, `proxy`, `redirect` |
+| `--fileSizeLimitMB` | `256` | Max upload file size |
+| `--minFreeSpace` | `1` (percent) | Min free disk space before marking volumes read-only |
+| `--securityFile` | | Path to `security.toml` for JWT keys and TLS certs |
+| `--metricsPort` | `0` (disabled) | Prometheus metrics endpoint port |
+| `--whiteList` | | Comma-separated IPs with write permission |
+| `--preStopSeconds` | `10` | Graceful drain period before shutdown |
+| `--compactionMBps` | `0` (unlimited) | Compaction I/O rate limit |
+| `--pprof` | `false` | Enable pprof HTTP handlers |
+
+Set `RUST_LOG=debug` (or `trace`, `info`, `warn`) for log level control.
+Set `SEAWEED_WRITE_QUEUE=1` to enable batched async write processing.
+
+## Features
+
+- **Binary compatible** -- reads and writes the same `.dat`/`.idx`/`.vif` files as the Go server; seamless migration with no data conversion.
+- **HTTP + gRPC** -- full implementation of the volume server HTTP API and all gRPC RPCs including streaming operations (copy, tail, incremental copy, vacuum).
+- **Master heartbeat** -- bidirectional streaming heartbeat with the Go master server; volume and EC shard registration, leader failover, graceful shutdown deregistration.
+- **JWT authentication** -- signing key configuration via `security.toml` with token source precedence (query > header > cookie), file_id claims validation, and separate read/write keys.
+- **TLS** -- HTTPS for the HTTP API and mTLS for gRPC, configured through `security.toml`.
+- **Erasure coding** -- Reed-Solomon EC shard management: mount/unmount, read, rebuild, copy, delete, and shard-to-volume reconstruction.
+- **S3 remote storage** -- `FetchAndWriteNeedle` reads from any S3-compatible backend (AWS, MinIO, Wasabi, Backblaze, etc.) and writes locally. Supports `VolumeTierMoveDatToRemote`/`FromRemote` for tiered storage.
+- **Needle map backends** -- in-memory HashMap, LevelDB (via `rusty-leveldb`), or redb (pure Rust disk-backed) needle maps.
+- **Image processing** -- on-the-fly resize/crop, JPEG EXIF orientation auto-fix, WebP support.
+- **Streaming reads** -- large files (>1MB) are streamed via `spawn_blocking` to avoid blocking the async runtime.
+- **Auto-compression** -- compressible file types (text, JSON, CSS, JS, SVG, etc.) are gzip-compressed on upload.
+- **Prometheus metrics** -- counters, histograms, and gauges exported at a dedicated metrics port; optional push gateway support.
+- **Graceful shutdown** -- SIGINT/SIGTERM handling with configurable `preStopSeconds` drain period.
+
+## Testing
+
+### Rust unit tests
+
+```bash
+cd seaweed-volume
+cargo test
+```
+
+### Go integration tests
+
+The Go test suite can target either the Go or Rust volume server via the `VOLUME_SERVER_IMPL` environment variable:
+
+```bash
+# Run all HTTP + gRPC integration tests against the Rust server
+VOLUME_SERVER_IMPL=rust go test -v -count=1 -timeout 1200s \
+  ./test/volume_server/grpc/... ./test/volume_server/http/...
+
+# Run a single test
+VOLUME_SERVER_IMPL=rust go test -v -count=1 -timeout 60s \
+  -run "TestName" ./test/volume_server/http/...
+
+# Run S3 remote storage tests
+VOLUME_SERVER_IMPL=rust go test -v -count=1 -timeout 180s \
+  -run "TestFetchAndWriteNeedle" ./test/volume_server/grpc/...
+```
+
+## Load testing
+
+A load test harness is available at `test/volume_server/loadtest/`. See that directory for usage instructions and scenarios.
+
+## Architecture
+
+The server runs three listeners concurrently:
+
+- **HTTP** (Axum 0.7) -- admin and public routers for file upload/download, status, and stats endpoints.
+- **gRPC** (Tonic 0.12) -- all `VolumeServer` RPCs from the SeaweedFS protobuf definition.
+- **Metrics** (optional) -- Prometheus scrape endpoint on a separate port.
+
+Key source modules:
+
+| Path | Description |
+|------|-------------|
+| `src/main.rs` | Entry point, server startup, signal handling |
+| `src/config.rs` | CLI parsing and configuration resolution |
+| `src/server/volume_server.rs` | HTTP router setup and middleware |
+| `src/server/handlers.rs` | HTTP request handlers (read, write, delete, status) |
+| `src/server/grpc_server.rs` | gRPC service implementation |
+| `src/server/heartbeat.rs` | Master heartbeat loop |
+| `src/storage/volume.rs` | Volume read/write/delete logic |
+| `src/storage/needle.rs` | Needle (file entry) serialization |
+| `src/storage/store.rs` | Multi-volume store management |
+| `src/security.rs` | JWT validation and IP whitelist guard |
+| `src/remote_storage/` | S3 remote storage backend |
+
+See [DEV_PLAN.md](DEV_PLAN.md) for the full development history and feature checklist.

seaweed-volume/build.rs

@@ -0,0 +1,33 @@
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    // Use the protoc that ships with protoc-bin-vendored rather than a system
+    // one, so the build needs no package manager and always sees the same
+    // version. An explicit PROTOC still wins, for packagers supplying their own.
+    if std::env::var_os("PROTOC").is_none() {
+        // SAFETY: a build script's main runs single-threaded before anything
+        // else in this process, so no other thread can be reading the
+        // environment concurrently.
+        unsafe {
+            std::env::set_var("PROTOC", protoc_bin_vendored::protoc_bin_path()?);
+        }
+    }
+
+    let out_dir = std::path::PathBuf::from(std::env::var("OUT_DIR")?);
+    tonic_prost_build::configure()
+        .build_server(true)
+        .build_client(true)
+        // filer.proto uses proto3 optional, which protoc rejects without this
+        // flag before 3.15. The vendored protoc is newer, but it still accepts
+        // the flag, so this keeps a build against an older PROTOC working.
+        .protoc_arg("--experimental_allow_proto3_optional")
+        .file_descriptor_set_path(out_dir.join("seaweed_descriptor.bin"))
+        .compile_protos(
+            &[
+                "proto/volume_server.proto",
+                "proto/master.proto",
+                "proto/remote.proto",
+                "../weed/pb/filer.proto",
+            ],
+            &["proto/", "../weed/pb/"],
+        )?;
+    Ok(())
+}

seaweed-volume/proto/master.proto

@@ -0,0 +1,611 @@
+syntax = "proto3";
+
+package master_pb;
+
+option go_package = "github.com/seaweedfs/seaweedfs/weed/pb/master_pb";
+
+import "volume_server.proto";
+
+//////////////////////////////////////////////////
+
+service Seaweed {
+  rpc SendHeartbeat (stream Heartbeat) returns (stream HeartbeatResponse) {
+  }
+  rpc KeepConnected (stream KeepConnectedRequest) returns (stream KeepConnectedResponse) {
+  }
+  rpc LookupVolume (LookupVolumeRequest) returns (LookupVolumeResponse) {
+  }
+  rpc Assign (AssignRequest) returns (AssignResponse) {
+  }
+  rpc StreamAssign (stream AssignRequest) returns (stream AssignResponse) {
+  }
+  rpc Statistics (StatisticsRequest) returns (StatisticsResponse) {
+  }
+  rpc CollectionList (CollectionListRequest) returns (CollectionListResponse) {
+  }
+  rpc CollectionDelete (CollectionDeleteRequest) returns (CollectionDeleteResponse) {
+  }
+  rpc VolumeList (VolumeListRequest) returns (VolumeListResponse) {
+  }
+  rpc VolumeListStream (VolumeListRequest) returns (stream VolumeListStreamResponse) {
+  }
+  rpc LookupEcVolume (LookupEcVolumeRequest) returns (LookupEcVolumeResponse) {
+  }
+  rpc VacuumVolume (VacuumVolumeRequest) returns (VacuumVolumeResponse) {
+  }
+  rpc DisableVacuum (DisableVacuumRequest) returns (DisableVacuumResponse) {
+  }
+  rpc EnableVacuum (EnableVacuumRequest) returns (EnableVacuumResponse) {
+  }
+  rpc VolumeMarkReadonly (VolumeMarkReadonlyRequest) returns (VolumeMarkReadonlyResponse) {
+  }
+  rpc GetMasterConfiguration (GetMasterConfigurationRequest) returns (GetMasterConfigurationResponse) {
+  }
+  rpc ListClusterNodes (ListClusterNodesRequest) returns (ListClusterNodesResponse) {
+  }
+  rpc LeaseAdminToken (LeaseAdminTokenRequest) returns (LeaseAdminTokenResponse) {
+  }
+  rpc ReleaseAdminToken (ReleaseAdminTokenRequest) returns (ReleaseAdminTokenResponse) {
+  }
+  rpc GetAdminLockStatus (GetAdminLockStatusRequest) returns (GetAdminLockStatusResponse) {
+  }
+  rpc Ping (PingRequest) returns (PingResponse) {
+  }
+  rpc RaftListClusterServers (RaftListClusterServersRequest) returns (RaftListClusterServersResponse) {
+  }
+  rpc RaftAddServer (RaftAddServerRequest) returns (RaftAddServerResponse) {
+  }
+  rpc RaftRemoveServer (RaftRemoveServerRequest) returns (RaftRemoveServerResponse) {
+  }
+  rpc RaftLeadershipTransfer (RaftLeadershipTransferRequest) returns (RaftLeadershipTransferResponse) {
+  }
+  rpc VolumeGrow (VolumeGrowRequest) returns (VolumeGrowResponse) {
+  }
+  rpc CollectionStatistics (CollectionStatisticsRequest) returns (CollectionStatisticsResponse) {
+  }
+}
+
+//////////////////////////////////////////////////
+
+message DiskTag {
+  uint32 disk_id = 1;
+  repeated string tags = 2;
+  // Physical disk descriptor, reported for every location including empty ones.
+  string type = 3;
+  int64 max_volume_count = 4;
+}
+
+message Heartbeat {
+  string ip = 1;
+  uint32 port = 2;
+  string public_url = 3;
+  uint64 max_file_key = 5;
+  string data_center = 6;
+  string rack = 7;
+  uint32 admin_port = 8;
+  repeated VolumeInformationMessage volumes = 9;
+  // delta volumes
+  repeated VolumeShortInformationMessage new_volumes = 10;
+  repeated VolumeShortInformationMessage deleted_volumes = 11;
+  bool has_no_volumes = 12;
+
+  // erasure coding
+  repeated VolumeEcShardInformationMessage ec_shards = 16;
+  // delta erasure coding shards
+  repeated VolumeEcShardInformationMessage new_ec_shards = 17;
+  repeated VolumeEcShardInformationMessage deleted_ec_shards = 18;
+  bool has_no_ec_shards = 19;
+
+  map<string, uint32> max_volume_counts = 4;
+  uint32 grpc_port = 20;
+  repeated string location_uuids = 21;
+  string id = 22; // volume server id, independent of ip:port for stable identification
+
+  // state flags
+  volume_server_pb.VolumeServerState state = 23;
+
+  repeated DiskTag disk_tags = 24;
+
+  // physical disk capacity per disk type, in bytes, from the underlying filesystem
+  map<string, uint64> disk_total_bytes = 25;
+  map<string, uint64> disk_free_bytes = 26;
+
+  // Digest of every volume in this heartbeat's view of the server, letting the
+  // master check its copy is current without being sent the whole list. Absent
+  // from servers that do not compute it, and distinct from a digest of 0, which
+  // is what a server holding no volumes reports.
+  optional uint64 volume_digest = 27;
+  // Volumes whose reported state changed since the last heartbeat, sent in
+  // place of `volumes`. A master that does not understand this never sets
+  // volume_digest_supported, so it keeps being sent the whole list.
+  repeated VolumeInformationMessage changed_volumes = 28;
+}
+
+message HeartbeatResponse {
+  uint64 volume_size_limit = 1;
+  string leader = 2;
+  string metrics_address = 3;
+  uint32 metrics_interval_seconds = 4;
+  repeated StorageBackend storage_backends = 5;
+  repeated string duplicated_uuids = 6;
+  bool preallocate = 7;
+  // The master's view of this server's volumes disagrees with the reported
+  // digest, so it needs the full volume list rather than changes alone.
+  bool resend_full_volume_list = 8;
+  // The master compares volume digests, so a server that reports one may send
+  // changed_volumes in place of its whole list.
+  bool volume_digest_supported = 9;
+}
+
+message VolumeInformationMessage {
+  uint32 id = 1;
+  uint64 size = 2;
+  string collection = 3;
+  uint64 file_count = 4;
+  uint64 delete_count = 5;
+  uint64 deleted_byte_count = 6;
+  bool read_only = 7;
+  bool read_only_can_delete = 17;
+  uint32 replica_placement = 8;
+  uint32 version = 9;
+  uint32 ttl = 10;
+  uint32 compact_revision = 11;
+  int64 modified_at_second = 12;
+  string remote_storage_name = 13;
+  string remote_storage_key = 14;
+  string disk_type = 15;
+  uint32 disk_id = 16;
+}
+
+message VolumeShortInformationMessage {
+  uint32 id = 1;
+  string collection = 3;
+  uint32 replica_placement = 8;
+  uint32 version = 9;
+  uint32 ttl = 10;
+  string disk_type = 15;
+  uint32 disk_id = 16;
+  bool read_only = 17;
+  bool read_only_can_delete = 18;
+}
+
+message VolumeEcShardInformationMessage {
+  uint32 id = 1;
+  string collection = 2;
+  uint32 ec_index_bits = 3;
+  string disk_type = 4;
+  uint64 expire_at_sec = 5; // used to record the destruction time of ec volume
+  uint32 disk_id = 6;
+  repeated int64 shard_sizes = 7; // optimized: sizes for shards in order of set bits in ec_index_bits
+  uint64 file_count = 8;   // total needles in the .ecx index (live + tombstoned)
+  uint64 delete_count = 9; // node-local tombstones in the .ecj deletion journal
+  // encode-run identity (unix nanos) from the .vif EcShardConfig; lets the admin
+  // group shards by encode generation. Numbered 14 (not 10) to skip the
+  // enterprise fork's reserved 10-13.
+  int64 encode_ts_ns = 14;
+  // fields 15-19 reserved for future upstream open-source additions.
+  // fields 20+ are owned by the enterprise fork (e.g. data_shards/parity_shards)
+  // and must not be used here without coordination.
+}
+
+message StorageBackend {
+  string type = 1;
+  string id = 2;
+  map<string, string> properties = 3;
+}
+
+message Empty {
+}
+
+message SuperBlockExtra {
+  message ErasureCoding {
+    uint32 data = 1;
+    uint32 parity = 2;
+    repeated uint32 volume_ids = 3;
+  }
+  ErasureCoding erasure_coding = 1;
+}
+
+message KeepConnectedRequest {
+  string client_type = 1;
+  string client_address = 3;
+  string version = 4;
+  string filer_group = 5;
+  string data_center = 6;
+  string rack = 7;
+}
+
+message VolumeLocation {
+  string url = 1;
+  string public_url = 2;
+  repeated uint32 new_vids = 3;
+  repeated uint32 deleted_vids = 4;
+  string leader = 5; // optional when leader is not itself
+  string data_center = 6; // optional when DataCenter is in use
+  uint32 grpc_port = 7;
+  repeated uint32 new_ec_vids = 8;
+  repeated uint32 deleted_ec_vids = 9;
+  repeated uint32 remote_vids = 10;
+  repeated uint32 read_only_vids = 11;
+  repeated uint32 read_only_can_delete_vids = 12;
+}
+
+message ClusterNodeUpdate {
+  string node_type = 1;
+  string address = 2;
+  bool is_add = 4;
+  string filer_group = 5;
+  int64 created_at_ns = 6;
+}
+
+message KeepConnectedResponse {
+  VolumeLocation volume_location = 1;
+  ClusterNodeUpdate cluster_node_update = 2;
+  LockRingUpdate lock_ring_update = 3;
+}
+
+// LockRingUpdate is sent by the master to all filers when the lock ring
+// membership changes. The master batches rapid changes (e.g., node drop + join)
+// and sends the complete member list atomically, avoiding intermediate ring
+// states that would cause unnecessary lock churn.
+message LockRingUpdate {
+  string filer_group = 1;
+  repeated string servers = 2;
+  int64 version = 3;
+}
+
+message LookupVolumeRequest {
+  repeated string volume_or_file_ids = 1;
+  string collection = 2; // optional, a bit faster if provided.
+}
+message LookupVolumeResponse {
+  message VolumeIdLocation {
+    string volume_or_file_id = 1;
+    repeated Location locations = 2;
+    string error = 3;
+    string auth = 4;
+  }
+  repeated VolumeIdLocation volume_id_locations = 1;
+}
+
+message Location {
+  string url = 1;
+  string public_url = 2;
+  uint32 grpc_port = 3;
+  string data_center = 4;
+  bool data_in_remote = 5;
+  bool read_only = 6;
+  bool read_only_can_delete = 7;
+}
+
+message AssignRequest {
+  uint64 count = 1;
+  string replication = 2;
+  string collection = 3;
+  string ttl = 4;
+  string data_center = 5;
+  string rack = 6;
+  string data_node = 7;
+  uint32 memory_map_max_size_mb = 8;
+  uint32 writable_volume_count = 9;
+  string disk_type = 10;
+  uint64 expected_data_size = 11; // hint for size-aware volume selection
+}
+
+message VolumeGrowRequest {
+  uint32 writable_volume_count = 1;
+  string replication = 2;
+  string collection = 3;
+  string ttl = 4;
+  string data_center = 5;
+  string rack = 6;
+  string data_node = 7;
+  uint32 memory_map_max_size_mb = 8;
+  string disk_type = 9;
+}
+
+message AssignResponse {
+  string fid = 1;
+  uint64 count = 4;
+  string error = 5;
+  string auth = 6;
+  repeated Location replicas = 7;
+  Location location = 8;
+}
+
+message StatisticsRequest {
+  string replication = 1;
+  string collection = 2;
+  string ttl = 3;
+  string disk_type = 4;
+}
+message StatisticsResponse {
+  uint64 total_size = 4;
+  uint64 used_size = 5;
+  uint64 file_count = 6;
+  // sizes counting one copy of the data: a single replica of a regular volume,
+  // the data shards of an ec volume. logical_total_size scales the free space
+  // by the copies the requested replication makes.
+  uint64 logical_total_size = 7;
+  uint64 logical_used_size = 8;
+}
+
+//
+// collection related
+//
+message Collection {
+  string name = 1;
+}
+message CollectionListRequest {
+  bool include_normal_volumes = 1;
+  bool include_ec_volumes = 2;
+}
+message CollectionListResponse {
+  repeated Collection collections = 1;
+}
+
+// Summarises what each collection holds, so a caller tracking usage does not
+// have to be sent every volume in the cluster to add it up itself.
+message CollectionStatisticsRequest {
+}
+message CollectionStatisticsResponse {
+  repeated CollectionStatistics collections = 1;
+}
+message CollectionStatistics {
+  string collection = 1;
+  uint64 file_count = 2;
+  uint64 delete_count = 3;
+  uint64 deleted_byte_count = 4;
+  // one copy of the data: a single replica of a regular volume, the data
+  // shards of an ec volume
+  uint64 size = 5;
+  // what is on disk: every replica, and parity shards
+  uint64 physical_size = 6;
+  uint64 volume_count = 7;
+}
+
+message CollectionDeleteRequest {
+  string name = 1;
+}
+message CollectionDeleteResponse {
+}
+
+//
+// volume related
+//
+message DiskInfo {
+  string type = 1;
+  int64 volume_count = 2;
+  int64 max_volume_count = 3;
+  int64 free_volume_count = 4;
+  int64 active_volume_count = 5;
+  repeated VolumeInformationMessage volume_infos = 6;
+  repeated VolumeEcShardInformationMessage ec_shard_infos = 7;
+  int64 remote_volume_count = 8;
+  // On a per-physical-disk DiskInfo (from SplitByPhysicalDisk) this is the disk's
+  // identity; on the type-keyed aggregate it is only a representative fallback
+  // (the first volume's disk id).
+  uint32 disk_id = 9;
+  repeated string tags = 10;
+  // Max volume count for every physical disk of this type, keyed by disk id,
+  // including disks with no volumes or EC shards; recovers empty disks that
+  // carry no per-volume/per-shard records.
+  map<uint32, int64> max_volume_count_by_disk = 11;
+  // physical disk capacity in bytes, from the underlying filesystem (0 if unknown)
+  uint64 disk_total_bytes = 12;
+  uint64 disk_free_bytes = 13;
+}
+message DataNodeInfo {
+  string id = 1;
+  map<string, DiskInfo> diskInfos = 2;
+  uint32 grpc_port = 3;
+  string address = 4; // ip:port for connecting to the volume server
+}
+message RackInfo {
+  string id = 1;
+  repeated DataNodeInfo data_node_infos = 2;
+  map<string, DiskInfo> diskInfos = 3;
+}
+message DataCenterInfo {
+  string id = 1;
+  repeated RackInfo rack_infos = 2;
+  map<string, DiskInfo> diskInfos = 3;
+}
+message TopologyInfo {
+  string id = 1;
+  repeated DataCenterInfo data_center_infos = 2;
+  map<string, DiskInfo> diskInfos = 3;
+}
+message VolumeListRequest {
+  // Empty and zero take everything. Only the volumes and ec shards listed
+  // under a disk are selected; the topology and its disk counters are always
+  // reported in full.
+  string collection = 1;
+  repeated uint32 volume_ids = 2;
+  // The one collection the empty string cannot name. A named collection wins.
+  bool default_collection_only = 3;
+  // Empty and zero take everything. Wildcards are supported.
+  string remote_storage_name = 4;
+  bool local_volume_only = 5;
+  // The topology, its disks and their counters alone, without the volumes
+  // and ec shards. Selecting volumes above contradicts this and is refused.
+  // A master that predates this field ignores it and answers in full.
+  bool topology_only = 6;
+}
+message VolumeListResponse {
+  TopologyInfo topology_info = 1;
+  uint64 volume_size_limit_mb = 2;
+}
+
+// VolumeListStream answers the same request as VolumeList without either end
+// holding every volume in the cluster at once. At 800k volumes the reply is
+// 36MB on the wire but 305MB as messages, which the master built in full
+// before sending any of it.
+message VolumeListStreamResponse {
+  // Sent once, first, listing no volumes: the topology, its disks and their
+  // counters. Every message after carries volumes for one of those disks.
+  VolumeListResponse header = 1;
+  // Which disk this batch is from. A disk arrives over as many batches as it
+  // takes, so append rather than assign.
+  string data_center = 2;
+  string rack = 3;
+  string data_node = 4;
+  string disk_type = 5;
+  repeated VolumeInformationMessage volume_infos = 6;
+  repeated VolumeEcShardInformationMessage ec_shard_infos = 7;
+}
+
+message LookupEcVolumeRequest {
+  uint32 volume_id = 1;
+}
+message LookupEcVolumeResponse {
+  uint32 volume_id = 1;
+  message EcShardIdLocation {
+    uint32 shard_id = 1;
+    repeated Location locations = 2;
+  }
+  repeated EcShardIdLocation shard_id_locations = 2;
+}
+
+message VacuumVolumeRequest {
+  float garbage_threshold = 1;
+  uint32 volume_id = 2;
+  string collection = 3;
+}
+message VacuumVolumeResponse {
+}
+
+message DisableVacuumRequest {
+    bool by_plugin = 1;
+}
+message DisableVacuumResponse {
+}
+
+message EnableVacuumRequest {
+    bool by_plugin = 1;
+}
+message EnableVacuumResponse {
+}
+
+message VolumeMarkReadonlyRequest {
+  string ip = 1;
+  uint32 port = 2;
+  uint32 volume_id = 4;
+  string collection = 5;
+  uint32 replica_placement = 6;
+  uint32 version = 7;
+  uint32 ttl = 8;
+  string disk_type = 9;
+  bool is_readonly = 10;
+}
+message VolumeMarkReadonlyResponse {
+}
+
+message GetMasterConfigurationRequest {
+}
+message GetMasterConfigurationResponse {
+  string metrics_address = 1;
+  uint32 metrics_interval_seconds = 2;
+  repeated StorageBackend storage_backends = 3;
+  string default_replication = 4;
+  string leader = 5;
+  uint32 volume_size_limit_m_b = 6;
+  bool volume_preallocate = 7;
+  // MIGRATION: fields 8-9 help migrate master.toml [master.maintenance] to admin script plugin. Remove after March 2027.
+  string maintenance_scripts = 8;
+  uint32 maintenance_sleep_minutes = 9;
+}
+
+message ListClusterNodesRequest {
+  string client_type = 1;
+  string filer_group = 2;
+  int32 limit = 4;
+}
+message ListClusterNodesResponse {
+  message ClusterNode {
+    string address = 1;
+    string version = 2;
+    int64 created_at_ns = 4;
+    string data_center = 5;
+    string rack = 6;
+  }
+  repeated ClusterNode cluster_nodes = 1;
+}
+
+message LeaseAdminTokenRequest {
+  int64 previous_token = 1;
+  int64 previous_lock_time = 2;
+  string lock_name = 3;
+  string client_name = 4;
+  string message = 5;
+}
+message LeaseAdminTokenResponse {
+  int64 token = 1;
+  int64 lock_ts_ns = 2;
+}
+
+message ReleaseAdminTokenRequest {
+  int64 previous_token = 1;
+  int64 previous_lock_time = 2;
+  string lock_name = 3;
+}
+message ReleaseAdminTokenResponse {
+}
+
+message GetAdminLockStatusRequest {
+  string lock_name = 1;
+}
+message GetAdminLockStatusResponse {
+  bool is_locked = 1;
+  string client_name = 2;
+  string message = 3;
+}
+
+message PingRequest {
+  string target = 1; // default to ping itself
+  string target_type = 2;
+}
+message PingResponse {
+  int64 start_time_ns = 1;
+  int64 remote_time_ns = 2;
+  int64 stop_time_ns = 3;
+}
+
+message RaftAddServerRequest {
+  string id = 1;
+  string address = 2;
+  bool voter = 3;
+}
+message RaftAddServerResponse {
+}
+
+message RaftRemoveServerRequest {
+  string id = 1;
+  bool force = 2;
+}
+message RaftRemoveServerResponse {
+}
+
+message RaftListClusterServersRequest {
+}
+message RaftListClusterServersResponse {
+  message ClusterServers {
+    string id = 1;
+    string address = 2;
+    string suffrage = 3;
+    bool   isLeader = 4;
+  }
+  repeated ClusterServers cluster_servers = 1;
+}
+
+message RaftLeadershipTransferRequest {
+  string target_id = 1;       // Optional: target server ID. If empty, transfers to any eligible follower
+  string target_address = 2;  // Optional: target server address. Required if target_id is specified
+}
+message RaftLeadershipTransferResponse {
+  string previous_leader = 1;
+  string new_leader = 2;
+}
+
+message VolumeGrowResponse {
+}

seaweed-volume/proto/remote.proto

@@ -0,0 +1,80 @@
+syntax = "proto3";
+
+package remote_pb;
+
+option go_package = "github.com/seaweedfs/seaweedfs/weed/pb/remote_pb";
+option java_package = "seaweedfs.client";
+option java_outer_classname = "FilerProto";
+
+/////////////////////////
+// Remote Storage related
+/////////////////////////
+message RemoteConf {
+  string type = 1;
+  string name = 2;
+  string s3_access_key = 4;
+  string s3_secret_key = 5;
+  string s3_region = 6;
+  string s3_endpoint = 7;
+  string s3_storage_class = 8;
+  bool s3_force_path_style = 9;
+  bool s3_support_tagging = 13;
+  bool s3_v4_signature = 11;
+
+  string gcs_google_application_credentials = 10;
+  string gcs_project_id = 12;
+
+  string azure_account_name = 15;
+  string azure_account_key = 16;
+  string azure_client_id = 17;
+  string azure_endpoint = 18;
+
+  string backblaze_key_id = 20;
+  string backblaze_application_key = 21;
+  string backblaze_endpoint = 22;
+  string backblaze_region = 23;
+
+  string aliyun_access_key = 25;
+  string aliyun_secret_key = 26;
+  string aliyun_endpoint = 27;
+  string aliyun_region = 28;
+
+  string tencent_secret_id = 30;
+  string tencent_secret_key = 31;
+  string tencent_endpoint = 32;
+
+  string baidu_access_key = 35;
+  string baidu_secret_key = 36;
+  string baidu_endpoint = 37;
+  string baidu_region = 38;
+
+  string wasabi_access_key = 40;
+  string wasabi_secret_key = 41;
+  string wasabi_endpoint = 42;
+  string wasabi_region = 43;
+
+  string filebase_access_key = 60;
+  string filebase_secret_key = 61;
+  string filebase_endpoint = 62;
+
+  string storj_access_key = 65;
+  string storj_secret_key = 66;
+  string storj_endpoint = 67;
+
+  string contabo_access_key = 68;
+  string contabo_secret_key = 69;
+  string contabo_endpoint = 70;
+  string contabo_region = 71;
+}
+
+message RemoteStorageMapping {
+  map<string,RemoteStorageLocation> mappings = 1;
+  string primary_bucket_storage_name = 2;
+}
+message RemoteStorageLocation {
+  string name = 1;
+  string bucket = 2;
+  string path = 3;
+  int32 listing_cache_ttl_seconds = 4; // 0 = disabled; >0 enables on-demand directory listing with this TTL in seconds
+  optional int32 cache_wait_ms = 5; // unset = size based default; 0 = read straight from the remote without caching
+}

seaweed-volume/proto/volume_server.proto

@@ -0,0 +1,829 @@
+syntax = "proto3";
+
+package volume_server_pb;
+option go_package = "github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb";
+
+import "remote.proto";
+
+//////////////////////////////////////////////////
+
+// Persistent state for volume servers.
+message VolumeServerState {
+    // whether the server is in maintenance (i.e. read-only) mode.
+    bool maintenance = 1;
+    // incremental version counter
+    uint32 version = 2;
+}
+
+//////////////////////////////////////////////////
+
+service VolumeServer {
+    //Experts only: takes multiple fid parameters. This function does not propagate deletes to replicas.
+    rpc BatchDelete (BatchDeleteRequest) returns (BatchDeleteResponse) {
+    }
+
+    rpc VacuumVolumeCheck (VacuumVolumeCheckRequest) returns (VacuumVolumeCheckResponse) {
+    }
+    rpc VacuumVolumeCompact (VacuumVolumeCompactRequest) returns (stream VacuumVolumeCompactResponse) {
+    }
+    rpc VacuumVolumeCommit (VacuumVolumeCommitRequest) returns (VacuumVolumeCommitResponse) {
+    }
+    rpc VacuumVolumeCleanup (VacuumVolumeCleanupRequest) returns (VacuumVolumeCleanupResponse) {
+    }
+
+    rpc DeleteCollection (DeleteCollectionRequest) returns (DeleteCollectionResponse) {
+    }
+    rpc AllocateVolume (AllocateVolumeRequest) returns (AllocateVolumeResponse) {
+    }
+
+    rpc VolumeSyncStatus (VolumeSyncStatusRequest) returns (VolumeSyncStatusResponse) {
+    }
+    rpc VolumeIncrementalCopy (VolumeIncrementalCopyRequest) returns (stream VolumeIncrementalCopyResponse) {
+    }
+
+    rpc VolumeMount (VolumeMountRequest) returns (VolumeMountResponse) {
+    }
+    rpc VolumeUnmount (VolumeUnmountRequest) returns (VolumeUnmountResponse) {
+    }
+    rpc VolumeConsolidateIndex (VolumeConsolidateIndexRequest) returns (VolumeConsolidateIndexResponse) {
+    }
+    rpc VolumeDelete (VolumeDeleteRequest) returns (VolumeDeleteResponse) {
+    }
+    rpc VolumeMarkReadonly (VolumeMarkReadonlyRequest) returns (VolumeMarkReadonlyResponse) {
+    }
+    rpc VolumeMarkWritable (VolumeMarkWritableRequest) returns (VolumeMarkWritableResponse) {
+    }
+    rpc VolumeConfigure (VolumeConfigureRequest) returns (VolumeConfigureResponse) {
+    }
+    rpc VolumeStatus (VolumeStatusRequest) returns (VolumeStatusResponse) {
+    }
+
+    rpc GetState (GetStateRequest) returns (GetStateResponse) {
+    }
+    rpc SetState (SetStateRequest) returns (SetStateResponse) {
+    }
+
+    // copy the .idx .dat files, and mount this volume
+    rpc VolumeCopy (VolumeCopyRequest) returns (stream VolumeCopyResponse) {
+    }
+    rpc ReadVolumeFileStatus (ReadVolumeFileStatusRequest) returns (ReadVolumeFileStatusResponse) {
+    }
+    rpc CopyFile (CopyFileRequest) returns (stream CopyFileResponse) {
+    }
+    rpc ReceiveFile (stream ReceiveFileRequest) returns (ReceiveFileResponse) {
+    }
+
+    rpc ReadNeedleBlob (ReadNeedleBlobRequest) returns (ReadNeedleBlobResponse) {
+    }
+    rpc ReadNeedleMeta (ReadNeedleMetaRequest) returns (ReadNeedleMetaResponse) {
+    }
+    rpc WriteNeedleBlob (WriteNeedleBlobRequest) returns (WriteNeedleBlobResponse) {
+    }
+    rpc ReadAllNeedles (ReadAllNeedlesRequest) returns (stream ReadAllNeedlesResponse) {
+    }
+
+    rpc VolumeTailSender (VolumeTailSenderRequest) returns (stream VolumeTailSenderResponse) {
+    }
+    rpc VolumeTailReceiver (VolumeTailReceiverRequest) returns (VolumeTailReceiverResponse) {
+    }
+
+    // erasure coding
+    rpc VolumeEcShardsGenerate (VolumeEcShardsGenerateRequest) returns (VolumeEcShardsGenerateResponse) {
+    }
+    rpc VolumeEcShardsRebuild (VolumeEcShardsRebuildRequest) returns (VolumeEcShardsRebuildResponse) {
+    }
+    rpc VolumeEcShardsCopy (VolumeEcShardsCopyRequest) returns (VolumeEcShardsCopyResponse) {
+    }
+    rpc VolumeEcShardsDelete (VolumeEcShardsDeleteRequest) returns (VolumeEcShardsDeleteResponse) {
+    }
+    rpc VolumeEcShardsMount (VolumeEcShardsMountRequest) returns (VolumeEcShardsMountResponse) {
+    }
+    rpc VolumeEcShardsUnmount (VolumeEcShardsUnmountRequest) returns (VolumeEcShardsUnmountResponse) {
+    }
+    rpc VolumeEcShardRead (VolumeEcShardReadRequest) returns (stream VolumeEcShardReadResponse) {
+    }
+    rpc VolumeEcBlobDelete (VolumeEcBlobDeleteRequest) returns (VolumeEcBlobDeleteResponse) {
+    }
+    rpc VolumeEcShardsToVolume (VolumeEcShardsToVolumeRequest) returns (VolumeEcShardsToVolumeResponse) {
+    }
+    rpc VolumeEcShardsInfo (VolumeEcShardsInfoRequest) returns (VolumeEcShardsInfoResponse) {
+    }
+
+    // tiered storage
+    rpc VolumeTierMoveDatToRemote (VolumeTierMoveDatToRemoteRequest) returns (stream VolumeTierMoveDatToRemoteResponse) {
+    }
+    rpc VolumeTierMoveDatFromRemote (VolumeTierMoveDatFromRemoteRequest) returns (stream VolumeTierMoveDatFromRemoteResponse) {
+    }
+
+    rpc VolumeServerStatus (VolumeServerStatusRequest) returns (VolumeServerStatusResponse) {
+    }
+    rpc VolumeServerLeave (VolumeServerLeaveRequest) returns (VolumeServerLeaveResponse) {
+    }
+
+    // remote storage
+    rpc FetchAndWriteNeedle (FetchAndWriteNeedleRequest) returns (FetchAndWriteNeedleResponse) {
+    }
+
+    // scrubbing
+    rpc ScrubVolume (ScrubVolumeRequest) returns (ScrubVolumeResponse) {
+    }
+    rpc ScrubEcVolume (ScrubEcVolumeRequest) returns (ScrubEcVolumeResponse) {
+    }
+
+    // <experimental> query
+    rpc Query (QueryRequest) returns (stream QueriedStripe) {
+    }
+
+    rpc VolumeNeedleStatus (VolumeNeedleStatusRequest) returns (VolumeNeedleStatusResponse) {
+    }
+
+    rpc Ping (PingRequest) returns (PingResponse) {
+    }
+
+}
+
+//////////////////////////////////////////////////
+
+message BatchDeleteRequest {
+    repeated string file_ids = 1;
+    bool skip_cookie_check = 2;
+}
+
+message BatchDeleteResponse {
+    repeated DeleteResult results = 1;
+}
+message DeleteResult {
+    string file_id = 1;
+    int32 status = 2;
+    string error = 3;
+    uint32 size = 4;
+    uint32 version = 5;
+}
+
+message Empty {
+}
+
+message VacuumVolumeCheckRequest {
+    uint32 volume_id = 1;
+}
+message VacuumVolumeCheckResponse {
+    double garbage_ratio = 1;
+}
+
+message VacuumVolumeCompactRequest {
+    uint32 volume_id = 1;
+    int64 preallocate = 2;
+}
+message VacuumVolumeCompactResponse {
+    int64 processed_bytes = 1;
+    float load_avg_1m = 2;
+}
+
+message VacuumVolumeCommitRequest {
+    uint32 volume_id = 1;
+}
+message VacuumVolumeCommitResponse {
+    bool is_read_only = 1;
+    uint64 volume_size = 2;
+}
+
+message VacuumVolumeCleanupRequest {
+    uint32 volume_id = 1;
+}
+message VacuumVolumeCleanupResponse {
+}
+
+message DeleteCollectionRequest {
+    string collection = 1;
+}
+message DeleteCollectionResponse {
+}
+
+message AllocateVolumeRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    int64 preallocate = 3;
+    string replication = 4;
+    string ttl = 5;
+    uint32 memory_map_max_size_mb = 6;
+    string disk_type = 7;
+    uint32 version = 8;
+}
+message AllocateVolumeResponse {
+}
+
+message VolumeSyncStatusRequest {
+    uint32 volume_id = 1;
+}
+message VolumeSyncStatusResponse {
+    uint32 volume_id = 1;
+    string collection = 2;
+    string replication = 4;
+    string ttl = 5;
+    uint64 tail_offset = 6;
+    uint32 compact_revision = 7;
+    uint64 idx_file_size = 8;
+    uint32 version = 9;
+}
+
+message VolumeIncrementalCopyRequest {
+    uint32 volume_id = 1;
+    uint64 since_ns = 2;
+}
+message VolumeIncrementalCopyResponse {
+    bytes file_content = 1;
+}
+
+message VolumeMountRequest {
+    uint32 volume_id = 1;
+    optional string collection = 2;
+}
+message VolumeMountResponse {
+}
+
+message VolumeUnmountRequest {
+    uint32 volume_id = 1;
+}
+message VolumeUnmountResponse {
+}
+
+message VolumeConsolidateIndexRequest {
+    uint32 volume_id = 1;
+}
+message VolumeConsolidateIndexResponse {
+}
+
+message VolumeDeleteRequest {
+    uint32 volume_id = 1;
+    bool only_empty = 2;
+    // when true, do not remove the cloud-tier object backing the volume.
+    // used for moves where another server is taking over the same .vif.
+    bool keep_remote_data = 3;
+}
+message VolumeDeleteResponse {
+}
+
+message VolumeMarkReadonlyRequest {
+    uint32 volume_id = 1;
+    bool persist = 2;
+    // reject writes but keep accepting deletes, so expiring data can drain the volume
+    bool can_delete = 3;
+}
+message VolumeMarkReadonlyResponse {
+}
+
+message VolumeMarkWritableRequest {
+    uint32 volume_id = 1;
+}
+message VolumeMarkWritableResponse {
+}
+
+message VolumeConfigureRequest {
+    uint32 volume_id = 1;
+    string replication = 2;
+}
+message VolumeConfigureResponse {
+    string error = 1;
+}
+
+message VolumeStatusRequest {
+    uint32 volume_id = 1;
+}
+message VolumeStatusResponse {
+    bool is_read_only = 1;
+    uint64 volume_size = 2;
+    uint64 file_count = 3;
+    uint64 file_deleted_count = 4;
+}
+
+message GetStateRequest {
+}
+message GetStateResponse {
+    VolumeServerState state = 1;
+}
+
+message SetStateRequest {
+    // SetState updates *all* volume server flags at once. Retrieve state with GetState(),
+    // modify individual flags as required, then call this RPC to update.
+    VolumeServerState state = 1;
+}
+message SetStateResponse {
+    VolumeServerState state = 1;
+}
+
+message VolumeCopyRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    string replication = 3;
+    string ttl = 4;
+    string source_data_node = 5;
+    string disk_type = 6;
+    int64 io_byte_per_second = 7;
+}
+message VolumeCopyResponse {
+    uint64 last_append_at_ns = 1;
+    int64 processed_bytes = 2;
+}
+
+message CopyFileRequest {
+    uint32 volume_id = 1;
+    string ext = 2;
+    uint32 compaction_revision = 3;
+    uint64 stop_offset = 4;
+    string collection = 5;
+    bool is_ec_volume = 6;
+    bool ignore_source_file_not_found = 7;
+}
+message CopyFileResponse {
+    bytes file_content = 1;
+    int64 modified_ts_ns = 2;
+}
+
+message ReceiveFileRequest {
+    oneof data {
+        ReceiveFileInfo info = 1;
+        bytes file_content = 2;
+    }
+}
+
+message ReceiveFileInfo {
+    uint32 volume_id = 1;
+    string ext = 2;
+    string collection = 3;
+    bool is_ec_volume = 4;
+    uint32 shard_id = 5;
+    uint64 file_size = 6;
+    uint32 disk_id = 7;  // EC shard disk; 0 = auto-select (see VolumeEcShardsCopyRequest.disk_id)
+
+    // Field numbers 8-11 are reserved for versioned-EC; disk_type stays at 12.
+    // Staged-new-volume mode (EC decode onto a clean peer): set on a non-EC push
+    // whose volume does not yet exist on this server. The server picks a disk
+    // location of this medium with a free slot and writes <base><ext>.copying,
+    // finalized by VolumeEcShardsToVolume(from_staged).
+    string disk_type = 12;
+}
+
+message ReceiveFileResponse {
+    uint64 bytes_written = 1;
+    string error = 2;
+}
+
+message ReadNeedleBlobRequest {
+    uint32 volume_id = 1;
+    int64 offset = 3; // actual offset
+    int32 size = 4;
+}
+message ReadNeedleBlobResponse {
+    bytes needle_blob = 1;
+}
+
+message ReadNeedleMetaRequest {
+    uint32 volume_id = 1;
+    uint64 needle_id = 2;
+    int64 offset = 3; // actual offset
+    int32 size = 4;
+}
+message ReadNeedleMetaResponse {
+    uint32 cookie = 1;
+    uint64 last_modified = 2;
+    uint32 crc = 3;
+    string ttl = 4;
+    uint64 append_at_ns = 5;
+}
+
+message WriteNeedleBlobRequest {
+    uint32 volume_id = 1;
+    uint64 needle_id = 2;
+    int32 size = 3;
+    bytes needle_blob = 4;
+}
+message WriteNeedleBlobResponse {
+}
+
+message ReadAllNeedlesRequest {
+    repeated uint32 volume_ids = 1;
+}
+message ReadAllNeedlesResponse {
+    uint32 volume_id = 1;
+    uint64 needle_id = 2;
+    uint32 cookie = 3;
+    bytes needle_blob = 5;
+    bool needle_blob_compressed = 6;
+    uint64 last_modified = 7;
+    uint32 crc = 8;
+    bytes name = 9;
+    bytes mime = 10;
+}
+
+message VolumeTailSenderRequest {
+    uint32 volume_id = 1;
+    uint64 since_ns = 2;
+    uint32 idle_timeout_seconds = 3;
+}
+message VolumeTailSenderResponse {
+    bytes needle_header = 1;
+    bytes needle_body = 2;
+    bool is_last_chunk = 3;
+    uint32 version = 4;
+}
+
+message VolumeTailReceiverRequest {
+    uint32 volume_id = 1;
+    uint64 since_ns = 2;
+    uint32 idle_timeout_seconds = 3;
+    string source_volume_server = 4;
+}
+message VolumeTailReceiverResponse {
+}
+
+message VolumeEcShardsGenerateRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+}
+message VolumeEcShardsGenerateResponse {
+}
+
+message VolumeEcShardsRebuildRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+}
+message VolumeEcShardsRebuildResponse {
+    repeated uint32 rebuilt_shard_ids = 1;
+}
+
+message VolumeEcShardsCopyRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    repeated uint32 shard_ids = 3;
+    bool copy_ecx_file = 4;
+    string source_data_node = 5;
+    bool copy_ecj_file = 6;
+    bool copy_vif_file = 7;
+    uint32 disk_id = 8;  // Target disk ID for storing EC shards
+    bool copy_ecsum_file = 9; // copy the bitrot checksum sidecar (.ecsum) when present; tolerant of a missing source (no-op), since this non-2PC path has no Prepare backstop
+}
+message VolumeEcShardsCopyResponse {
+}
+
+message VolumeEcShardsDeleteRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    repeated uint32 shard_ids = 3;
+    bool full_teardown = 4; // pre-encode cleanup: wipe every EC artifact + generation for this volume, not just shard_ids
+    int64 encode_ts_ns = 5; // full_teardown generation fence: delete only a disk whose .vif generation is strictly OLDER than this; preserve same-or-newer, generation 0, and an unreadable .vif. 0 => wipe-all (shell pre-encode / pre-upgrade)
+}
+message VolumeEcShardsDeleteResponse {
+    bool full_teardown_done = 1; // set by a new server that performed full_teardown; absent from an old server lets the caller detect the silent no-op
+}
+
+message VolumeEcShardsMountRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    repeated uint32 shard_ids = 3;
+    string source_disk_type = 4; // disk type of the source volume, applied to the in-memory EC volume so heartbeats report under it (#9423)
+    bool recover_missing_index = 5; // first fetch a missing .ecx index from a peer so on-disk shards without a local index become mountable (#10104)
+}
+message VolumeEcShardsMountResponse {
+}
+
+message VolumeEcShardsUnmountRequest {
+    uint32 volume_id = 1;
+    repeated uint32 shard_ids = 3;
+    int64 encode_ts_ns = 4; // generation fence: skip a disk whose mounted EC volume is this generation or newer (0 = unfenced, unmount all)
+}
+message VolumeEcShardsUnmountResponse {
+}
+
+message VolumeEcShardReadRequest {
+    uint32 volume_id = 1;
+    uint32 shard_id = 2;
+    int64 offset = 3;
+    int64 size = 4;
+    uint64 file_key = 5;
+    reserved 6;
+    int64 encode_ts_ns = 7; // caller's expected encode time; the server rejects a shard from a different encode run
+}
+message VolumeEcShardReadResponse {
+    bytes data = 1;
+    bool is_deleted = 2;
+    int64 encode_ts_ns = 3; // identity of the shard actually served; client rejects a mismatch (0 = pre-upgrade server)
+}
+
+message VolumeEcBlobDeleteRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    uint64 file_key = 3;
+    uint32 version = 4;
+}
+message VolumeEcBlobDeleteResponse {
+}
+
+message VolumeEcShardsToVolumeRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    // Staged mode: the caller already decoded the EC shards off-box and pushed
+    // .dat/.idx/.vif as <base><ext>.copying to this server (ReceiveFile
+    // staged-new-volume mode). Adopt them as a normal volume instead of decoding
+    // local EC shards in place, so <vid> is never registered as both EC and
+    // normal on one disk.
+    bool from_staged = 3;
+    string disk_type = 4; // target medium's disk location for the normal volume (staged mode)
+}
+message VolumeEcShardsToVolumeResponse {
+}
+
+message VolumeEcShardsInfoRequest {
+    uint32 volume_id = 1;
+}
+message VolumeEcShardsInfoResponse {
+    repeated EcShardInfo ec_shard_infos = 1;
+    uint64 volume_size = 2;
+    uint64 file_count = 3;
+    uint64 file_deleted_count = 4;
+    EcShardConfig ec_shard_config = 10;  // the layout this holder serves reads through; a binary predating a field reports it as unset
+}
+
+message EcShardInfo {
+    uint32 shard_id = 1;
+    int64 size = 2;
+    string collection = 3;
+    uint32 volume_id = 4;
+}
+
+message ReadVolumeFileStatusRequest {
+    uint32 volume_id = 1;
+}
+message ReadVolumeFileStatusResponse {
+    uint32 volume_id = 1;
+    uint64 idx_file_timestamp_seconds = 2;
+    uint64 idx_file_size = 3;
+    uint64 dat_file_timestamp_seconds = 4;
+    uint64 dat_file_size = 5;
+    uint64 file_count = 6;
+    uint32 compaction_revision = 7;
+    string collection = 8;
+    string disk_type = 9;
+    VolumeInfo volume_info = 10;
+    uint32 version = 11;
+}
+
+message DiskStatus {
+    string dir = 1;
+    uint64 all = 2;
+    uint64 used = 3;
+    uint64 free = 4;
+    float percent_free = 5;
+    float percent_used = 6;
+    string disk_type = 7;
+}
+
+message MemStatus {
+    int32 goroutines = 1;
+    uint64 all = 2;
+    uint64 used = 3;
+    uint64 free = 4;
+    uint64 self = 5;
+    uint64 heap = 6;
+    uint64 stack = 7;
+}
+
+// tired storage on volume servers
+message RemoteFile {
+    string backend_type = 1;
+    string backend_id = 2;
+    string key = 3;
+    uint64 offset = 4;
+    uint64 file_size = 5;
+    uint64 modified_time = 6;
+    string extension = 7;
+}
+message VolumeInfo {
+    repeated RemoteFile files = 1;
+    uint32 version = 2;
+    string replication = 3;
+    uint32 bytes_offset = 4;
+    int64 dat_file_size = 5; // store the original dat file size
+    uint64 expire_at_sec = 6; // expiration time of ec volume
+    bool read_only = 7;
+    EcShardConfig ec_shard_config = 8; // EC shard configuration (optional, null = use default 10+4)
+    bool read_only_can_delete = 9; // with read_only: writes are rejected but deletes still land
+}
+
+// EcShardConfig specifies erasure coding shard configuration
+message EcShardConfig {
+    uint32 data_shards = 1;   // Number of data shards (e.g., 10)
+    uint32 parity_shards = 2; // Number of parity shards (e.g., 4)
+    int64 encode_ts_ns = 3;   // encode time (unix nanos); a read served from a shard of a different encode run is rejected
+    int64 block_size = 4;     // uniform block layout: each shard is a single contiguous block of this many bytes; 0 = legacy 1GiB/1MiB two-tier layout
+}
+// EcBitrotProtection is the entire content of a bitrot checksum sidecar
+// (<base>.ecsum for the legacy generation, <base>.ecsum.v<N> for vacuum
+// generation N): per-shard, per-block CRC32C so a CHECKSUM scrub can detect
+// silent bit rot in any shard (including parity) without decoding. Field
+// numbers and types match weed/pb/volume_server.proto byte-for-byte so the
+// serialized sidecar payload is wire-identical across the Go and Rust binaries.
+message EcBitrotProtection {
+    ChecksumAlgorithm algorithm = 1;       // CRC32C (Castagnoli)
+    uint32 block_size = 2;                  // bytes per checksum block; default 16777216 (16 MiB), a power-of-two multiple of 1 MiB
+    uint32 generation = 3;                  // EC vacuum generation these checksums describe (0 = legacy/fresh); must match the sidecar filename version
+    EcShardConfig ec_shard_config = 4;      // data/parity shard counts at encode time
+    repeated EcShardChecksums shards = 5;   // one entry per shard id in the active layout
+    bytes encode_uuid = 6;                  // random per-encode identity, for stale-sidecar detection across in-place re-encodes
+}
+message EcShardChecksums {
+    uint32 shard_id = 1;     // 0..MaxShardCount-1 (custom EC ratios go up to 32)
+    int64 covered_size = 2;  // shard byte length these checksums cover (must equal the on-disk shard length)
+    bytes block_crc32c = 3;  // packed little-endian uint32[] = ceil(covered_size/block_size) entries
+}
+enum ChecksumAlgorithm {
+    CHECKSUM_NONE = 0;
+    CHECKSUM_CRC32C = 1;
+}
+message OldVersionVolumeInfo {
+    repeated RemoteFile files = 1;
+    uint32 version = 2;
+    string replication = 3;
+    uint32 BytesOffset = 4;
+    int64 dat_file_size = 5; // store the original dat file size
+    uint64 DestroyTime = 6; // expiration time of ec volume
+    bool read_only = 7;
+}
+
+// tiered storage
+message VolumeTierMoveDatToRemoteRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    string destination_backend_name = 3;
+    bool keep_local_dat_file = 4;
+}
+message VolumeTierMoveDatToRemoteResponse {
+    int64 processed = 1;
+    float processedPercentage = 2;
+}
+
+message VolumeTierMoveDatFromRemoteRequest {
+    uint32 volume_id = 1;
+    string collection = 2;
+    bool keep_remote_dat_file = 3;
+}
+message VolumeTierMoveDatFromRemoteResponse {
+    int64 processed = 1;
+    float processedPercentage = 2;
+}
+
+message VolumeServerStatusRequest {
+
+}
+message VolumeServerStatusResponse {
+    repeated DiskStatus disk_statuses = 1;
+    MemStatus memory_status = 2;
+    string version = 3;
+    string data_center = 4;
+    string rack = 5;
+    VolumeServerState state = 6;
+}
+
+message VolumeServerLeaveRequest {
+}
+message VolumeServerLeaveResponse {
+}
+
+// remote storage
+message FetchAndWriteNeedleRequest {
+    uint32 volume_id = 1;
+    uint64 needle_id = 2;
+    uint32 cookie = 3;
+    int64 offset = 4;
+    int64 size = 5;
+    message Replica {
+        string url = 1;
+        string public_url = 2;
+        int32 grpc_port = 3;
+    }
+    repeated Replica replicas = 6;
+    string auth = 7;
+    int32 download_concurrency = 8; // multipart download concurrency if supported by the remote storage client; for S3, 0 = default (5)
+    // remote conf
+    remote_pb.RemoteConf remote_conf = 15;
+    remote_pb.RemoteStorageLocation remote_location = 16;
+}
+message FetchAndWriteNeedleResponse {
+    string e_tag = 1;
+}
+
+enum VolumeScrubMode {
+    UNKNOWN = 0;
+    INDEX = 1;
+    FULL = 2;
+    LOCAL = 3;
+    CHECKSUM = 4; // EC only: verify each local shard's raw bytes against the bitrot checksum sidecar
+    READS = 5;    // like FULL, but EC intervals no shard can serve are reconstructed from parity
+}
+
+message ScrubVolumeRequest {
+    VolumeScrubMode mode = 1;
+    // optional list of volume IDs to scrub. if empty, all volumes for the server are scrubbed.
+    repeated uint32 volume_ids = 2;
+    bool mark_broken_volumes_readonly = 3;
+}
+message ScrubVolumeResponse {
+    uint64 total_volumes = 1;
+    uint64 total_files = 2;
+    repeated uint32 broken_volume_ids = 3;
+    repeated string details = 4;
+}
+
+message ScrubEcVolumeRequest {
+    VolumeScrubMode mode = 1;
+    // optional list of volume IDs to scrub. if empty, all EC volumes for the server are scrubbed.
+    repeated uint32 volume_ids = 2;
+    bool force_deleted_needles_check = 3; // FULL mode only; may report false positives when EC indexes disagree
+}
+message ScrubEcVolumeResponse {
+    uint64 total_volumes = 1;
+    uint64 total_files = 2;
+    repeated uint32 broken_volume_ids = 3;
+    repeated EcShardInfo broken_shard_infos = 4;
+    repeated string details = 5;
+}
+
+// select on volume servers
+message QueryRequest {
+    repeated string selections = 1;
+    repeated string from_file_ids = 2;
+    message Filter {
+        string field = 1;
+        string operand = 2;
+        string value = 3;
+    }
+    Filter filter = 3;
+
+    message InputSerialization {
+        // NONE | GZIP | BZIP2
+        string compression_type = 1;
+        message CSVInput {
+            string file_header_info = 1; // Valid values: NONE | USE | IGNORE
+            string record_delimiter = 2; // Default: \n
+            string field_delimiter = 3; // Default: ,
+            string quote_character = 4; // Default: "
+            string quote_escape_character = 5; // Default: "
+            string comments = 6; // Default: #
+            // If true, records might contain record delimiters within quote characters
+            bool allow_quoted_record_delimiter = 7; // default False.
+        }
+        message JSONInput {
+            string type = 1; // Valid values: DOCUMENT | LINES
+        }
+        message ParquetInput {
+        }
+
+        CSVInput csv_input = 2;
+        JSONInput json_input = 3;
+        ParquetInput parquet_input = 4;
+    }
+    InputSerialization input_serialization = 4;
+
+    message OutputSerialization {
+        message CSVOutput {
+            string quote_fields = 1; // Valid values: ALWAYS | ASNEEDED
+            string record_delimiter = 2; // Default: \n
+            string field_delimiter = 3; // Default: ,
+            string quote_character = 4; // Default: "
+            string quote_escape_character = 5; // Default: "
+        }
+        message JSONOutput {
+            string record_delimiter = 1;
+        }
+
+        CSVOutput csv_output = 2;
+        JSONOutput json_output = 3;
+    }
+
+    OutputSerialization output_serialization = 5;
+}
+message QueriedStripe {
+    bytes records = 1;
+}
+
+message VolumeNeedleStatusRequest {
+    uint32 volume_id = 1;
+    uint64 needle_id = 2;
+}
+message VolumeNeedleStatusResponse {
+    uint64 needle_id = 1;
+    uint32 cookie = 2;
+    uint32 size = 3;
+    uint64 last_modified = 4;
+    uint32 crc = 5;
+    string ttl = 6;
+}
+
+message PingRequest {
+    string target = 1; // default to ping itself
+    string target_type = 2;
+}
+message PingResponse {
+    int64 start_time_ns = 1;
+    int64 remote_time_ns = 2;
+    int64 stop_time_ns = 3;
+}

seaweed-volume/src/images.rs

@@ -0,0 +1,275 @@
+//! JPEG EXIF orientation auto-fix, matching Go's `FixJpgOrientation`.
+//!
+//! Reads the EXIF orientation tag from JPEG data and rotates/flips the image
+//! to normalize it to orientation 1 (top-left). If EXIF parsing fails or
+//! orientation is already normal, returns the original data unchanged.
+
+use std::io::Cursor;
+
+use image::{DynamicImage, GenericImageView, ImageFormat, RgbaImage};
+
+/// EXIF orientation tag values.
+/// See: <http://sylvana.net/jpegcrop/exif_orientation.html>
+const TOP_LEFT_SIDE: u32 = 1;
+const TOP_RIGHT_SIDE: u32 = 2;
+const BOTTOM_RIGHT_SIDE: u32 = 3;
+const BOTTOM_LEFT_SIDE: u32 = 4;
+const LEFT_SIDE_TOP: u32 = 5;
+const RIGHT_SIDE_TOP: u32 = 6;
+const RIGHT_SIDE_BOTTOM: u32 = 7;
+const LEFT_SIDE_BOTTOM: u32 = 8;
+
+/// Fix JPEG orientation based on EXIF data.
+///
+/// Reads the EXIF orientation tag and applies the appropriate rotation/flip
+/// to normalize the image to orientation 1 (top-left). Re-encodes as JPEG.
+///
+/// Returns the original data unchanged if:
+/// - EXIF data cannot be parsed
+/// - No orientation tag is present
+/// - Orientation is already 1 (normal)
+/// - Image decoding or re-encoding fails
+pub fn fix_jpg_orientation(data: &[u8]) -> Vec<u8> {
+    // Parse EXIF data
+    let orientation = match read_exif_orientation(data) {
+        Some(o) => o,
+        None => return data.to_vec(),
+    };
+
+    // Orientation 1 means normal — no transformation needed
+    if orientation == TOP_LEFT_SIDE {
+        return data.to_vec();
+    }
+
+    // Determine rotation angle and flip mode
+    let (angle, flip_horizontal) = match orientation {
+        TOP_RIGHT_SIDE => (0, true),
+        BOTTOM_RIGHT_SIDE => (180, false),
+        BOTTOM_LEFT_SIDE => (180, true),
+        LEFT_SIDE_TOP => (-90, true),
+        RIGHT_SIDE_TOP => (-90, false),
+        RIGHT_SIDE_BOTTOM => (90, true),
+        LEFT_SIDE_BOTTOM => (90, false),
+        _ => return data.to_vec(),
+    };
+
+    // Decode the image
+    let src_image = match image::load_from_memory_with_format(data, ImageFormat::Jpeg) {
+        Ok(img) => img,
+        Err(_) => return data.to_vec(),
+    };
+
+    // Apply rotation then flip (matching Go's flip(rotate(img, angle), flipMode))
+    let transformed = flip_horizontal_if(rotate(src_image, angle), flip_horizontal);
+
+    // Re-encode as JPEG
+    let mut buf = Cursor::new(Vec::new());
+    match transformed.write_to(&mut buf, ImageFormat::Jpeg) {
+        Ok(_) => buf.into_inner(),
+        Err(_) => data.to_vec(),
+    }
+}
+
+/// Read the EXIF orientation tag from JPEG data.
+/// Returns None if EXIF cannot be parsed or orientation tag is not present.
+fn read_exif_orientation(data: &[u8]) -> Option<u32> {
+    let exif_reader = exif::Reader::new();
+    let mut cursor = Cursor::new(data);
+    let exif_data = exif_reader.read_from_container(&mut cursor).ok()?;
+
+    let orientation_field = exif_data.get_field(exif::Tag::Orientation, exif::In::PRIMARY)?;
+    match orientation_field.value {
+        exif::Value::Short(ref v) if !v.is_empty() => Some(v[0] as u32),
+        _ => orientation_field.value.get_uint(0),
+    }
+}
+
+/// Rotate an image by the given angle (counter-clockwise, in degrees).
+/// Matches Go's rotate function.
+fn rotate(img: DynamicImage, angle: i32) -> DynamicImage {
+    let (width, height) = img.dimensions();
+
+    match angle {
+        90 => {
+            // 90 degrees counter-clockwise
+            let new_w = height;
+            let new_h = width;
+            let mut out = RgbaImage::new(new_w, new_h);
+            for y in 0..new_h {
+                for x in 0..new_w {
+                    out.put_pixel(x, y, img.get_pixel(new_h - 1 - y, x));
+                }
+            }
+            DynamicImage::ImageRgba8(out)
+        }
+        -90 => {
+            // 90 degrees clockwise (or 270 counter-clockwise)
+            let new_w = height;
+            let new_h = width;
+            let mut out = RgbaImage::new(new_w, new_h);
+            for y in 0..new_h {
+                for x in 0..new_w {
+                    out.put_pixel(x, y, img.get_pixel(y, new_w - 1 - x));
+                }
+            }
+            DynamicImage::ImageRgba8(out)
+        }
+        180 | -180 => {
+            let mut out = RgbaImage::new(width, height);
+            for y in 0..height {
+                for x in 0..width {
+                    out.put_pixel(x, y, img.get_pixel(width - 1 - x, height - 1 - y));
+                }
+            }
+            DynamicImage::ImageRgba8(out)
+        }
+        _ => img,
+    }
+}
+
+/// Flip the image horizontally if requested.
+/// In Go, flipMode 2 == FlipHorizontal. We simplify since only horizontal flip is used.
+fn flip_horizontal_if(img: DynamicImage, do_flip: bool) -> DynamicImage {
+    if !do_flip {
+        return img;
+    }
+    let (width, height) = img.dimensions();
+    let mut out = RgbaImage::new(width, height);
+    for y in 0..height {
+        for x in 0..width {
+            out.put_pixel(x, y, img.get_pixel(width - 1 - x, y));
+        }
+    }
+    DynamicImage::ImageRgba8(out)
+}
+
+/// Returns true if the given MIME type or file path extension indicates a JPEG file.
+pub fn is_jpeg(mime_type: &str, path: &str) -> bool {
+    if mime_type == "image/jpeg" {
+        return true;
+    }
+    let lower = path.to_lowercase();
+    lower.ends_with(".jpg") || lower.ends_with(".jpeg")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_non_jpeg_data_returned_unchanged() {
+        let data = b"not a jpeg file at all";
+        let result = fix_jpg_orientation(data);
+        assert_eq!(result, data);
+    }
+
+    #[test]
+    fn test_jpeg_without_exif_returned_unchanged() {
+        // Create a minimal JPEG without EXIF data
+        let img = DynamicImage::ImageRgba8(RgbaImage::new(2, 2));
+        let mut buf = Cursor::new(Vec::new());
+        img.write_to(&mut buf, ImageFormat::Jpeg).unwrap();
+        let jpeg_data = buf.into_inner();
+
+        let result = fix_jpg_orientation(&jpeg_data);
+        // Should return data unchanged (no EXIF orientation tag)
+        // Just verify it's still valid JPEG
+        assert!(!result.is_empty());
+        assert_eq!(&result[0..2], &[0xFF, 0xD8]); // JPEG magic bytes
+    }
+
+    #[test]
+    fn test_is_jpeg() {
+        assert!(is_jpeg("image/jpeg", ""));
+        assert!(is_jpeg("", "/3,abc.jpg"));
+        assert!(is_jpeg("", "/3,abc.JPEG"));
+        assert!(is_jpeg("application/octet-stream", "/3,abc.JPG"));
+        assert!(!is_jpeg("image/png", "/3,abc.png"));
+        assert!(!is_jpeg("", "/3,abc.png"));
+    }
+
+    #[test]
+    fn test_rotate_180() {
+        // Create a 2x2 image with distinct pixel colors
+        let mut img = RgbaImage::new(2, 2);
+        img.put_pixel(0, 0, image::Rgba([255, 0, 0, 255])); // red top-left
+        img.put_pixel(1, 0, image::Rgba([0, 255, 0, 255])); // green top-right
+        img.put_pixel(0, 1, image::Rgba([0, 0, 255, 255])); // blue bottom-left
+        img.put_pixel(1, 1, image::Rgba([255, 255, 0, 255])); // yellow bottom-right
+        let dynamic = DynamicImage::ImageRgba8(img);
+
+        let rotated = rotate(dynamic, 180);
+        let (w, h) = rotated.dimensions();
+        assert_eq!((w, h), (2, 2));
+        // After 180 rotation: top-left should be yellow, top-right should be blue
+        assert_eq!(rotated.get_pixel(0, 0), image::Rgba([255, 255, 0, 255]));
+        assert_eq!(rotated.get_pixel(1, 0), image::Rgba([0, 0, 255, 255]));
+        assert_eq!(rotated.get_pixel(0, 1), image::Rgba([0, 255, 0, 255]));
+        assert_eq!(rotated.get_pixel(1, 1), image::Rgba([255, 0, 0, 255]));
+    }
+
+    #[test]
+    fn test_rotate_90_ccw() {
+        // Create 3x2 image (width=3, height=2)
+        let mut img = RgbaImage::new(3, 2);
+        img.put_pixel(0, 0, image::Rgba([1, 0, 0, 255]));
+        img.put_pixel(1, 0, image::Rgba([2, 0, 0, 255]));
+        img.put_pixel(2, 0, image::Rgba([3, 0, 0, 255]));
+        img.put_pixel(0, 1, image::Rgba([4, 0, 0, 255]));
+        img.put_pixel(1, 1, image::Rgba([5, 0, 0, 255]));
+        img.put_pixel(2, 1, image::Rgba([6, 0, 0, 255]));
+        let dynamic = DynamicImage::ImageRgba8(img);
+
+        let rotated = rotate(dynamic, 90);
+        let (w, h) = rotated.dimensions();
+        // 90 CCW: width=3,height=2 -> new_w=2, new_h=3
+        assert_eq!((w, h), (2, 3));
+        // Top-right (2,0) should move to top-left (0,0) in CCW 90
+        assert_eq!(rotated.get_pixel(0, 0)[0], 3);
+        assert_eq!(rotated.get_pixel(1, 0)[0], 6);
+    }
+
+    #[test]
+    fn test_rotate_neg90_cw() {
+        // Create 3x2 image
+        let mut img = RgbaImage::new(3, 2);
+        img.put_pixel(0, 0, image::Rgba([1, 0, 0, 255]));
+        img.put_pixel(1, 0, image::Rgba([2, 0, 0, 255]));
+        img.put_pixel(2, 0, image::Rgba([3, 0, 0, 255]));
+        img.put_pixel(0, 1, image::Rgba([4, 0, 0, 255]));
+        img.put_pixel(1, 1, image::Rgba([5, 0, 0, 255]));
+        img.put_pixel(2, 1, image::Rgba([6, 0, 0, 255]));
+        let dynamic = DynamicImage::ImageRgba8(img);
+
+        let rotated = rotate(dynamic, -90);
+        let (w, h) = rotated.dimensions();
+        assert_eq!((w, h), (2, 3));
+        // -90 (CW 90): top-left (0,0) should go to top-right
+        assert_eq!(rotated.get_pixel(0, 0)[0], 4);
+        assert_eq!(rotated.get_pixel(1, 0)[0], 1);
+    }
+
+    #[test]
+    fn test_flip_horizontal() {
+        let mut img = RgbaImage::new(2, 1);
+        img.put_pixel(0, 0, image::Rgba([10, 0, 0, 255]));
+        img.put_pixel(1, 0, image::Rgba([20, 0, 0, 255]));
+        let dynamic = DynamicImage::ImageRgba8(img);
+
+        let flipped = flip_horizontal_if(dynamic, true);
+        assert_eq!(flipped.get_pixel(0, 0)[0], 20);
+        assert_eq!(flipped.get_pixel(1, 0)[0], 10);
+    }
+
+    #[test]
+    fn test_flip_horizontal_noop() {
+        let mut img = RgbaImage::new(2, 1);
+        img.put_pixel(0, 0, image::Rgba([10, 0, 0, 255]));
+        img.put_pixel(1, 0, image::Rgba([20, 0, 0, 255]));
+        let dynamic = DynamicImage::ImageRgba8(img);
+
+        let not_flipped = flip_horizontal_if(dynamic, false);
+        assert_eq!(not_flipped.get_pixel(0, 0)[0], 10);
+        assert_eq!(not_flipped.get_pixel(1, 0)[0], 20);
+    }
+}

seaweed-volume/src/lib.rs

@@ -0,0 +1,28 @@
+pub mod config;
+pub mod images;
+pub mod malloc_tuning;
+pub mod metrics;
+pub mod remote_storage;
+pub mod security;
+pub mod server;
+pub mod storage;
+pub mod version;
+
+/// Generated protobuf modules.
+pub mod pb {
+    pub const FILE_DESCRIPTOR_SET: &[u8] =
+        tonic::include_file_descriptor_set!("seaweed_descriptor");
+
+    pub mod remote_pb {
+        tonic::include_proto!("remote_pb");
+    }
+    pub mod volume_server_pb {
+        tonic::include_proto!("volume_server_pb");
+    }
+    pub mod master_pb {
+        tonic::include_proto!("master_pb");
+    }
+    pub mod filer_pb {
+        tonic::include_proto!("filer_pb");
+    }
+}

seaweed-volume/src/malloc_tuning.rs

@@ -0,0 +1,467 @@
+//! Keep glibc from silently converting large short-lived buffers into heap the
+//! process never gives back.
+//!
+//! glibc serves an allocation with `mmap` when it is at least
+//! `M_MMAP_THRESHOLD` (128 KiB by default), and `munmap`s it on free, so the
+//! pages go straight back to the OS. That threshold is **adaptive**: whenever a
+//! block that came from `mmap` is freed, glibc raises the threshold to that
+//! block's size — up to 32 MiB — on the theory that a workload repeatedly
+//! allocating buffers of that size is better served from the heap.
+//!
+//! For a volume server that theory is wrong in a specific, expensive way. EC
+//! reconstruction and needle reassembly allocate large, short-lived buffers.
+//! The first few are mmap'd and freed, which trains the threshold upward; every
+//! later buffer of that size is then carved out of the heap instead. Heap
+//! memory is only returned to the OS from the top of the arena, so those pages
+//! stay resident as anonymous memory for the life of the process. They are
+//! still *reusable* — this is not a leak, and a repeat workload does not grow
+//! the footprint further — but under a hard cgroup `MemoryMax` they are
+//! indistinguishable from a leak, because anonymous pages cannot be reclaimed
+//! under pressure the way page cache can. The retained footprint eats exactly
+//! the headroom that a burst of maintenance work needs, and the process is
+//! OOM-killed while most of its resident memory is free-but-unreturned.
+//!
+//! Measured on a 17-node cluster (EC 10+4, `--index=redb`), one node, two
+//! identical `ec.scrub -mode full` rounds over 10912 EC files each, comparing
+//! the same unit restarted with and without a pinned threshold:
+//!
+//! | | baseline | round 1 | round 2 | 60s idle |
+//! |---|---|---|---|---|
+//! | default (adaptive) | 10 MB | 84 MB | 88 MB | **88 MB** |
+//! | pinned threshold   | 10 MB | 13 MB | 14 MB | **14 MB** |
+//!
+//! 78 MB retained versus 4 MB for identical work. On that cluster's heavier
+//! mixed scrub workloads the same effect reached ~600 MB of retained anonymous
+//! memory per volume server, against a 3 GiB cap.
+//!
+//! Calling `mallopt(M_MMAP_THRESHOLD, ...)` sets the threshold *and* disables
+//! the dynamic adjustment, which is the documented behaviour of setting it
+//! explicitly. We pin it to glibc's own default rather than inventing a value:
+//! the goal is to stop the adaptation, not to second-guess the default.
+
+/// glibc's own default `M_MMAP_THRESHOLD`. Pinning to this value changes
+/// nothing about which allocations use `mmap` on a freshly started process; it
+/// only prevents the threshold from drifting upward later.
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+const DEFAULT_MMAP_THRESHOLD: libc::c_int = 128 * 1024;
+
+/// Legacy environment variable glibc reads for the same setting. If an operator
+/// has set it, honour their value and do not override it.
+pub const MMAP_THRESHOLD_ENV: &str = "MALLOC_MMAP_THRESHOLD_";
+
+/// Modern glibc tunables environment variable. Operators may set the threshold
+/// via `GLIBC_TUNABLES=glibc.malloc.mmap_threshold=...` instead of the legacy
+/// variable; that override is honoured too.
+pub const GLIBC_TUNABLES_ENV: &str = "GLIBC_TUNABLES";
+
+/// The tunable name within `GLIBC_TUNABLES` that maps to `M_MMAP_THRESHOLD`.
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+const MMAP_THRESHOLD_TUNABLE: &str = "glibc.malloc.mmap_threshold";
+
+/// Outcome of the tuning attempt, so the caller can log it and tests can assert
+/// on it without inspecting global allocator state.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum MallocTuning {
+    /// Threshold pinned to `DEFAULT_MMAP_THRESHOLD`; dynamic adjustment is off.
+    Pinned(i32),
+    /// An allocator override (`MALLOC_MMAP_THRESHOLD_` or
+    /// `GLIBC_TUNABLES=glibc.malloc.mmap_threshold=...`) was set, so the
+    /// operator's value wins.
+    DeferredToEnv,
+    /// `mallopt` reported failure. Not fatal — the server runs, it just keeps
+    /// glibc's adaptive behaviour.
+    Failed,
+    /// Not glibc, so there is no adaptive threshold to pin.
+    NotApplicable,
+}
+
+/// Pin glibc's mmap threshold unless the operator has set an allocator override.
+/// Safe to call more than once; call it before serving traffic, since the point
+/// is to prevent the threshold from being trained upward by early allocations.
+pub fn pin_mmap_threshold() -> MallocTuning {
+    pin_mmap_threshold_inner()
+}
+
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+fn pin_mmap_threshold_inner() -> MallocTuning {
+    if operator_mmap_threshold_override_active() {
+        return MallocTuning::DeferredToEnv;
+    }
+    // SAFETY: `mallopt` is a libc entry point that takes two ints and mutates
+    // only allocator-internal tunables. It has no preconditions and no effect
+    // on memory this process already owns.
+    let rc = unsafe { libc::mallopt(libc::M_MMAP_THRESHOLD, DEFAULT_MMAP_THRESHOLD) };
+    if rc == 1 {
+        MallocTuning::Pinned(DEFAULT_MMAP_THRESHOLD)
+    } else {
+        MallocTuning::Failed
+    }
+}
+
+#[cfg(not(all(target_os = "linux", target_env = "gnu")))]
+fn pin_mmap_threshold_inner() -> MallocTuning {
+    MallocTuning::NotApplicable
+}
+
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+fn operator_mmap_threshold_override_active() -> bool {
+    // MALLOC_MMAP_THRESHOLD_: glibc calls atoi(value) then mallopt, which
+    // always sets the threshold and disables dynamic adjustment — even for
+    // empty, negative, or non-numeric values (atoi returns 0). So any presence
+    // of the variable means the operator's override is in effect.
+    std::env::var_os(MMAP_THRESHOLD_ENV).is_some()
+        || usable_glibc_tunable_threshold(std::env::var_os(GLIBC_TUNABLES_ENV))
+}
+
+/// Look for `glibc.malloc.mmap_threshold=<value>` among the colon-separated
+/// tunables in `GLIBC_TUNABLES`. glibc's `parse_tunables_string` (elf/dl-tunables.c)
+/// rejects the **entire** string (returns -1) if it reaches `\0` before finding
+/// `=` in a name (last entry has no `=`), or if any entry's value contains a
+/// duplicate `=`. When `parse_tunables_string` returns -1, `parse_tunables`
+/// prints a warning and returns immediately without applying ANY tunable —
+/// including ones already parsed into the tunables array. We match that by
+/// returning `false` for the entire string on any of those conditions.
+///
+/// glibc parses tunable values with `_dl_strtoul`, which accepts decimal,
+/// `0x` hex, `0` octal, an optional sign (negatives wrap to `unsigned long`),
+/// and requires the entire value to be consumed; we match that with
+/// `dl_strtoul_consumes_all`.
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+fn usable_glibc_tunable_threshold(tunables: Option<std::ffi::OsString>) -> bool {
+    let s = match tunables.and_then(|v| v.into_string().ok()) {
+        Some(s) => s,
+        None => return false,
+    };
+    if s.is_empty() {
+        return false;
+    }
+
+    // Parse the string character-by-character, matching glibc's
+    // parse_tunables_string logic exactly. Using split(':') would lose the
+    // distinction between an entry terminated by ':' (skip) and one terminated
+    // by '\0' with no '=' (reject entire string).
+    let bytes = s.as_bytes();
+    let mut pos = 0;
+    let mut found_threshold = false;
+
+    loop {
+        // Find where the name ends ('=', ':', or end of string).
+        let name_start = pos;
+        while pos < bytes.len() && bytes[pos] != b'=' && bytes[pos] != b':' {
+            pos += 1;
+        }
+
+        // End of string before '=' → glibc returns -1 (reject entire string).
+        if pos >= bytes.len() {
+            return false;
+        }
+
+        // ':' before '=' → glibc skips this entry and continues.
+        if bytes[pos] == b':' {
+            pos += 1;
+            continue;
+        }
+
+        // Skip the '='.
+        let name_end = pos;
+        pos += 1;
+
+        // Find where the value ends ('=', ':', or end of string).
+        let val_start = pos;
+        while pos < bytes.len() && bytes[pos] != b'=' && bytes[pos] != b':' {
+            pos += 1;
+        }
+
+        // '=' in value → glibc returns -1 (reject entire string).
+        if pos < bytes.len() && bytes[pos] == b'=' {
+            return false;
+        }
+
+        let key = &s[name_start..name_end];
+        let val = &s[val_start..pos];
+
+        if key == MMAP_THRESHOLD_TUNABLE && dl_strtoul_consumes_all(val) {
+            found_threshold = true;
+        }
+
+        // End of string → done.
+        if pos >= bytes.len() {
+            break;
+        }
+
+        // Skip the ':'.
+        pos += 1;
+    }
+
+    found_threshold
+}
+
+/// Replicate glibc's `_dl_strtoul` (elf/dl-misc.c) just enough to determine
+/// whether it would consume the entire string — which is what
+/// `tunable_parse_num` checks (`endptr == strval + len`). Returns `true` if
+/// glibc would accept the value and apply it.
+///
+/// `_dl_strtoul` skips leading spaces/tabs, accepts an optional `+`/`-` sign,
+/// and parses `0x`-prefixed hex, `0`-prefixed octal, or plain decimal. A
+/// negative result wraps to `unsigned long` (`-1` → `SIZE_MAX`). If no digit is
+/// found after the sign, the end pointer stays at the current position — which
+/// still counts as "consumed" when the string is empty or whitespace-only
+/// (value 0). On overflow, `_dl_strtoul` stops at the overflowing digit (endptr
+/// does not reach the end), so `tunable_parse_num` rejects the value.
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+fn dl_strtoul_consumes_all(s: &str) -> bool {
+    let bytes = s.as_bytes();
+    let mut pos = 0;
+
+    // Skip leading whitespace (spaces and tabs, matching _dl_strtoul).
+    while pos < bytes.len() && (bytes[pos] == b' ' || bytes[pos] == b'\t') {
+        pos += 1;
+    }
+
+    // Optional sign.
+    if pos < bytes.len() && (bytes[pos] == b'-' || bytes[pos] == b'+') {
+        pos += 1;
+    }
+
+    // Must have at least one digit (0-9) to start parsing, unless we're already
+    // at the end (empty / whitespace-only / sign-only → value 0, consumed).
+    if pos >= bytes.len() {
+        return true;
+    }
+    if bytes[pos] < b'0' || bytes[pos] > b'9' {
+        return false;
+    }
+
+    // Determine base: 0x → hex, 0 → octal, else decimal. _dl_strtoul unconditionally
+    // advances past "0x"/"0X" when the first char is '0' and the next is 'x'/'X',
+    // even if no hex digit follows — in that case the digit loop breaks immediately,
+    // endptr reaches the end, and the value is 0.
+    let base: u32 = if bytes[pos] == b'0'
+        && pos + 1 < bytes.len()
+        && (bytes[pos + 1] == b'x' || bytes[pos + 1] == b'X')
+    {
+        pos += 2; // skip "0x"
+        16
+    } else if bytes[pos] == b'0' {
+        8
+    } else {
+        10
+    };
+
+    // Parse digits with overflow detection, matching _dl_strtoul's cutoff/cutlim
+    // logic. On overflow, _dl_strtoul sets endptr to the overflowing digit and
+    // returns UINT64_MAX — so the value is NOT fully consumed and
+    // tunable_parse_num rejects it.
+    let mut result: u64 = 0;
+    let cutoff = u64::MAX / base as u64;
+    let cutlim = u64::MAX % base as u64;
+
+    while pos < bytes.len() {
+        let b = bytes[pos];
+        let digval: u32 = match digit_value(b, base) {
+            Some(v) => v,
+            None => break,
+        };
+        if result > cutoff || (result == cutoff && digval as u64 > cutlim) {
+            // Overflow: _dl_strtoul stops here, endptr points at this digit.
+            return false;
+        }
+        result *= base as u64;
+        result += digval as u64;
+        pos += 1;
+    }
+
+    // The entire string must be consumed (matching tunable_parse_num's check).
+    pos == bytes.len()
+}
+
+/// Returns the numeric value of a digit byte in the given base, or `None` if
+/// the byte is not a valid digit in that base.
+#[cfg(all(target_os = "linux", target_env = "gnu"))]
+fn digit_value(b: u8, base: u32) -> Option<u32> {
+    if (b'0'..=b'0' + (base - 1).min(9) as u8).contains(&b) {
+        return Some((b - b'0') as u32);
+    }
+    if base == 16 {
+        if (b'a'..=b'f').contains(&b) {
+            return Some((b - b'a' + 10) as u32);
+        }
+        if (b'A'..=b'F').contains(&b) {
+            return Some((b - b'A' + 10) as u32);
+        }
+    }
+    None
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn env_override_constants_match_glibc_names() {
+        // Verified against the real accessor rather than a copy of the name, so
+        // renaming the constant cannot silently break the override contract.
+        assert_eq!(MMAP_THRESHOLD_ENV, "MALLOC_MMAP_THRESHOLD_");
+        assert_eq!(GLIBC_TUNABLES_ENV, "GLIBC_TUNABLES");
+    }
+
+    #[test]
+    fn calling_twice_is_stable() {
+        // Startup paths get re-entered in tests and in `weed mini`; the second
+        // call must not report a different outcome from the first.
+        let first = pin_mmap_threshold();
+        let second = pin_mmap_threshold();
+        assert_eq!(first, second);
+    }
+
+    #[cfg(all(target_os = "linux", target_env = "gnu"))]
+    #[test]
+    fn pins_threshold_on_glibc_when_no_override_is_set() {
+        // The env override is not set in the test process, so this exercises the
+        // mallopt path. If an override happens to be present, defer to it.
+        if operator_mmap_threshold_override_active() {
+            assert_eq!(pin_mmap_threshold(), MallocTuning::DeferredToEnv);
+            return;
+        }
+        assert_eq!(
+            pin_mmap_threshold(),
+            MallocTuning::Pinned(DEFAULT_MMAP_THRESHOLD),
+            "mallopt(M_MMAP_THRESHOLD) should succeed on glibc"
+        );
+    }
+
+    #[cfg(not(all(target_os = "linux", target_env = "gnu")))]
+    #[test]
+    fn is_a_noop_off_glibc() {
+        // No glibc adaptive threshold exists off glibc, so there is nothing to
+        // pin regardless of any environment variables that happen to be set.
+        assert_eq!(pin_mmap_threshold(), MallocTuning::NotApplicable);
+    }
+
+    #[cfg(all(target_os = "linux", target_env = "gnu"))]
+    #[test]
+    fn dl_strtoul_consumes_all_matches_glibc_parser() {
+        // Decimal — any non-empty decimal integer is accepted, including
+        // negative (wraps to unsigned) and zero.
+        assert!(dl_strtoul_consumes_all("131072"));
+        assert!(dl_strtoul_consumes_all("0"));
+        assert!(dl_strtoul_consumes_all("-1"));
+        assert!(dl_strtoul_consumes_all("-131072"));
+        // Values above i64::MAX are valid for glibc's unsigned parser.
+        assert!(dl_strtoul_consumes_all("9223372036854775808"));
+        // Hex with 0x prefix.
+        assert!(dl_strtoul_consumes_all("0x20000"));
+        assert!(dl_strtoul_consumes_all("0X20000"));
+        assert!(dl_strtoul_consumes_all("0x0"));
+        // Octal with leading 0.
+        assert!(dl_strtoul_consumes_all("010"));
+        // Leading whitespace (spaces and tabs) is skipped.
+        assert!(dl_strtoul_consumes_all("  131072"));
+        assert!(dl_strtoul_consumes_all("\t0x20000"));
+        // Empty and whitespace-only strings are accepted (value 0).
+        assert!(dl_strtoul_consumes_all(""));
+        assert!(dl_strtoul_consumes_all("  "));
+        assert!(dl_strtoul_consumes_all("\t"));
+        // Sign-only strings are accepted: _dl_strtoul skips the sign, finds no
+        // digit, sets endptr to the position after the sign (== end of string),
+        // and returns 0. tunable_parse_num sees endptr == strval + len → true.
+        assert!(dl_strtoul_consumes_all("-"));
+        assert!(dl_strtoul_consumes_all("+"));
+
+        // Trailing garbage is rejected — _dl_strtoul stops at the first
+        // non-digit and tunable_parse_num requires the entire string consumed.
+        assert!(!dl_strtoul_consumes_all("131072abc"));
+        // In hex mode, a-f are digits, so "0x20000abc" is a valid hex number.
+        // Use a non-hex character like 'g' to test trailing garbage in hex.
+        assert!(!dl_strtoul_consumes_all("0x20000g"));
+        assert!(!dl_strtoul_consumes_all("128K"));
+        // Non-numeric strings are rejected.
+        assert!(!dl_strtoul_consumes_all("abc"));
+        // "0x" with no hex digits: _dl_strtoul advances past "0x", the digit loop
+        // breaks immediately (no hex digit), endptr reaches the end, value is 0.
+        // tunable_parse_num accepts it.
+        assert!(dl_strtoul_consumes_all("0x"));
+        assert!(dl_strtoul_consumes_all("0X"));
+
+        // Overflow: _dl_strtoul stops at the overflowing digit (endptr points
+        // there, not at the end), so tunable_parse_num rejects the value.
+        assert!(!dl_strtoul_consumes_all("18446744073709551616")); // u64::MAX + 1
+        assert!(!dl_strtoul_consumes_all("99999999999999999999")); // 20 nines
+        assert!(!dl_strtoul_consumes_all("0x10000000000000000")); // 2^64
+        // u64::MAX itself is accepted: the last digit (5) equals cutlim (=5),
+        // so the overflow check (digval > cutlim) is false.
+        assert!(dl_strtoul_consumes_all("18446744073709551615")); // u64::MAX
+    }
+
+    #[cfg(all(target_os = "linux", target_env = "gnu"))]
+    #[test]
+    fn usable_glibc_tunable_threshold_detects_mmap_threshold() {
+        // Decimal, hex, octal, negative, and zero values are all accepted by
+        // glibc's _dl_strtoul and cause the threshold to be pinned.
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=131072".into()
+        )));
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=0x20000".into()
+        )));
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=0".into()
+        )));
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=-1".into()
+        )));
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=9223372036854775808".into()
+        )));
+        // u64::MAX is accepted by _dl_strtoul (last digit == cutlim, no overflow).
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=18446744073709551615".into()
+        )));
+        // Appears alongside other tunables.
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.cpu.x=1:glibc.malloc.mmap_threshold=131072".into()
+        )));
+        // Leading ':' is accepted — glibc skips the empty entry and continues.
+        assert!(usable_glibc_tunable_threshold(Some(
+            ":glibc.malloc.mmap_threshold=131072".into()
+        )));
+        // Empty value is accepted by _dl_strtoul (value 0).
+        assert!(usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=".into()
+        )));
+
+        // Non-numeric values are rejected by _dl_strtoul.
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=abc".into()
+        )));
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=128K".into()
+        )));
+        // A malformed sibling entry (duplicate '=') makes glibc reject the
+        // entire string, so we must not accept the threshold entry either.
+        // This applies regardless of whether the threshold is before or after
+        // the malformed entry — parse_tunables_string returns -1, and
+        // parse_tunables discards all tunables without applying any.
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.check=2=2:glibc.malloc.mmap_threshold=131072".into()
+        )));
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=262144:glibc.malloc.check=2=2".into()
+        )));
+        // A trailing entry with no '=' makes glibc reject the entire string
+        // (parse_tunables_string hits '\0' before '=' and returns -1).
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=262144:glibc.cpu.x".into()
+        )));
+        // A trailing ':' makes glibc reject the entire string (the empty entry
+        // after ':' hits '\0' before '=' and returns -1).
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.malloc.mmap_threshold=262144:".into()
+        )));
+        // Unrelated tunables do not count.
+        assert!(!usable_glibc_tunable_threshold(Some(
+            "glibc.cpu.x=1".into()
+        )));
+        assert!(!usable_glibc_tunable_threshold(None));
+    }
+}

seaweed-volume/src/metrics.rs

@@ -0,0 +1,621 @@
+//! Prometheus metrics for the volume server.
+//!
+//! Mirrors the Go SeaweedFS volume server metrics.
+
+use prometheus::{
+    self, Encoder, GaugeVec, HistogramOpts, HistogramVec, IntCounter, IntCounterVec, IntGauge,
+    IntGaugeVec, Opts, Registry, TextEncoder,
+};
+use std::sync::{LazyLock, Once};
+
+use crate::version;
+
+#[derive(Clone, Debug, Default, PartialEq, Eq)]
+pub struct PushGatewayConfig {
+    pub address: String,
+    pub interval_seconds: u32,
+}
+
+pub static REGISTRY: LazyLock<Registry> = LazyLock::new(Registry::new);
+
+// ---- Request metrics (Go: VolumeServerRequestCounter, VolumeServerRequestHistogram) ----
+
+/// Request counter with labels `type` (HTTP method) and `code` (HTTP status).
+pub static REQUEST_COUNTER: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_request_total",
+            "Volume server requests",
+        ),
+        &["type", "code"],
+    )
+    .expect("metric can be created")
+});
+
+/// Request duration histogram with label `type` (HTTP method).
+pub static REQUEST_DURATION: LazyLock<HistogramVec> = LazyLock::new(|| {
+    HistogramVec::new(
+        HistogramOpts::new(
+            "SeaweedFS_volumeServer_request_seconds",
+            "Volume server request duration in seconds",
+        )
+        .buckets(exponential_buckets(0.0001, 2.0, 24)),
+        &["type"],
+    )
+    .expect("metric can be created")
+});
+
+// ---- Handler counters (Go: VolumeServerHandlerCounter) ----
+
+/// Handler-level operation counter with label `type`.
+pub static HANDLER_COUNTER: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_handler_total",
+            "Volume server handler counters",
+        ),
+        &["type"],
+    )
+    .expect("metric can be created")
+});
+
+// ---- Vacuuming metrics (Go: VolumeServerVacuuming*) ----
+
+/// Vacuuming compact counter with label `success` (true/false).
+pub static VACUUMING_COMPACT_COUNTER: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_vacuuming_compact_count",
+            "Counter of volume vacuuming Compact counter",
+        ),
+        &["success"],
+    )
+    .expect("metric can be created")
+});
+
+/// Vacuuming commit counter with label `success` (true/false).
+pub static VACUUMING_COMMIT_COUNTER: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_vacuuming_commit_count",
+            "Counter of volume vacuuming commit counter",
+        ),
+        &["success"],
+    )
+    .expect("metric can be created")
+});
+
+/// Vacuuming duration histogram with label `type` (compact/commit).
+pub static VACUUMING_HISTOGRAM: LazyLock<HistogramVec> = LazyLock::new(|| {
+    HistogramVec::new(
+        HistogramOpts::new(
+            "SeaweedFS_volumeServer_vacuuming_seconds",
+            "Volume vacuuming duration in seconds",
+        )
+        .buckets(exponential_buckets(0.0001, 2.0, 24)),
+        &["type"],
+    )
+    .expect("metric can be created")
+});
+
+// ---- Volume gauges (Go: VolumeServerVolumeGauge, VolumeServerReadOnlyVolumeGauge) ----
+
+/// Volumes per collection and type (volume/ec_shards).
+pub static VOLUME_GAUGE: LazyLock<GaugeVec> = LazyLock::new(|| {
+    GaugeVec::new(
+        Opts::new("SeaweedFS_volumeServer_volumes", "Number of volumes"),
+        &["collection", "type"],
+    )
+    .expect("metric can be created")
+});
+
+/// Read-only volumes per collection and type.
+pub static READ_ONLY_VOLUME_GAUGE: LazyLock<GaugeVec> = LazyLock::new(|| {
+    GaugeVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_read_only_volumes",
+            "Number of read-only volumes.",
+        ),
+        &["collection", "type"],
+    )
+    .expect("metric can be created")
+});
+
+/// Maximum number of volumes this server can hold.
+pub static MAX_VOLUMES: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "SeaweedFS_volumeServer_max_volumes",
+        "Maximum number of volumes",
+    )
+    .expect("metric can be created")
+});
+
+// ---- Disk size gauges (Go: VolumeServerDiskSizeGauge) ----
+
+/// Actual disk size used by volumes per collection and type (normal/deleted_bytes/ec).
+pub static DISK_SIZE_GAUGE: LazyLock<GaugeVec> = LazyLock::new(|| {
+    GaugeVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_total_disk_size",
+            "Actual disk size used by volumes",
+        ),
+        &["collection", "type"],
+    )
+    .expect("metric can be created")
+});
+
+// ---- Resource gauges (Go: VolumeServerResourceGauge) ----
+
+/// Disk resource usage per directory and type (all/used/free/avail).
+pub static RESOURCE_GAUGE: LazyLock<GaugeVec> = LazyLock::new(|| {
+    GaugeVec::new(
+        Opts::new("SeaweedFS_volumeServer_resource", "Server resource usage"),
+        &["name", "type"],
+    )
+    .expect("metric can be created")
+});
+
+// ---- In-flight gauges (Go: VolumeServerInFlightRequestsGauge, InFlightDownload/UploadSize) ----
+
+/// In-flight requests per HTTP method.
+pub static INFLIGHT_REQUESTS_GAUGE: LazyLock<IntGaugeVec> = LazyLock::new(|| {
+    IntGaugeVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_in_flight_requests",
+            "Current number of in-flight requests being handled by volume server.",
+        ),
+        &["type"],
+    )
+    .expect("metric can be created")
+});
+
+/// Concurrent download limit in bytes.
+pub static CONCURRENT_DOWNLOAD_LIMIT: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "SeaweedFS_volumeServer_concurrent_download_limit",
+        "Limit for total concurrent download size in bytes",
+    )
+    .expect("metric can be created")
+});
+
+/// Concurrent upload limit in bytes.
+pub static CONCURRENT_UPLOAD_LIMIT: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "SeaweedFS_volumeServer_concurrent_upload_limit",
+        "Limit for total concurrent upload size in bytes",
+    )
+    .expect("metric can be created")
+});
+
+/// Current in-flight download bytes.
+pub static INFLIGHT_DOWNLOAD_SIZE: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "SeaweedFS_volumeServer_in_flight_download_size",
+        "In flight total download size.",
+    )
+    .expect("metric can be created")
+});
+
+/// Current in-flight upload bytes.
+pub static INFLIGHT_UPLOAD_SIZE: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "SeaweedFS_volumeServer_in_flight_upload_size",
+        "In flight total upload size.",
+    )
+    .expect("metric can be created")
+});
+
+/// Upload error counter by HTTP status code. Code "0" = transport error (no response).
+pub static UPLOAD_ERROR_COUNTER: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+    Opts::new("SeaweedFS_upload_error_total",
+        "Counter of upload errors by HTTP status code. Code 0 means transport error (no response received)."),
+    &["code"],
+).expect("metric can be created")
+});
+
+// ---- Scrubbing metrics (Go: VolumeServerScrub*) ----
+
+/// Last scrub execution time, as seconds since UNIX epoch, with label `mode`.
+pub static SCRUB_LAST_TIME_SECONDS: LazyLock<GaugeVec> = LazyLock::new(|| {
+    GaugeVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_scrub_last_time_seconds",
+            "Last scrub execution time, as seconds since UNIX epoch.",
+        ),
+        &["mode"],
+    )
+    .expect("metric can be created")
+});
+
+/// Counter of overall volumes with issues detected during scrubbing, with label `mode`.
+pub static SCRUB_VOLUME_FAILURES: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_scrub_volume_failures",
+            "Counter of overall volumes with issues detected during scrubbing.",
+        ),
+        &["mode"],
+    )
+    .expect("metric can be created")
+});
+
+/// Counter of overall EC shards with issues detected during scrubbing, with label `mode`.
+pub static SCRUB_SHARD_FAILURES: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    IntCounterVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_scrub_shard_failures",
+            "Counter of overall EC shards with issues detected during scrubbing.",
+        ),
+        &["mode"],
+    )
+    .expect("metric can be created")
+});
+
+/// Counter of storage read/write EIO errors on volumes and EC shards.
+/// Mirrors Go's VolumeServerStorageIoErrorCounter.
+pub static STORAGE_IO_ERROR_COUNTER: LazyLock<IntCounter> = LazyLock::new(|| {
+    IntCounter::new(
+        "SeaweedFS_volumeServer_storage_io_error_total",
+        "Counter of storage read/write EIO errors on volumes and EC shards.",
+    )
+    .expect("metric can be created")
+});
+
+/// Number of volumes quarantined due to storage IO errors.
+/// Mirrors Go's VolumeServerIoQuarantineGauge.
+pub static IO_QUARANTINE_GAUGE: LazyLock<IntGaugeVec> = LazyLock::new(|| {
+    IntGaugeVec::new(
+        Opts::new(
+            "SeaweedFS_volumeServer_io_quarantine",
+            "Number of volumes or EC shards quarantined due to storage IO errors.",
+        ),
+        &["kind"],
+    )
+    .expect("metric can be created")
+});
+
+// ---- Legacy aliases for backward compat with existing code ----
+
+/// Total number of volumes on this server (flat gauge).
+pub static VOLUMES_TOTAL: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new("volume_server_volumes_total", "Total number of volumes")
+        .expect("metric can be created")
+});
+
+/// Disk size in bytes per directory.
+pub static DISK_SIZE_BYTES: LazyLock<IntGaugeVec> = LazyLock::new(|| {
+    IntGaugeVec::new(
+        Opts::new("volume_server_disk_size_bytes", "Disk size in bytes"),
+        &["dir"],
+    )
+    .expect("metric can be created")
+});
+
+/// Disk free bytes per directory.
+pub static DISK_FREE_BYTES: LazyLock<IntGaugeVec> = LazyLock::new(|| {
+    IntGaugeVec::new(
+        Opts::new("volume_server_disk_free_bytes", "Disk free space in bytes"),
+        &["dir"],
+    )
+    .expect("metric can be created")
+});
+
+/// Current number of in-flight requests (flat gauge).
+pub static INFLIGHT_REQUESTS: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "volume_server_inflight_requests",
+        "Current number of in-flight requests",
+    )
+    .expect("metric can be created")
+});
+
+/// Total number of files stored across all volumes.
+pub static VOLUME_FILE_COUNT: LazyLock<IntGauge> = LazyLock::new(|| {
+    IntGauge::new(
+        "volume_server_volume_file_count",
+        "Total number of files stored across all volumes",
+    )
+    .expect("metric can be created")
+});
+
+// ---- Build info (Go: BuildInfo) ----
+
+/// Build information gauge, always set to 1. Matches Go:
+/// Namespace="SeaweedFS", Subsystem="build", Name="info",
+/// labels: version, commit, sizelimit, goos, goarch.
+pub static BUILD_INFO: LazyLock<GaugeVec> = LazyLock::new(|| {
+    GaugeVec::new(
+    Opts::new("SeaweedFS_build_info", "A metric with a constant '1' value labeled by version, commit, sizelimit, goos, and goarch from which SeaweedFS was built."),
+    &["version", "commit", "sizelimit", "goos", "goarch"],
+).expect("metric can be created")
+});
+
+/// Generate exponential bucket boundaries for histograms.
+fn exponential_buckets(start: f64, factor: f64, count: usize) -> Vec<f64> {
+    let mut buckets = Vec::with_capacity(count);
+    let mut val = start;
+    for _ in 0..count {
+        buckets.push(val);
+        val *= factor;
+    }
+    buckets
+}
+
+// Handler counter type constants (matches Go's metrics_names.go).
+pub const WRITE_TO_LOCAL_DISK: &str = "writeToLocalDisk";
+pub const WRITE_TO_REPLICAS: &str = "writeToReplicas";
+pub const DOWNLOAD_LIMIT_COND: &str = "downloadLimitCondition";
+pub const UPLOAD_LIMIT_COND: &str = "uploadLimitCondition";
+pub const READ_PROXY_REQ: &str = "readProxyRequest";
+pub const READ_REDIRECT_REQ: &str = "readRedirectRequest";
+pub const EMPTY_READ_PROXY_LOC: &str = "emptyReadProxyLocaction";
+pub const FAILED_READ_PROXY_REQ: &str = "failedReadProxyRequest";
+
+// Error metric name constants.
+pub const ERROR_SIZE_MISMATCH_OFFSET_SIZE: &str = "errorSizeMismatchOffsetSize";
+pub const ERROR_SIZE_MISMATCH: &str = "errorSizeMismatch";
+pub const ERROR_CRC: &str = "errorCRC";
+pub const ERROR_INDEX_OUT_OF_RANGE: &str = "errorIndexOutOfRange";
+pub const ERROR_GET_NOT_FOUND: &str = "errorGetNotFound";
+pub const ERROR_GET_INTERNAL: &str = "errorGetInternal";
+pub const ERROR_WRITE_TO_LOCAL_DISK: &str = "errorWriteToLocalDisk";
+pub const ERROR_UNMARSHAL_PAIRS: &str = "errorUnmarshalPairs";
+pub const ERROR_WRITE_TO_REPLICAS: &str = "errorWriteToReplicas";
+
+// Go volume heartbeat metric label values.
+pub const READ_ONLY_LABEL_IS_READ_ONLY: &str = "IsReadOnly";
+pub const READ_ONLY_LABEL_NO_WRITE_OR_DELETE: &str = "noWriteOrDelete";
+pub const READ_ONLY_LABEL_NO_WRITE_CAN_DELETE: &str = "noWriteCanDelete";
+pub const READ_ONLY_LABEL_IS_DISK_SPACE_LOW: &str = "isDiskSpaceLow";
+pub const DISK_SIZE_LABEL_NORMAL: &str = "normal";
+pub const DISK_SIZE_LABEL_DELETED_BYTES: &str = "deleted_bytes";
+pub const DISK_SIZE_LABEL_EC: &str = "ec";
+
+static REGISTER_METRICS: Once = Once::new();
+
+/// Register all metrics with the custom registry.
+/// Call this once at startup.
+pub fn register_metrics() {
+    REGISTER_METRICS.call_once(|| {
+        let metrics: Vec<Box<dyn prometheus::core::Collector>> = vec![
+            // New Go-compatible metrics
+            Box::new(REQUEST_COUNTER.clone()),
+            Box::new(REQUEST_DURATION.clone()),
+            Box::new(HANDLER_COUNTER.clone()),
+            Box::new(VACUUMING_COMPACT_COUNTER.clone()),
+            Box::new(VACUUMING_COMMIT_COUNTER.clone()),
+            Box::new(VACUUMING_HISTOGRAM.clone()),
+            Box::new(VOLUME_GAUGE.clone()),
+            Box::new(READ_ONLY_VOLUME_GAUGE.clone()),
+            Box::new(MAX_VOLUMES.clone()),
+            Box::new(DISK_SIZE_GAUGE.clone()),
+            Box::new(RESOURCE_GAUGE.clone()),
+            Box::new(INFLIGHT_REQUESTS_GAUGE.clone()),
+            Box::new(CONCURRENT_DOWNLOAD_LIMIT.clone()),
+            Box::new(CONCURRENT_UPLOAD_LIMIT.clone()),
+            Box::new(INFLIGHT_DOWNLOAD_SIZE.clone()),
+            Box::new(INFLIGHT_UPLOAD_SIZE.clone()),
+            Box::new(UPLOAD_ERROR_COUNTER.clone()),
+            Box::new(SCRUB_LAST_TIME_SECONDS.clone()),
+            Box::new(SCRUB_VOLUME_FAILURES.clone()),
+            Box::new(SCRUB_SHARD_FAILURES.clone()),
+            Box::new(STORAGE_IO_ERROR_COUNTER.clone()),
+            Box::new(IO_QUARANTINE_GAUGE.clone()),
+            // Legacy metrics
+            Box::new(VOLUMES_TOTAL.clone()),
+            Box::new(DISK_SIZE_BYTES.clone()),
+            Box::new(DISK_FREE_BYTES.clone()),
+            Box::new(INFLIGHT_REQUESTS.clone()),
+            Box::new(VOLUME_FILE_COUNT.clone()),
+            // Build info
+            Box::new(BUILD_INFO.clone()),
+        ];
+        for m in metrics {
+            REGISTRY.register(m).expect("metric registered");
+        }
+
+        // Set build info gauge to 1 with version/commit/sizelimit/os/arch labels (matches Go).
+        BUILD_INFO
+            .with_label_values(&[
+                version::version(),
+                version::commit(),
+                version::size_limit(),
+                std::env::consts::OS,
+                std::env::consts::ARCH,
+            ])
+            .set(1.0);
+    });
+}
+
+/// Gather all metrics and encode them in Prometheus text exposition format.
+pub fn gather_metrics() -> String {
+    let encoder = TextEncoder::new();
+    let metric_families = REGISTRY.gather();
+    let mut buffer = Vec::new();
+    encoder
+        .encode(&metric_families, &mut buffer)
+        .expect("encoding metrics");
+    String::from_utf8(buffer).expect("metrics are valid UTF-8")
+}
+
+pub fn delete_collection_metrics(collection: &str) {
+    // Mirrors Go's DeletePartialMatch(prometheus.Labels{"collection": collection})
+    // which removes ALL metric entries matching the collection label, regardless
+    // of other label values (like "type"). We gather the metric families to discover
+    // all type values dynamically, matching Go's partial-match behavior.
+    delete_partial_match_collection(&VOLUME_GAUGE, collection);
+    delete_partial_match_collection(&READ_ONLY_VOLUME_GAUGE, collection);
+    delete_partial_match_collection(&DISK_SIZE_GAUGE, collection);
+}
+
+/// Drop a collection's volume server series once its last volume leaves this
+/// server. These gauges are only ever set for collections still present, so the
+/// values from the heartbeat that saw the last volume would otherwise stand
+/// until the process restarts.
+pub fn delete_volume_server_collection_metrics(collection: &str) {
+    let _ = DISK_SIZE_GAUGE.remove_label_values(&[collection, DISK_SIZE_LABEL_NORMAL]);
+    let _ = DISK_SIZE_GAUGE.remove_label_values(&[collection, DISK_SIZE_LABEL_DELETED_BYTES]);
+    delete_partial_match_collection(&READ_ONLY_VOLUME_GAUGE, collection);
+}
+
+/// Remove all metric entries from a GaugeVec where the "collection" label matches.
+/// This emulates Go's `DeletePartialMatch(prometheus.Labels{"collection": collection})`.
+fn delete_partial_match_collection(gauge: &GaugeVec, collection: &str) {
+    use prometheus::core::Collector;
+    let families = gauge.collect();
+    for family in &families {
+        for metric in family.get_metric() {
+            let labels = metric.get_label();
+            let mut matches_collection = false;
+            let mut type_value = None;
+            for label in labels {
+                if label.get_name() == "collection" && label.get_value() == collection {
+                    matches_collection = true;
+                }
+                if label.get_name() == "type" {
+                    type_value = Some(label.get_value().to_string());
+                }
+            }
+            if matches_collection && let Some(ref tv) = type_value {
+                let _ = gauge.remove_label_values(&[collection, tv]);
+            }
+        }
+    }
+}
+
+pub fn build_pushgateway_url(address: &str, job: &str, instance: &str) -> String {
+    let base = if address.starts_with("http://") || address.starts_with("https://") {
+        address.to_string()
+    } else {
+        format!("http://{}", address)
+    };
+    let base = base.trim_end_matches('/');
+    format!("{}/metrics/job/{}/instance/{}", base, job, instance)
+}
+
+pub async fn push_metrics_once(
+    client: &reqwest::Client,
+    address: &str,
+    job: &str,
+    instance: &str,
+) -> Result<(), String> {
+    let url = build_pushgateway_url(address, job, instance);
+    let response = client
+        .put(&url)
+        .header(
+            reqwest::header::CONTENT_TYPE,
+            "text/plain; version=0.0.4; charset=utf-8",
+        )
+        .body(gather_metrics())
+        .send()
+        .await
+        .map_err(|e| format!("push metrics request failed: {}", e))?;
+
+    if response.status().is_success() {
+        Ok(())
+    } else {
+        Err(format!(
+            "push metrics failed with status {}",
+            response.status()
+        ))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use axum::{Router, routing::put};
+    use std::sync::{Arc, Mutex};
+
+    #[test]
+    fn test_gather_metrics_returns_text() {
+        register_metrics();
+        REQUEST_COUNTER.with_label_values(&["GET", "200"]).inc();
+        let output = gather_metrics();
+        assert!(output.contains("SeaweedFS_volumeServer_request_total"));
+    }
+
+    #[test]
+    fn test_build_pushgateway_url() {
+        assert_eq!(
+            build_pushgateway_url("localhost:9091", "volumeServer", "test-instance"),
+            "http://localhost:9091/metrics/job/volumeServer/instance/test-instance"
+        );
+        assert_eq!(
+            build_pushgateway_url("https://push.example", "volumeServer", "node-a"),
+            "https://push.example/metrics/job/volumeServer/instance/node-a"
+        );
+    }
+
+    #[tokio::test]
+    async fn test_push_metrics_once() {
+        register_metrics();
+        // A CounterVec with no children emits nothing, so create a labelset
+        // instead of depending on another test having touched the counter
+        // first (test order is nondeterministic).
+        REQUEST_COUNTER.with_label_values(&["GET", "200"]).inc();
+
+        let captured = Arc::new(Mutex::new(None::<String>));
+        let captured_clone = captured.clone();
+
+        let app = Router::new().route(
+            "/metrics/job/volumeServer/instance/test-instance",
+            put(move |body: String| {
+                let captured = captured_clone.clone();
+                async move {
+                    *captured.lock().unwrap() = Some(body);
+                    "ok"
+                }
+            }),
+        );
+
+        let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
+        let addr = listener.local_addr().unwrap();
+        let server = tokio::spawn(async move {
+            axum::serve(listener, app).await.unwrap();
+        });
+
+        let client = reqwest::Client::new();
+        push_metrics_once(
+            &client,
+            &format!("127.0.0.1:{}", addr.port()),
+            "volumeServer",
+            "test-instance",
+        )
+        .await
+        .unwrap();
+
+        let body = captured.lock().unwrap().clone().unwrap();
+        assert!(body.contains("SeaweedFS_volumeServer_request_total"));
+
+        server.abort();
+    }
+
+    #[test]
+    fn test_delete_collection_metrics_removes_collection_labelsets() {
+        register_metrics();
+
+        VOLUME_GAUGE.with_label_values(&["pics", "volume"]).set(2.0);
+        VOLUME_GAUGE
+            .with_label_values(&["pics", "ec_shards"])
+            .set(3.0);
+        READ_ONLY_VOLUME_GAUGE
+            .with_label_values(&["pics", "volume"])
+            .set(1.0);
+        DISK_SIZE_GAUGE
+            .with_label_values(&["pics", "normal"])
+            .set(10.0);
+        DISK_SIZE_GAUGE
+            .with_label_values(&["pics", "deleted_bytes"])
+            .set(4.0);
+
+        delete_collection_metrics("pics");
+
+        let output = gather_metrics();
+        assert!(!output.contains("collection=\"pics\",type=\"volume\""));
+        assert!(!output.contains("collection=\"pics\",type=\"ec_shards\""));
+        assert!(!output.contains("collection=\"pics\",type=\"normal\""));
+        assert!(!output.contains("collection=\"pics\",type=\"deleted_bytes\""));
+    }
+}

seaweed-volume/src/remote_storage/endpoint_guard.rs

@@ -0,0 +1,654 @@
+//! SSRF guard for caller-supplied remote storage endpoints.
+//!
+//! `FetchAndWriteNeedle` accepts an S3 endpoint URL chosen by the caller and
+//! dials it directly from the volume server, which typically has network
+//! access to cluster-internal hosts. Without validation a caller could point
+//! the server at loopback / link-local / RFC 1918 / cloud-metadata addresses
+//! and read internal services (SSRF). This mirrors the Go volume server's
+//! `validateRemoteEndpoint` (`weed/server/volume_grpc_remote.go`); operators
+//! that legitimately fetch from private hosts opt out with
+//! `-volume.allowUntrustedRemoteEndpoints`.
+//!
+//! NOTE: the Go server additionally pins the validated endpoint against DNS
+//! rebinding with a custom dialer that re-checks the resolved IP at TCP connect
+//! time (`guardedDialer`). The `aws-sdk-s3` client used here builds its own
+//! connector, so that connect-time re-validation is not yet ported; the
+//! up-front resolve-and-check below still blocks the common SSRF vectors. A
+//! narrow TOCTOU window (a hostname that resolves to a public IP here and then
+//! flips to a blocked one when the SDK dials) remains as a follow-up.
+
+use std::net::{IpAddr, Ipv4Addr, Ipv6Addr};
+
+/// AWS/Azure/GCP IPv4 instance-metadata-service (IMDS) address. It is
+/// link-local and thus already covered by [`is_link_local`], but is named
+/// explicitly so the rejection reason is unambiguous in logs.
+const IMDS_IPV4: IpAddr = IpAddr::V4(Ipv4Addr::new(169, 254, 169, 254));
+
+/// Hostnames that target cloud instance metadata services, blocked regardless
+/// of how they resolve because some environments alias the IMDS address under
+/// a name.
+fn is_blocked_imds_host(host: &str) -> bool {
+    matches!(host, "metadata.google.internal" | "metadata")
+}
+
+/// Whether `ip` is in a link-local range: IPv4 169.254.0.0/16, IPv6 fe80::/10
+/// (unicast) and interface-local / link-local multicast (scope 1 and 2).
+fn is_link_local(ip: IpAddr) -> bool {
+    match ip {
+        IpAddr::V4(v4) => v4.is_link_local(),
+        IpAddr::V6(v6) => {
+            let seg0 = v6.segments()[0];
+            if (seg0 & 0xffc0) == 0xfe80 {
+                return true; // fe80::/10 link-local unicast
+            }
+            if (seg0 & 0xff00) == 0xff00 {
+                // ffXS:: multicast; reject interface-local (scope 1) and
+                // link-local (scope 2) the way Go's IsInterfaceLocalMulticast /
+                // IsLinkLocalMulticast do.
+                let scope = seg0 & 0x000f;
+                return scope == 1 || scope == 2;
+            }
+            false
+        }
+    }
+}
+
+/// Whether `ip` is in a private range: IPv4 RFC 1918, IPv6 fc00::/7 unique-local
+/// (matching Go's `net.IP.IsPrivate`).
+fn is_private(ip: IpAddr) -> bool {
+    match ip {
+        IpAddr::V4(v4) => v4.is_private(),
+        IpAddr::V6(v6) => (v6.segments()[0] & 0xfe00) == 0xfc00,
+    }
+}
+
+/// Whether `ip` is in the RFC 6598 carrier-grade NAT range (100.64.0.0/10).
+/// The IPv4 private check does not cover CGNAT, so check it explicitly.
+fn is_cgnat(ip: IpAddr) -> bool {
+    match ip {
+        IpAddr::V4(v4) => {
+            let o = v4.octets();
+            o[0] == 100 && (o[1] & 0xc0) == 0x40 // second octet 64..=127
+        }
+        IpAddr::V6(_) => false,
+    }
+}
+
+/// Returns the IPv4 address carried by an IPv6 transition address -- NAT64
+/// 64:ff9b::/96 (RFC 6052), 6to4 2002::/16 (RFC 3056), Teredo 2001:0000::/32
+/// (RFC 4380), and the deprecated IPv4-compatible ::/96 (RFC 4291) -- or None
+/// when it is not one of those. IPv4-mapped ::ffff:0:0/96 is excluded; it is
+/// already normalized via `to_ipv4_mapped`.
+fn embedded_transition_ipv4(v6: Ipv6Addr) -> Option<Ipv4Addr> {
+    let o = v6.octets();
+    if o[0] == 0x00
+        && o[1] == 0x64
+        && o[2] == 0xff
+        && o[3] == 0x9b
+        && o[4..12].iter().all(|&b| b == 0)
+    {
+        return Some(Ipv4Addr::new(o[12], o[13], o[14], o[15]));
+    }
+    if o[0] == 0x20 && o[1] == 0x02 {
+        return Some(Ipv4Addr::new(o[2], o[3], o[4], o[5]));
+    }
+    if o[0] == 0x20 && o[1] == 0x01 && o[2] == 0x00 && o[3] == 0x00 {
+        // Teredo obfuscates the client IPv4 as its ones' complement.
+        return Some(Ipv4Addr::new(
+            o[12] ^ 0xff,
+            o[13] ^ 0xff,
+            o[14] ^ 0xff,
+            o[15] ^ 0xff,
+        ));
+    }
+    if o[..12].iter().all(|&b| b == 0) {
+        // IPv4-compatible ::a.b.c.d; :: and ::1 are already handled by the
+        // unspecified / loopback checks before extraction runs.
+        return Some(Ipv4Addr::new(o[12], o[13], o[14], o[15]));
+    }
+    None
+}
+
+/// Returns an error if `ip` is not safe to dial from a server that can reach
+/// cluster-internal hosts. Mirrors Go's `checkBlockedIP`.
+pub fn check_blocked_ip(endpoint: &str, ip: IpAddr) -> Result<(), String> {
+    check_blocked_ip_policy(endpoint, ip, false)
+}
+
+/// Like [`check_blocked_ip`], but `allow_private` keeps RFC 1918 / CGNAT
+/// reachable for callers whose target legitimately sits on an internal network
+/// (peer volume servers), while still blocking loopback, link-local (IMDS) and
+/// unspecified. Mirrors Go's `checkBlockedIPPolicy`.
+pub fn check_blocked_ip_policy(
+    endpoint: &str,
+    ip: IpAddr,
+    allow_private: bool,
+) -> Result<(), String> {
+    // Normalize IPv4-mapped IPv6 (`::ffff:a.b.c.d`) to its IPv4 form so the
+    // IPv4 deny rules apply. The OS routes these to the embedded IPv4 address,
+    // so without this `::ffff:127.0.0.1` / `::ffff:169.254.169.254` would slip
+    // past the IPv6-only checks. Mirrors Go's reliance on net.IP.To4().
+    let ip = match ip {
+        IpAddr::V6(v6) => match v6.to_ipv4_mapped() {
+            Some(v4) => IpAddr::V4(v4),
+            None => IpAddr::V6(v6),
+        },
+        other => other,
+    };
+    if ip == IMDS_IPV4 {
+        return Err(format!(
+            "remote endpoint {:?} targets instance metadata service {}",
+            endpoint, ip
+        ));
+    }
+    if ip.is_loopback() {
+        return Err(format!(
+            "remote endpoint {:?} resolves to loopback address {}",
+            endpoint, ip
+        ));
+    }
+    if ip.is_unspecified() {
+        return Err(format!(
+            "remote endpoint {:?} resolves to unspecified address {}",
+            endpoint, ip
+        ));
+    }
+    if is_link_local(ip) {
+        return Err(format!(
+            "remote endpoint {:?} resolves to link-local address {}",
+            endpoint, ip
+        ));
+    }
+    if !allow_private {
+        if is_private(ip) {
+            return Err(format!(
+                "remote endpoint {:?} resolves to private address {}",
+                endpoint, ip
+            ));
+        }
+        if is_cgnat(ip) {
+            return Err(format!(
+                "remote endpoint {:?} resolves to CGNAT address {}",
+                endpoint, ip
+            ));
+        }
+    }
+    // IPv6 transition addresses embed an IPv4 destination that routes to the
+    // same host wherever the matching relay exists (common in IPv6-only cloud).
+    // to_ipv4_mapped above only covers ::ffff: mapped addresses, so pull the
+    // embedded IPv4 out of the other forms and re-check it against the rules.
+    if let IpAddr::V6(v6) = ip
+        && let Some(v4) = embedded_transition_ipv4(v6)
+    {
+        return check_blocked_ip_policy(endpoint, IpAddr::V4(v4), allow_private);
+    }
+    Ok(())
+}
+
+/// Outcome of the synchronous, DNS-free portion of endpoint validation.
+#[derive(Debug)]
+enum HostCheck {
+    /// Host was a literal IP and already passed [`check_blocked_ip`].
+    Validated,
+    /// Host is a name that must be resolved and each address checked.
+    NeedsResolution(String),
+}
+
+/// Validate the scheme and host without touching DNS. Pure and unit-testable.
+fn precheck_endpoint(endpoint: &str) -> Result<HostCheck, String> {
+    let trimmed = endpoint.trim();
+    if trimmed.is_empty() {
+        return Err("remote endpoint is empty".to_string());
+    }
+
+    let Some(scheme_end) = trimmed.find("://") else {
+        return Err(format!(
+            "remote endpoint {:?} must use http or https",
+            endpoint
+        ));
+    };
+    let scheme = trimmed[..scheme_end].to_ascii_lowercase();
+    if scheme != "http" && scheme != "https" {
+        return Err(format!(
+            "remote endpoint {:?} must use http or https, got {:?}",
+            endpoint,
+            &trimmed[..scheme_end]
+        ));
+    }
+
+    // Authority is everything up to the first '/', '?', or '#'.
+    let after = &trimmed[scheme_end + 3..];
+    let authority_end = after.find(['/', '?', '#']).unwrap_or(after.len());
+    let authority = &after[..authority_end];
+
+    // Strip optional userinfo ("user:pass@").
+    let host_port = match authority.rfind('@') {
+        Some(at) => &authority[at + 1..],
+        None => authority,
+    };
+
+    // Extract the host, handling bracketed IPv6 literals and host:port.
+    let host = if let Some(rest) = host_port.strip_prefix('[') {
+        match rest.find(']') {
+            Some(end) => &rest[..end],
+            None => {
+                return Err(format!(
+                    "remote endpoint {:?} has a malformed IPv6 host",
+                    endpoint
+                ));
+            }
+        }
+    } else {
+        match host_port.find(':') {
+            Some(colon) => &host_port[..colon],
+            None => host_port,
+        }
+    };
+
+    if host.is_empty() {
+        return Err(format!("remote endpoint {:?} has no host", endpoint));
+    }
+
+    if is_blocked_imds_host(&host.to_ascii_lowercase()) {
+        return Err(format!(
+            "remote endpoint {:?} targets instance metadata service",
+            endpoint
+        ));
+    }
+
+    if let Ok(ip) = host.parse::<IpAddr>() {
+        check_blocked_ip(endpoint, ip)?;
+        return Ok(HostCheck::Validated);
+    }
+
+    Ok(HostCheck::NeedsResolution(host.to_string()))
+}
+
+/// Resolve `host` to its addresses with a short timeout, matching Go's 2s
+/// resolver deadline.
+async fn resolve_host(host: &str) -> Result<Vec<IpAddr>, String> {
+    let lookup = tokio::net::lookup_host((host, 0u16));
+    let addrs = tokio::time::timeout(std::time::Duration::from_secs(2), lookup)
+        .await
+        .map_err(|_| format!("resolve remote endpoint host {:?}: timed out", host))?
+        .map_err(|e| format!("resolve remote endpoint host {:?}: {}", host, e))?;
+    Ok(addrs.map(|sock| sock.ip()).collect())
+}
+
+/// Returns an error if `endpoint` is not safe to dial from a server with access
+/// to cluster-internal hosts. Rejects empty / non-http(s) endpoints,
+/// loopback / unspecified / link-local / RFC 1918 / CGNAT addresses, and
+/// well-known IMDS hostnames. Hostnames are resolved and every returned address
+/// is checked. Mirrors Go's `validateRemoteEndpoint`.
+pub async fn validate_remote_endpoint(endpoint: &str) -> Result<(), String> {
+    match precheck_endpoint(endpoint)? {
+        HostCheck::Validated => Ok(()),
+        HostCheck::NeedsResolution(host) => {
+            let addrs = resolve_host(&host).await?;
+            if addrs.is_empty() {
+                return Err(format!(
+                    "resolve remote endpoint host {:?}: no addresses",
+                    host
+                ));
+            }
+            for ip in addrs {
+                check_blocked_ip(endpoint, ip)?;
+            }
+            Ok(())
+        }
+    }
+}
+
+/// Returns an error if `target` could redirect a replica upload away from a peer
+/// volume server. The target must be a bare `host:port` -- a scheme, userinfo,
+/// path, query or fragment can smuggle a different destination into the
+/// formatted upload URL -- whose host is not loopback, link-local (IMDS) or
+/// unspecified. Cluster peers legitimately sit on private networks, so RFC 1918
+/// / CGNAT are allowed. Mirrors Go's `validateReplicaTarget`.
+pub async fn validate_replica_target(target: &str) -> Result<(), String> {
+    let trimmed = target.trim();
+    if trimmed.is_empty() {
+        return Err("replica target is empty".to_string());
+    }
+    if trimmed.contains("://") || trimmed.contains(['/', '?', '#', '@', '\\']) {
+        return Err(format!(
+            "replica target {:?} must be a bare host:port",
+            target
+        ));
+    }
+
+    // Require an explicit host:port, handling `[IPv6]:port`. A bracketless IPv6
+    // literal (which carries its own colons) is rejected; peers are addressed as
+    // `[ipv6]:port`, matching Go's net.SplitHostPort.
+    let host = if let Some(rest) = trimmed.strip_prefix('[') {
+        match rest.split_once(']') {
+            Some((h, port)) if port.starts_with(':') && port.len() > 1 => h,
+            _ => {
+                return Err(format!(
+                    "replica target {:?} must be a bare host:port",
+                    target
+                ));
+            }
+        }
+    } else {
+        match trimmed.rsplit_once(':') {
+            Some((h, port)) if !port.is_empty() && !h.contains(':') => h,
+            _ => {
+                return Err(format!(
+                    "replica target {:?} must be a bare host:port",
+                    target
+                ));
+            }
+        }
+    };
+
+    if host.is_empty() {
+        return Err(format!("replica target {:?} has no host", target));
+    }
+    if is_blocked_imds_host(&host.to_ascii_lowercase()) {
+        return Err(format!(
+            "replica target {:?} targets instance metadata service",
+            target
+        ));
+    }
+    if let Ok(ip) = host.parse::<IpAddr>() {
+        return check_blocked_ip_policy(target, ip, true);
+    }
+
+    let addrs = resolve_host(host).await?;
+    if addrs.is_empty() {
+        return Err(format!(
+            "resolve replica target host {:?}: no addresses",
+            host
+        ));
+    }
+    for ip in addrs {
+        check_blocked_ip_policy(target, ip, true)?;
+    }
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn ip(s: &str) -> IpAddr {
+        s.parse().unwrap()
+    }
+
+    #[test]
+    fn rejects_blocked_literal_endpoints() {
+        let cases = [
+            ("http://127.0.0.1:8080", "loopback"),
+            ("http://[::1]:8080", "loopback"),
+            ("http://169.254.169.254/", "metadata"),
+            ("http://0.0.0.0/", "unspecified"),
+            ("http://[fe80::1]/", "link-local"),
+            ("http://10.0.0.1/", "private"),
+            ("http://172.16.5.5/", "private"),
+            ("http://192.168.0.1/", "private"),
+            ("http://100.64.0.1/", "CGNAT"),
+        ];
+        for (endpoint, want) in cases {
+            let err = precheck_endpoint(endpoint).expect_err(endpoint);
+            assert!(err.contains(want), "{endpoint}: {err:?} missing {want:?}");
+        }
+    }
+
+    #[test]
+    fn rejects_empty_and_bad_scheme() {
+        assert!(precheck_endpoint("").unwrap_err().contains("empty"));
+        assert!(
+            precheck_endpoint("ftp://example.com/")
+                .unwrap_err()
+                .contains("http or https")
+        );
+        assert!(
+            precheck_endpoint("example.com/")
+                .unwrap_err()
+                .contains("http or https")
+        );
+    }
+
+    #[test]
+    fn rejects_imds_hostnames() {
+        assert!(
+            precheck_endpoint("http://metadata.google.internal/")
+                .unwrap_err()
+                .contains("metadata service")
+        );
+        assert!(
+            precheck_endpoint("http://metadata/")
+                .unwrap_err()
+                .contains("metadata service")
+        );
+    }
+
+    #[test]
+    fn accepts_public_literal_and_defers_hostname() {
+        assert!(matches!(
+            precheck_endpoint("https://52.216.10.10/"),
+            Ok(HostCheck::Validated)
+        ));
+        match precheck_endpoint("https://s3.us-east-1.amazonaws.com/") {
+            Ok(HostCheck::NeedsResolution(host)) => {
+                assert_eq!(host, "s3.us-east-1.amazonaws.com")
+            }
+            other => panic!("expected resolution, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn check_blocked_ip_matches_resolved_categories() {
+        // Mirror Go's "host resolves to X" cases at the address level.
+        assert!(
+            check_blocked_ip("e", ip("127.0.0.1"))
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            check_blocked_ip("e", ip("169.254.10.20"))
+                .unwrap_err()
+                .contains("link-local")
+        );
+        assert!(
+            check_blocked_ip("e", ip("10.1.2.3"))
+                .unwrap_err()
+                .contains("private")
+        );
+        assert!(
+            check_blocked_ip("e", ip("172.20.0.5"))
+                .unwrap_err()
+                .contains("private")
+        );
+        assert!(
+            check_blocked_ip("e", ip("192.168.1.1"))
+                .unwrap_err()
+                .contains("private")
+        );
+        assert!(
+            check_blocked_ip("e", ip("100.64.0.42"))
+                .unwrap_err()
+                .contains("CGNAT")
+        );
+        assert!(
+            check_blocked_ip("e", ip("fc00::1"))
+                .unwrap_err()
+                .contains("private")
+        );
+        assert!(check_blocked_ip("e", ip("52.216.10.10")).is_ok());
+        assert!(check_blocked_ip("e", ip("2606:4700:4700::1111")).is_ok());
+    }
+
+    #[test]
+    fn rejects_ipv6_transition_addresses() {
+        // Every transition form encoding an internal IPv4 must be blocked.
+        assert!(
+            check_blocked_ip("e", ip("64:ff9b::a9fe:a9fe")) // NAT64 -> IMDS
+                .unwrap_err()
+                .contains("metadata")
+        );
+        assert!(
+            check_blocked_ip("e", ip("64:ff9b::7f00:1")) // NAT64 -> loopback
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            check_blocked_ip("e", ip("2002:a00:1::")) // 6to4 -> 10.0.0.1
+                .unwrap_err()
+                .contains("private")
+        );
+        assert!(
+            check_blocked_ip("e", ip("2001:0:4136:e378:8000:63bf:80ff:fffe")) // Teredo -> loopback
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            check_blocked_ip("e", ip("::7f00:1")) // IPv4-compatible -> loopback
+                .unwrap_err()
+                .contains("loopback")
+        );
+        // A NAT64 address outside the 64:ff9b::/96 well-known prefix is not
+        // decoded (its embedded IPv4 lives elsewhere), so it is left as-is.
+        assert!(check_blocked_ip("e", ip("64:ff9b:1::a9fe:a9fe")).is_ok());
+        // Every transition form embedding a public IPv4 (8.8.8.8) still passes:
+        // NAT64, 6to4, Teredo, IPv4-compatible.
+        assert!(check_blocked_ip("e", ip("64:ff9b::808:808")).is_ok());
+        assert!(check_blocked_ip("e", ip("2002:808:808::")).is_ok());
+        assert!(check_blocked_ip("e", ip("2001::f7f7:f7f7")).is_ok());
+        assert!(check_blocked_ip("e", ip("::808:808")).is_ok());
+        // Bracketed transition literal via the full endpoint path.
+        assert!(
+            precheck_endpoint("http://[64:ff9b::a9fe:a9fe]/")
+                .unwrap_err()
+                .contains("metadata")
+        );
+    }
+
+    #[test]
+    fn rejects_ipv4_mapped_ipv6() {
+        // IPv4-mapped IPv6 must be unmapped so the IPv4 rules catch it.
+        assert!(
+            check_blocked_ip("e", ip("::ffff:127.0.0.1"))
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            check_blocked_ip("e", ip("::ffff:169.254.169.254"))
+                .unwrap_err()
+                .contains("metadata")
+        );
+        assert!(
+            check_blocked_ip("e", ip("::ffff:10.0.0.1"))
+                .unwrap_err()
+                .contains("private")
+        );
+        // A mapped public address still passes, and genuine IPv6 loopback is
+        // still caught by the V6 path.
+        assert!(check_blocked_ip("e", ip("::ffff:52.216.10.10")).is_ok());
+        assert!(
+            check_blocked_ip("e", ip("::1"))
+                .unwrap_err()
+                .contains("loopback")
+        );
+        // Bracketed mapped literal via the full endpoint path.
+        assert!(
+            precheck_endpoint("http://[::ffff:127.0.0.1]/")
+                .unwrap_err()
+                .contains("loopback")
+        );
+    }
+
+    #[test]
+    fn check_blocked_ip_policy_allows_private_peers() {
+        // The replica leg targets peer volume servers, which may be private.
+        assert!(check_blocked_ip_policy("e", ip("10.0.0.7"), true).is_ok());
+        assert!(check_blocked_ip_policy("e", ip("192.168.1.5"), true).is_ok());
+        assert!(check_blocked_ip_policy("e", ip("100.64.0.42"), true).is_ok());
+        // Loopback / IMDS / unspecified stay blocked even when private is allowed.
+        assert!(
+            check_blocked_ip_policy("e", ip("127.0.0.1"), true)
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            check_blocked_ip_policy("e", ip("169.254.169.254"), true)
+                .unwrap_err()
+                .contains("metadata")
+        );
+        assert!(
+            check_blocked_ip_policy("e", ip("0.0.0.0"), true)
+                .unwrap_err()
+                .contains("unspecified")
+        );
+    }
+
+    #[tokio::test]
+    async fn validate_replica_target_rejects_and_allows() {
+        // A path plus a trailing ?a= would otherwise swallow ?type=replicate.
+        assert!(
+            validate_replica_target("127.0.0.1:7000/status/x/?a=")
+                .await
+                .unwrap_err()
+                .contains("bare host:port")
+        );
+        assert!(
+            validate_replica_target("http://10.0.0.7:8080")
+                .await
+                .unwrap_err()
+                .contains("bare host:port")
+        );
+        assert!(
+            validate_replica_target("user@10.0.0.7:8080")
+                .await
+                .unwrap_err()
+                .contains("bare host:port")
+        );
+        assert!(
+            validate_replica_target("10.0.0.7")
+                .await
+                .unwrap_err()
+                .contains("bare host:port")
+        );
+        assert!(
+            validate_replica_target("peer.example.com")
+                .await
+                .unwrap_err()
+                .contains("bare host:port")
+        );
+        assert!(
+            validate_replica_target("127.0.0.1:8080")
+                .await
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            validate_replica_target("[::1]:8080")
+                .await
+                .unwrap_err()
+                .contains("loopback")
+        );
+        assert!(
+            validate_replica_target("169.254.169.254:80")
+                .await
+                .unwrap_err()
+                .contains("metadata")
+        );
+        assert!(
+            validate_replica_target("metadata:80")
+                .await
+                .unwrap_err()
+                .contains("metadata")
+        );
+        assert!(
+            validate_replica_target("")
+                .await
+                .unwrap_err()
+                .contains("empty")
+        );
+        // Legitimate peer volume servers on private networks pass.
+        assert!(validate_replica_target("10.0.0.7:8080").await.is_ok());
+        assert!(validate_replica_target("192.168.1.5:8080").await.is_ok());
+        assert!(validate_replica_target("[fd00::1]:8080").await.is_ok());
+    }
+}

seaweed-volume/src/remote_storage/mod.rs

@@ -0,0 +1,246 @@
+//! Remote storage backends for tiered storage support.
+//!
+//! Provides a trait-based abstraction over cloud storage providers (S3, GCS, Azure, etc.)
+//! and a registry to create clients from protobuf RemoteConf messages.
+
+pub mod endpoint_guard;
+pub mod s3;
+pub mod s3_tier;
+
+pub use endpoint_guard::{validate_remote_endpoint, validate_replica_target};
+
+use crate::pb::remote_pb::{RemoteConf, RemoteStorageLocation};
+
+/// Error type for remote storage operations.
+#[derive(Debug, thiserror::Error)]
+pub enum RemoteStorageError {
+    #[error("remote storage type {0} not found")]
+    TypeNotFound(String),
+    #[error("remote object not found: {0}")]
+    ObjectNotFound(String),
+    #[error("remote storage error: {0}")]
+    Other(String),
+    #[error("io error: {0}")]
+    Io(#[from] std::io::Error),
+}
+
+/// Metadata about a remote file entry.
+#[derive(Debug, Clone)]
+pub struct RemoteEntry {
+    pub size: i64,
+    pub last_modified_at: i64, // Unix seconds
+    pub e_tag: String,
+    pub storage_name: String,
+}
+
+/// Trait for remote storage clients. Matches Go's RemoteStorageClient interface.
+#[async_trait::async_trait]
+pub trait RemoteStorageClient: Send + Sync {
+    /// Read (part of) a file from remote storage.
+    async fn read_file(
+        &self,
+        loc: &RemoteStorageLocation,
+        offset: i64,
+        size: i64,
+    ) -> Result<Vec<u8>, RemoteStorageError>;
+
+    /// Write a file to remote storage.
+    async fn write_file(
+        &self,
+        loc: &RemoteStorageLocation,
+        data: &[u8],
+    ) -> Result<RemoteEntry, RemoteStorageError>;
+
+    /// Get metadata for a file in remote storage.
+    async fn stat_file(
+        &self,
+        loc: &RemoteStorageLocation,
+    ) -> Result<RemoteEntry, RemoteStorageError>;
+
+    /// Delete a file from remote storage.
+    async fn delete_file(&self, loc: &RemoteStorageLocation) -> Result<(), RemoteStorageError>;
+
+    /// List all buckets.
+    async fn list_buckets(&self) -> Result<Vec<String>, RemoteStorageError>;
+
+    /// The RemoteConf used to create this client.
+    fn remote_conf(&self) -> &RemoteConf;
+}
+
+/// Create a new remote storage client from a RemoteConf.
+pub fn make_remote_storage_client(
+    conf: &RemoteConf,
+) -> Result<Box<dyn RemoteStorageClient>, RemoteStorageError> {
+    match conf.r#type.as_str() {
+        // All S3-compatible backends use the same client with different credentials
+        "s3" | "wasabi" | "backblaze" | "aliyun" | "tencent" | "baidu" | "filebase" | "storj"
+        | "contabo" => {
+            let (access_key, secret_key, endpoint, region) = extract_s3_credentials(conf);
+            Ok(Box::new(s3::S3RemoteStorageClient::new(
+                conf.clone(),
+                &access_key,
+                &secret_key,
+                &region,
+                &endpoint,
+                conf.s3_force_path_style,
+            )))
+        }
+        other => Err(RemoteStorageError::TypeNotFound(other.to_string())),
+    }
+}
+
+/// Endpoint URL that the volume server would dial directly for `conf`, or
+/// `None` for non-S3-compatible backends. Every type handled here routes
+/// through [`s3::S3RemoteStorageClient`] with a caller-supplied endpoint, so
+/// the SSRF guard must validate each one. Keep this match in sync with
+/// [`make_remote_storage_client`].
+pub fn s3_compatible_endpoint(conf: &RemoteConf) -> Option<&str> {
+    match conf.r#type.as_str() {
+        "s3" => Some(&conf.s3_endpoint),
+        "wasabi" => Some(&conf.wasabi_endpoint),
+        "backblaze" => Some(&conf.backblaze_endpoint),
+        "aliyun" => Some(&conf.aliyun_endpoint),
+        "tencent" => Some(&conf.tencent_endpoint),
+        "baidu" => Some(&conf.baidu_endpoint),
+        "filebase" => Some(&conf.filebase_endpoint),
+        "storj" => Some(&conf.storj_endpoint),
+        "contabo" => Some(&conf.contabo_endpoint),
+        _ => None,
+    }
+}
+
+/// Extract S3-compatible credentials from a RemoteConf based on its type.
+fn extract_s3_credentials(conf: &RemoteConf) -> (String, String, String, String) {
+    match conf.r#type.as_str() {
+        "s3" => (
+            conf.s3_access_key.clone(),
+            conf.s3_secret_key.clone(),
+            conf.s3_endpoint.clone(),
+            if conf.s3_region.is_empty() {
+                "us-east-1".to_string()
+            } else {
+                conf.s3_region.clone()
+            },
+        ),
+        "wasabi" => (
+            conf.wasabi_access_key.clone(),
+            conf.wasabi_secret_key.clone(),
+            conf.wasabi_endpoint.clone(),
+            conf.wasabi_region.clone(),
+        ),
+        "backblaze" => (
+            conf.backblaze_key_id.clone(),
+            conf.backblaze_application_key.clone(),
+            conf.backblaze_endpoint.clone(),
+            conf.backblaze_region.clone(),
+        ),
+        "aliyun" => (
+            conf.aliyun_access_key.clone(),
+            conf.aliyun_secret_key.clone(),
+            conf.aliyun_endpoint.clone(),
+            conf.aliyun_region.clone(),
+        ),
+        "tencent" => (
+            conf.tencent_secret_id.clone(),
+            conf.tencent_secret_key.clone(),
+            conf.tencent_endpoint.clone(),
+            String::new(),
+        ),
+        "baidu" => (
+            conf.baidu_access_key.clone(),
+            conf.baidu_secret_key.clone(),
+            conf.baidu_endpoint.clone(),
+            conf.baidu_region.clone(),
+        ),
+        "filebase" => (
+            conf.filebase_access_key.clone(),
+            conf.filebase_secret_key.clone(),
+            conf.filebase_endpoint.clone(),
+            String::new(),
+        ),
+        "storj" => (
+            conf.storj_access_key.clone(),
+            conf.storj_secret_key.clone(),
+            conf.storj_endpoint.clone(),
+            String::new(),
+        ),
+        "contabo" => (
+            conf.contabo_access_key.clone(),
+            conf.contabo_secret_key.clone(),
+            conf.contabo_endpoint.clone(),
+            conf.contabo_region.clone(),
+        ),
+        _ => (
+            conf.s3_access_key.clone(),
+            conf.s3_secret_key.clone(),
+            conf.s3_endpoint.clone(),
+            conf.s3_region.clone(),
+        ),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn s3_compatible_endpoint_covers_all_s3_backends() {
+        let s3 = RemoteConf {
+            r#type: "s3".to_string(),
+            s3_endpoint: "http://s3.internal".to_string(),
+            ..Default::default()
+        };
+        assert_eq!(s3_compatible_endpoint(&s3), Some("http://s3.internal"));
+
+        // A non-"s3" S3-compatible type still surfaces its own endpoint, so the
+        // SSRF guard cannot be bypassed by picking a different alias.
+        let wasabi = RemoteConf {
+            r#type: "wasabi".to_string(),
+            wasabi_endpoint: "http://wasabi.internal".to_string(),
+            ..Default::default()
+        };
+        assert_eq!(
+            s3_compatible_endpoint(&wasabi),
+            Some("http://wasabi.internal")
+        );
+
+        // Non-S3 backends do not dial a caller-supplied URL directly.
+        let gcs = RemoteConf {
+            r#type: "gcs".to_string(),
+            ..Default::default()
+        };
+        assert_eq!(s3_compatible_endpoint(&gcs), None);
+    }
+
+    #[test]
+    fn azure_endpoint_has_no_ssrf_path() {
+        // The Go volume server guards the caller-supplied azure endpoint against
+        // SSRF. This server has no azure backend, so there is nothing to dial:
+        // azure is not S3-compatible (the endpoint guard does not apply) and
+        // make_remote_storage_client rejects the type before building a client.
+        let azure = RemoteConf {
+            r#type: "azure".to_string(),
+            azure_endpoint: "https://169.254.169.254/".to_string(),
+            ..Default::default()
+        };
+        assert_eq!(s3_compatible_endpoint(&azure), None);
+        assert!(make_remote_storage_client(&azure).is_err());
+    }
+
+    #[test]
+    fn gcs_credentials_have_no_ssrf_path() {
+        // The Go volume server accepts only static-key gcs credentials and puts
+        // their token endpoint behind the SSRF guard, because the SDK dials
+        // whatever url, file or executable the credentials name. This server has
+        // no gcs backend, so make_remote_storage_client rejects the type before
+        // any credentials are parsed. Anyone adding one must carry both guards
+        // over with it.
+        let gcs = RemoteConf {
+            r#type: "gcs".to_string(),
+            gcs_google_application_credentials: r#"{"type":"external_account","credential_source":{"url":"http://169.254.169.254/"}}"#.to_string(),
+            ..Default::default()
+        };
+        assert_eq!(s3_compatible_endpoint(&gcs), None);
+        assert!(make_remote_storage_client(&gcs).is_err());
+    }
+}

seaweed-volume/src/remote_storage/s3.rs

@@ -0,0 +1,186 @@
+//! S3-compatible remote storage client.
+//!
+//! Works with AWS S3, MinIO, SeaweedFS S3, and all S3-compatible providers.
+
+use aws_sdk_s3::Client;
+use aws_sdk_s3::config::{BehaviorVersion, Credentials, Region};
+use aws_sdk_s3::primitives::ByteStream;
+
+use super::{RemoteEntry, RemoteStorageClient, RemoteStorageError};
+use crate::pb::remote_pb::{RemoteConf, RemoteStorageLocation};
+
+/// S3-compatible remote storage client.
+pub struct S3RemoteStorageClient {
+    client: Client,
+    conf: RemoteConf,
+}
+
+impl S3RemoteStorageClient {
+    /// Create a new S3 client from credentials and endpoint configuration.
+    pub fn new(
+        conf: RemoteConf,
+        access_key: &str,
+        secret_key: &str,
+        region: &str,
+        endpoint: &str,
+        force_path_style: bool,
+    ) -> Self {
+        let region = if region.is_empty() {
+            "us-east-1"
+        } else {
+            region
+        };
+
+        let credentials = Credentials::new(
+            access_key,
+            secret_key,
+            None, // session token
+            None, // expiry
+            "seaweedfs-volume",
+        );
+
+        let mut s3_config = aws_sdk_s3::Config::builder()
+            .behavior_version(BehaviorVersion::latest())
+            .region(Region::new(region.to_string()))
+            .credentials_provider(credentials)
+            .force_path_style(force_path_style);
+
+        if !endpoint.is_empty() {
+            s3_config = s3_config.endpoint_url(endpoint);
+        }
+
+        let client = Client::from_conf(s3_config.build());
+
+        S3RemoteStorageClient { client, conf }
+    }
+}
+
+#[async_trait::async_trait]
+impl RemoteStorageClient for S3RemoteStorageClient {
+    async fn read_file(
+        &self,
+        loc: &RemoteStorageLocation,
+        offset: i64,
+        size: i64,
+    ) -> Result<Vec<u8>, RemoteStorageError> {
+        let key = loc.path.trim_start_matches('/');
+
+        let mut req = self.client.get_object().bucket(&loc.bucket).key(key);
+
+        // Set byte range if specified
+        if size > 0 {
+            let end = offset + size - 1;
+            req = req.range(format!("bytes={}-{}", offset, end));
+        } else if offset > 0 {
+            req = req.range(format!("bytes={}-", offset));
+        }
+
+        let resp = req.send().await.map_err(|e| {
+            let msg = format!("{}", e);
+            if msg.contains("NoSuchKey") || msg.contains("404") {
+                RemoteStorageError::ObjectNotFound(format!("{}/{}", loc.bucket, key))
+            } else {
+                RemoteStorageError::Other(format!("s3 get object: {}", e))
+            }
+        })?;
+
+        let data = resp
+            .body
+            .collect()
+            .await
+            .map_err(|e| RemoteStorageError::Other(format!("s3 read body: {}", e)))?;
+
+        Ok(data.into_bytes().to_vec())
+    }
+
+    async fn write_file(
+        &self,
+        loc: &RemoteStorageLocation,
+        data: &[u8],
+    ) -> Result<RemoteEntry, RemoteStorageError> {
+        let key = loc.path.trim_start_matches('/');
+
+        let resp = self
+            .client
+            .put_object()
+            .bucket(&loc.bucket)
+            .key(key)
+            .body(ByteStream::from(data.to_vec()))
+            .send()
+            .await
+            .map_err(|e| RemoteStorageError::Other(format!("s3 put object: {}", e)))?;
+
+        Ok(RemoteEntry {
+            size: data.len() as i64,
+            last_modified_at: std::time::SystemTime::now()
+                .duration_since(std::time::UNIX_EPOCH)
+                .unwrap_or_default()
+                .as_secs() as i64,
+            e_tag: resp.e_tag().unwrap_or_default().to_string(),
+            storage_name: loc.name.clone(),
+        })
+    }
+
+    async fn stat_file(
+        &self,
+        loc: &RemoteStorageLocation,
+    ) -> Result<RemoteEntry, RemoteStorageError> {
+        let key = loc.path.trim_start_matches('/');
+
+        let resp = self
+            .client
+            .head_object()
+            .bucket(&loc.bucket)
+            .key(key)
+            .send()
+            .await
+            .map_err(|e| {
+                let msg = format!("{}", e);
+                if msg.contains("404") || msg.contains("NotFound") {
+                    RemoteStorageError::ObjectNotFound(format!("{}/{}", loc.bucket, key))
+                } else {
+                    RemoteStorageError::Other(format!("s3 head object: {}", e))
+                }
+            })?;
+
+        Ok(RemoteEntry {
+            size: resp.content_length().unwrap_or(0),
+            last_modified_at: resp.last_modified().map(|t| t.secs()).unwrap_or(0),
+            e_tag: resp.e_tag().unwrap_or_default().to_string(),
+            storage_name: loc.name.clone(),
+        })
+    }
+
+    async fn delete_file(&self, loc: &RemoteStorageLocation) -> Result<(), RemoteStorageError> {
+        let key = loc.path.trim_start_matches('/');
+
+        self.client
+            .delete_object()
+            .bucket(&loc.bucket)
+            .key(key)
+            .send()
+            .await
+            .map_err(|e| RemoteStorageError::Other(format!("s3 delete object: {}", e)))?;
+
+        Ok(())
+    }
+
+    async fn list_buckets(&self) -> Result<Vec<String>, RemoteStorageError> {
+        let resp = self
+            .client
+            .list_buckets()
+            .send()
+            .await
+            .map_err(|e| RemoteStorageError::Other(format!("s3 list buckets: {}", e)))?;
+
+        Ok(resp
+            .buckets()
+            .iter()
+            .filter_map(|b| b.name().map(String::from))
+            .collect())
+    }
+
+    fn remote_conf(&self) -> &RemoteConf {
+        &self.conf
+    }
+}

seaweed-volume/src/remote_storage/s3_tier.rs

@@ -0,0 +1,572 @@
+//! S3-compatible tiered storage backend for volume .dat file upload/download.
+//!
+//! Provides multipart upload and concurrent download with progress callbacks,
+//! matching the Go SeaweedFS S3 backend behavior.
+
+use std::collections::HashMap;
+use std::future::Future;
+use std::sync::{Arc, OnceLock, RwLock};
+
+use aws_sdk_s3::Client;
+use aws_sdk_s3::config::{BehaviorVersion, Credentials, Region};
+use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
+use tokio::io::{AsyncReadExt, AsyncSeekExt, AsyncWriteExt};
+use tokio::sync::Semaphore;
+
+/// Concurrency limit for multipart upload/download (matches Go's s3manager).
+const CONCURRENCY: usize = 5;
+
+/// Configuration for an S3 tier backend.
+#[derive(Debug, Clone)]
+pub struct S3TierConfig {
+    pub access_key: String,
+    pub secret_key: String,
+    pub region: String,
+    pub bucket: String,
+    pub endpoint: String,
+    pub storage_class: String,
+    pub force_path_style: bool,
+}
+
+/// S3 tier backend for uploading/downloading volume .dat files.
+pub struct S3TierBackend {
+    client: Client,
+    pub bucket: String,
+    pub storage_class: String,
+}
+
+impl S3TierBackend {
+    /// Create a new S3 tier backend from configuration.
+    pub fn new(config: &S3TierConfig) -> Self {
+        let region = if config.region.is_empty() {
+            "us-east-1"
+        } else {
+            &config.region
+        };
+
+        let credentials = Credentials::new(
+            &config.access_key,
+            &config.secret_key,
+            None,
+            None,
+            "seaweedfs-volume-tier",
+        );
+
+        let mut s3_config = aws_sdk_s3::Config::builder()
+            .behavior_version(BehaviorVersion::latest())
+            .region(Region::new(region.to_string()))
+            .credentials_provider(credentials)
+            .force_path_style(config.force_path_style);
+
+        if !config.endpoint.is_empty() {
+            s3_config = s3_config.endpoint_url(&config.endpoint);
+        }
+
+        let client = Client::from_conf(s3_config.build());
+
+        S3TierBackend {
+            client,
+            bucket: config.bucket.clone(),
+            storage_class: if config.storage_class.is_empty() {
+                "STANDARD_IA".to_string()
+            } else {
+                config.storage_class.clone()
+            },
+        }
+    }
+
+    /// Upload a local file to S3 using multipart upload with concurrent parts
+    /// and progress reporting.
+    ///
+    /// Returns (s3_key, file_size) on success.
+    /// The progress callback receives (bytes_uploaded, percentage).
+    /// Uses 64MB part size and 5 concurrent uploads (matches Go s3manager).
+    /// `progress_fn` returning `Err` aborts the upload, mirroring Go's
+    /// `fn(progressed, percentage) error` in `s3_upload.go`, where the error
+    /// surfaces out of `ReadAt` and fails the transfer. It is what lets a
+    /// caller that has hung up stop the work it is no longer waiting for.
+    pub async fn upload_file<F>(
+        &self,
+        file_path: &str,
+        progress_fn: F,
+    ) -> Result<(String, u64), String>
+    where
+        F: FnMut(i64, f32) -> Result<(), String> + Send + Sync + 'static,
+    {
+        let key = uuid::Uuid::new_v4().to_string();
+
+        let metadata = tokio::fs::metadata(file_path)
+            .await
+            .map_err(|e| format!("failed to stat file {}: {}", file_path, e))?;
+        let file_size = metadata.len();
+
+        // Calculate part size: start at 64MB, scale up for very large files (matches Go)
+        let mut part_size: u64 = 64 * 1024 * 1024;
+        while part_size * 1000 < file_size {
+            part_size *= 4;
+        }
+
+        // Initiate multipart upload
+        let create_resp = self
+            .client
+            .create_multipart_upload()
+            .bucket(&self.bucket)
+            .key(&key)
+            .storage_class(
+                self.storage_class
+                    .parse()
+                    .unwrap_or(aws_sdk_s3::types::StorageClass::StandardIa),
+            )
+            .send()
+            .await
+            .map_err(|e| format!("failed to create multipart upload: {}", e))?;
+
+        let upload_id = create_resp
+            .upload_id()
+            .ok_or_else(|| "no upload_id in multipart upload response".to_string())?
+            .to_string();
+
+        // Build list of (part_number, offset, size) for all parts
+        let mut parts_plan: Vec<(i32, u64, usize)> = Vec::new();
+        let mut offset: u64 = 0;
+        let mut part_number: i32 = 1;
+        while offset < file_size {
+            let remaining = file_size - offset;
+            let this_part_size = std::cmp::min(part_size, remaining) as usize;
+            parts_plan.push((part_number, offset, this_part_size));
+            offset += this_part_size as u64;
+            part_number += 1;
+        }
+
+        // Upload parts concurrently with a semaphore limiting to CONCURRENCY
+        let semaphore = Arc::new(Semaphore::new(CONCURRENCY));
+        let client = &self.client;
+        let bucket = &self.bucket;
+        let file_path_owned = file_path.to_string();
+        let progress = Arc::new(std::sync::Mutex::new((0u64, progress_fn)));
+
+        let mut handles = Vec::with_capacity(parts_plan.len());
+        for (pn, off, size) in parts_plan {
+            let sem = semaphore.clone();
+            let client = client.clone();
+            let bucket = bucket.clone();
+            let key = key.clone();
+            let upload_id = upload_id.clone();
+            let fp = file_path_owned.clone();
+            let progress = progress.clone();
+
+            handles.push(tokio::spawn(async move {
+                let _permit = sem
+                    .acquire()
+                    .await
+                    .map_err(|e| format!("semaphore error: {}", e))?;
+
+                // Read this part's data from the file at the correct offset
+                let mut file = tokio::fs::File::open(&fp)
+                    .await
+                    .map_err(|e| format!("failed to open file {}: {}", fp, e))?;
+                file.seek(std::io::SeekFrom::Start(off))
+                    .await
+                    .map_err(|e| format!("failed to seek to offset {}: {}", off, e))?;
+                let mut buf = vec![0u8; size];
+                file.read_exact(&mut buf)
+                    .await
+                    .map_err(|e| format!("failed to read file at offset {}: {}", off, e))?;
+
+                let upload_part_resp = client
+                    .upload_part()
+                    .bucket(&bucket)
+                    .key(&key)
+                    .upload_id(&upload_id)
+                    .part_number(pn)
+                    .body(buf.into())
+                    .send()
+                    .await
+                    .map_err(|e| {
+                        format!("failed to upload part {} at offset {}: {}", pn, off, e)
+                    })?;
+
+                let e_tag = upload_part_resp.e_tag().unwrap_or_default().to_string();
+
+                // Report progress. The lock is released before the result is
+                // propagated so an aborting callback cannot poison the mutex
+                // for the other parts still in flight.
+                let progress_result = {
+                    let mut guard = progress.lock().unwrap();
+                    guard.0 += size as u64;
+                    let uploaded = guard.0;
+                    let pct = if file_size > 0 {
+                        (uploaded as f32 * 100.0) / file_size as f32
+                    } else {
+                        100.0
+                    };
+                    (guard.1)(uploaded as i64, pct)
+                };
+                progress_result?;
+
+                Ok::<_, String>(
+                    CompletedPart::builder()
+                        .e_tag(e_tag)
+                        .part_number(pn)
+                        .build(),
+                )
+            }));
+        }
+
+        let finish = async {
+            // Collect results, preserving part order
+            let mut completed_parts = Vec::with_capacity(handles.len());
+            for handle in handles {
+                let part = handle
+                    .await
+                    .map_err(|e| format!("upload task panicked: {}", e))??;
+                completed_parts.push(part);
+            }
+
+            // Complete multipart upload
+            let completed_upload = CompletedMultipartUpload::builder()
+                .set_parts(Some(completed_parts))
+                .build();
+
+            self.client
+                .complete_multipart_upload()
+                .bucket(&self.bucket)
+                .key(&key)
+                .upload_id(&upload_id)
+                .multipart_upload(completed_upload)
+                .send()
+                .await
+                .map_err(|e| format!("failed to complete multipart upload: {}", e))?;
+
+            Ok::<(), String>(())
+        }
+        .await;
+
+        if let Err(e) = finish {
+            // An abandoned multipart upload does not appear in an ordinary
+            // object listing but still accrues storage charges until a
+            // lifecycle rule reaps it. Now that a departing caller aborts the
+            // transfer this is a routine path, not a rare one.
+            if let Err(abort_err) = self
+                .client
+                .abort_multipart_upload()
+                .bucket(&self.bucket)
+                .key(&key)
+                .upload_id(&upload_id)
+                .send()
+                .await
+            {
+                tracing::warn!(
+                    "failed to abort multipart upload {} for key {}: {}",
+                    upload_id,
+                    key,
+                    abort_err
+                );
+            }
+            return Err(e);
+        }
+
+        Ok((key, file_size))
+    }
+
+    /// Download a file from S3 to a local path with concurrent range requests
+    /// and progress reporting.
+    ///
+    /// Returns the file size on success.
+    /// Uses 64MB part size and 5 concurrent downloads (matches Go s3manager).
+    /// `progress_fn` returning `Err` aborts the download, mirroring Go's
+    /// `fn(progressed, percentage) error` in `s3_download.go`.
+    pub async fn download_file<F>(
+        &self,
+        dest_path: &str,
+        key: &str,
+        progress_fn: F,
+    ) -> Result<u64, String>
+    where
+        F: FnMut(i64, f32) -> Result<(), String> + Send + Sync + 'static,
+    {
+        // Get file size first
+        let head_resp = self
+            .client
+            .head_object()
+            .bucket(&self.bucket)
+            .key(key)
+            .send()
+            .await
+            .map_err(|e| format!("failed to head object {}: {}", key, e))?;
+
+        let file_size = head_resp.content_length().unwrap_or(0) as u64;
+
+        // Pre-allocate file to full size so concurrent WriteAt-style writes work
+        {
+            let file = tokio::fs::OpenOptions::new()
+                .write(true)
+                .create(true)
+                .truncate(true)
+                .open(dest_path)
+                .await
+                .map_err(|e| format!("failed to open dest file {}: {}", dest_path, e))?;
+            file.set_len(file_size)
+                .await
+                .map_err(|e| format!("failed to set file length: {}", e))?;
+        }
+
+        let part_size: u64 = 64 * 1024 * 1024;
+
+        // Build list of (offset, size) for all parts
+        let mut parts_plan: Vec<(u64, u64)> = Vec::new();
+        let mut offset: u64 = 0;
+        while offset < file_size {
+            let remaining = file_size - offset;
+            let this_part_size = std::cmp::min(part_size, remaining);
+            parts_plan.push((offset, this_part_size));
+            offset += this_part_size;
+        }
+
+        // Download parts concurrently with a semaphore limiting to CONCURRENCY
+        let semaphore = Arc::new(Semaphore::new(CONCURRENCY));
+        let client = &self.client;
+        let bucket = &self.bucket;
+        let dest_path_owned = dest_path.to_string();
+        let key_owned = key.to_string();
+        let progress = Arc::new(std::sync::Mutex::new((0u64, progress_fn)));
+
+        let mut handles = Vec::with_capacity(parts_plan.len());
+        for (off, size) in parts_plan {
+            let sem = semaphore.clone();
+            let client = client.clone();
+            let bucket = bucket.clone();
+            let key = key_owned.clone();
+            let dp = dest_path_owned.clone();
+            let progress = progress.clone();
+
+            handles.push(tokio::spawn(async move {
+                let _permit = sem
+                    .acquire()
+                    .await
+                    .map_err(|e| format!("semaphore error: {}", e))?;
+
+                let end = off + size - 1;
+                let range = format!("bytes={}-{}", off, end);
+
+                let get_resp = client
+                    .get_object()
+                    .bucket(&bucket)
+                    .key(&key)
+                    .range(&range)
+                    .send()
+                    .await
+                    .map_err(|e| format!("failed to get object {} range {}: {}", key, range, e))?;
+
+                let body = get_resp
+                    .body
+                    .collect()
+                    .await
+                    .map_err(|e| format!("failed to read body: {}", e))?;
+                let bytes = body.into_bytes();
+
+                // Write at the correct offset (like Go's WriteAt)
+                let mut file = tokio::fs::OpenOptions::new()
+                    .write(true)
+                    .open(&dp)
+                    .await
+                    .map_err(|e| format!("failed to open dest file {}: {}", dp, e))?;
+                file.seek(std::io::SeekFrom::Start(off))
+                    .await
+                    .map_err(|e| format!("failed to seek to offset {}: {}", off, e))?;
+                file.write_all(&bytes)
+                    .await
+                    .map_err(|e| format!("failed to write to {}: {}", dp, e))?;
+
+                // Report progress. The lock is released before the result is
+                // propagated so an aborting callback cannot poison the mutex
+                // for the other parts still in flight.
+                let progress_result = {
+                    let mut guard = progress.lock().unwrap();
+                    guard.0 += bytes.len() as u64;
+                    let downloaded = guard.0;
+                    let pct = if file_size > 0 {
+                        (downloaded as f32 * 100.0) / file_size as f32
+                    } else {
+                        100.0
+                    };
+                    (guard.1)(downloaded as i64, pct)
+                };
+                progress_result?;
+
+                Ok::<_, String>(())
+            }));
+        }
+
+        // Wait for all download tasks
+        for handle in handles {
+            handle
+                .await
+                .map_err(|e| format!("download task panicked: {}", e))??;
+        }
+
+        // fsync the file so its content is durable before the caller trims the .vif
+        // and deletes the remote object (matches Go's DownloadFile f.Sync()).
+        let synced = tokio::fs::OpenOptions::new()
+            .write(true)
+            .open(dest_path)
+            .await
+            .map_err(|e| format!("failed to open {} for fsync: {}", dest_path, e))?;
+        synced
+            .sync_all()
+            .await
+            .map_err(|e| format!("failed to fsync {}: {}", dest_path, e))?;
+
+        Ok(file_size)
+    }
+
+    pub async fn read_range(&self, key: &str, offset: u64, size: usize) -> Result<Vec<u8>, String> {
+        let end = offset + (size as u64).saturating_sub(1);
+        let range = format!("bytes={}-{}", offset, end);
+        let resp = self
+            .client
+            .get_object()
+            .bucket(&self.bucket)
+            .key(key)
+            .range(&range)
+            .send()
+            .await
+            .map_err(|e| format!("failed to get object {} range {}: {}", key, range, e))?;
+
+        let body = resp
+            .body
+            .collect()
+            .await
+            .map_err(|e| format!("failed to read object {} body: {}", key, e))?;
+        Ok(body.into_bytes().to_vec())
+    }
+
+    /// Delete a file from S3.
+    pub async fn delete_file(&self, key: &str) -> Result<(), String> {
+        self.client
+            .delete_object()
+            .bucket(&self.bucket)
+            .key(key)
+            .send()
+            .await
+            .map_err(|e| format!("failed to delete object {}: {}", key, e))?;
+        Ok(())
+    }
+
+    pub fn delete_file_blocking(&self, key: &str) -> Result<(), String> {
+        let client = self.client.clone();
+        let bucket = self.bucket.clone();
+        let key = key.to_string();
+        block_on_tier_future(async move {
+            client
+                .delete_object()
+                .bucket(&bucket)
+                .key(&key)
+                .send()
+                .await
+                .map_err(|e| format!("failed to delete object {}: {}", key, e))?;
+            Ok(())
+        })
+    }
+
+    pub fn read_range_blocking(
+        &self,
+        key: &str,
+        offset: u64,
+        size: usize,
+    ) -> Result<Vec<u8>, String> {
+        let client = self.client.clone();
+        let bucket = self.bucket.clone();
+        let key = key.to_string();
+        block_on_tier_future(async move {
+            let end = offset + (size as u64).saturating_sub(1);
+            let range = format!("bytes={}-{}", offset, end);
+            let resp = client
+                .get_object()
+                .bucket(&bucket)
+                .key(&key)
+                .range(&range)
+                .send()
+                .await
+                .map_err(|e| format!("failed to get object {} range {}: {}", key, range, e))?;
+
+            let body = resp
+                .body
+                .collect()
+                .await
+                .map_err(|e| format!("failed to read object {} body: {}", key, e))?;
+            Ok(body.into_bytes().to_vec())
+        })
+    }
+}
+
+/// Parse a backend name like "s3" or "s3.default" into (backend_type, backend_id).
+/// Matches Go's `BackendNameToTypeId`.
+pub fn backend_name_to_type_id(backend_name: &str) -> (String, String) {
+    let parts: Vec<&str> = backend_name.split('.').collect();
+    match parts.len() {
+        1 => (backend_name.to_string(), "default".to_string()),
+        2 => (parts[0].to_string(), parts[1].to_string()),
+        _ => (String::new(), String::new()),
+    }
+}
+
+/// A registry of configured S3 tier backends, keyed by backend name (e.g., "s3.default").
+#[derive(Default)]
+pub struct S3TierRegistry {
+    backends: HashMap<String, Arc<S3TierBackend>>,
+}
+
+impl S3TierRegistry {
+    pub fn new() -> Self {
+        Self {
+            backends: HashMap::new(),
+        }
+    }
+
+    /// Register a backend with the given name.
+    pub fn register(&mut self, name: String, backend: S3TierBackend) {
+        self.backends.insert(name, Arc::new(backend));
+    }
+
+    /// Look up a backend by name.
+    pub fn get(&self, name: &str) -> Option<Arc<S3TierBackend>> {
+        self.backends.get(name).cloned()
+    }
+
+    /// List all registered backend names.
+    pub fn names(&self) -> Vec<String> {
+        self.backends.keys().cloned().collect()
+    }
+
+    /// Remove a backend by name.
+    pub fn remove(&mut self, name: &str) {
+        self.backends.remove(name);
+    }
+
+    pub fn clear(&mut self) {
+        self.backends.clear();
+    }
+}
+
+static GLOBAL_S3_TIER_REGISTRY: OnceLock<RwLock<S3TierRegistry>> = OnceLock::new();
+
+pub fn global_s3_tier_registry() -> &'static RwLock<S3TierRegistry> {
+    GLOBAL_S3_TIER_REGISTRY.get_or_init(|| RwLock::new(S3TierRegistry::new()))
+}
+
+fn block_on_tier_future<F, T>(future: F) -> Result<T, String>
+where
+    F: Future<Output = Result<T, String>> + Send + 'static,
+    T: Send + 'static,
+{
+    std::thread::spawn(move || {
+        let runtime = tokio::runtime::Builder::new_current_thread()
+            .enable_all()
+            .build()
+            .map_err(|e| format!("failed to build tokio runtime: {}", e))?;
+        runtime.block_on(future)
+    })
+    .join()
+    .map_err(|_| "tier runtime thread panicked".to_string())?
+}

seaweed-volume/src/security.rs

@@ -0,0 +1,547 @@
+//! Security: JWT validation and IP whitelist checking.
+//!
+//! Matches Go's security/guard.go and security/jwt.go.
+//! - Guard: combines whitelist IP checking with JWT token validation
+//! - JWT: HS256 HMAC signing with file-id claims
+
+pub mod tls;
+
+use std::collections::HashSet;
+use std::net::IpAddr;
+use std::time::{SystemTime, UNIX_EPOCH};
+
+use jsonwebtoken::{Algorithm, DecodingKey, EncodingKey, Header, Validation, decode, encode};
+use serde::{Deserialize, Serialize};
+
+// ============================================================================
+// JWT Claims
+// ============================================================================
+
+/// Claims for volume server file access tokens.
+/// Matches Go's `SeaweedFileIdClaims`.
+#[derive(Debug, Serialize, Deserialize)]
+pub struct FileIdClaims {
+    /// File ID this token grants access to (e.g., "3,01637037d6").
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub fid: Option<String>,
+
+    /// Expiration time (Unix timestamp).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub exp: Option<u64>,
+
+    /// Not before (Unix timestamp).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub nbf: Option<u64>,
+}
+
+/// Signing key wrapper (empty = security disabled).
+#[derive(Clone)]
+pub struct SigningKey(pub Vec<u8>);
+
+impl SigningKey {
+    pub fn is_empty(&self) -> bool {
+        self.0.is_empty()
+    }
+
+    pub fn from_string(s: &str) -> Self {
+        SigningKey(s.as_bytes().to_vec())
+    }
+}
+
+/// Generate a JWT token for file access.
+pub fn gen_jwt(
+    signing_key: &SigningKey,
+    expires_after_sec: i64,
+    file_id: &str,
+) -> Result<String, JwtError> {
+    if signing_key.is_empty() {
+        return Err(JwtError::NoSigningKey);
+    }
+
+    let now = SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs();
+
+    let claims = FileIdClaims {
+        fid: Some(file_id.to_string()),
+        exp: if expires_after_sec > 0 {
+            Some(now + expires_after_sec as u64)
+        } else {
+            None
+        },
+        nbf: None,
+    };
+
+    let token = encode(
+        &Header::new(Algorithm::HS256),
+        &claims,
+        &EncodingKey::from_secret(&signing_key.0),
+    )?;
+
+    Ok(token)
+}
+
+/// Decode and validate a JWT token.
+pub fn decode_jwt(signing_key: &SigningKey, token: &str) -> Result<FileIdClaims, JwtError> {
+    if signing_key.is_empty() {
+        return Err(JwtError::NoSigningKey);
+    }
+
+    let mut validation = Validation::new(Algorithm::HS256);
+    // Match Go behavior: tokens without exp are accepted (Go's jwt-go does not require exp)
+    // But if exp IS present, it must be valid (not expired).
+    validation.required_spec_claims.clear();
+    validation.validate_exp = true;
+    // Go's jwt-go/v5 validates nbf when present
+    validation.validate_nbf = true;
+    validation.leeway = 0;
+
+    let data = decode::<FileIdClaims>(
+        token,
+        &DecodingKey::from_secret(&signing_key.0),
+        &validation,
+    )?;
+
+    Ok(data.claims)
+}
+
+// ============================================================================
+// Guard
+// ============================================================================
+
+/// Security guard: IP whitelist + JWT token validation.
+pub struct Guard {
+    whitelist_ips: HashSet<String>,
+    whitelist_cidrs: Vec<(IpAddr, u8)>, // (network, prefix_len)
+    pub signing_key: SigningKey,
+    pub expires_after_sec: i64,
+    pub read_signing_key: SigningKey,
+    pub read_expires_after_sec: i64,
+    /// Combined flag: true when whitelist is non-empty OR signing key is present.
+    /// Matches Go's `isWriteActive = !isEmptyWhiteList || len(SigningKey) != 0`.
+    is_write_active: bool,
+}
+
+impl Guard {
+    pub fn new(
+        whitelist: &[String],
+        signing_key: SigningKey,
+        expires_after_sec: i64,
+        read_signing_key: SigningKey,
+        read_expires_after_sec: i64,
+    ) -> Self {
+        let mut guard = Guard {
+            whitelist_ips: HashSet::new(),
+            whitelist_cidrs: Vec::new(),
+            signing_key,
+            expires_after_sec,
+            read_signing_key,
+            read_expires_after_sec,
+            is_write_active: false,
+        };
+        guard.update_whitelist(whitelist);
+        guard
+    }
+
+    /// Update the IP whitelist.
+    pub fn update_whitelist(&mut self, entries: &[String]) {
+        self.whitelist_ips.clear();
+        self.whitelist_cidrs.clear();
+
+        for entry in entries {
+            let entry = entry.trim();
+            if entry.is_empty() {
+                continue;
+            }
+            if entry.contains('/') {
+                // CIDR range
+                if let Some((ip, prefix)) = parse_cidr(entry) {
+                    self.whitelist_cidrs.push((ip, prefix));
+                } else {
+                    tracing::error!("Parse CIDR {} in whitelist failed", entry);
+                }
+            } else {
+                // Exact IP/hostname
+                self.whitelist_ips.insert(entry.to_string());
+            }
+        }
+
+        // Match Go: isWriteActive = !isEmptyWhiteList || len(SigningKey) != 0
+        let is_empty_whitelist = self.whitelist_ips.is_empty() && self.whitelist_cidrs.is_empty();
+        self.is_write_active = !is_empty_whitelist || !self.signing_key.is_empty();
+    }
+
+    /// Refresh the JWT signing keys and their expirations in place so operators
+    /// can rotate keys via SIGHUP without restarting the process. Mirrors Go's
+    /// `Guard::UpdateSigningKeys`: it swaps the fields and recomputes
+    /// `is_write_active`.
+    pub fn update_signing_keys(
+        &mut self,
+        signing_key: SigningKey,
+        expires_after_sec: i64,
+        read_signing_key: SigningKey,
+        read_expires_after_sec: i64,
+    ) {
+        self.signing_key = signing_key;
+        self.expires_after_sec = expires_after_sec;
+        self.read_signing_key = read_signing_key;
+        self.read_expires_after_sec = read_expires_after_sec;
+
+        let is_empty_whitelist = self.whitelist_ips.is_empty() && self.whitelist_cidrs.is_empty();
+        self.is_write_active = !is_empty_whitelist || !self.signing_key.is_empty();
+    }
+
+    /// Whether any write-side control is configured: a non-empty whitelist or a
+    /// signing key. Mirrors Go's non-nil `guard` -- when this is false, every
+    /// caller is allowed and admin RPCs need not carry peer info.
+    pub fn is_write_active(&self) -> bool {
+        self.is_write_active
+    }
+
+    /// Check if a remote IP is in the whitelist.
+    /// Returns true if write security is inactive (no whitelist and no signing key),
+    /// if the whitelist is empty, or if the IP matches.
+    pub fn check_whitelist(&self, remote_addr: &str) -> bool {
+        if !self.is_write_active {
+            return true;
+        }
+        if self.whitelist_ips.is_empty() && self.whitelist_cidrs.is_empty() {
+            return true;
+        }
+
+        let host = extract_host(remote_addr);
+
+        // Check exact match
+        if self.whitelist_ips.contains(&host) {
+            return true;
+        }
+
+        // Check CIDR ranges
+        if let Ok(ip) = host.parse::<IpAddr>() {
+            for &(ref network, prefix_len) in &self.whitelist_cidrs {
+                if ip_in_cidr(&ip, network, prefix_len) {
+                    return true;
+                }
+            }
+        }
+
+        false
+    }
+
+    /// Check if a read signing key is configured.
+    pub fn has_read_signing_key(&self) -> bool {
+        !self.read_signing_key.is_empty()
+    }
+
+    /// Validate a request's JWT token.
+    /// `is_write` determines which signing key to use.
+    /// Returns Ok(()) if valid, or if security is disabled.
+    pub fn check_jwt(&self, token: Option<&str>, is_write: bool) -> Result<(), JwtError> {
+        let key = if is_write {
+            &self.signing_key
+        } else {
+            &self.read_signing_key
+        };
+
+        if key.is_empty() {
+            return Ok(()); // Security disabled for this operation type
+        }
+
+        let token = token.ok_or(JwtError::MissingToken)?;
+        decode_jwt(key, token)?;
+        Ok(())
+    }
+
+    /// Check JWT and validate the file ID claim matches.
+    pub fn check_jwt_for_file(
+        &self,
+        token: Option<&str>,
+        expected_fid: &str,
+        is_write: bool,
+    ) -> Result<(), JwtError> {
+        let key = if is_write {
+            &self.signing_key
+        } else {
+            &self.read_signing_key
+        };
+
+        if key.is_empty() {
+            return Ok(());
+        }
+
+        let token = token.ok_or(JwtError::MissingToken)?;
+        let claims = decode_jwt(key, token)?;
+
+        match claims.fid {
+            None => {
+                return Err(JwtError::MissingFileIdClaim);
+            }
+            Some(ref fid) if fid != expected_fid => {
+                return Err(JwtError::FileIdMismatch {
+                    expected: expected_fid.to_string(),
+                    got: fid.to_string(),
+                });
+            }
+            _ => {}
+        }
+
+        Ok(())
+    }
+}
+
+// ============================================================================
+// Helpers
+// ============================================================================
+
+/// Extract host from "host:port" or "[::1]:port" format.
+fn extract_host(addr: &str) -> String {
+    // Handle IPv6 with brackets
+    if addr.starts_with('[')
+        && let Some(end) = addr.find(']')
+    {
+        return addr[1..end].to_string();
+    }
+    // Handle host:port
+    if let Some(pos) = addr.rfind(':') {
+        return addr[..pos].to_string();
+    }
+    addr.to_string()
+}
+
+/// Parse CIDR notation "192.168.1.0/24" into (IpAddr, prefix_len).
+fn parse_cidr(cidr: &str) -> Option<(IpAddr, u8)> {
+    let parts: Vec<&str> = cidr.split('/').collect();
+    if parts.len() != 2 {
+        return None;
+    }
+    let ip: IpAddr = parts[0].parse().ok()?;
+    let prefix: u8 = parts[1].parse().ok()?;
+    Some((ip, prefix))
+}
+
+/// Check if an IP is within a CIDR range.
+fn ip_in_cidr(ip: &IpAddr, network: &IpAddr, prefix_len: u8) -> bool {
+    match (ip, network) {
+        (IpAddr::V4(ip), IpAddr::V4(net)) => {
+            let ip_bits = u32::from(*ip);
+            let net_bits = u32::from(*net);
+            let mask = if prefix_len == 0 {
+                0
+            } else if prefix_len >= 32 {
+                u32::MAX
+            } else {
+                u32::MAX << (32 - prefix_len)
+            };
+            (ip_bits & mask) == (net_bits & mask)
+        }
+        (IpAddr::V6(ip), IpAddr::V6(net)) => {
+            let ip_bits = u128::from(*ip);
+            let net_bits = u128::from(*net);
+            let mask = if prefix_len == 0 {
+                0
+            } else if prefix_len >= 128 {
+                u128::MAX
+            } else {
+                u128::MAX << (128 - prefix_len)
+            };
+            (ip_bits & mask) == (net_bits & mask)
+        }
+        _ => false, // V4/V6 mismatch
+    }
+}
+
+// ============================================================================
+// Errors
+// ============================================================================
+
+#[derive(Debug, thiserror::Error)]
+pub enum JwtError {
+    #[error("no signing key configured")]
+    NoSigningKey,
+
+    #[error("missing JWT token")]
+    MissingToken,
+
+    #[error("JWT error: {0}")]
+    Jwt(#[from] jsonwebtoken::errors::Error),
+
+    #[error("JWT token missing required fid claim")]
+    MissingFileIdClaim,
+
+    #[error("file ID mismatch: expected {expected}, got {got}")]
+    FileIdMismatch { expected: String, got: String },
+}
+
+// ============================================================================
+// Tests
+// ============================================================================
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_jwt_round_trip() {
+        let key = SigningKey::from_string("test-secret-key");
+        let token = gen_jwt(&key, 3600, "3,01637037d6").unwrap();
+        let claims = decode_jwt(&key, &token).unwrap();
+        assert_eq!(claims.fid, Some("3,01637037d6".to_string()));
+    }
+
+    #[test]
+    fn test_jwt_no_signing_key() {
+        let key = SigningKey(vec![]);
+        assert!(gen_jwt(&key, 3600, "1,abc").is_err());
+    }
+
+    #[test]
+    fn test_jwt_invalid_token() {
+        let key = SigningKey::from_string("secret");
+        let result = decode_jwt(&key, "invalid.token.here");
+        assert!(result.is_err());
+    }
+
+    #[test]
+    fn test_jwt_wrong_key() {
+        let key1 = SigningKey::from_string("secret1");
+        let key2 = SigningKey::from_string("secret2");
+        let token = gen_jwt(&key1, 3600, "1,abc").unwrap();
+        assert!(decode_jwt(&key2, &token).is_err());
+    }
+
+    #[test]
+    fn test_guard_empty_whitelist() {
+        let guard = Guard::new(&[], SigningKey(vec![]), 0, SigningKey(vec![]), 0);
+        assert!(guard.check_whitelist("192.168.1.1:8080"));
+    }
+
+    #[test]
+    fn test_guard_whitelist_exact() {
+        let guard = Guard::new(
+            &["192.168.1.1".to_string(), "10.0.0.1".to_string()],
+            SigningKey(vec![]),
+            0,
+            SigningKey(vec![]),
+            0,
+        );
+        assert!(guard.check_whitelist("192.168.1.1:8080"));
+        assert!(guard.check_whitelist("10.0.0.1:1234"));
+        assert!(!guard.check_whitelist("172.16.0.1:8080"));
+    }
+
+    #[test]
+    fn test_guard_whitelist_cidr() {
+        let guard = Guard::new(
+            &["10.0.0.0/8".to_string()],
+            SigningKey(vec![]),
+            0,
+            SigningKey(vec![]),
+            0,
+        );
+        assert!(guard.check_whitelist("10.1.2.3:8080"));
+        assert!(guard.check_whitelist("10.255.255.255:80"));
+        assert!(!guard.check_whitelist("11.0.0.1:80"));
+    }
+
+    #[test]
+    fn test_guard_check_jwt_disabled() {
+        let guard = Guard::new(&[], SigningKey(vec![]), 0, SigningKey(vec![]), 0);
+        // No signing key = security disabled
+        assert!(guard.check_jwt(None, true).is_ok());
+        assert!(guard.check_jwt(None, false).is_ok());
+    }
+
+    #[test]
+    fn test_guard_check_jwt_enabled() {
+        let key = SigningKey::from_string("write-secret");
+        let read_key = SigningKey::from_string("read-secret");
+        let guard = Guard::new(&[], key.clone(), 3600, read_key.clone(), 3600);
+
+        // Missing token
+        assert!(guard.check_jwt(None, true).is_err());
+
+        // Valid write token
+        let token = gen_jwt(&key, 3600, "1,abc").unwrap();
+        assert!(guard.check_jwt(Some(&token), true).is_ok());
+
+        // Write token for read should fail (different key)
+        assert!(guard.check_jwt(Some(&token), false).is_err());
+
+        // Valid read token
+        let read_token = gen_jwt(&read_key, 3600, "1,abc").unwrap();
+        assert!(guard.check_jwt(Some(&read_token), false).is_ok());
+    }
+
+    #[test]
+    fn test_guard_check_jwt_file_id() {
+        let key = SigningKey::from_string("secret");
+        let guard = Guard::new(&[], key.clone(), 3600, SigningKey(vec![]), 0);
+
+        let token = gen_jwt(&key, 3600, "3,01637037d6").unwrap();
+
+        // Correct file ID
+        assert!(
+            guard
+                .check_jwt_for_file(Some(&token), "3,01637037d6", true)
+                .is_ok()
+        );
+
+        // Wrong file ID
+        let err = guard.check_jwt_for_file(Some(&token), "4,deadbeef", true);
+        assert!(matches!(err, Err(JwtError::FileIdMismatch { .. })));
+    }
+
+    #[test]
+    fn test_update_signing_keys_rotates() {
+        let mut guard = Guard::new(
+            &[],
+            SigningKey::from_string("old-write"),
+            10,
+            SigningKey::from_string("old-read"),
+            60,
+        );
+
+        guard.update_signing_keys(
+            SigningKey::from_string("new-write"),
+            11,
+            SigningKey::from_string("new-read"),
+            61,
+        );
+
+        let token = gen_jwt(&guard.signing_key, 60, "3,01637037d6").unwrap();
+        // Validates with the rotated key, not the old one.
+        assert!(decode_jwt(&guard.signing_key, &token).is_ok());
+        assert!(decode_jwt(&SigningKey::from_string("old-write"), &token).is_err());
+        assert_eq!(guard.expires_after_sec, 11);
+        assert_eq!(guard.read_expires_after_sec, 61);
+    }
+
+    #[test]
+    fn test_update_signing_keys_toggles_write_active() {
+        let mut guard = Guard::new(&[], SigningKey(vec![]), 0, SigningKey(vec![]), 0);
+        assert!(guard.check_jwt(None, true).is_ok());
+
+        guard.update_signing_keys(SigningKey::from_string("write"), 10, SigningKey(vec![]), 0);
+        assert!(guard.is_write_active);
+
+        guard.update_signing_keys(SigningKey(vec![]), 0, SigningKey(vec![]), 0);
+        assert!(!guard.is_write_active);
+    }
+
+    #[test]
+    fn test_extract_host() {
+        assert_eq!(extract_host("192.168.1.1:8080"), "192.168.1.1");
+        assert_eq!(extract_host("[::1]:8080"), "::1");
+        assert_eq!(extract_host("localhost"), "localhost");
+    }
+
+    #[test]
+    fn test_ip_in_cidr() {
+        let net: IpAddr = "10.0.0.0".parse().unwrap();
+        let ip1: IpAddr = "10.1.2.3".parse().unwrap();
+        let ip2: IpAddr = "11.0.0.1".parse().unwrap();
+        assert!(ip_in_cidr(&ip1, &net, 8));
+        assert!(!ip_in_cidr(&ip2, &net, 8));
+    }
+}

seaweed-volume/src/security/tls.rs

@@ -0,0 +1,444 @@
+use std::collections::HashSet;
+use std::fmt;
+use std::sync::Arc;
+
+use rustls::client::danger::HandshakeSignatureValid;
+use rustls::crypto::CryptoProvider;
+use rustls::crypto::aws_lc_rs;
+use rustls::pki_types::UnixTime;
+use rustls::pki_types::{CertificateDer, PrivateKeyDer};
+use rustls::server::WebPkiClientVerifier;
+use rustls::server::danger::{ClientCertVerified, ClientCertVerifier};
+use rustls::{
+    CipherSuite, DigitallySignedStruct, DistinguishedName, RootCertStore, ServerConfig,
+    SignatureScheme, SupportedCipherSuite, SupportedProtocolVersion,
+};
+use x509_parser::prelude::{FromDer, X509Certificate};
+
+#[derive(Clone, Debug, Default, PartialEq, Eq)]
+pub struct TlsPolicy {
+    pub min_version: String,
+    pub max_version: String,
+    pub cipher_suites: String,
+}
+
+#[derive(Clone, Debug, Default, PartialEq, Eq)]
+pub struct GrpcClientAuthPolicy {
+    pub allowed_common_names: Vec<String>,
+    pub allowed_wildcard_domain: String,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct TlsPolicyError(String);
+
+impl fmt::Display for TlsPolicyError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.write_str(&self.0)
+    }
+}
+
+impl std::error::Error for TlsPolicyError {}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
+enum GoTlsVersion {
+    Ssl3,
+    Tls10,
+    Tls11,
+    Tls12,
+    Tls13,
+}
+
+#[derive(Debug)]
+struct CommonNameVerifier {
+    inner: Arc<dyn ClientCertVerifier>,
+    allowed_common_names: HashSet<String>,
+    allowed_wildcard_domain: String,
+}
+
+impl ClientCertVerifier for CommonNameVerifier {
+    fn offer_client_auth(&self) -> bool {
+        self.inner.offer_client_auth()
+    }
+
+    fn client_auth_mandatory(&self) -> bool {
+        self.inner.client_auth_mandatory()
+    }
+
+    fn root_hint_subjects(&self) -> &[DistinguishedName] {
+        self.inner.root_hint_subjects()
+    }
+
+    fn verify_client_cert(
+        &self,
+        end_entity: &CertificateDer<'_>,
+        intermediates: &[CertificateDer<'_>],
+        now: UnixTime,
+    ) -> Result<ClientCertVerified, rustls::Error> {
+        self.inner
+            .verify_client_cert(end_entity, intermediates, now)?;
+        let common_name = parse_common_name(end_entity).map_err(|e| {
+            rustls::Error::General(format!(
+                "parse client certificate common name failed: {}",
+                e
+            ))
+        })?;
+        if common_name_is_allowed(
+            &common_name,
+            &self.allowed_common_names,
+            &self.allowed_wildcard_domain,
+        ) {
+            return Ok(ClientCertVerified::assertion());
+        }
+        Err(rustls::Error::General(format!(
+            "Authenticate: invalid subject client common name: {}",
+            common_name
+        )))
+    }
+
+    fn verify_tls12_signature(
+        &self,
+        message: &[u8],
+        cert: &CertificateDer<'_>,
+        dss: &DigitallySignedStruct,
+    ) -> Result<HandshakeSignatureValid, rustls::Error> {
+        self.inner.verify_tls12_signature(message, cert, dss)
+    }
+
+    fn verify_tls13_signature(
+        &self,
+        message: &[u8],
+        cert: &CertificateDer<'_>,
+        dss: &DigitallySignedStruct,
+    ) -> Result<HandshakeSignatureValid, rustls::Error> {
+        self.inner.verify_tls13_signature(message, cert, dss)
+    }
+
+    fn supported_verify_schemes(&self) -> Vec<SignatureScheme> {
+        self.inner.supported_verify_schemes()
+    }
+}
+
+// aws-lc-rs and ring both get linked transitively, so rustls can't auto-select
+// a provider and tonic's client TLS panics on first use. Pin the default to
+// aws-lc-rs, matching the server config. Idempotent.
+pub fn install_default_crypto_provider() {
+    let _ = aws_lc_rs::default_provider().install_default();
+}
+
+pub fn build_rustls_server_config(
+    cert_path: &str,
+    key_path: &str,
+    ca_path: &str,
+    policy: &TlsPolicy,
+) -> Result<ServerConfig, TlsPolicyError> {
+    build_rustls_server_config_with_client_auth(cert_path, key_path, ca_path, policy, None)
+}
+
+pub fn build_rustls_server_config_with_grpc_client_auth(
+    cert_path: &str,
+    key_path: &str,
+    ca_path: &str,
+    policy: &TlsPolicy,
+    client_auth_policy: &GrpcClientAuthPolicy,
+) -> Result<ServerConfig, TlsPolicyError> {
+    build_rustls_server_config_with_client_auth(
+        cert_path,
+        key_path,
+        ca_path,
+        policy,
+        Some(client_auth_policy),
+    )
+}
+
+fn build_rustls_server_config_with_client_auth(
+    cert_path: &str,
+    key_path: &str,
+    ca_path: &str,
+    policy: &TlsPolicy,
+    client_auth_policy: Option<&GrpcClientAuthPolicy>,
+) -> Result<ServerConfig, TlsPolicyError> {
+    let cert_chain = read_cert_chain(cert_path)?;
+    let private_key = read_private_key(key_path)?;
+    let provider = build_crypto_provider(policy)?;
+    let versions = build_supported_versions(policy)?;
+
+    let builder = ServerConfig::builder_with_provider(provider.clone())
+        .with_protocol_versions(&versions)
+        .map_err(|e| TlsPolicyError(format!("invalid TLS version policy: {}", e)))?;
+
+    let builder = if ca_path.is_empty() {
+        builder.with_no_client_auth()
+    } else {
+        let roots = read_root_store(ca_path)?;
+        let verifier =
+            WebPkiClientVerifier::builder_with_provider(Arc::new(roots), provider.clone())
+                .build()
+                .map_err(|e| TlsPolicyError(format!("build client verifier failed: {}", e)))?;
+        let verifier: Arc<dyn ClientCertVerifier> = if let Some(client_auth_policy) =
+            client_auth_policy.filter(|policy| {
+                !policy.allowed_common_names.is_empty()
+                    || !policy.allowed_wildcard_domain.is_empty()
+            }) {
+            Arc::new(CommonNameVerifier {
+                inner: verifier,
+                allowed_common_names: client_auth_policy
+                    .allowed_common_names
+                    .iter()
+                    .cloned()
+                    .collect(),
+                allowed_wildcard_domain: client_auth_policy.allowed_wildcard_domain.clone(),
+            })
+        } else {
+            verifier
+        };
+        builder.with_client_cert_verifier(verifier)
+    };
+
+    builder
+        .with_single_cert(cert_chain, private_key)
+        .map_err(|e| TlsPolicyError(format!("build rustls server config failed: {}", e)))
+}
+
+fn read_cert_chain(cert_path: &str) -> Result<Vec<CertificateDer<'static>>, TlsPolicyError> {
+    let cert_pem = std::fs::read(cert_path).map_err(|e| {
+        TlsPolicyError(format!(
+            "Failed to read TLS cert file '{}': {}",
+            cert_path, e
+        ))
+    })?;
+    rustls_pemfile::certs(&mut &cert_pem[..])
+        .collect::<Result<Vec<_>, _>>()
+        .map_err(|e| {
+            TlsPolicyError(format!(
+                "Failed to parse TLS cert PEM '{}': {}",
+                cert_path, e
+            ))
+        })
+}
+
+fn read_private_key(key_path: &str) -> Result<PrivateKeyDer<'static>, TlsPolicyError> {
+    let key_pem = std::fs::read(key_path).map_err(|e| {
+        TlsPolicyError(format!("Failed to read TLS key file '{}': {}", key_path, e))
+    })?;
+    rustls_pemfile::private_key(&mut &key_pem[..])
+        .map_err(|e| TlsPolicyError(format!("Failed to parse TLS key PEM '{}': {}", key_path, e)))?
+        .ok_or_else(|| TlsPolicyError(format!("No private key found in '{}'", key_path)))
+}
+
+fn read_root_store(ca_path: &str) -> Result<RootCertStore, TlsPolicyError> {
+    let ca_pem = std::fs::read(ca_path)
+        .map_err(|e| TlsPolicyError(format!("Failed to read TLS CA file '{}': {}", ca_path, e)))?;
+    let ca_certs = rustls_pemfile::certs(&mut &ca_pem[..])
+        .collect::<Result<Vec<_>, _>>()
+        .map_err(|e| TlsPolicyError(format!("Failed to parse TLS CA PEM '{}': {}", ca_path, e)))?;
+    let mut roots = RootCertStore::empty();
+    for cert in ca_certs {
+        roots
+            .add(cert)
+            .map_err(|e| TlsPolicyError(format!("Failed to add CA cert '{}': {}", ca_path, e)))?;
+    }
+    Ok(roots)
+}
+
+fn build_crypto_provider(policy: &TlsPolicy) -> Result<Arc<CryptoProvider>, TlsPolicyError> {
+    let mut provider = aws_lc_rs::default_provider();
+    let cipher_suites = parse_cipher_suites(&provider.cipher_suites, &policy.cipher_suites)?;
+    if !cipher_suites.is_empty() {
+        provider.cipher_suites = cipher_suites;
+    }
+    Ok(Arc::new(provider))
+}
+
+pub fn build_supported_versions(
+    policy: &TlsPolicy,
+) -> Result<Vec<&'static SupportedProtocolVersion>, TlsPolicyError> {
+    let min_version = parse_go_tls_version(&policy.min_version)?;
+    let max_version = parse_go_tls_version(&policy.max_version)?;
+    let versions = [&rustls::version::TLS13, &rustls::version::TLS12]
+        .into_iter()
+        .filter(|version| {
+            let current = go_tls_version_for_supported(version);
+            min_version.map(|min| current >= min).unwrap_or(true)
+                && max_version.map(|max| current <= max).unwrap_or(true)
+        })
+        .collect::<Vec<_>>();
+
+    if versions.is_empty() {
+        return Err(TlsPolicyError(format!(
+            "TLS version range min='{}' max='{}' is unsupported by rustls",
+            policy.min_version, policy.max_version
+        )));
+    }
+
+    Ok(versions)
+}
+
+fn parse_go_tls_version(value: &str) -> Result<Option<GoTlsVersion>, TlsPolicyError> {
+    match value.trim() {
+        "" => Ok(None),
+        "SSLv3" => Ok(Some(GoTlsVersion::Ssl3)),
+        "TLS 1.0" => Ok(Some(GoTlsVersion::Tls10)),
+        "TLS 1.1" => Ok(Some(GoTlsVersion::Tls11)),
+        "TLS 1.2" => Ok(Some(GoTlsVersion::Tls12)),
+        "TLS 1.3" => Ok(Some(GoTlsVersion::Tls13)),
+        other => Err(TlsPolicyError(format!("invalid TLS version {}", other))),
+    }
+}
+
+fn parse_cipher_suites(
+    available: &[SupportedCipherSuite],
+    value: &str,
+) -> Result<Vec<SupportedCipherSuite>, TlsPolicyError> {
+    let trimmed = value.trim();
+    if trimmed.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    trimmed
+        .split(',')
+        .map(|name| {
+            let suite = parse_cipher_suite_name(name.trim())?;
+            available
+                .iter()
+                .copied()
+                .find(|candidate| candidate.suite() == suite)
+                .ok_or_else(|| {
+                    TlsPolicyError(format!(
+                        "TLS cipher suite '{}' is unsupported by the Rust implementation",
+                        name.trim()
+                    ))
+                })
+        })
+        .collect()
+}
+
+fn parse_cipher_suite_name(value: &str) -> Result<CipherSuite, TlsPolicyError> {
+    match value {
+        "TLS_AES_128_GCM_SHA256" => Ok(CipherSuite::TLS13_AES_128_GCM_SHA256),
+        "TLS_AES_256_GCM_SHA384" => Ok(CipherSuite::TLS13_AES_256_GCM_SHA384),
+        "TLS_CHACHA20_POLY1305_SHA256" => Ok(CipherSuite::TLS13_CHACHA20_POLY1305_SHA256),
+        "TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256" => {
+            Ok(CipherSuite::TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256)
+        }
+        "TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384" => {
+            Ok(CipherSuite::TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384)
+        }
+        "TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256" => {
+            Ok(CipherSuite::TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256)
+        }
+        "TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256" => {
+            Ok(CipherSuite::TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256)
+        }
+        "TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384" => {
+            Ok(CipherSuite::TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384)
+        }
+        "TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256" => {
+            Ok(CipherSuite::TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256)
+        }
+        other => Err(TlsPolicyError(format!(
+            "TLS cipher suite '{}' is unsupported by the Rust implementation",
+            other
+        ))),
+    }
+}
+
+fn parse_common_name(cert: &CertificateDer<'_>) -> Result<String, TlsPolicyError> {
+    let (_, certificate) = X509Certificate::from_der(cert.as_ref())
+        .map_err(|e| TlsPolicyError(format!("parse X.509 certificate failed: {}", e)))?;
+    let common_name = certificate
+        .subject()
+        .iter_common_name()
+        .next()
+        .and_then(|common_name| common_name.as_str().ok())
+        .map(str::to_string);
+    match common_name {
+        Some(common_name) => Ok(common_name),
+        None => Ok(String::new()),
+    }
+}
+
+fn common_name_is_allowed(
+    common_name: &str,
+    allowed_common_names: &HashSet<String>,
+    allowed_wildcard_domain: &str,
+) -> bool {
+    (!allowed_wildcard_domain.is_empty() && common_name.ends_with(allowed_wildcard_domain))
+        || allowed_common_names.contains(common_name)
+}
+
+fn go_tls_version_for_supported(version: &SupportedProtocolVersion) -> GoTlsVersion {
+    match version.version {
+        rustls::ProtocolVersion::TLSv1_2 => GoTlsVersion::Tls12,
+        rustls::ProtocolVersion::TLSv1_3 => GoTlsVersion::Tls13,
+        _ => unreachable!("rustls only exposes TLS 1.2 and 1.3"),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{TlsPolicy, build_supported_versions, common_name_is_allowed, parse_cipher_suites};
+    use rustls::crypto::aws_lc_rs;
+    use std::collections::HashSet;
+
+    #[test]
+    fn test_build_supported_versions_defaults_to_tls12_and_tls13() {
+        let versions = build_supported_versions(&TlsPolicy::default()).unwrap();
+        assert_eq!(
+            versions,
+            vec![&rustls::version::TLS13, &rustls::version::TLS12]
+        );
+    }
+
+    #[test]
+    fn test_build_supported_versions_filters_to_tls13() {
+        let versions = build_supported_versions(&TlsPolicy {
+            min_version: "TLS 1.3".to_string(),
+            max_version: "TLS 1.3".to_string(),
+            cipher_suites: String::new(),
+        })
+        .unwrap();
+        assert_eq!(versions, vec![&rustls::version::TLS13]);
+    }
+
+    #[test]
+    fn test_build_supported_versions_rejects_unsupported_legacy_range() {
+        let err = build_supported_versions(&TlsPolicy {
+            min_version: "TLS 1.0".to_string(),
+            max_version: "TLS 1.1".to_string(),
+            cipher_suites: String::new(),
+        })
+        .unwrap_err();
+        assert!(err.to_string().contains("unsupported by rustls"));
+    }
+
+    #[test]
+    fn test_parse_cipher_suites_accepts_go_names() {
+        let cipher_suites = parse_cipher_suites(
+            &aws_lc_rs::default_provider().cipher_suites,
+            "TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256,TLS_AES_128_GCM_SHA256",
+        )
+        .unwrap();
+        assert_eq!(cipher_suites.len(), 2);
+    }
+
+    #[test]
+    fn test_common_name_is_allowed_matches_exact_and_wildcard() {
+        let allowed_common_names =
+            HashSet::from([String::from("volume-a.internal"), String::from("worker-7")]);
+        assert!(common_name_is_allowed(
+            "volume-a.internal",
+            &allowed_common_names,
+            "",
+        ));
+        assert!(common_name_is_allowed(
+            "node.prod.example.com",
+            &allowed_common_names,
+            ".example.com",
+        ));
+        assert!(!common_name_is_allowed(
+            "node.prod.other.net",
+            &allowed_common_names,
+            ".example.com",
+        ));
+    }
+}

seaweed-volume/src/server/debug.rs

@@ -0,0 +1,159 @@
+use axum::Router;
+use axum::body::Body;
+use axum::extract::Query;
+use axum::http::{StatusCode, header};
+use axum::response::{IntoResponse, Response};
+use axum::routing::{any, get};
+use pprof::protos::Message;
+use serde::Deserialize;
+
+#[derive(Deserialize, Default)]
+struct ProfileQuery {
+    seconds: Option<u64>,
+}
+
+pub fn build_debug_router() -> Router {
+    Router::new()
+        .route("/debug/pprof/", get(pprof_index_handler))
+        .route("/debug/pprof/cmdline", get(pprof_cmdline_handler))
+        .route("/debug/pprof/profile", get(pprof_profile_handler))
+        .route("/debug/pprof/symbol", any(pprof_symbol_handler))
+        .route("/debug/pprof/trace", get(pprof_trace_handler))
+}
+
+async fn pprof_index_handler() -> Response {
+    let body = concat!(
+        "<html><head><title>/debug/pprof/</title></head><body>",
+        "<a href=\"cmdline\">cmdline</a><br>",
+        "<a href=\"profile\">profile</a><br>",
+        "<a href=\"symbol\">symbol</a><br>",
+        "<a href=\"trace\">trace</a><br>",
+        "</body></html>",
+    );
+    (
+        StatusCode::OK,
+        [(header::CONTENT_TYPE, "text/html; charset=utf-8")],
+        body,
+    )
+        .into_response()
+}
+
+async fn pprof_cmdline_handler() -> Response {
+    let body = std::env::args().collect::<Vec<_>>().join("\0");
+    (
+        StatusCode::OK,
+        [(header::CONTENT_TYPE, "text/plain; charset=utf-8")],
+        body,
+    )
+        .into_response()
+}
+
+async fn pprof_profile_handler(Query(query): Query<ProfileQuery>) -> Response {
+    let seconds = query.seconds.unwrap_or(30).clamp(1, 300);
+    let guard = match pprof::ProfilerGuard::new(100) {
+        Ok(guard) => guard,
+        Err(e) => {
+            return (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                format!("failed to start profiler: {}", e),
+            )
+                .into_response();
+        }
+    };
+
+    tokio::time::sleep(std::time::Duration::from_secs(seconds)).await;
+
+    let report = match guard.report().build() {
+        Ok(report) => report,
+        Err(e) => {
+            return (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                format!("failed to build profile report: {}", e),
+            )
+                .into_response();
+        }
+    };
+
+    let profile = match report.pprof() {
+        Ok(profile) => profile,
+        Err(e) => {
+            return (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                format!("failed to encode profile: {}", e),
+            )
+                .into_response();
+        }
+    };
+
+    let mut bytes = Vec::new();
+    if let Err(e) = profile.encode(&mut bytes) {
+        return (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            format!("failed to serialize profile: {}", e),
+        )
+            .into_response();
+    }
+
+    (
+        StatusCode::OK,
+        [(header::CONTENT_TYPE, "application/octet-stream")],
+        bytes,
+    )
+        .into_response()
+}
+
+async fn pprof_symbol_handler() -> Response {
+    (
+        StatusCode::OK,
+        [(header::CONTENT_TYPE, "text/plain; charset=utf-8")],
+        "num_symbols: 0\n",
+    )
+        .into_response()
+}
+
+async fn pprof_trace_handler(Query(query): Query<ProfileQuery>) -> Response {
+    let seconds = query.seconds.unwrap_or(1).clamp(1, 30);
+    tokio::time::sleep(std::time::Duration::from_secs(seconds)).await;
+    Response::builder()
+        .status(StatusCode::OK)
+        .header(header::CONTENT_TYPE, "application/octet-stream")
+        .body(Body::from(Vec::<u8>::new()))
+        .unwrap()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use axum::http::Request;
+    use tower::ServiceExt;
+
+    #[tokio::test]
+    async fn test_debug_index_route() {
+        let app = build_debug_router();
+        let response = app
+            .oneshot(
+                Request::builder()
+                    .uri("/debug/pprof/")
+                    .body(Body::empty())
+                    .unwrap(),
+            )
+            .await
+            .unwrap();
+        assert_eq!(response.status(), StatusCode::OK);
+    }
+
+    #[tokio::test]
+    async fn test_debug_cmdline_route() {
+        let app = build_debug_router();
+        let response = app
+            .oneshot(
+                Request::builder()
+                    .uri("/debug/pprof/cmdline")
+                    .body(Body::empty())
+                    .unwrap(),
+            )
+            .await
+            .unwrap();
+        assert_eq!(response.status(), StatusCode::OK);
+    }
+}

seaweed-volume/src/server/grpc_client.rs

@@ -0,0 +1,370 @@
+use std::error::Error;
+use std::fmt;
+use std::time::Duration;
+
+use hyper::http::Uri;
+use tonic::transport::{Certificate, Channel, ClientTlsConfig, Endpoint, Identity};
+
+use crate::config::VolumeServerConfig;
+
+pub const GRPC_MAX_MESSAGE_SIZE: usize = 1 << 30;
+const GRPC_KEEPALIVE_INTERVAL: Duration = Duration::from_secs(60);
+const GRPC_KEEPALIVE_TIMEOUT: Duration = Duration::from_secs(20);
+const GRPC_INITIAL_WINDOW_SIZE: u32 = 16 * 1024 * 1024;
+
+#[derive(Clone, Debug)]
+pub struct OutgoingGrpcTlsConfig {
+    cert_pem: String,
+    key_pem: String,
+    ca_pem: String,
+}
+
+#[derive(Debug)]
+pub struct GrpcClientError(String);
+
+impl fmt::Display for GrpcClientError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.write_str(&self.0)
+    }
+}
+
+impl Error for GrpcClientError {}
+
+pub fn load_outgoing_grpc_tls(
+    config: &VolumeServerConfig,
+) -> Result<Option<OutgoingGrpcTlsConfig>, GrpcClientError> {
+    // prefer a dedicated client certificate: CAs may issue certs with only one of the serverAuth/clientAuth EKUs
+    let (cert_file, key_file) = if !config.grpc_client_cert_file.is_empty()
+        && !config.grpc_client_key_file.is_empty()
+    {
+        (&config.grpc_client_cert_file, &config.grpc_client_key_file)
+    } else {
+        if !config.grpc_client_cert_file.is_empty() || !config.grpc_client_key_file.is_empty() {
+            tracing::warn!(
+                "grpc.volume.client_cert and grpc.volume.client_key must both be set, falling back to grpc.volume.cert and grpc.volume.key"
+            );
+        }
+        (&config.grpc_cert_file, &config.grpc_key_file)
+    };
+    if cert_file.is_empty() || key_file.is_empty() || config.grpc_ca_file.is_empty() {
+        return Ok(None);
+    }
+
+    let cert_pem = std::fs::read_to_string(cert_file).map_err(|e| {
+        GrpcClientError(format!(
+            "Failed to read outgoing gRPC cert '{}': {}",
+            cert_file, e
+        ))
+    })?;
+    let key_pem = std::fs::read_to_string(key_file).map_err(|e| {
+        GrpcClientError(format!(
+            "Failed to read outgoing gRPC key '{}': {}",
+            key_file, e
+        ))
+    })?;
+    let ca_pem = std::fs::read_to_string(&config.grpc_ca_file).map_err(|e| {
+        GrpcClientError(format!(
+            "Failed to read outgoing gRPC CA '{}': {}",
+            config.grpc_ca_file, e
+        ))
+    })?;
+
+    Ok(Some(OutgoingGrpcTlsConfig {
+        cert_pem,
+        key_pem,
+        ca_pem,
+    }))
+}
+
+pub fn grpc_endpoint_uri(grpc_host_port: &str, tls: Option<&OutgoingGrpcTlsConfig>) -> String {
+    let scheme = if tls.is_some() { "https" } else { "http" };
+    format!("{}://{}", scheme, grpc_host_port)
+}
+
+pub fn build_grpc_endpoint(
+    grpc_host_port: &str,
+    tls: Option<&OutgoingGrpcTlsConfig>,
+) -> Result<Endpoint, GrpcClientError> {
+    let uri = grpc_endpoint_uri(grpc_host_port, tls);
+    let mut endpoint = Channel::from_shared(uri.clone())
+        .map_err(|e| GrpcClientError(format!("invalid gRPC endpoint {}: {}", uri, e)))?
+        .http2_keep_alive_interval(GRPC_KEEPALIVE_INTERVAL)
+        .keep_alive_timeout(GRPC_KEEPALIVE_TIMEOUT)
+        .keep_alive_while_idle(false)
+        .initial_stream_window_size(Some(GRPC_INITIAL_WINDOW_SIZE))
+        .initial_connection_window_size(Some(GRPC_INITIAL_WINDOW_SIZE))
+        .http2_adaptive_window(false);
+
+    if let Some(tls) = tls {
+        let parsed = uri
+            .parse::<Uri>()
+            .map_err(|e| GrpcClientError(format!("invalid gRPC endpoint {}: {}", uri, e)))?;
+        let host = parsed
+            .host()
+            .ok_or_else(|| GrpcClientError(format!("missing host in gRPC endpoint {}", uri)))?;
+        let tls_config = ClientTlsConfig::new()
+            .identity(Identity::from_pem(
+                tls.cert_pem.clone(),
+                tls.key_pem.clone(),
+            ))
+            .ca_certificate(Certificate::from_pem(tls.ca_pem.clone()))
+            .domain_name(host.to_string());
+        endpoint = endpoint.tls_config(tls_config).map_err(|e| {
+            GrpcClientError(format!("configure gRPC TLS for {} failed: {}", uri, e))
+        })?;
+    }
+
+    Ok(endpoint)
+}
+
+/// Parse a SeaweedFS server address (`"ip:port.grpcPort"` or
+/// `"ip:port"`) into the `host:grpcPort` form `build_grpc_endpoint`
+/// expects. With the trailing `.grpcPort` segment, that segment IS
+/// the gRPC port; without it, the gRPC port is `port + 10000`
+/// (SeaweedFS's HTTP↔gRPC port-offset convention).
+///
+/// Shared between `grpc_server.rs` and the distributed-EC-read path
+/// in `store_ec.rs` — keep this as the single source of truth so the
+/// HTTP↔gRPC port translation can't drift between callers.
+pub fn parse_grpc_address(source: &str) -> Result<String, String> {
+    let colon_idx = source
+        .rfind(':')
+        .ok_or_else(|| format!("cannot parse address: {}", source))?;
+    let host = &source[..colon_idx];
+    let port_part = &source[colon_idx + 1..];
+
+    if let Some(dot_idx) = port_part.rfind('.') {
+        // Format: "ip:port.grpcPort". Validate BOTH ports as u16
+        // so a malformed HTTP port (e.g. `host:abc.18080`) is
+        // rejected here rather than tripping a downstream
+        // `build_grpc_endpoint` URI parse failure with a less
+        // useful error.
+        let http_port = &port_part[..dot_idx];
+        let grpc_port = &port_part[dot_idx + 1..];
+        http_port
+            .parse::<u16>()
+            .map_err(|e| format!("invalid http port {:?}: {}", http_port, e))?;
+        grpc_port
+            .parse::<u16>()
+            .map_err(|e| format!("invalid grpc port {:?}: {}", grpc_port, e))?;
+        return Ok(format!("{}:{}", host, grpc_port));
+    }
+
+    // Format: "ip:port" → grpc = port + 10000. Reject inputs whose
+    // implicit grpc port would overflow the TCP port range (e.g.
+    // `host:60000` produces 70000 — invalid). Without this check
+    // the cast silently wraps and the endpoint call later fails
+    // with an opaque connection error.
+    let port: u16 = port_part
+        .parse()
+        .map_err(|e| format!("invalid port {:?}: {}", port_part, e))?;
+    let grpc_port = port as u32 + 10000;
+    if grpc_port > u16::MAX as u32 {
+        return Err(format!(
+            "implicit grpc port out of range: {} + 10000 = {}",
+            port, grpc_port
+        ));
+    }
+    Ok(format!("{}:{}", host, grpc_port))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{build_grpc_endpoint, grpc_endpoint_uri, load_outgoing_grpc_tls};
+    use crate::config::{NeedleMapKind, ReadMode, VolumeServerConfig};
+    use crate::security::tls::TlsPolicy;
+
+    const TEST_CERT_PEM: &str = "-----BEGIN CERTIFICATE-----\nMIIBPDCB76ADAgECAhRuRPQgeAu43BT/M7EfAWSdapVdYDAFBgMrZXAwFDESMBAG\nA1UEAwwJbG9jYWxob3N0MB4XDTI2MDcwNTE2MTUyOVoXDTM2MDcwMjE2MTUyOVow\nFDESMBAGA1UEAwwJbG9jYWxob3N0MCowBQYDK2VwAyEAr/3bNIFI+8V32oCiY6y+\nXRFmZpdNQ2g//VtRkT+nQg+jUzBRMB0GA1UdDgQWBBTsy9tLf1zPiXCQfgci6zNi\ndEzRSjAfBgNVHSMEGDAWgBTsy9tLf1zPiXCQfgci6zNidEzRSjAPBgNVHRMBAf8E\nBTADAQH/MAUGAytlcANBAIvsdw0IbvOBBkb9cd7BfMJfIP9pQQrAL03pCRWJFnFh\nSysaLVgFXI4T078IiaM874oO+iB+5vNbWEpc7CkGow4=\n-----END CERTIFICATE-----\n";
+    const TEST_KEY_PEM: &str = "-----BEGIN PRIVATE KEY-----\nMC4CAQAwBQYDK2VwBCIEIHbyn71Kk+Y7KT3sBctit7uZpErpoH6qDbFj6P8qGaZH\n-----END PRIVATE KEY-----\n";
+
+    #[test]
+    fn test_build_grpc_endpoint_with_tls_resolves_crypto_provider() {
+        crate::security::tls::install_default_crypto_provider();
+        let tls = super::OutgoingGrpcTlsConfig {
+            cert_pem: TEST_CERT_PEM.to_string(),
+            key_pem: TEST_KEY_PEM.to_string(),
+            ca_pem: TEST_CERT_PEM.to_string(),
+        };
+        let endpoint = build_grpc_endpoint("127.0.0.1:19333", Some(&tls)).unwrap();
+        assert_eq!(endpoint.uri().scheme_str(), Some("https"));
+    }
+
+    fn sample_config() -> VolumeServerConfig {
+        VolumeServerConfig {
+            port: 8080,
+            grpc_port: 18080,
+            public_port: 8080,
+            ip: "127.0.0.1".to_string(),
+            bind_ip: String::new(),
+            public_url: "127.0.0.1:8080".to_string(),
+            id: String::new(),
+            masters: vec![],
+            pre_stop_seconds: 0,
+            idle_timeout: 0,
+            data_center: String::new(),
+            rack: String::new(),
+            index_type: NeedleMapKind::InMemory,
+            disk_type: String::new(),
+            folders: vec![],
+            folder_max_limits: vec![],
+            folder_tags: vec![],
+            min_free_spaces: vec![],
+            disk_types: vec![],
+            idx_folder: String::new(),
+            white_list: vec![],
+            fix_jpg_orientation: false,
+            read_mode: ReadMode::Local,
+            allow_untrusted_remote_endpoints: false,
+            cpu_profile: String::new(),
+            mem_profile: String::new(),
+            compaction_byte_per_second: 0,
+            maintenance_byte_per_second: 0,
+            file_size_limit_bytes: 0,
+            concurrent_upload_limit: 0,
+            concurrent_download_limit: 0,
+            inflight_upload_data_timeout: std::time::Duration::from_secs(0),
+            inflight_download_data_timeout: std::time::Duration::from_secs(0),
+            has_slow_read: false,
+            read_buffer_size_mb: 0,
+            ldb_timeout: 0,
+            pprof: false,
+            metrics_port: 0,
+            metrics_ip: String::new(),
+            debug: false,
+            debug_port: 0,
+            ui_enabled: false,
+            jwt_signing_key: vec![],
+            jwt_signing_expires_seconds: 0,
+            jwt_read_signing_key: vec![],
+            jwt_read_signing_expires_seconds: 0,
+            https_cert_file: String::new(),
+            https_key_file: String::new(),
+            https_ca_file: String::new(),
+            https_client_enabled: false,
+            https_client_cert_file: String::new(),
+            https_client_key_file: String::new(),
+            https_client_ca_file: String::new(),
+            grpc_cert_file: String::new(),
+            grpc_key_file: String::new(),
+            grpc_client_cert_file: String::new(),
+            grpc_client_key_file: String::new(),
+            grpc_ca_file: String::new(),
+            grpc_allowed_wildcard_domain: String::new(),
+            grpc_volume_allowed_common_names: vec![],
+            tls_policy: TlsPolicy::default(),
+            enable_write_queue: false,
+            security_file: String::new(),
+        }
+    }
+
+    #[test]
+    fn test_grpc_endpoint_uri_uses_https_when_tls_enabled() {
+        let tls = super::OutgoingGrpcTlsConfig {
+            cert_pem: "cert".to_string(),
+            key_pem: "key".to_string(),
+            ca_pem: "ca".to_string(),
+        };
+        assert_eq!(
+            grpc_endpoint_uri("master.example.com:19333", Some(&tls)),
+            "https://master.example.com:19333"
+        );
+    }
+
+    #[test]
+    fn test_load_outgoing_grpc_tls_requires_cert_key_and_ca() {
+        let mut config = sample_config();
+        config.grpc_cert_file = "/tmp/client.pem".to_string();
+        assert!(load_outgoing_grpc_tls(&config).unwrap().is_none());
+    }
+
+    fn write_pem_files(dir: &tempfile::TempDir, config: &mut VolumeServerConfig) {
+        let write = |name: &str, content: &str| {
+            let path = dir.path().join(name);
+            std::fs::write(&path, content).unwrap();
+            path.to_str().unwrap().to_string()
+        };
+        config.grpc_cert_file = write("server.pem", "server-cert");
+        config.grpc_key_file = write("server.key", "server-key");
+        config.grpc_ca_file = write("ca.pem", "ca");
+    }
+
+    #[test]
+    fn test_load_outgoing_grpc_tls_prefers_client_cert() {
+        let dir = tempfile::TempDir::new().unwrap();
+        let mut config = sample_config();
+        write_pem_files(&dir, &mut config);
+        let client_cert = dir.path().join("client.pem");
+        let client_key = dir.path().join("client.key");
+        std::fs::write(&client_cert, "client-cert").unwrap();
+        std::fs::write(&client_key, "client-key").unwrap();
+        config.grpc_client_cert_file = client_cert.to_str().unwrap().to_string();
+        config.grpc_client_key_file = client_key.to_str().unwrap().to_string();
+
+        let tls = load_outgoing_grpc_tls(&config).unwrap().unwrap();
+        assert_eq!(tls.cert_pem, "client-cert");
+        assert_eq!(tls.key_pem, "client-key");
+    }
+
+    #[test]
+    fn test_load_outgoing_grpc_tls_falls_back_to_server_cert() {
+        let dir = tempfile::TempDir::new().unwrap();
+        let mut config = sample_config();
+        write_pem_files(&dir, &mut config);
+
+        let tls = load_outgoing_grpc_tls(&config).unwrap().unwrap();
+        assert_eq!(tls.cert_pem, "server-cert");
+        assert_eq!(tls.key_pem, "server-key");
+    }
+
+    #[test]
+    fn test_build_grpc_endpoint_without_tls_uses_http_scheme() {
+        let endpoint = build_grpc_endpoint("127.0.0.1:19333", None).unwrap();
+        assert_eq!(endpoint.uri().scheme_str(), Some("http"));
+    }
+
+    #[test]
+    fn test_parse_grpc_address_dotted_form() {
+        use super::parse_grpc_address;
+        assert_eq!(
+            parse_grpc_address("127.0.0.1:8080.18080").unwrap(),
+            "127.0.0.1:18080"
+        );
+    }
+
+    #[test]
+    fn test_parse_grpc_address_implicit_form_adds_10000() {
+        use super::parse_grpc_address;
+        assert_eq!(
+            parse_grpc_address("127.0.0.1:8080").unwrap(),
+            "127.0.0.1:18080"
+        );
+    }
+
+    #[test]
+    fn test_parse_grpc_address_rejects_non_numeric_http_port_in_dotted_form() {
+        use super::parse_grpc_address;
+        let err = parse_grpc_address("host:abc.18080").unwrap_err();
+        assert!(err.contains("invalid http port"), "{}", err);
+    }
+
+    #[test]
+    fn test_parse_grpc_address_rejects_non_numeric_grpc_port_in_dotted_form() {
+        use super::parse_grpc_address;
+        let err = parse_grpc_address("host:8080.xyz").unwrap_err();
+        assert!(err.contains("invalid grpc port"), "{}", err);
+    }
+
+    #[test]
+    fn test_parse_grpc_address_rejects_implicit_port_that_overflows() {
+        use super::parse_grpc_address;
+        let err = parse_grpc_address("127.0.0.1:60000").unwrap_err();
+        assert!(err.contains("out of range"), "{}", err);
+    }
+
+    #[test]
+    fn test_parse_grpc_address_rejects_input_without_colon() {
+        use super::parse_grpc_address;
+        let err = parse_grpc_address("hostname").unwrap_err();
+        assert!(err.contains("cannot parse"), "{}", err);
+    }
+}

seaweed-volume/src/server/memory_status.rs

@@ -0,0 +1,102 @@
+use crate::pb::volume_server_pb;
+
+pub fn collect_mem_status() -> volume_server_pb::MemStatus {
+    #[allow(unused_mut)]
+    let mut mem = volume_server_pb::MemStatus {
+        goroutines: 1,
+        ..Default::default()
+    };
+
+    #[cfg(target_os = "linux")]
+    {
+        if let Some((all, free)) = get_system_memory_linux() {
+            mem.all = all;
+            mem.free = free;
+            mem.used = all.saturating_sub(free);
+        }
+
+        if let Some(status) = read_process_status_linux() {
+            if status.threads > 0 {
+                mem.goroutines = status.threads as i32;
+            }
+            if let Some(rss) = status.rss {
+                mem.self_ = rss;
+            }
+            if let Some(heap) = status.data.or(status.rss) {
+                mem.heap = heap;
+            }
+            if let Some(stack) = status.stack {
+                mem.stack = stack;
+            }
+        }
+    }
+
+    mem
+}
+
+#[cfg(target_os = "linux")]
+fn get_system_memory_linux() -> Option<(u64, u64)> {
+    unsafe {
+        let mut info: libc::sysinfo = std::mem::zeroed();
+        if libc::sysinfo(&mut info) == 0 {
+            let unit = info.mem_unit as u64;
+            let total = info.totalram as u64 * unit;
+            let free = info.freeram as u64 * unit;
+            return Some((total, free));
+        }
+    }
+    None
+}
+
+#[cfg(target_os = "linux")]
+#[derive(Default)]
+struct ProcessStatus {
+    threads: u64,
+    rss: Option<u64>,
+    data: Option<u64>,
+    stack: Option<u64>,
+}
+
+#[cfg(target_os = "linux")]
+fn read_process_status_linux() -> Option<ProcessStatus> {
+    let status = std::fs::read_to_string("/proc/self/status").ok()?;
+    let mut out = ProcessStatus::default();
+
+    for line in status.lines() {
+        if let Some(value) = line.strip_prefix("Threads:") {
+            out.threads = value.trim().parse().ok()?;
+            continue;
+        }
+        if let Some(value) = parse_proc_status_kib_field(line, "VmRSS:") {
+            out.rss = Some(value);
+            continue;
+        }
+        if let Some(value) = parse_proc_status_kib_field(line, "VmData:") {
+            out.data = Some(value);
+            continue;
+        }
+        if let Some(value) = parse_proc_status_kib_field(line, "VmStk:") {
+            out.stack = Some(value);
+        }
+    }
+
+    Some(out)
+}
+
+#[cfg(target_os = "linux")]
+fn parse_proc_status_kib_field(line: &str, prefix: &str) -> Option<u64> {
+    let raw = line.strip_prefix(prefix)?.trim();
+    let value = raw.strip_suffix(" kB").unwrap_or(raw).trim();
+    value.parse::<u64>().ok().map(|kib| kib * 1024)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::collect_mem_status;
+
+    #[test]
+    fn test_collect_mem_status_reports_live_process_state() {
+        let mem = collect_mem_status();
+        assert!(mem.goroutines > 0);
+    }
+}

seaweed-volume/src/server/mod.rs

@@ -0,0 +1,15 @@
+#[cfg(unix)]
+pub mod debug;
+pub mod grpc_client;
+pub mod grpc_server;
+pub mod handlers;
+pub mod heartbeat;
+pub mod memory_status;
+#[cfg(unix)]
+pub mod profiling;
+pub mod request_id;
+pub mod server_stats;
+pub mod store_ec;
+pub mod ui;
+pub mod volume_server;
+pub mod write_queue;

seaweed-volume/src/server/profiling.rs

@@ -0,0 +1,190 @@
+use std::fs::File;
+use std::io::Write;
+use std::path::PathBuf;
+
+use pprof::protos::Message;
+
+use crate::config::VolumeServerConfig;
+
+const GO_CPU_PROFILE_FREQUENCY: i32 = 100;
+const GO_PPROF_BLOCKLIST: [&str; 4] = ["libc", "libgcc", "pthread", "vdso"];
+
+pub struct CpuProfileSession {
+    output_path: PathBuf,
+    guard: pprof::ProfilerGuard<'static>,
+}
+
+impl CpuProfileSession {
+    pub fn start(config: &VolumeServerConfig) -> Result<Option<Self>, String> {
+        if config.cpu_profile.is_empty() {
+            if !config.mem_profile.is_empty() && !config.pprof {
+                tracing::warn!(
+                    "--memprofile is not yet supported in the Rust volume server; ignoring '{}'",
+                    config.mem_profile
+                );
+            }
+            return Ok(None);
+        }
+
+        if config.pprof {
+            tracing::info!(
+                "--pprof is enabled; ignoring --cpuprofile '{}' and --memprofile '{}'",
+                config.cpu_profile,
+                config.mem_profile
+            );
+            return Ok(None);
+        }
+
+        if !config.mem_profile.is_empty() {
+            tracing::warn!(
+                "--memprofile is not yet supported in the Rust volume server; only --cpuprofile '{}' will be written",
+                config.cpu_profile
+            );
+        }
+
+        let guard = pprof::ProfilerGuardBuilder::default()
+            .frequency(GO_CPU_PROFILE_FREQUENCY)
+            .blocklist(&GO_PPROF_BLOCKLIST)
+            .build()
+            .map_err(|e| {
+                format!(
+                    "Failed to start CPU profiler '{}': {}",
+                    config.cpu_profile, e
+                )
+            })?;
+
+        Ok(Some(Self {
+            output_path: PathBuf::from(&config.cpu_profile),
+            guard,
+        }))
+    }
+
+    pub fn finish(self) -> Result<(), String> {
+        let report = self
+            .guard
+            .report()
+            .build()
+            .map_err(|e| format!("Failed to build CPU profile report: {}", e))?;
+        let profile = report
+            .pprof()
+            .map_err(|e| format!("Failed to encode CPU profile report: {}", e))?;
+
+        let mut bytes = Vec::new();
+        profile
+            .encode(&mut bytes)
+            .map_err(|e| format!("Failed to serialize CPU profile report: {}", e))?;
+
+        let mut file = File::create(&self.output_path).map_err(|e| {
+            format!(
+                "Failed to create CPU profile '{}': {}",
+                self.output_path.display(),
+                e
+            )
+        })?;
+        file.write_all(&bytes).map_err(|e| {
+            format!(
+                "Failed to write CPU profile '{}': {}",
+                self.output_path.display(),
+                e
+            )
+        })?;
+        file.flush().map_err(|e| {
+            format!(
+                "Failed to flush CPU profile '{}': {}",
+                self.output_path.display(),
+                e
+            )
+        })?;
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::CpuProfileSession;
+    use crate::config::{NeedleMapKind, ReadMode, VolumeServerConfig};
+    use crate::security::tls::TlsPolicy;
+
+    fn sample_config() -> VolumeServerConfig {
+        VolumeServerConfig {
+            port: 8080,
+            grpc_port: 18080,
+            public_port: 8080,
+            ip: "127.0.0.1".to_string(),
+            bind_ip: "127.0.0.1".to_string(),
+            public_url: "127.0.0.1:8080".to_string(),
+            id: "127.0.0.1:8080".to_string(),
+            masters: vec![],
+            pre_stop_seconds: 0,
+            idle_timeout: 0,
+            data_center: String::new(),
+            rack: String::new(),
+            index_type: NeedleMapKind::InMemory,
+            disk_type: String::new(),
+            folders: vec!["/tmp".to_string()],
+            folder_max_limits: vec![8],
+            folder_tags: vec![vec![]],
+            min_free_spaces: vec![],
+            disk_types: vec![String::new()],
+            idx_folder: String::new(),
+            white_list: vec![],
+            fix_jpg_orientation: false,
+            read_mode: ReadMode::Local,
+            allow_untrusted_remote_endpoints: false,
+            cpu_profile: String::new(),
+            mem_profile: String::new(),
+            compaction_byte_per_second: 0,
+            maintenance_byte_per_second: 0,
+            file_size_limit_bytes: 0,
+            concurrent_upload_limit: 0,
+            concurrent_download_limit: 0,
+            inflight_upload_data_timeout: std::time::Duration::from_secs(0),
+            inflight_download_data_timeout: std::time::Duration::from_secs(0),
+            has_slow_read: false,
+            read_buffer_size_mb: 4,
+            ldb_timeout: 0,
+            pprof: false,
+            metrics_port: 0,
+            metrics_ip: String::new(),
+            debug: false,
+            debug_port: 0,
+            ui_enabled: false,
+            jwt_signing_key: vec![],
+            jwt_signing_expires_seconds: 0,
+            jwt_read_signing_key: vec![],
+            jwt_read_signing_expires_seconds: 0,
+            https_cert_file: String::new(),
+            https_key_file: String::new(),
+            https_ca_file: String::new(),
+            https_client_enabled: false,
+            https_client_cert_file: String::new(),
+            https_client_key_file: String::new(),
+            https_client_ca_file: String::new(),
+            grpc_cert_file: String::new(),
+            grpc_key_file: String::new(),
+            grpc_client_cert_file: String::new(),
+            grpc_client_key_file: String::new(),
+            grpc_ca_file: String::new(),
+            grpc_allowed_wildcard_domain: String::new(),
+            grpc_volume_allowed_common_names: vec![],
+            tls_policy: TlsPolicy::default(),
+            enable_write_queue: false,
+            security_file: String::new(),
+        }
+    }
+
+    #[test]
+    fn test_cpu_profile_session_skips_when_disabled() {
+        let config = sample_config();
+        assert!(CpuProfileSession::start(&config).unwrap().is_none());
+    }
+
+    #[test]
+    fn test_cpu_profile_session_skips_when_pprof_enabled() {
+        let mut config = sample_config();
+        config.cpu_profile = "/tmp/cpu.pb".to_string();
+        config.pprof = true;
+        assert!(CpuProfileSession::start(&config).unwrap().is_none());
+    }
+}

seaweed-volume/src/server/request_id.rs

@@ -0,0 +1,137 @@
+use std::future::Future;
+use std::pin::Pin;
+use std::task::{Context, Poll};
+
+use hyper::http::{self, HeaderValue};
+use tonic::metadata::MetadataValue;
+use tonic::{Request, Status};
+use tower::{Layer, Service};
+
+tokio::task_local! {
+    static CURRENT_REQUEST_ID: String;
+}
+
+#[derive(Clone, Debug, Default)]
+pub struct GrpcRequestIdLayer;
+
+#[derive(Clone, Debug)]
+pub struct GrpcRequestIdService<S> {
+    inner: S,
+}
+
+impl<S> Layer<S> for GrpcRequestIdLayer {
+    type Service = GrpcRequestIdService<S>;
+
+    fn layer(&self, inner: S) -> Self::Service {
+        GrpcRequestIdService { inner }
+    }
+}
+
+impl<S, B> Service<http::Request<B>> for GrpcRequestIdService<S>
+where
+    S: Service<http::Request<B>, Response = http::Response<tonic::body::Body>> + Send + 'static,
+    S::Future: Send + 'static,
+    B: Send + 'static,
+{
+    type Response = http::Response<tonic::body::Body>;
+    type Error = S::Error;
+    type Future = Pin<Box<dyn Future<Output = Result<Self::Response, Self::Error>> + Send>>;
+
+    fn poll_ready(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
+        self.inner.poll_ready(cx)
+    }
+
+    fn call(&mut self, mut request: http::Request<B>) -> Self::Future {
+        let request_id = match request.headers().get("x-amz-request-id") {
+            Some(value) => match value.to_str() {
+                Ok(value) if !value.is_empty() => value.to_owned(),
+                _ => generate_grpc_request_id(),
+            },
+            None => generate_grpc_request_id(),
+        };
+
+        if let Ok(value) = HeaderValue::from_str(&request_id) {
+            request.headers_mut().insert("x-amz-request-id", value);
+        }
+
+        let future = self.inner.call(request);
+
+        Box::pin(async move {
+            let mut response: http::Response<tonic::body::Body> =
+                scope_request_id(request_id.clone(), future).await?;
+            if let Ok(value) = HeaderValue::from_str(&request_id) {
+                response.headers_mut().insert("x-amz-request-id", value);
+            }
+            Ok(response)
+        })
+    }
+}
+
+pub async fn scope_request_id<F, T>(request_id: String, future: F) -> T
+where
+    F: Future<Output = T>,
+{
+    CURRENT_REQUEST_ID.scope(request_id, future).await
+}
+
+pub fn current_request_id() -> Option<String> {
+    CURRENT_REQUEST_ID.try_with(Clone::clone).ok()
+}
+
+pub fn outgoing_request_id_interceptor(mut request: Request<()>) -> Result<Request<()>, Status> {
+    if let Some(request_id) = current_request_id() {
+        let value = MetadataValue::try_from(request_id.as_str())
+            .map_err(|_| Status::internal("invalid scoped request id"))?;
+        request.metadata_mut().insert("x-amz-request-id", value);
+    }
+    Ok(request)
+}
+
+pub fn generate_http_request_id() -> String {
+    use rand::RngExt;
+
+    let nanos = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_nanos() as u64;
+    let rand_val: u32 = rand::rng().random();
+    format!("{:X}{:08X}", nanos, rand_val)
+}
+
+fn generate_grpc_request_id() -> String {
+    uuid::Uuid::new_v4().to_string()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{current_request_id, outgoing_request_id_interceptor, scope_request_id};
+    use tonic::Request;
+
+    #[tokio::test]
+    async fn test_scope_request_id_exposes_current_value() {
+        let request_id = "req-123".to_string();
+        let current = scope_request_id(
+            request_id.clone(),
+            async move { current_request_id().unwrap() },
+        )
+        .await;
+        assert_eq!(current, request_id);
+    }
+
+    #[tokio::test]
+    async fn test_outgoing_request_id_interceptor_propagates_scope() {
+        let request = scope_request_id("req-456".to_string(), async move {
+            outgoing_request_id_interceptor(Request::new(())).unwrap()
+        })
+        .await;
+        assert_eq!(
+            request
+                .metadata()
+                .get("x-amz-request-id")
+                .unwrap()
+                .to_str()
+                .unwrap(),
+            "req-456"
+        );
+    }
+}

seaweed-volume/src/server/server_stats.rs

@@ -0,0 +1,248 @@
+use chrono::{Datelike, Local, Timelike};
+use serde::Serialize;
+use std::sync::{LazyLock, Mutex};
+use std::time::Instant;
+
+static START_TIME: LazyLock<Instant> = LazyLock::new(Instant::now);
+static SERVER_STATS: LazyLock<ServerStats> = LazyLock::new(ServerStats::default);
+
+#[derive(Default)]
+pub struct ServerStats {
+    inner: Mutex<ServerStatsInner>,
+}
+
+#[derive(Default)]
+struct ServerStatsInner {
+    requests: DurationCounter,
+    connections: DurationCounter,
+    assign_requests: DurationCounter,
+    read_requests: DurationCounter,
+    write_requests: DurationCounter,
+    delete_requests: DurationCounter,
+    bytes_in: DurationCounter,
+    bytes_out: DurationCounter,
+}
+
+#[derive(Clone, Serialize)]
+#[serde(rename_all = "PascalCase")]
+pub struct ServerStatsSnapshot {
+    pub requests: DurationCounterSnapshot,
+    pub connections: DurationCounterSnapshot,
+    pub assign_requests: DurationCounterSnapshot,
+    pub read_requests: DurationCounterSnapshot,
+    pub write_requests: DurationCounterSnapshot,
+    pub delete_requests: DurationCounterSnapshot,
+    pub bytes_in: DurationCounterSnapshot,
+    pub bytes_out: DurationCounterSnapshot,
+}
+
+#[derive(Clone, Serialize)]
+#[serde(rename_all = "PascalCase")]
+pub struct DurationCounterSnapshot {
+    pub minute_counter: RoundRobinCounterSnapshot,
+    pub hour_counter: RoundRobinCounterSnapshot,
+    pub day_counter: RoundRobinCounterSnapshot,
+    pub week_counter: RoundRobinCounterSnapshot,
+}
+
+#[derive(Clone, Serialize)]
+#[serde(rename_all = "PascalCase")]
+pub struct RoundRobinCounterSnapshot {
+    pub last_index: i32,
+    pub values: Vec<i64>,
+    pub counts: Vec<i64>,
+}
+
+#[derive(Clone)]
+struct DurationCounter {
+    minute_counter: RoundRobinCounter,
+    hour_counter: RoundRobinCounter,
+    day_counter: RoundRobinCounter,
+    week_counter: RoundRobinCounter,
+}
+
+#[derive(Clone)]
+struct RoundRobinCounter {
+    last_index: i32,
+    values: Vec<i64>,
+    counts: Vec<i64>,
+}
+
+impl Default for DurationCounter {
+    fn default() -> Self {
+        Self {
+            minute_counter: RoundRobinCounter::new(60),
+            hour_counter: RoundRobinCounter::new(60),
+            day_counter: RoundRobinCounter::new(24),
+            week_counter: RoundRobinCounter::new(7),
+        }
+    }
+}
+
+impl RoundRobinCounter {
+    fn new(slots: usize) -> Self {
+        Self {
+            last_index: -1,
+            values: vec![0; slots],
+            counts: vec![0; slots],
+        }
+    }
+
+    fn add(&mut self, index: usize, val: i64) {
+        if index >= self.values.len() {
+            return;
+        }
+        while self.last_index != index as i32 {
+            self.last_index = (self.last_index + 1).rem_euclid(self.values.len() as i32);
+            self.values[self.last_index as usize] = 0;
+            self.counts[self.last_index as usize] = 0;
+        }
+        self.values[index] += val;
+        self.counts[index] += 1;
+    }
+
+    fn snapshot(&self) -> RoundRobinCounterSnapshot {
+        RoundRobinCounterSnapshot {
+            last_index: self.last_index,
+            values: self.values.clone(),
+            counts: self.counts.clone(),
+        }
+    }
+}
+
+impl DurationCounter {
+    fn add_now(&mut self, val: i64) {
+        let now = Local::now();
+        self.minute_counter.add(now.second() as usize, val);
+        self.hour_counter.add(now.minute() as usize, val);
+        self.day_counter.add(now.hour() as usize, val);
+        self.week_counter
+            .add(now.weekday().num_days_from_sunday() as usize, val);
+    }
+
+    fn snapshot(&self) -> DurationCounterSnapshot {
+        DurationCounterSnapshot {
+            minute_counter: self.minute_counter.snapshot(),
+            hour_counter: self.hour_counter.snapshot(),
+            day_counter: self.day_counter.snapshot(),
+            week_counter: self.week_counter.snapshot(),
+        }
+    }
+}
+
+impl ServerStatsInner {
+    fn snapshot(&self) -> ServerStatsSnapshot {
+        ServerStatsSnapshot {
+            requests: self.requests.snapshot(),
+            connections: self.connections.snapshot(),
+            assign_requests: self.assign_requests.snapshot(),
+            read_requests: self.read_requests.snapshot(),
+            write_requests: self.write_requests.snapshot(),
+            delete_requests: self.delete_requests.snapshot(),
+            bytes_in: self.bytes_in.snapshot(),
+            bytes_out: self.bytes_out.snapshot(),
+        }
+    }
+}
+
+impl ServerStats {
+    fn update<F>(&self, update: F)
+    where
+        F: FnOnce(&mut ServerStatsInner),
+    {
+        let mut inner = self.inner.lock().unwrap();
+        update(&mut inner);
+    }
+
+    fn snapshot(&self) -> ServerStatsSnapshot {
+        self.inner.lock().unwrap().snapshot()
+    }
+}
+
+impl RoundRobinCounterSnapshot {
+    pub fn to_list(&self) -> Vec<i64> {
+        if self.values.is_empty() {
+            return Vec::new();
+        }
+        let mut ret = Vec::with_capacity(self.values.len());
+        let mut index = self.last_index;
+        let mut step = self.values.len();
+        while step > 0 {
+            step -= 1;
+            index += 1;
+            if index >= self.values.len() as i32 {
+                index = 0;
+            }
+            ret.push(self.values[index as usize]);
+        }
+        ret
+    }
+}
+
+pub fn init_process_start() {
+    LazyLock::force(&START_TIME);
+    LazyLock::force(&SERVER_STATS);
+}
+
+pub fn uptime_string() -> String {
+    let secs = START_TIME.elapsed().as_secs();
+    let hours = secs / 3600;
+    let minutes = (secs % 3600) / 60;
+    let seconds = secs % 60;
+    let mut out = String::new();
+    if hours > 0 {
+        out.push_str(&format!("{}h", hours));
+    }
+    if hours > 0 || minutes > 0 {
+        out.push_str(&format!("{}m", minutes));
+    }
+    out.push_str(&format!("{}s", seconds));
+    out
+}
+
+pub fn snapshot() -> ServerStatsSnapshot {
+    SERVER_STATS.snapshot()
+}
+
+pub fn record_request_open() {
+    SERVER_STATS.update(|inner| inner.requests.add_now(1));
+}
+
+pub fn record_request_close() {
+    SERVER_STATS.update(|inner| inner.requests.add_now(-1));
+}
+
+pub fn record_connection_open() {
+    SERVER_STATS.update(|inner| inner.connections.add_now(1));
+}
+
+pub fn record_connection_close() {
+    SERVER_STATS.update(|inner| inner.connections.add_now(-1));
+}
+
+pub fn record_read_request() {
+    SERVER_STATS.update(|inner| inner.read_requests.add_now(1));
+}
+
+pub fn record_write_request() {
+    SERVER_STATS.update(|inner| inner.write_requests.add_now(1));
+}
+
+pub fn record_delete_request() {
+    SERVER_STATS.update(|inner| inner.delete_requests.add_now(1));
+}
+
+pub fn record_bytes_in(bytes: i64) {
+    SERVER_STATS.update(|inner| inner.bytes_in.add_now(bytes));
+}
+
+pub fn record_bytes_out(bytes: i64) {
+    SERVER_STATS.update(|inner| inner.bytes_out.add_now(bytes));
+}
+
+#[cfg(test)]
+pub fn reset_for_tests() {
+    LazyLock::force(&START_TIME);
+    let mut inner = SERVER_STATS.inner.lock().unwrap();
+    *inner = ServerStatsInner::default();
+}

seaweed-volume/src/server/ui.rs

@@ -0,0 +1,506 @@
+use std::fmt::Write as _;
+
+use crate::server::server_stats;
+use crate::server::volume_server::VolumeServerState;
+use crate::storage::store::Store;
+
+pub struct EmbeddedAsset {
+    pub content_type: &'static str,
+    pub bytes: &'static [u8],
+}
+
+struct UiDiskRow {
+    dir: String,
+    disk_type: String,
+    all: u64,
+    free: u64,
+    used: u64,
+}
+
+struct UiVolumeRow {
+    id: u32,
+    collection: String,
+    disk_type: String,
+    size: u64,
+    file_count: i64,
+    delete_count: i64,
+    deleted_byte_count: u64,
+    ttl: String,
+    read_only: bool,
+    version: u32,
+    remote_storage_name: String,
+    remote_storage_key: String,
+}
+
+struct UiEcShardRow {
+    shard_id: u8,
+    size: u64,
+}
+
+struct UiEcVolumeRow {
+    volume_id: u32,
+    collection: String,
+    size: u64,
+    shards: Vec<UiEcShardRow>,
+    created_at: String,
+}
+
+pub fn favicon_asset() -> EmbeddedAsset {
+    EmbeddedAsset {
+        content_type: "image/x-icon",
+        bytes: include_bytes!(concat!(
+            env!("CARGO_MANIFEST_DIR"),
+            "/../weed/static/favicon.ico"
+        )),
+    }
+}
+
+pub fn lookup_static_asset(path: &str) -> Option<EmbeddedAsset> {
+    let path = path.trim_start_matches('/');
+    let asset = match path {
+        "bootstrap/3.3.1/css/bootstrap.min.css" => EmbeddedAsset {
+            content_type: "text/css; charset=utf-8",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/bootstrap/3.3.1/css/bootstrap.min.css"
+            )),
+        },
+        "bootstrap/3.3.1/fonts/glyphicons-halflings-regular.eot" => EmbeddedAsset {
+            content_type: "application/vnd.ms-fontobject",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/bootstrap/3.3.1/fonts/glyphicons-halflings-regular.eot"
+            )),
+        },
+        "bootstrap/3.3.1/fonts/glyphicons-halflings-regular.svg" => EmbeddedAsset {
+            content_type: "image/svg+xml",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/bootstrap/3.3.1/fonts/glyphicons-halflings-regular.svg"
+            )),
+        },
+        "bootstrap/3.3.1/fonts/glyphicons-halflings-regular.ttf" => EmbeddedAsset {
+            content_type: "font/ttf",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/bootstrap/3.3.1/fonts/glyphicons-halflings-regular.ttf"
+            )),
+        },
+        "bootstrap/3.3.1/fonts/glyphicons-halflings-regular.woff" => EmbeddedAsset {
+            content_type: "font/woff",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/bootstrap/3.3.1/fonts/glyphicons-halflings-regular.woff"
+            )),
+        },
+        "images/folder.gif" => EmbeddedAsset {
+            content_type: "image/gif",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/images/folder.gif"
+            )),
+        },
+        "javascript/jquery-3.6.0.min.js" => EmbeddedAsset {
+            content_type: "application/javascript; charset=utf-8",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/javascript/jquery-3.6.0.min.js"
+            )),
+        },
+        "javascript/jquery-sparklines/2.1.2/jquery.sparkline.min.js" => EmbeddedAsset {
+            content_type: "application/javascript; charset=utf-8",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/javascript/jquery-sparklines/2.1.2/jquery.sparkline.min.js"
+            )),
+        },
+        "seaweed50x50.png" => EmbeddedAsset {
+            content_type: "image/png",
+            bytes: include_bytes!(concat!(
+                env!("CARGO_MANIFEST_DIR"),
+                "/../weed/static/seaweed50x50.png"
+            )),
+        },
+        _ => return None,
+    };
+    Some(asset)
+}
+
+pub fn render_volume_server_html(state: &VolumeServerState) -> String {
+    let counters = server_stats::snapshot();
+    let (disk_rows, volume_rows, remote_volume_rows, ec_volume_rows) = {
+        let store = state.store.read().unwrap();
+        collect_ui_data(&store)
+    };
+
+    let masters = if state.master_urls.is_empty() {
+        "[]".to_string()
+    } else {
+        format!("[{}]", state.master_urls.join(" "))
+    };
+    let uptime = server_stats::uptime_string();
+    let read_week = join_i64(&counters.read_requests.week_counter.to_list());
+    let read_day = join_i64(&counters.read_requests.day_counter.to_list());
+    let read_hour = join_i64(&counters.read_requests.hour_counter.to_list());
+    let read_minute = join_i64(&counters.read_requests.minute_counter.to_list());
+
+    let mut disk_rows_html = String::new();
+    for disk in &disk_rows {
+        let _ = write!(
+            disk_rows_html,
+            "<tr><td>{}</td><td>{}</td><td>{}</td><td>{}</td><td>{:.2}%</td></tr>",
+            escape_html(&disk.dir),
+            escape_html(&disk.disk_type),
+            bytes_to_human_readable(disk.all),
+            bytes_to_human_readable(disk.free),
+            percent_from(disk.all, disk.used),
+        );
+    }
+
+    let mut volume_rows_html = String::new();
+    for volume in &volume_rows {
+        let _ = write!(
+            volume_rows_html,
+            "<tr><td><code>{}</code></td><td>{}</td><td>{}</td><td>{}</td><td>{}</td><td>{} / {}</td><td>{}</td><td>{}</td><td>{}</td></tr>",
+            volume.id,
+            escape_html(&volume.collection),
+            escape_html(&volume.disk_type),
+            bytes_to_human_readable(volume.size),
+            volume.file_count,
+            volume.delete_count,
+            bytes_to_human_readable(volume.deleted_byte_count),
+            escape_html(&volume.ttl),
+            volume.read_only,
+            volume.version,
+        );
+    }
+
+    let remote_section = if remote_volume_rows.is_empty() {
+        String::new()
+    } else {
+        let mut remote_rows_html = String::new();
+        for volume in &remote_volume_rows {
+            let _ = write!(
+                remote_rows_html,
+                "<tr><td><code>{}</code></td><td>{}</td><td>{}</td><td>{}</td><td>{} / {}</td><td>{}</td><td>{}</td></tr>",
+                volume.id,
+                escape_html(&volume.collection),
+                bytes_to_human_readable(volume.size),
+                volume.file_count,
+                volume.delete_count,
+                bytes_to_human_readable(volume.deleted_byte_count),
+                escape_html(&volume.remote_storage_name),
+                escape_html(&volume.remote_storage_key),
+            );
+        }
+        format!(
+            r#"<div class="row">
+            <h2>Remote Volumes</h2>
+            <table class="table table-striped">
+                <thead>
+                    <tr>
+                        <th>Id</th>
+                        <th>Collection</th>
+                        <th>Size</th>
+                        <th>Files</th>
+                        <th>Trash</th>
+                        <th>Remote</th>
+                        <th>Key</th>
+                    </tr>
+                </thead>
+                <tbody>{}</tbody>
+            </table>
+        </div>"#,
+            remote_rows_html
+        )
+    };
+
+    let ec_section = if ec_volume_rows.is_empty() {
+        String::new()
+    } else {
+        let mut ec_rows_html = String::new();
+        for ec in &ec_volume_rows {
+            let mut shard_labels = String::new();
+            for shard in &ec.shards {
+                let _ = write!(
+                    shard_labels,
+                    "<span class=\"label label-info\" style=\"margin-right: 5px;\">{}: {}</span>",
+                    shard.shard_id,
+                    bytes_to_human_readable(shard.size)
+                );
+            }
+            let _ = write!(
+                ec_rows_html,
+                "<tr><td><code>{}</code></td><td>{}</td><td>{}</td><td>{}</td><td>{}</td></tr>",
+                ec.volume_id,
+                escape_html(&ec.collection),
+                bytes_to_human_readable(ec.size),
+                shard_labels,
+                escape_html(&ec.created_at),
+            );
+        }
+        format!(
+            r#"<div class="row">
+            <h2>Erasure Coding Shards</h2>
+            <table class="table table-striped">
+                <thead>
+                    <tr>
+                        <th>Id</th>
+                        <th>Collection</th>
+                        <th>Total Size</th>
+                        <th>Shard Details</th>
+                        <th>CreatedAt</th>
+                    </tr>
+                </thead>
+                <tbody>{}</tbody>
+            </table>
+        </div>"#,
+            ec_rows_html
+        )
+    };
+
+    format!(
+        r#"<!DOCTYPE html>
+<html>
+<head>
+    <title>SeaweedFS {version}</title>
+    <link rel="stylesheet" href="/seaweedfsstatic/bootstrap/3.3.1/css/bootstrap.min.css">
+    <script type="text/javascript" src="/seaweedfsstatic/javascript/jquery-3.6.0.min.js"></script>
+    <script type="text/javascript" src="/seaweedfsstatic/javascript/jquery-sparklines/2.1.2/jquery.sparkline.min.js"></script>
+    <script type="text/javascript">
+        $(function () {{
+            var periods = ['second', 'minute', 'hour', 'day'];
+            for (var i = 0; i < periods.length; i++) {{
+                var period = periods[i];
+                $('.inlinesparkline-' + period).sparkline('html', {{
+                    type: 'line',
+                    barColor: 'red',
+                    tooltipSuffix: ' request per ' + period
+                }});
+            }}
+        }});
+    </script>
+    <style>
+        #jqstooltip {{
+            height: 28px !important;
+            width: 150px !important;
+        }}
+    </style>
+</head>
+<body>
+    <div class="container">
+        <div class="page-header">
+            <h1>
+                <a href="https://github.com/seaweedfs/seaweedfs"><img src="/seaweedfsstatic/seaweed50x50.png" alt="SeaweedFS"></a>
+                SeaweedFS <small>{version}</small>
+            </h1>
+        </div>
+
+        <div class="row">
+            <div class="col-sm-6">
+                <h2>Disk Stats</h2>
+                <table class="table table-striped">
+                    <thead>
+                        <tr>
+                            <th>Path</th>
+                            <th>Disk</th>
+                            <th>Total</th>
+                            <th>Free</th>
+                            <th>Usage</th>
+                        </tr>
+                    </thead>
+                    <tbody>{disk_rows_html}</tbody>
+                </table>
+            </div>
+
+            <div class="col-sm-6">
+                <h2>System Stats</h2>
+                <table class="table table-condensed table-striped">
+                    <tr><th>Masters</th><td>{masters}</td></tr>
+                    <tr><th>Weekly # ReadRequests</th><td><span class="inlinesparkline-day">{read_week}</span></td></tr>
+                    <tr><th>Daily # ReadRequests</th><td><span class="inlinesparkline-hour">{read_day}</span></td></tr>
+                    <tr><th>Hourly # ReadRequests</th><td><span class="inlinesparkline-minute">{read_hour}</span></td></tr>
+                    <tr><th>Last Minute # ReadRequests</th><td><span class="inlinesparkline-second">{read_minute}</span></td></tr>
+                    <tr><th>Up Time</th><td>{uptime}</td></tr>
+                </table>
+            </div>
+        </div>
+
+        <div class="row">
+            <h2>Volumes</h2>
+            <table class="table table-striped">
+                <thead>
+                    <tr>
+                        <th>Id</th>
+                        <th>Collection</th>
+                        <th>Disk</th>
+                        <th>Data Size</th>
+                        <th>Files</th>
+                        <th>Trash</th>
+                        <th>TTL</th>
+                        <th>ReadOnly</th>
+                        <th>Version</th>
+                    </tr>
+                </thead>
+                <tbody>{volume_rows_html}</tbody>
+            </table>
+        </div>
+
+        {remote_section}
+        {ec_section}
+    </div>
+</body>
+</html>"#,
+        version = escape_html(crate::version::version()),
+        disk_rows_html = disk_rows_html,
+        masters = escape_html(&masters),
+        read_week = read_week,
+        read_day = read_day,
+        read_hour = read_hour,
+        read_minute = read_minute,
+        uptime = escape_html(&uptime),
+        volume_rows_html = volume_rows_html,
+        remote_section = remote_section,
+        ec_section = ec_section,
+    )
+}
+
+fn collect_ui_data(
+    store: &Store,
+) -> (
+    Vec<UiDiskRow>,
+    Vec<UiVolumeRow>,
+    Vec<UiVolumeRow>,
+    Vec<UiEcVolumeRow>,
+) {
+    let mut disk_rows = Vec::new();
+    let mut volumes = Vec::new();
+    let mut remote_volumes = Vec::new();
+    let mut ec_volumes = Vec::new();
+
+    for loc in &store.locations {
+        let dir = absolute_display_path(&loc.directory);
+        let (all, free) = crate::storage::disk_location::get_disk_stats(&dir);
+        disk_rows.push(UiDiskRow {
+            dir,
+            disk_type: loc.disk_type.to_string(),
+            all,
+            free,
+            used: all.saturating_sub(free),
+        });
+
+        for (_, volume) in loc.volumes() {
+            let (remote_storage_name, remote_storage_key) = volume.remote_storage_name_key();
+            let row = UiVolumeRow {
+                id: volume.id.0,
+                collection: volume.collection.clone(),
+                disk_type: loc.disk_type.to_string(),
+                size: volume.content_size(),
+                file_count: volume.file_count(),
+                delete_count: volume.deleted_count(),
+                deleted_byte_count: volume.deleted_size(),
+                ttl: volume.super_block.ttl.to_string(),
+                read_only: volume.is_read_only(),
+                version: volume.version().0 as u32,
+                remote_storage_name,
+                remote_storage_key,
+            };
+            if row.remote_storage_name.is_empty() {
+                volumes.push(row);
+            } else {
+                remote_volumes.push(row);
+            }
+        }
+
+        for (_, ec_volume) in loc.ec_volumes() {
+            let mut shards = Vec::new();
+            let mut total_size = 0u64;
+            let mut created_at = String::from("-");
+            for shard in ec_volume.shards.iter().flatten() {
+                let shard_size = shard.file_size().max(0) as u64;
+                total_size = total_size.saturating_add(shard_size);
+                shards.push(UiEcShardRow {
+                    shard_id: shard.shard_id,
+                    size: shard_size,
+                });
+                if created_at == "-"
+                    && let Ok(metadata) = std::fs::metadata(shard.file_name())
+                    && let Ok(modified) = metadata.modified()
+                {
+                    let ts: chrono::DateTime<chrono::Local> = modified.into();
+                    created_at = ts.format("%Y-%m-%d %H:%M").to_string();
+                }
+            }
+            let preferred_size = ec_volume.dat_file_size.max(0) as u64;
+            ec_volumes.push(UiEcVolumeRow {
+                volume_id: ec_volume.volume_id.0,
+                collection: ec_volume.collection.clone(),
+                size: preferred_size.max(total_size),
+                shards,
+                created_at,
+            });
+        }
+    }
+
+    disk_rows.sort_by(|left, right| left.dir.cmp(&right.dir));
+    volumes.sort_by_key(|row| row.id);
+    remote_volumes.sort_by_key(|row| row.id);
+    ec_volumes.sort_by_key(|row| row.volume_id);
+
+    (disk_rows, volumes, remote_volumes, ec_volumes)
+}
+
+fn absolute_display_path(path: &str) -> String {
+    let p = std::path::Path::new(path);
+    if p.is_absolute() {
+        return path.to_string();
+    }
+    std::env::current_dir()
+        .map(|cwd| cwd.join(p).to_string_lossy().to_string())
+        .unwrap_or_else(|_| path.to_string())
+}
+
+fn join_i64(values: &[i64]) -> String {
+    values
+        .iter()
+        .map(std::string::ToString::to_string)
+        .collect::<Vec<_>>()
+        .join(",")
+}
+
+fn percent_from(total: u64, part: u64) -> f64 {
+    if total == 0 {
+        return 0.0;
+    }
+    (part as f64 / total as f64) * 100.0
+}
+
+fn bytes_to_human_readable(bytes: u64) -> String {
+    const UNIT: u64 = 1024;
+    if bytes < UNIT {
+        return format!("{} B", bytes);
+    }
+
+    let mut div = UNIT;
+    let mut exp = 0usize;
+    let mut n = bytes / UNIT;
+    while n >= UNIT {
+        div *= UNIT;
+        n /= UNIT;
+        exp += 1;
+    }
+
+    format!(
+        "{:.2} {}iB",
+        bytes as f64 / div as f64,
+        ["K", "M", "G", "T", "P", "E"][exp]
+    )
+}
+
+fn escape_html(input: &str) -> String {
+    input
+        .replace('&', "&amp;")
+        .replace('<', "&lt;")
+        .replace('>', "&gt;")
+        .replace('"', "&quot;")
+}

seaweed-volume/src/server/volume_server.rs

@@ -0,0 +1,542 @@
+//! VolumeServer: the main HTTP server for volume operations.
+//!
+//! Routes:
+//!   GET/HEAD /{vid},{fid}     — read a file
+//!   POST/PUT /{vid},{fid}     — write a file
+//!   DELETE   /{vid},{fid}     — delete a file
+//!   GET      /status          — server status
+//!   GET      /healthz         — health check
+//!
+//! Matches Go's server/volume_server.go.
+
+use std::net::SocketAddr;
+use std::sync::atomic::{AtomicBool, AtomicI64, AtomicU32, Ordering};
+use std::sync::{Arc, RwLock};
+
+use axum::{
+    Router,
+    extract::{Request, State, connect_info::ConnectInfo},
+    http::{HeaderValue, Method, StatusCode, header},
+    middleware::{self, Next},
+    response::{IntoResponse, Response},
+    routing::{any, get},
+};
+
+use crate::config::ReadMode;
+use crate::security::Guard;
+use crate::storage::store::Store;
+
+use super::grpc_client::OutgoingGrpcTlsConfig;
+use super::handlers;
+use super::write_queue::WriteQueue;
+
+#[derive(Clone, Debug, Default)]
+pub struct RuntimeMetricsConfig {
+    pub push_gateway: crate::metrics::PushGatewayConfig,
+}
+
+/// Shared state for the volume server.
+pub struct VolumeServerState {
+    pub store: RwLock<Store>,
+    pub guard: RwLock<Guard>,
+    pub is_stopping: RwLock<bool>,
+    /// Maintenance mode flag.
+    pub maintenance: AtomicBool,
+    /// State version — incremented on each SetState call.
+    pub state_version: AtomicU32,
+    /// Throttling: concurrent upload/download limits (in bytes, 0 = disabled).
+    pub concurrent_upload_limit: i64,
+    pub concurrent_download_limit: i64,
+    pub inflight_upload_data_timeout: std::time::Duration,
+    pub inflight_download_data_timeout: std::time::Duration,
+    /// Current in-flight upload/download bytes.
+    pub inflight_upload_bytes: AtomicI64,
+    pub inflight_download_bytes: AtomicI64,
+    /// Notify waiters when inflight bytes decrease.
+    pub upload_notify: tokio::sync::Notify,
+    pub download_notify: tokio::sync::Notify,
+    /// Data center name from config.
+    pub data_center: String,
+    /// Rack name from config.
+    pub rack: String,
+    /// File size limit in bytes (0 = no limit).
+    pub file_size_limit_bytes: i64,
+    /// Default IO rate limit for maintenance copy/replication work.
+    pub maintenance_byte_per_second: i64,
+    /// Whether the server is connected to master (heartbeat active).
+    pub is_heartbeating: AtomicBool,
+    /// Whether master addresses are configured.
+    pub has_master: bool,
+    /// Seconds to wait before shutting down servers (graceful drain).
+    pub pre_stop_seconds: u32,
+    /// Notify heartbeat to send an immediate update when volume state changes.
+    pub volume_state_notify: tokio::sync::Notify,
+    /// Optional batched write queue for improved throughput under load.
+    pub write_queue: std::sync::OnceLock<WriteQueue>,
+    /// Registry of S3 tier backends for tiered storage operations.
+    pub s3_tier_registry: std::sync::RwLock<crate::remote_storage::s3_tier::S3TierRegistry>,
+    /// Read mode: local, proxy, or redirect for non-local volumes.
+    pub read_mode: ReadMode,
+    /// If true, FetchAndWriteNeedle skips remote S3 endpoint validation,
+    /// allowing arbitrary (incl. loopback / link-local / metadata) hosts.
+    pub allow_untrusted_remote_endpoints: bool,
+    /// First master address for volume lookups (e.g., "localhost:9333").
+    pub master_url: String,
+    /// Seed master addresses for UI rendering.
+    pub master_urls: Vec<String>,
+    /// Canonical http `host:port` form of every configured seed master.
+    /// Built once at construction so Ping admission stays O(1). Mirrors
+    /// Go's `seedMasterSet` on `VolumeServer`.
+    pub seed_master_set: std::collections::HashSet<String>,
+    /// Current master this server is heartbeating with, in canonical http
+    /// `host:port` form. Empty when no heartbeat connection is active. The
+    /// heartbeat goroutine writes; admission reads — the lock keeps them
+    /// from racing on a leader change. Mirrors Go's `currentMaster` plus
+    /// `currentMasterLock`.
+    pub current_master_url: tokio::sync::RwLock<String>,
+    /// This server's own address (ip:port) for filtering self from lookup results.
+    pub self_url: String,
+    /// HTTP client for proxy requests and master lookups.
+    pub http_client: reqwest::Client,
+    /// Scheme used for outgoing master and peer HTTP requests ("http" or "https").
+    pub outgoing_http_scheme: String,
+    /// Optional client TLS material for outgoing gRPC connections.
+    pub outgoing_grpc_tls: Option<OutgoingGrpcTlsConfig>,
+    /// Metrics push settings learned from master heartbeat responses.
+    pub metrics_runtime: std::sync::RwLock<RuntimeMetricsConfig>,
+    pub metrics_notify: tokio::sync::Notify,
+    /// Whether JPEG uploads should be normalized using EXIF orientation.
+    pub fix_jpg_orientation: bool,
+    /// Read tuning flags for large-file streaming.
+    pub has_slow_read: bool,
+    pub read_buffer_size_bytes: usize,
+    /// Path to security.toml — stored for SIGHUP reload.
+    pub security_file: String,
+    /// Original CLI whitelist entries — stored for SIGHUP reload.
+    pub cli_white_list: Vec<String>,
+    /// Path to state.pb file for persisting VolumeServerState across restarts.
+    pub state_file_path: String,
+}
+
+impl VolumeServerState {
+    /// Check if the server is in maintenance mode; return gRPC error if so.
+    pub fn check_maintenance(&self) -> Result<(), tonic::Status> {
+        if self.maintenance.load(Ordering::Relaxed) {
+            let id = self.store.read().unwrap().id.clone();
+            return Err(tonic::Status::unavailable(format!(
+                "volume server {} is in maintenance mode",
+                id
+            )));
+        }
+        Ok(())
+    }
+
+    /// Build the seed master set from a list of raw `host:port[.grpcPort]`
+    /// addresses, normalised the same way Go's `pb.ServerAddress.ToHttpAddress`
+    /// does (drop the `.grpcPort` suffix, preserve everything else).
+    pub fn build_seed_master_set(master_urls: &[String]) -> std::collections::HashSet<String> {
+        master_urls
+            .iter()
+            .map(|m| to_http_address(m).into_owned())
+            .collect()
+    }
+
+    /// Returns true iff `target` (normalised to canonical http `host:port`)
+    /// is a master this server already knows about. Volume servers do not
+    /// keep a peer-volume or peer-filer list, so Ping is scoped to masters.
+    /// Mirrors Go's `VolumeServer.isKnownPingTarget`.
+    pub async fn is_known_ping_target(&self, target: &str, target_type: &str) -> bool {
+        if target_type != "master" {
+            return false;
+        }
+        let key = to_http_address(target).into_owned();
+        if key.is_empty() {
+            return false;
+        }
+        let current = self.current_master_url.read().await.clone();
+        if !current.is_empty() && current == key {
+            return true;
+        }
+        self.seed_master_set.contains(&key)
+    }
+}
+
+pub fn build_metrics_router() -> Router {
+    Router::new().route("/metrics", get(handlers::metrics_handler))
+}
+
+pub fn normalize_outgoing_http_url(scheme: &str, raw_target: &str) -> Result<String, String> {
+    if raw_target.starts_with("http://") || raw_target.starts_with("https://") {
+        let mut url = reqwest::Url::parse(raw_target)
+            .map_err(|e| format!("invalid url {}: {}", raw_target, e))?;
+        url.set_scheme(scheme)
+            .map_err(|_| format!("invalid scheme {}", scheme))?;
+        return Ok(url.to_string());
+    }
+    Ok(format!("{}://{}", scheme, raw_target))
+}
+
+/// Convert a SeaweedFS server address to its HTTP `host:port` form.
+///
+/// Mirrors Go's `pb.ServerAddress.ToHttpAddress()`: SeaweedFS encodes a
+/// server's gRPC port by appending `.grpcPort` to the HTTP port (e.g.
+/// `host:9333.19333`). For HTTP requests we want only the HTTP `host:port`.
+/// - `host:port.grpcPort` -> `host:port`
+/// - `host:port` -> `host:port` (unchanged)
+/// - Anything that does not look like `host:port[.grpcPort]` is returned unchanged.
+///
+/// Returns a `Cow<str>` so the common (no-suffix) case borrows from `addr`
+/// without allocating; only the rewrite branch produces a new `String`.
+pub fn to_http_address(addr: &str) -> std::borrow::Cow<'_, str> {
+    let Some(ports_sep_index) = addr.rfind(':') else {
+        return std::borrow::Cow::Borrowed(addr);
+    };
+    let ports = &addr[ports_sep_index + 1..];
+    if let Some(dot_idx) = ports.rfind('.') {
+        let http_port = &ports[..dot_idx];
+        let grpc_port = &ports[dot_idx + 1..];
+        // Only strip the suffix when both parts parse as real ports — leave
+        // anything else (e.g. "host:abc.def") untouched so bad config surfaces
+        // rather than being silently rewritten. Mirrors the validation already
+        // done in `to_grpc_address` for the inverse direction.
+        if let (Ok(_), Ok(_)) = (http_port.parse::<u16>(), grpc_port.parse::<u16>()) {
+            return std::borrow::Cow::Owned(addr[..ports_sep_index + 1 + dot_idx].to_string());
+        }
+    }
+    std::borrow::Cow::Borrowed(addr)
+}
+
+fn request_remote_addr(request: &Request) -> Option<SocketAddr> {
+    request
+        .extensions()
+        .get::<ConnectInfo<SocketAddr>>()
+        .map(|info| info.0)
+}
+
+fn request_is_whitelisted(state: &VolumeServerState, request: &Request) -> bool {
+    request_remote_addr(request)
+        .map(|remote_addr| {
+            state
+                .guard
+                .read()
+                .unwrap()
+                .check_whitelist(&remote_addr.to_string())
+        })
+        .unwrap_or(true)
+}
+
+/// Middleware: set Server header, echo x-amz-request-id, set CORS if Origin present.
+async fn common_headers_middleware(request: Request, next: Next) -> Response {
+    let origin = request.headers().get("origin").cloned();
+    let request_id = super::request_id::generate_http_request_id();
+
+    let mut response =
+        super::request_id::scope_request_id(
+            request_id.clone(),
+            async move { next.run(request).await },
+        )
+        .await;
+
+    let headers = response.headers_mut();
+    if let Ok(val) = HeaderValue::from_str(crate::version::server_header()) {
+        headers.insert("Server", val);
+    }
+
+    if let Ok(val) = HeaderValue::from_str(&request_id) {
+        headers.insert("X-Request-Id", val.clone());
+        headers.insert("x-amz-request-id", val);
+    }
+
+    if origin.is_some() {
+        headers.insert("Access-Control-Allow-Origin", HeaderValue::from_static("*"));
+        headers.insert(
+            "Access-Control-Allow-Credentials",
+            HeaderValue::from_static("true"),
+        );
+    }
+
+    response
+}
+
+/// Admin store handler — dispatches based on HTTP method.
+/// Matches Go's privateStoreHandler: GET/HEAD → read, POST/PUT → write,
+/// DELETE → delete, OPTIONS → CORS headers, anything else → 400.
+async fn admin_store_handler(state: State<Arc<VolumeServerState>>, request: Request) -> Response {
+    let start = std::time::Instant::now();
+    let method = request.method().clone();
+    let mut method_str = method.as_str().to_string();
+    let request_bytes = request
+        .headers()
+        .get(header::CONTENT_LENGTH)
+        .and_then(|value| value.to_str().ok())
+        .and_then(|value| value.parse::<i64>().ok())
+        .filter(|value| *value > 0)
+        .unwrap_or(0);
+    super::server_stats::record_request_open();
+    crate::metrics::INFLIGHT_REQUESTS_GAUGE
+        .with_label_values(&[&method_str])
+        .inc();
+    let whitelist_rejected = matches!(method, Method::POST | Method::PUT | Method::DELETE)
+        && !request_is_whitelisted(&state, &request);
+    let response = match method.clone() {
+        _ if whitelist_rejected => StatusCode::UNAUTHORIZED.into_response(),
+        Method::GET | Method::HEAD => {
+            super::server_stats::record_read_request();
+            handlers::get_or_head_handler_from_request(state, request).await
+        }
+        Method::POST | Method::PUT => {
+            super::server_stats::record_write_request();
+            if request_bytes > 0 {
+                super::server_stats::record_bytes_in(request_bytes);
+            }
+            handlers::post_handler(state, request).await
+        }
+        Method::DELETE => {
+            super::server_stats::record_delete_request();
+            handlers::delete_handler(state, request).await
+        }
+        Method::OPTIONS => {
+            super::server_stats::record_read_request();
+            admin_options_response()
+        }
+        _ => {
+            let method_name = request.method().to_string();
+            let query = request.uri().query().map(|q| q.to_string());
+            method_str = "INVALID".to_string();
+            handlers::json_error_with_query(
+                StatusCode::BAD_REQUEST,
+                format!("unsupported method {}", method_name),
+                query.as_deref(),
+            )
+        }
+    };
+    if method == Method::GET
+        && let Some(response_bytes) = response
+            .headers()
+            .get(header::CONTENT_LENGTH)
+            .and_then(|value| value.to_str().ok())
+            .and_then(|value| value.parse::<i64>().ok())
+            .filter(|value| *value > 0)
+    {
+        super::server_stats::record_bytes_out(response_bytes);
+    }
+    super::server_stats::record_request_close();
+    crate::metrics::INFLIGHT_REQUESTS_GAUGE
+        .with_label_values(&[&method_str])
+        .dec();
+    crate::metrics::REQUEST_COUNTER
+        .with_label_values(&[&method_str, response.status().as_str()])
+        .inc();
+    crate::metrics::REQUEST_DURATION
+        .with_label_values(&[&method_str])
+        .observe(start.elapsed().as_secs_f64());
+    response
+}
+
+/// Public store handler — dispatches based on HTTP method.
+/// Matches Go's publicReadOnlyHandler: GET/HEAD → read, OPTIONS → CORS,
+/// anything else → 200 (passthrough no-op).
+async fn public_store_handler(state: State<Arc<VolumeServerState>>, request: Request) -> Response {
+    let start = std::time::Instant::now();
+    let method = request.method().clone();
+    let method_str = method.as_str().to_string();
+    super::server_stats::record_request_open();
+    crate::metrics::INFLIGHT_REQUESTS_GAUGE
+        .with_label_values(&[&method_str])
+        .inc();
+    let response = match method.clone() {
+        Method::GET | Method::HEAD => {
+            super::server_stats::record_read_request();
+            handlers::get_or_head_handler_from_request(state, request).await
+        }
+        Method::OPTIONS => {
+            super::server_stats::record_read_request();
+            public_options_response()
+        }
+        _ => StatusCode::OK.into_response(),
+    };
+    if method == Method::GET
+        && let Some(response_bytes) = response
+            .headers()
+            .get(header::CONTENT_LENGTH)
+            .and_then(|value| value.to_str().ok())
+            .and_then(|value| value.parse::<i64>().ok())
+            .filter(|value| *value > 0)
+    {
+        super::server_stats::record_bytes_out(response_bytes);
+    }
+    super::server_stats::record_request_close();
+    crate::metrics::INFLIGHT_REQUESTS_GAUGE
+        .with_label_values(&[&method_str])
+        .dec();
+    crate::metrics::REQUEST_COUNTER
+        .with_label_values(&[&method_str, response.status().as_str()])
+        .inc();
+    crate::metrics::REQUEST_DURATION
+        .with_label_values(&[&method_str])
+        .observe(start.elapsed().as_secs_f64());
+    response
+}
+
+/// Build OPTIONS response for admin port.
+fn admin_options_response() -> Response {
+    let mut response = StatusCode::OK.into_response();
+    let headers = response.headers_mut();
+    headers.insert(
+        "Access-Control-Allow-Methods",
+        HeaderValue::from_static("PUT, POST, GET, DELETE, OPTIONS"),
+    );
+    headers.insert(
+        "Access-Control-Allow-Headers",
+        HeaderValue::from_static("*"),
+    );
+    response
+}
+
+/// Build OPTIONS response for public port.
+fn public_options_response() -> Response {
+    let mut response = StatusCode::OK.into_response();
+    let headers = response.headers_mut();
+    headers.insert(
+        "Access-Control-Allow-Methods",
+        HeaderValue::from_static("GET, OPTIONS"),
+    );
+    headers.insert(
+        "Access-Control-Allow-Headers",
+        HeaderValue::from_static("*"),
+    );
+    response
+}
+
+/// Build the admin (private) HTTP router — supports all operations.
+/// UI route is only registered when no signing keys are configured,
+/// matching Go's `if signingKey == "" || enableUiAccess` check.
+pub fn build_admin_router(state: Arc<VolumeServerState>) -> Router {
+    let guard = state.guard.read().unwrap();
+    // This helper can only derive the default Go behavior from the guard state:
+    // UI stays enabled when the write signing key is empty. The explicit
+    // `access.ui` override is handled by `build_admin_router_with_ui(...)`.
+    let ui_enabled = guard.signing_key.0.is_empty();
+    drop(guard);
+    build_admin_router_with_ui(state, ui_enabled)
+}
+
+/// Build the admin router with an explicit UI exposure flag.
+pub fn build_admin_router_with_ui(state: Arc<VolumeServerState>, ui_enabled: bool) -> Router {
+    let mut router = Router::new()
+        .route("/status", get(handlers::status_handler))
+        .route("/healthz", get(handlers::healthz_handler))
+        .route("/favicon.ico", get(handlers::favicon_handler))
+        .route(
+            "/seaweedfsstatic/{*path}",
+            get(handlers::static_asset_handler),
+        )
+        .route("/", any(admin_store_handler))
+        .route("/{path}", any(admin_store_handler))
+        .route("/{vid}/{fid}", any(admin_store_handler))
+        .route("/{vid}/{fid}/{filename}", any(admin_store_handler))
+        .fallback(admin_store_handler);
+    if ui_enabled {
+        // Note: /stats/* endpoints are commented out in Go's volume_server.go (L130-134).
+        // Only the UI endpoint is registered when UI access is enabled.
+        router = router.route("/ui/index.html", get(handlers::ui_handler));
+    }
+    router
+        .layer(middleware::from_fn(common_headers_middleware))
+        .with_state(state)
+}
+
+/// Build the public (read-only) HTTP router — only GET/HEAD.
+pub fn build_public_router(state: Arc<VolumeServerState>) -> Router {
+    Router::new()
+        .route("/favicon.ico", get(handlers::favicon_handler))
+        .route(
+            "/seaweedfsstatic/{*path}",
+            get(handlers::static_asset_handler),
+        )
+        .route("/", any(public_store_handler))
+        .route("/{path}", any(public_store_handler))
+        .route("/{vid}/{fid}", any(public_store_handler))
+        .route("/{vid}/{fid}/{filename}", any(public_store_handler))
+        .fallback(public_store_handler)
+        .layer(middleware::from_fn(common_headers_middleware))
+        .with_state(state)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_to_http_address_strips_grpc_port_suffix() {
+        assert_eq!(to_http_address("10.0.0.1:9333.19333"), "10.0.0.1:9333");
+        assert_eq!(
+            to_http_address("master.local:9333.19333"),
+            "master.local:9333"
+        );
+        assert_eq!(to_http_address("10.85.183.6:5300.6300"), "10.85.183.6:5300");
+    }
+
+    #[test]
+    fn test_to_http_address_passthrough_without_grpc_suffix() {
+        assert_eq!(to_http_address("10.0.0.1:9333"), "10.0.0.1:9333");
+        assert_eq!(to_http_address("master.local:9333"), "master.local:9333");
+    }
+
+    #[test]
+    fn test_to_http_address_returns_input_when_unparseable() {
+        assert_eq!(to_http_address(""), "");
+        assert_eq!(to_http_address("no-port"), "no-port");
+        // Trailing colon: nothing after the separator, treat as unparseable.
+        assert_eq!(to_http_address("host:"), "host:");
+    }
+
+    #[test]
+    fn test_to_http_address_borrows_when_unchanged_and_owns_when_stripped() {
+        // The common case (no suffix) must not allocate.
+        let result = to_http_address("10.0.0.1:9333");
+        assert!(matches!(result, std::borrow::Cow::Borrowed(_)));
+        // Stripping requires a new string.
+        let result = to_http_address("10.0.0.1:9333.19333");
+        assert!(matches!(result, std::borrow::Cow::Owned(_)));
+        assert_eq!(result, "10.0.0.1:9333");
+        // Unparseable / passthrough also borrows.
+        let result = to_http_address("host:abc.def");
+        assert!(matches!(result, std::borrow::Cow::Borrowed(_)));
+    }
+
+    #[test]
+    fn test_to_http_address_keeps_non_numeric_dotted_suffix() {
+        // The dotted form is only valid when both sides are real port numbers.
+        // Otherwise the address is malformed config (e.g. a hostname like
+        // "host:abc.def"), and silently rewriting it would just hide the bug.
+        assert_eq!(to_http_address("host:abc.def"), "host:abc.def");
+        assert_eq!(to_http_address("host:9333.notaport"), "host:9333.notaport");
+        assert_eq!(
+            to_http_address("host:notaport.19333"),
+            "host:notaport.19333"
+        );
+        // Out-of-range ports must not be silently truncated either.
+        assert_eq!(to_http_address("host:99999.19333"), "host:99999.19333");
+    }
+
+    #[test]
+    fn test_to_http_address_handles_bracketed_ipv6_literals() {
+        // The function uses `rfind(':')`, so for bracketed IPv6 the port
+        // separator is correctly identified as the colon AFTER the closing
+        // bracket — making IPv4 and IPv6 behave the same.
+        assert_eq!(to_http_address("[::1]:9333.19333"), "[::1]:9333");
+        assert_eq!(
+            to_http_address("[2001:db8::10]:5300.6300"),
+            "[2001:db8::10]:5300"
+        );
+        // Plain bracketed IPv6 without a dotted suffix is borrowed unchanged.
+        let result = to_http_address("[2001:db8::1]:9333");
+        assert!(matches!(result, std::borrow::Cow::Borrowed(_)));
+        assert_eq!(result, "[2001:db8::1]:9333");
+        // Non-numeric / out-of-range / missing suffix all preserve the input.
+        assert_eq!(to_http_address("[::1]:"), "[::1]:");
+        assert_eq!(to_http_address("[::1]:abc.def"), "[::1]:abc.def");
+        assert_eq!(to_http_address("[::1]:99999.19333"), "[::1]:99999.19333");
+    }
+}

seaweed-volume/src/server/write_queue.rs

@@ -0,0 +1,339 @@
+//! Async batched write processing for the volume server.
+//!
+//! Instead of each upload handler directly calling `write_needle`, writes are
+//! submitted to a queue. A background worker drains the queue in batches (up to
+//! 128 entries), groups them by volume ID, and processes them together under a
+//! single store lock. Requests that asked for `fsync` are flushed by
+//! `write_needle` itself, one flush per durable write.
+
+use std::sync::Arc;
+
+use tokio::sync::{mpsc, oneshot};
+use tracing::debug;
+
+use crate::storage::needle::needle::Needle;
+use crate::storage::types::{Size, VolumeId};
+use crate::storage::volume::VolumeError;
+
+use super::volume_server::VolumeServerState;
+
+/// Result of a single write operation: (offset, size, is_unchanged).
+pub type WriteResult = Result<(u64, Size, bool), VolumeError>;
+
+/// The needles queued for one volume, each with the durability it asked for.
+type VolumeBatch = Vec<(Needle, bool, oneshot::Sender<WriteResult>)>;
+
+/// A request to write a needle, submitted to the write queue.
+pub struct WriteRequest {
+    pub volume_id: VolumeId,
+    pub needle: Needle,
+    pub fsync: bool,
+    pub response_tx: oneshot::Sender<WriteResult>,
+}
+
+/// Maximum number of write requests to batch together.
+const MAX_BATCH_SIZE: usize = 128;
+
+/// Maximum bytes to accumulate per batch before breaking (matches Go's 4MB limit).
+/// This prevents large writes from accumulating unbounded latency.
+const MAX_BATCH_BYTES: usize = 4 * 1024 * 1024;
+
+/// Handle for submitting write requests to the background worker.
+#[derive(Clone)]
+pub struct WriteQueue {
+    tx: mpsc::Sender<WriteRequest>,
+}
+
+impl WriteQueue {
+    /// Create a new write queue and spawn the background worker.
+    ///
+    /// `capacity` controls the channel buffer size (backpressure kicks in when full).
+    /// The worker holds a reference to `state` for accessing the store.
+    pub fn new(state: Arc<VolumeServerState>, capacity: usize) -> Self {
+        let (tx, rx) = mpsc::channel(capacity);
+        let worker = WriteQueueWorker { rx, state };
+        tokio::spawn(worker.run());
+        WriteQueue { tx }
+    }
+
+    /// Submit a write request and wait for the result.
+    ///
+    /// Returns `Err` if the worker has shut down or the response channel was dropped.
+    pub async fn submit(&self, volume_id: VolumeId, needle: Needle, fsync: bool) -> WriteResult {
+        let (response_tx, response_rx) = oneshot::channel();
+        let request = WriteRequest {
+            volume_id,
+            needle,
+            fsync,
+            response_tx,
+        };
+
+        // Send to queue; this awaits if the channel is full (backpressure).
+        if self.tx.send(request).await.is_err() {
+            return Err(VolumeError::Io(std::io::Error::new(
+                std::io::ErrorKind::BrokenPipe,
+                "write queue worker has shut down",
+            )));
+        }
+
+        // Wait for the worker to process our request.
+        match response_rx.await {
+            Ok(result) => result,
+            Err(_) => Err(VolumeError::Io(std::io::Error::new(
+                std::io::ErrorKind::BrokenPipe,
+                "write queue worker dropped response channel",
+            ))),
+        }
+    }
+}
+
+/// Background worker that drains write requests and processes them in batches.
+struct WriteQueueWorker {
+    rx: mpsc::Receiver<WriteRequest>,
+    state: Arc<VolumeServerState>,
+}
+
+impl WriteQueueWorker {
+    async fn run(mut self) {
+        debug!("write queue worker started");
+
+        loop {
+            // Wait for the first request (blocks until one arrives or channel closes).
+            let first = match self.rx.recv().await {
+                Some(req) => req,
+                None => {
+                    debug!("write queue channel closed, worker exiting");
+                    return;
+                }
+            };
+
+            // Drain as many additional requests as available, up to MAX_BATCH_SIZE
+            // or MAX_BATCH_BYTES (matches Go: 128 requests or 4MB, whichever comes first).
+            let mut batch = Vec::with_capacity(MAX_BATCH_SIZE);
+            let mut batch_bytes: usize = first.needle.data.len();
+            batch.push(first);
+
+            while batch.len() < MAX_BATCH_SIZE && batch_bytes < MAX_BATCH_BYTES {
+                match self.rx.try_recv() {
+                    Ok(req) => {
+                        batch_bytes += req.needle.data.len();
+                        batch.push(req);
+                    }
+                    Err(_) => break,
+                }
+            }
+
+            let batch_size = batch.len();
+            debug!("processing write batch of {} requests", batch_size);
+
+            // Process the batch in spawn_blocking since write_needle does file I/O.
+            let state = self.state.clone();
+            let _ = tokio::task::spawn_blocking(move || {
+                process_batch(state, batch);
+            })
+            .await;
+        }
+    }
+}
+
+/// Process a batch of write requests, grouped by volume ID.
+///
+/// Groups writes by volume to minimize the number of store lock acquisitions,
+/// then sends results back via each request's oneshot channel.
+fn process_batch(state: Arc<VolumeServerState>, batch: Vec<WriteRequest>) {
+    // Group requests by volume ID for efficient processing.
+    // We use a Vec of (VolumeId, Vec<(Needle, Sender)>) to preserve order
+    // and avoid requiring Hash on VolumeId.
+    let mut groups: Vec<(VolumeId, VolumeBatch)> = Vec::new();
+
+    for req in batch {
+        let vid = req.volume_id;
+        if let Some(group) = groups.iter_mut().find(|(v, _)| *v == vid) {
+            group.1.push((req.needle, req.fsync, req.response_tx));
+        } else {
+            groups.push((vid, vec![(req.needle, req.fsync, req.response_tx)]));
+        }
+    }
+
+    // Process each volume group under a single store lock.
+    let mut store = state.store.write().unwrap();
+
+    for (vid, entries) in groups {
+        for (mut needle, fsync, response_tx) in entries {
+            let result = store.write_volume_needle(vid, &mut needle, fsync);
+            // Send result back; ignore error if receiver dropped.
+            let _ = response_tx.send(result);
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::storage::types::VolumeId;
+
+    /// Helper to create a minimal VolumeServerState for testing.
+    fn make_test_state() -> Arc<VolumeServerState> {
+        use crate::security::{Guard, SigningKey};
+        use crate::server::volume_server::RuntimeMetricsConfig;
+        use crate::storage::needle_map::NeedleMapKind;
+        use crate::storage::store::Store;
+        use std::sync::RwLock;
+        use std::sync::atomic::{AtomicBool, AtomicI64, AtomicU32};
+
+        let store = Store::new(NeedleMapKind::InMemory);
+        let guard = Guard::new(&[], SigningKey(vec![]), 0, SigningKey(vec![]), 0);
+
+        Arc::new(VolumeServerState {
+            store: RwLock::new(store),
+            guard: RwLock::new(guard),
+            is_stopping: RwLock::new(false),
+            maintenance: AtomicBool::new(false),
+            state_version: AtomicU32::new(0),
+            concurrent_upload_limit: 0,
+            concurrent_download_limit: 0,
+            inflight_upload_data_timeout: std::time::Duration::ZERO,
+            inflight_download_data_timeout: std::time::Duration::ZERO,
+            inflight_upload_bytes: AtomicI64::new(0),
+            inflight_download_bytes: AtomicI64::new(0),
+            upload_notify: tokio::sync::Notify::new(),
+            download_notify: tokio::sync::Notify::new(),
+            data_center: String::new(),
+            rack: String::new(),
+            file_size_limit_bytes: 0,
+            maintenance_byte_per_second: 0,
+            is_heartbeating: AtomicBool::new(false),
+            has_master: false,
+            pre_stop_seconds: 0,
+            volume_state_notify: tokio::sync::Notify::new(),
+            write_queue: std::sync::OnceLock::new(),
+            s3_tier_registry: std::sync::RwLock::new(
+                crate::remote_storage::s3_tier::S3TierRegistry::new(),
+            ),
+            read_mode: crate::config::ReadMode::Local,
+            allow_untrusted_remote_endpoints: false,
+            master_url: String::new(),
+            master_urls: Vec::new(),
+            seed_master_set: std::collections::HashSet::new(),
+            current_master_url: tokio::sync::RwLock::new(String::new()),
+            self_url: String::new(),
+            http_client: reqwest::Client::new(),
+            outgoing_http_scheme: "http".to_string(),
+            outgoing_grpc_tls: None,
+            metrics_runtime: std::sync::RwLock::new(RuntimeMetricsConfig::default()),
+            metrics_notify: tokio::sync::Notify::new(),
+            fix_jpg_orientation: false,
+            has_slow_read: true,
+            read_buffer_size_bytes: 4 * 1024 * 1024,
+            security_file: String::new(),
+            cli_white_list: vec![],
+            state_file_path: String::new(),
+        })
+    }
+
+    #[tokio::test]
+    async fn test_write_queue_submit_no_volume() {
+        // Submit a write to a non-existent volume -- should return VolumeError::NotFound.
+        let state = make_test_state();
+        let queue = WriteQueue::new(state, MAX_BATCH_SIZE);
+
+        let needle = Needle {
+            id: 1.into(),
+            cookie: 0x12345678.into(),
+            data: vec![1, 2, 3],
+            data_size: 3,
+            ..Needle::default()
+        };
+
+        let result = queue.submit(VolumeId(999), needle, false).await;
+        assert!(result.is_err());
+        match result {
+            Err(VolumeError::NotFound) => {} // expected
+            other => panic!("expected NotFound, got {:?}", other),
+        }
+    }
+
+    #[tokio::test]
+    async fn test_write_queue_concurrent_submissions() {
+        // Submit multiple concurrent writes -- all should complete (with errors since no volume).
+        let state = make_test_state();
+        let queue = WriteQueue::new(state, MAX_BATCH_SIZE);
+
+        let mut handles = Vec::new();
+        for i in 0..10u64 {
+            let q = queue.clone();
+            handles.push(tokio::spawn(async move {
+                let needle = Needle {
+                    id: i.into(),
+                    cookie: 0xABCD.into(),
+                    data: vec![i as u8; 10],
+                    data_size: 10,
+                    ..Needle::default()
+                };
+                q.submit(VolumeId(1), needle, false).await
+            }));
+        }
+
+        for handle in handles {
+            let result = handle.await.unwrap();
+            // All should fail with NotFound since there's no volume 1
+            assert!(matches!(result, Err(VolumeError::NotFound)));
+        }
+    }
+
+    #[tokio::test]
+    async fn test_write_queue_batching() {
+        // Verify that many concurrent writes get processed (testing the batching path).
+        let state = make_test_state();
+        let queue = WriteQueue::new(state, MAX_BATCH_SIZE);
+
+        // Submit MAX_BATCH_SIZE requests concurrently
+        let mut handles = Vec::new();
+        for i in 0..MAX_BATCH_SIZE as u64 {
+            let q = queue.clone();
+            handles.push(tokio::spawn(async move {
+                let needle = Needle {
+                    id: i.into(),
+                    cookie: 0x1111.into(),
+                    data: vec![0u8; 4],
+                    data_size: 4,
+                    ..Needle::default()
+                };
+                q.submit(VolumeId(42), needle, false).await
+            }));
+        }
+
+        let mut results = Vec::new();
+        for handle in handles {
+            results.push(handle.await.unwrap());
+        }
+
+        // All should complete (with NotFound errors since no volume exists)
+        assert_eq!(results.len(), MAX_BATCH_SIZE);
+        for r in results {
+            assert!(matches!(r, Err(VolumeError::NotFound)));
+        }
+    }
+
+    #[tokio::test]
+    async fn test_write_queue_dropped_sender() {
+        // When the queue is dropped, subsequent submits should fail gracefully.
+        let state = make_test_state();
+        let queue = WriteQueue::new(state, 1);
+
+        // Clone then drop the original -- the worker keeps running via its rx handle.
+        let queue2 = queue.clone();
+        drop(queue);
+
+        // This should still work since the worker is alive.
+        let needle = Needle {
+            id: 1.into(),
+            cookie: 0.into(),
+            data: vec![],
+            data_size: 0,
+            ..Needle::default()
+        };
+        let result = queue2.submit(VolumeId(1), needle, false).await;
+        assert!(result.is_err()); // NotFound is fine -- the point is it doesn't panic
+    }
+}

seaweed-volume/src/storage/erasure_coding/ec_bitrot.rs

@@ -0,0 +1,961 @@
+//! EC bitrot detection — checksum sidecar.
+//!
+//! A per-volume sidecar file stores a CRC32C (Castagnoli) checksum for every
+//! fixed-size block of every EC shard, so a scrub (and the reconstruction path)
+//! can detect silent disk corruption in any shard — including cold parity shards
+//! that are never read during normal serving.
+//!
+//! The sidecar is OPTIONAL: an absent or generation-mismatched sidecar simply
+//! means "feature off" for that generation, so old binaries, JSON-only nodes,
+//! and rollback deployments ignore it and degrade gracefully.
+//!
+//! On-disk layout of `<base>.ecsum` (legacy/generation 0) and `<base>.ecsum.v<N>`:
+//!
+//! ```text
+//! [ magic(4) | format_version(2) | payload_len(4) | payload_crc32c(4) ] [ proto payload ]
+//! ```
+//!
+//! All header fields are BIG-ENDIAN. The header's `payload_crc32c` lets a loader
+//! detect corruption of the sidecar itself BEFORE trusting any contents, so a
+//! rotted sidecar can never be mistaken for shard corruption. This format is
+//! byte-identical to the Go implementation in
+//! `weed/storage/erasure_coding/ec_bitrot.go`.
+
+use std::fs::File;
+use std::io::{self, Read, Write};
+
+use prost::Message;
+
+use crate::pb::volume_server_pb::{
+    ChecksumAlgorithm, EcBitrotProtection, EcShardChecksums, EcShardConfig,
+};
+use crate::storage::erasure_coding::ec_shard::MAX_SHARD_COUNT;
+use crate::storage::needle::crc::CRC;
+
+/// Canonical extension for the checksum sidecar. Generation 0 (legacy/fresh
+/// encode) uses `<base>.ecsum`; vacuum generation N uses `<base>.ecsum.v<N>`,
+/// mirroring the `.vif`/`.ecx` versioned convention.
+pub const BITROT_SIDECAR_EXT: &str = ".ecsum";
+
+/// Default checksum granularity (16 MiB). It is a power-of-two multiple of
+/// `ERASURE_CODING_SMALL_BLOCK_SIZE` (1 MiB) and keeps the sidecar tiny
+/// (~11 KB for a 30 GB volume) while localizing corruption to a 16 MiB region.
+pub const DEFAULT_BITROT_BLOCK_SIZE: usize = 16 * 1024 * 1024;
+
+/// Caps the block granularity so a loaded sidecar cannot force a huge
+/// scrub/verify scratch buffer. Power-of-two multiple of 1 MiB.
+pub const MAX_BITROT_BLOCK_SIZE: u32 = 64 * 1024 * 1024;
+
+/// Magic "ECSU".
+pub const BITROT_MAGIC: u32 = 0x4543_5355;
+/// On-disk format version.
+pub const BITROT_FORMAT_VERSION: u16 = 1;
+/// Header size: magic(4) + version(2) + payload_len(4) + payload_crc32c(4).
+pub const BITROT_HEADER_SIZE: usize = 14;
+
+/// Resolved protection state of an EC volume's active generation after loading
+/// and validating its sidecar.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum BitrotStatus {
+    /// No sidecar, or a sidecar that does not describe the active generation.
+    /// The generation is unprotected; this is NOT corruption.
+    Off,
+    /// A complete, well-formed, generation-matching sidecar is loaded.
+    On,
+    /// A generation-matching sidecar that is malformed, incomplete, or
+    /// self-integrity-failed. The generation is unprotected pending repair of
+    /// the sidecar, and an integrity alarm should fire. The rebuild path treats
+    /// this as fail-closed.
+    Invalid,
+}
+
+impl std::fmt::Display for BitrotStatus {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        let s = match self {
+            BitrotStatus::On => "on",
+            BitrotStatus::Invalid => "invalid",
+            BitrotStatus::Off => "off",
+        };
+        f.write_str(s)
+    }
+}
+
+/// Error returned by [`load_bitrot_sidecar`]. A self-integrity failure is a
+/// sidecar-integrity problem (the caller maps it to [`BitrotStatus::Invalid`]),
+/// never a shard-corruption signal.
+#[derive(Debug)]
+pub enum BitrotLoadError {
+    /// File missing or other underlying I/O error.
+    Io(io::Error),
+    /// File shorter than the fixed header.
+    TooShort(usize),
+    /// Header magic did not match.
+    BadMagic(u32),
+    /// Header format version is unsupported.
+    UnsupportedVersion(u16),
+    /// Header payload_len disagrees with the actual payload length.
+    LengthMismatch { header: u32, actual: usize },
+    /// Header payload_crc32c disagrees with the computed CRC32C of the payload.
+    CrcMismatch { header: u32, computed: u32 },
+    /// Protobuf payload failed to decode.
+    Decode(prost::DecodeError),
+}
+
+impl std::fmt::Display for BitrotLoadError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            BitrotLoadError::Io(e) => write!(f, "bitrot sidecar io error: {}", e),
+            BitrotLoadError::TooShort(n) => {
+                write!(f, "bitrot sidecar too short ({} bytes)", n)
+            }
+            BitrotLoadError::BadMagic(m) => write!(f, "bitrot sidecar bad magic {:#x}", m),
+            BitrotLoadError::UnsupportedVersion(v) => {
+                write!(f, "bitrot sidecar unsupported format version {}", v)
+            }
+            BitrotLoadError::LengthMismatch { header, actual } => write!(
+                f,
+                "bitrot sidecar length mismatch: header {}, actual {}",
+                header, actual
+            ),
+            BitrotLoadError::CrcMismatch { header, computed } => write!(
+                f,
+                "bitrot sidecar self-integrity CRC mismatch: header {:#x}, computed {:#x}",
+                header, computed
+            ),
+            BitrotLoadError::Decode(e) => write!(f, "unmarshal bitrot sidecar: {}", e),
+        }
+    }
+}
+
+impl std::error::Error for BitrotLoadError {}
+
+impl From<io::Error> for BitrotLoadError {
+    fn from(e: io::Error) -> Self {
+        BitrotLoadError::Io(e)
+    }
+}
+
+/// Returns the sidecar path for a base file name and EC generation. Generation
+/// 0 is the un-suffixed legacy path; generation N>0 is the versioned path,
+/// consistent with how `.vif`/`.ecx` are versioned by the 2PC switch.
+pub fn bitrot_sidecar_path(base: &str, generation: u32) -> String {
+    if generation == 0 {
+        format!("{}{}", base, BITROT_SIDECAR_EXT)
+    } else {
+        format!("{}{}.v{}", base, BITROT_SIDECAR_EXT, generation)
+    }
+}
+
+/// Remove the bitrot checksum sidecars for a base file name: the legacy
+/// `<base>.ecsum` (generation 0) and any versioned `<base>.ecsum.v<N>`.
+///
+/// Already-gone is success; returns the first real removal failure (and surfaces
+/// a directory-scan error) so a stale sidecar left behind is not reported as
+/// cleaned. Mirrors Go's `RemoveBitrotSidecars` / `removeBitrotSidecars`.
+pub fn remove_bitrot_sidecars(base: &str) -> io::Result<()> {
+    use std::fs;
+    use std::path::Path;
+
+    let rm = |path: std::path::PathBuf| -> io::Result<()> {
+        match fs::remove_file(&path) {
+            Err(e) if e.kind() != io::ErrorKind::NotFound => Err(e),
+            _ => Ok(()),
+        }
+    };
+    let mut first_err: Option<io::Error> = None;
+    let mut record = |res: io::Result<()>| {
+        if let Err(e) = res
+            && first_err.is_none()
+        {
+            first_err = Some(e);
+        }
+    };
+    record(rm(format!("{}{}", base, BITROT_SIDECAR_EXT).into()));
+    let path = Path::new(base);
+    if let (Some(parent), Some(fname)) = (path.parent(), path.file_name()) {
+        let prefix = format!("{}{}.v", fname.display(), BITROT_SIDECAR_EXT);
+        match fs::read_dir(parent) {
+            Ok(entries) => {
+                for entry in entries.flatten() {
+                    if entry.file_name().to_string_lossy().starts_with(&prefix) {
+                        record(rm(entry.path()));
+                    }
+                }
+            }
+            Err(e) if e.kind() != io::ErrorKind::NotFound => record(Err(e)),
+            Err(_) => {}
+        }
+    }
+    match first_err {
+        Some(e) => Err(e),
+        None => Ok(()),
+    }
+}
+
+/// Returns a fresh random per-encode identity used to detect a stale sidecar
+/// left behind by an in-place re-encode.
+pub fn new_encode_uuid() -> Vec<u8> {
+    use rand::Rng;
+    let mut b = vec![0u8; 16];
+    rand::rng().fill_bytes(&mut b);
+    b
+}
+
+/// Reports whether `block_size` is a power of two in [1 MiB, MAX_BITROT_BLOCK_SIZE].
+pub fn is_pow2_multiple_of_1mib(block_size: u32) -> bool {
+    ((1 << 20)..=MAX_BITROT_BLOCK_SIZE).contains(&block_size) && block_size.count_ones() == 1
+}
+
+/// Returns ceil(covered_size / block_size).
+fn expected_block_count(covered_size: i64, block_size: i64) -> usize {
+    if block_size <= 0 {
+        return 0;
+    }
+    ((covered_size + block_size - 1) / block_size) as usize
+}
+
+/// Packs a slice of u32 into little-endian bytes.
+fn pack_u32_le(vals: &[u32]) -> Vec<u8> {
+    let mut out = Vec::with_capacity(vals.len() * 4);
+    for v in vals {
+        out.extend_from_slice(&v.to_le_bytes());
+    }
+    out
+}
+
+/// Unpacks little-endian bytes into a Vec<u32>. Trailing bytes (len % 4) are
+/// ignored, matching the Go implementation's `len(b)/4` truncation.
+fn unpack_u32_le(b: &[u8]) -> Vec<u32> {
+    let n = b.len() / 4;
+    let mut out = Vec::with_capacity(n);
+    for i in 0..n {
+        out.push(u32::from_le_bytes([
+            b[i * 4],
+            b[i * 4 + 1],
+            b[i * 4 + 2],
+            b[i * 4 + 3],
+        ]));
+    }
+    out
+}
+
+/// Accumulates the per-block CRC32C of a single shard's byte stream as it is
+/// written. Tolerates arbitrary chunk sizes that cross block boundaries, so it
+/// works for both small encode buffers and the larger rebuild buffers.
+pub struct ShardChecksumBuilder {
+    block_size: i64,
+    cur: CRC,
+    cur_len: i64,
+    total: i64,
+    blocks: Vec<u32>,
+}
+
+impl ShardChecksumBuilder {
+    /// Creates a builder over a given block size in bytes.
+    pub fn new(block_size: i64) -> Self {
+        ShardChecksumBuilder {
+            block_size,
+            cur: CRC(0),
+            cur_len: 0,
+            total: 0,
+            blocks: Vec::new(),
+        }
+    }
+
+    /// Feeds a chunk of arbitrary size, splitting it across block boundaries.
+    pub fn write(&mut self, mut p: &[u8]) {
+        while !p.is_empty() {
+            let room = self.block_size - self.cur_len;
+            let n = (p.len() as i64).min(room) as usize;
+            self.cur = self.cur.update(&p[..n]);
+            self.cur_len += n as i64;
+            self.total += n as i64;
+            p = &p[n..];
+            if self.cur_len == self.block_size {
+                self.blocks.push(self.cur.0);
+                self.cur = CRC(0);
+                self.cur_len = 0;
+            }
+        }
+    }
+
+    /// Flushes any partial last block and returns the covered size and the
+    /// packed little-endian u32 CRC array.
+    pub fn finalize(mut self) -> (i64, Vec<u8>) {
+        if self.cur_len > 0 {
+            self.blocks.push(self.cur.0);
+            self.cur = CRC(0);
+            self.cur_len = 0;
+        }
+        (self.total, pack_u32_le(&self.blocks))
+    }
+}
+
+/// Atomically writes `prot` to `path`, wrapped in the on-disk header with a
+/// CRC32C over the serialized payload (temp file + rename).
+pub fn save_bitrot_sidecar(path: &str, prot: &EcBitrotProtection) -> io::Result<()> {
+    // The header records payload_len as a uint32 and the buffer allocation below
+    // adds it to a constant. Bound the payload well under any overflow (a real
+    // manifest is a few KB) so neither the length field nor the buffer can wrap.
+    // Check the encoded length BEFORE serializing so an oversized manifest never
+    // allocates a huge buffer. Mirrors Go's SaveBitrotSidecar maxBitrotPayloadSize.
+    const MAX_BITROT_PAYLOAD_SIZE: usize = 1 << 30; // 1 GiB, vastly above any real sidecar
+    let payload_len = prot.encoded_len();
+    if payload_len > MAX_BITROT_PAYLOAD_SIZE {
+        return Err(io::Error::new(
+            io::ErrorKind::InvalidData,
+            format!("bitrot sidecar payload too large: {} bytes", payload_len),
+        ));
+    }
+    let payload = prot.encode_to_vec();
+    let mut buf = Vec::with_capacity(BITROT_HEADER_SIZE + payload.len());
+    buf.extend_from_slice(&BITROT_MAGIC.to_be_bytes());
+    buf.extend_from_slice(&BITROT_FORMAT_VERSION.to_be_bytes());
+    buf.extend_from_slice(&(payload.len() as u32).to_be_bytes());
+    buf.extend_from_slice(&CRC::new(&payload).0.to_be_bytes());
+    buf.extend_from_slice(&payload);
+
+    let tmp = format!("{}.tmp", path);
+    {
+        let mut f = File::create(&tmp)?;
+        f.write_all(&buf)?;
+        f.sync_all()?;
+    }
+    if let Err(e) = std::fs::rename(&tmp, path) {
+        let _ = std::fs::remove_file(&tmp);
+        return Err(e);
+    }
+    Ok(())
+}
+
+/// Reads and self-integrity-checks a sidecar file. Returns the parsed message,
+/// or an error if the file is missing, truncated, has a bad magic/version, or
+/// fails the payload CRC. A self-integrity failure is a sidecar-integrity
+/// problem (caller maps it to [`BitrotStatus::Invalid`]), never a shard
+/// corruption signal.
+pub fn load_bitrot_sidecar(path: &str) -> Result<EcBitrotProtection, BitrotLoadError> {
+    let mut data = Vec::new();
+    File::open(path)?.read_to_end(&mut data)?;
+    if data.len() < BITROT_HEADER_SIZE {
+        return Err(BitrotLoadError::TooShort(data.len()));
+    }
+    let magic = u32::from_be_bytes([data[0], data[1], data[2], data[3]]);
+    if magic != BITROT_MAGIC {
+        return Err(BitrotLoadError::BadMagic(magic));
+    }
+    let ver = u16::from_be_bytes([data[4], data[5]]);
+    if ver != BITROT_FORMAT_VERSION {
+        return Err(BitrotLoadError::UnsupportedVersion(ver));
+    }
+    let payload_len = u32::from_be_bytes([data[6], data[7], data[8], data[9]]);
+    let want_crc = u32::from_be_bytes([data[10], data[11], data[12], data[13]]);
+    let payload = &data[BITROT_HEADER_SIZE..];
+    if payload_len as usize != payload.len() {
+        return Err(BitrotLoadError::LengthMismatch {
+            header: payload_len,
+            actual: payload.len(),
+        });
+    }
+    let got = CRC::new(payload).0;
+    if got != want_crc {
+        return Err(BitrotLoadError::CrcMismatch {
+            header: want_crc,
+            computed: got,
+        });
+    }
+    EcBitrotProtection::decode(payload).map_err(BitrotLoadError::Decode)
+}
+
+/// Performs the disk-free manifest/syntax checks that every loader runs:
+/// supported algorithm, valid block size, exactly one entry per shard id in the
+/// active layout (no duplicates, no out-of-range ids), positive covered_size,
+/// and a packed-CRC count consistent with covered_size.
+///
+/// It does NOT compare covered_size against on-disk shard lengths — that is a
+/// per-node physical check done only for locally-held shards.
+pub fn validate_manifest(
+    prot: &EcBitrotProtection,
+    data_shards: usize,
+    parity_shards: usize,
+) -> Result<(), String> {
+    if prot.algorithm != ChecksumAlgorithm::ChecksumCrc32c as i32 {
+        return Err(format!("unsupported checksum algorithm {}", prot.algorithm));
+    }
+    if !is_pow2_multiple_of_1mib(prot.block_size) {
+        return Err(format!(
+            "invalid block_size {} (must be a power-of-two multiple of 1 MiB, at most {})",
+            prot.block_size, MAX_BITROT_BLOCK_SIZE
+        ));
+    }
+    let bs = prot.block_size as i64;
+    let total = data_shards + parity_shards;
+    if total == 0 || total > MAX_SHARD_COUNT {
+        return Err(format!(
+            "invalid active layout: data={} parity={}",
+            data_shards, parity_shards
+        ));
+    }
+    if prot.shards.len() != total {
+        return Err(format!(
+            "incomplete manifest: {} shard entries, expected {}",
+            prot.shards.len(),
+            total
+        ));
+    }
+    let mut seen = [false; MAX_SHARD_COUNT];
+    for s in &prot.shards {
+        if s.shard_id >= total as u32 {
+            return Err(format!(
+                "shard id {} out of range [0,{})",
+                s.shard_id, total
+            ));
+        }
+        if seen[s.shard_id as usize] {
+            return Err(format!("duplicate shard id {}", s.shard_id));
+        }
+        seen[s.shard_id as usize] = true;
+        if s.covered_size <= 0 {
+            return Err(format!(
+                "shard {} has non-positive covered_size {}",
+                s.shard_id, s.covered_size
+            ));
+        }
+        let want_count = expected_block_count(s.covered_size, bs);
+        if s.block_crc32c.len() != want_count * 4 {
+            return Err(format!(
+                "shard {} crc count mismatch: {} bytes, expected {} (covered_size={} block_size={})",
+                s.shard_id,
+                s.block_crc32c.len(),
+                want_count * 4,
+                s.covered_size,
+                prot.block_size
+            ));
+        }
+    }
+    Ok(())
+}
+
+/// Resolves the protection status of a loaded-or-missing sidecar against the
+/// active generation and layout. `loaded` is the result of attempting to load
+/// the sidecar at the active generation's path.
+///
+/// - missing sidecar (NotFound) => [`BitrotStatus::Off`]
+/// - load/self-integrity failure => [`BitrotStatus::Invalid`]
+/// - generation mismatch => [`BitrotStatus::Off`]
+/// - manifest validation failure => [`BitrotStatus::Invalid`]
+/// - otherwise => [`BitrotStatus::On`]
+pub fn resolve_status(
+    loaded: &Result<EcBitrotProtection, BitrotLoadError>,
+    active_generation: u32,
+    data_shards: usize,
+    parity_shards: usize,
+) -> BitrotStatus {
+    match loaded {
+        Err(BitrotLoadError::Io(e)) if e.kind() == io::ErrorKind::NotFound => BitrotStatus::Off,
+        Err(_) => BitrotStatus::Invalid,
+        Ok(prot) => {
+            if prot.generation != active_generation {
+                return BitrotStatus::Off;
+            }
+            if validate_manifest(prot, data_shards, parity_shards).is_err() {
+                return BitrotStatus::Invalid;
+            }
+            BitrotStatus::On
+        }
+    }
+}
+
+/// Whether a generation-matching sidecar agrees with the geometry the volume is
+/// mounted with. Both files record the layout the generation was encoded with,
+/// so a disagreement means one of them is wrong and reads through the other
+/// would land at the wrong shard offsets — the caller fails the mount rather
+/// than merely dropping protection. A sidecar that records no EC config has
+/// nothing to contradict.
+pub fn geometry_matches(
+    prot: &EcBitrotProtection,
+    data_shards: usize,
+    parity_shards: usize,
+    block_size: i64,
+) -> bool {
+    match &prot.ec_shard_config {
+        None => true,
+        Some(cfg) => {
+            cfg.data_shards as usize == data_shards
+                && cfg.parity_shards as usize == parity_shards
+                && cfg.block_size == block_size
+        }
+    }
+}
+
+/// Returns the [`EcShardChecksums`] entry for a shard id, or `None`.
+pub fn shard_checksums(prot: &EcBitrotProtection, shard_id: u32) -> Option<&EcShardChecksums> {
+    prot.shards.iter().find(|s| s.shard_id == shard_id)
+}
+
+/// Reads a shard file at `path` in `block_size` chunks and compares each block's
+/// CRC32C against the manifest entry. Returns the list of mismatching block
+/// indices (empty == clean), or a fatal `io::Error` for genuine I/O problems.
+///
+/// A length mismatch (truncation or unexpected trailing bytes) is itself shard
+/// corruption: every block index is reported as mismatched so the caller treats
+/// the shard as bad. It does not interpret the result — the caller (scrub /
+/// rebuild) arbitrates shard-vs-sidecar via Reed-Solomon before acting.
+pub fn verify_shard_file_blocks(
+    path: &str,
+    entry: &EcShardChecksums,
+    block_size: i64,
+) -> io::Result<Vec<usize>> {
+    verify_shard_blocks(&File::open(path)?, entry, block_size)
+}
+
+/// Same verification against an ALREADY-OPEN shard handle.
+///
+/// Go's `ChecksumScrub` reads through `shard.ReadAt`, i.e. the handle the
+/// EcVolumeShard already holds, so a concurrent teardown that unlinks the shard
+/// cannot turn an intentional removal into a scrub read error. A scrub that
+/// runs with the store lock released has to read the same way — see
+/// `EcChecksumScrubPlan`.
+pub fn verify_shard_blocks(
+    f: &File,
+    entry: &EcShardChecksums,
+    block_size: i64,
+) -> io::Result<Vec<usize>> {
+    let file_size = f.metadata()?.len() as i64;
+    let want = unpack_u32_le(&entry.block_crc32c);
+
+    if file_size != entry.covered_size {
+        // Length drift is shard corruption: report every block as mismatched.
+        return Ok((0..want.len()).collect());
+    }
+
+    let mut mismatched = Vec::new();
+    let mut buf = vec![0u8; block_size.max(1) as usize];
+    let mut offset: i64 = 0;
+    for (i, want_crc) in want.iter().enumerate() {
+        let to_read = (entry.covered_size - offset).min(block_size);
+        if to_read <= 0 {
+            break;
+        }
+        let to_read = to_read as usize;
+        read_full_at(f, &mut buf[..to_read], offset as u64)?;
+        if CRC::new(&buf[..to_read]).0 != *want_crc {
+            mismatched.push(i);
+        }
+        offset += to_read as i64;
+    }
+    Ok(mismatched)
+}
+
+/// Reads exactly `buf.len()` bytes from `f` at `offset`, erroring on early EOF.
+fn read_full_at(f: &File, buf: &mut [u8], offset: u64) -> io::Result<()> {
+    let mut total = 0usize;
+    while total < buf.len() {
+        #[cfg(unix)]
+        let n = {
+            use std::os::unix::fs::FileExt;
+            f.read_at(&mut buf[total..], offset + total as u64)?
+        };
+        #[cfg(not(unix))]
+        let n = {
+            use std::io::{Read, Seek, SeekFrom};
+            let mut fc = f.try_clone()?;
+            fc.seek(SeekFrom::Start(offset + total as u64))?;
+            fc.read(&mut buf[total..])?
+        };
+        if n == 0 {
+            return Err(io::Error::new(
+                io::ErrorKind::UnexpectedEof,
+                "short read on shard block",
+            ));
+        }
+        total += n;
+    }
+    Ok(())
+}
+
+/// Builds the `EcShardConfig` proto for the given layout. The bitrot sidecar
+/// carries its own top-level encode_uuid, so the nested config leaves it empty.
+pub fn ec_shard_config(data_shards: u32, parity_shards: u32, block_size: i64) -> EcShardConfig {
+    EcShardConfig {
+        data_shards,
+        parity_shards,
+        encode_ts_ns: 0,
+        block_size,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// Cross-binary byte-exact proof: CANONICAL_HEX equals `canonicalInteropHex`
+    /// in weed/storage/erasure_coding/ec_bitrot_interop_test.go (the Go reference
+    /// asserts the same constant). Both binaries port the same format, so a
+    /// sidecar written by either must be byte-identical; this pins the Rust side
+    /// of that guarantee. If you change the format, regenerate and update BOTH.
+    #[test]
+    fn test_byte_exact_go_interop() {
+        const CANONICAL_HEX: &str = "45435355000100000039cc1b826a080110808080082204080a10042a0a108080401a04040302012a0c0801108080401a04080706053210000102030405060708090a0b0c0d0e0f";
+        let prot = EcBitrotProtection {
+            algorithm: ChecksumAlgorithm::ChecksumCrc32c as i32,
+            block_size: DEFAULT_BITROT_BLOCK_SIZE as u32,
+            generation: 0,
+            ec_shard_config: Some(EcShardConfig {
+                data_shards: 10,
+                parity_shards: 4,
+                encode_ts_ns: 0,
+                block_size: 0,
+            }),
+            shards: vec![
+                EcShardChecksums {
+                    shard_id: 0,
+                    covered_size: 1024 * 1024,
+                    block_crc32c: pack_u32_le(&[0x0102_0304]),
+                },
+                EcShardChecksums {
+                    shard_id: 1,
+                    covered_size: 1024 * 1024,
+                    block_crc32c: pack_u32_le(&[0x0506_0708]),
+                },
+            ],
+            encode_uuid: (0..16u8).collect(),
+        };
+        let dir = std::env::temp_dir().join(format!("ecsum_interop_{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let path = dir.join("v1.ecsum");
+        let path = path.to_str().unwrap();
+        save_bitrot_sidecar(path, &prot).unwrap();
+        let bytes = std::fs::read(path).unwrap();
+        let hex: String = bytes.iter().map(|b| format!("{:02x}", b)).collect();
+        assert_eq!(
+            hex, CANONICAL_HEX,
+            "Rust .ecsum bytes drifted from the Go canonical form"
+        );
+        let _ = std::fs::remove_file(path);
+    }
+
+    #[test]
+    fn test_sidecar_path_generations() {
+        assert_eq!(bitrot_sidecar_path("/d/1", 0), "/d/1.ecsum");
+        assert_eq!(bitrot_sidecar_path("/d/1", 3), "/d/1.ecsum.v3");
+    }
+
+    /// Mirrors Go's TestRemoveBitrotSidecars: legacy + versioned sidecars go,
+    /// unrelated files (a shard, a longer-vid sidecar) survive, absent is success.
+    #[test]
+    fn test_remove_bitrot_sidecars() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        let base = tmp.path().join("5").to_str().unwrap().to_string();
+        for p in [
+            format!("{}.ecsum", base),
+            format!("{}.ecsum.v1", base),
+            format!("{}.ecsum.v7", base),
+        ] {
+            std::fs::write(&p, b"x").unwrap();
+        }
+        let keep_shard = format!("{}.ec00", base);
+        let keep_other_vid = format!("{}0.ecsum", base);
+        std::fs::write(&keep_shard, b"x").unwrap();
+        std::fs::write(&keep_other_vid, b"x").unwrap();
+
+        remove_bitrot_sidecars(&base).unwrap();
+
+        for p in [
+            format!("{}.ecsum", base),
+            format!("{}.ecsum.v1", base),
+            format!("{}.ecsum.v7", base),
+        ] {
+            assert!(
+                !std::path::Path::new(&p).exists(),
+                "{} should be removed",
+                p
+            );
+        }
+        assert!(std::path::Path::new(&keep_shard).exists());
+        assert!(std::path::Path::new(&keep_other_vid).exists());
+
+        // Already-gone is success.
+        remove_bitrot_sidecars(&base).unwrap();
+    }
+
+    #[test]
+    fn test_is_pow2_multiple_of_1mib() {
+        assert!(is_pow2_multiple_of_1mib(1 << 20)); // 1 MiB
+        assert!(is_pow2_multiple_of_1mib(16 * 1024 * 1024)); // 16 MiB default
+        assert!(is_pow2_multiple_of_1mib(1 << 25));
+        assert!(is_pow2_multiple_of_1mib(MAX_BITROT_BLOCK_SIZE)); // 64 MiB boundary
+        assert!(!is_pow2_multiple_of_1mib(0));
+        assert!(!is_pow2_multiple_of_1mib(1 << 19)); // 512 KiB, too small
+        assert!(!is_pow2_multiple_of_1mib(3 << 20)); // 3 MiB, not pow2
+        assert!(!is_pow2_multiple_of_1mib(128 * 1024 * 1024)); // pow2 but > MAX_BITROT_BLOCK_SIZE
+        assert!(!is_pow2_multiple_of_1mib(
+            DEFAULT_BITROT_BLOCK_SIZE as u32 + 1
+        ));
+    }
+
+    #[test]
+    fn test_expected_block_count() {
+        assert_eq!(expected_block_count(0, 16), 0);
+        assert_eq!(expected_block_count(1, 16), 1);
+        assert_eq!(expected_block_count(16, 16), 1);
+        assert_eq!(expected_block_count(17, 16), 2);
+        assert_eq!(expected_block_count(32, 16), 2);
+        assert_eq!(expected_block_count(100, 0), 0);
+    }
+
+    #[test]
+    fn test_pack_unpack_roundtrip() {
+        let vals = vec![0x0102_0304u32, 0xdead_beef, 0, u32::MAX];
+        let packed = pack_u32_le(&vals);
+        assert_eq!(packed.len(), 16);
+        // Verify little-endian byte order of the first entry.
+        assert_eq!(&packed[0..4], &[0x04, 0x03, 0x02, 0x01]);
+        assert_eq!(unpack_u32_le(&packed), vals);
+    }
+
+    #[test]
+    fn test_builder_block_boundaries() {
+        // block_size = 4; feed 10 bytes in chunks that cross boundaries.
+        let mut b = ShardChecksumBuilder::new(4);
+        let data = b"0123456789";
+        b.write(&data[0..3]); // partial block 0
+        b.write(&data[3..7]); // completes block 0 (idx 3), fills block 1
+        b.write(&data[7..10]); // partial block 2
+        let (covered, packed) = b.finalize();
+        assert_eq!(covered, 10);
+        let crcs = unpack_u32_le(&packed);
+        // ceil(10/4) = 3 blocks
+        assert_eq!(crcs.len(), 3);
+        // Compare against direct per-block CRCs.
+        assert_eq!(crcs[0], CRC::new(&data[0..4]).0);
+        assert_eq!(crcs[1], CRC::new(&data[4..8]).0);
+        assert_eq!(crcs[2], CRC::new(&data[8..10]).0);
+    }
+
+    #[test]
+    fn test_builder_exact_block_multiple() {
+        let mut b = ShardChecksumBuilder::new(4);
+        b.write(b"01234567"); // exactly 2 blocks, no partial
+        let (covered, packed) = b.finalize();
+        assert_eq!(covered, 8);
+        assert_eq!(unpack_u32_le(&packed).len(), 2);
+    }
+
+    #[test]
+    fn test_save_load_roundtrip() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        let path = tmp.path().join("vol.ecsum").to_str().unwrap().to_string();
+
+        let mut builder = ShardChecksumBuilder::new(DEFAULT_BITROT_BLOCK_SIZE as i64);
+        builder.write(b"hello world");
+        let (covered, packed) = builder.finalize();
+
+        let prot = EcBitrotProtection {
+            algorithm: ChecksumAlgorithm::ChecksumCrc32c as i32,
+            block_size: DEFAULT_BITROT_BLOCK_SIZE as u32,
+            generation: 0,
+            ec_shard_config: Some(ec_shard_config(10, 4, 0)),
+            shards: vec![EcShardChecksums {
+                shard_id: 0,
+                covered_size: covered,
+                block_crc32c: packed,
+            }],
+            encode_uuid: new_encode_uuid(),
+        };
+
+        save_bitrot_sidecar(&path, &prot).unwrap();
+        let loaded = load_bitrot_sidecar(&path).unwrap();
+        assert_eq!(loaded, prot);
+    }
+
+    #[test]
+    fn test_load_rejects_bad_magic() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        let path = tmp.path().join("bad.ecsum").to_str().unwrap().to_string();
+        std::fs::write(&path, vec![0u8; BITROT_HEADER_SIZE + 4]).unwrap();
+        match load_bitrot_sidecar(&path) {
+            Err(BitrotLoadError::BadMagic(_)) => {}
+            other => panic!("expected BadMagic, got {:?}", other),
+        }
+    }
+
+    #[test]
+    fn test_load_rejects_corrupted_payload() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        let path = tmp
+            .path()
+            .join("corrupt.ecsum")
+            .to_str()
+            .unwrap()
+            .to_string();
+        let prot = EcBitrotProtection {
+            algorithm: ChecksumAlgorithm::ChecksumCrc32c as i32,
+            block_size: DEFAULT_BITROT_BLOCK_SIZE as u32,
+            generation: 0,
+            ec_shard_config: Some(ec_shard_config(10, 4, 0)),
+            shards: vec![EcShardChecksums {
+                shard_id: 0,
+                covered_size: 5,
+                block_crc32c: pack_u32_le(&[CRC::new(b"hello").0]),
+            }],
+            encode_uuid: vec![0u8; 16],
+        };
+        save_bitrot_sidecar(&path, &prot).unwrap();
+
+        // Flip a byte in the payload (after the 14-byte header).
+        let mut data = std::fs::read(&path).unwrap();
+        let last = data.len() - 1;
+        data[last] ^= 0xff;
+        std::fs::write(&path, &data).unwrap();
+
+        match load_bitrot_sidecar(&path) {
+            Err(BitrotLoadError::CrcMismatch { .. }) => {}
+            other => panic!("expected CrcMismatch, got {:?}", other),
+        }
+    }
+
+    #[test]
+    fn test_load_missing_is_notfound() {
+        let res = load_bitrot_sidecar("/nonexistent/path/x.ecsum");
+        match res {
+            Err(BitrotLoadError::Io(e)) => assert_eq!(e.kind(), io::ErrorKind::NotFound),
+            other => panic!("expected Io(NotFound), got {:?}", other),
+        }
+    }
+
+    fn good_manifest() -> EcBitrotProtection {
+        let mut shards = Vec::new();
+        for id in 0..14u32 {
+            shards.push(EcShardChecksums {
+                shard_id: id,
+                covered_size: 5,
+                block_crc32c: pack_u32_le(&[CRC::new(b"hello").0]),
+            });
+        }
+        EcBitrotProtection {
+            algorithm: ChecksumAlgorithm::ChecksumCrc32c as i32,
+            block_size: DEFAULT_BITROT_BLOCK_SIZE as u32,
+            generation: 0,
+            ec_shard_config: Some(ec_shard_config(10, 4, 0)),
+            shards,
+            encode_uuid: vec![0u8; 16],
+        }
+    }
+
+    #[test]
+    fn test_validate_manifest_ok() {
+        assert!(validate_manifest(&good_manifest(), 10, 4).is_ok());
+    }
+
+    #[test]
+    fn test_validate_rejects_wrong_algorithm() {
+        let mut m = good_manifest();
+        m.algorithm = ChecksumAlgorithm::ChecksumNone as i32;
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_validate_rejects_bad_block_size() {
+        let mut m = good_manifest();
+        m.block_size = 3 << 20;
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_validate_rejects_incomplete() {
+        let mut m = good_manifest();
+        m.shards.pop();
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_validate_rejects_duplicate_shard_id() {
+        let mut m = good_manifest();
+        m.shards[1].shard_id = 0;
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_validate_rejects_out_of_range_id() {
+        let mut m = good_manifest();
+        m.shards[13].shard_id = 14;
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_validate_rejects_nonpositive_covered_size() {
+        let mut m = good_manifest();
+        m.shards[0].covered_size = 0;
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_validate_rejects_crc_count_mismatch() {
+        let mut m = good_manifest();
+        m.shards[0].block_crc32c = vec![0u8; 8]; // 2 entries but covered_size=5 => want 1
+        assert!(validate_manifest(&m, 10, 4).is_err());
+    }
+
+    #[test]
+    fn test_resolve_status() {
+        // Missing => Off.
+        let notfound: Result<EcBitrotProtection, BitrotLoadError> = Err(BitrotLoadError::Io(
+            io::Error::new(io::ErrorKind::NotFound, "x"),
+        ));
+        assert_eq!(resolve_status(&notfound, 0, 10, 4), BitrotStatus::Off);
+
+        // Integrity failure => Invalid.
+        let bad: Result<EcBitrotProtection, BitrotLoadError> = Err(BitrotLoadError::BadMagic(0));
+        assert_eq!(resolve_status(&bad, 0, 10, 4), BitrotStatus::Invalid);
+
+        // Generation mismatch => Off.
+        let mut m = good_manifest();
+        m.generation = 7;
+        let ok: Result<EcBitrotProtection, BitrotLoadError> = Ok(m);
+        assert_eq!(resolve_status(&ok, 0, 10, 4), BitrotStatus::Off);
+
+        // Matching + valid => On.
+        let ok2: Result<EcBitrotProtection, BitrotLoadError> = Ok(good_manifest());
+        assert_eq!(resolve_status(&ok2, 0, 10, 4), BitrotStatus::On);
+
+        // Matching generation but invalid manifest => Invalid.
+        let mut bad_m = good_manifest();
+        bad_m.shards.pop();
+        let ok3: Result<EcBitrotProtection, BitrotLoadError> = Ok(bad_m);
+        assert_eq!(resolve_status(&ok3, 0, 10, 4), BitrotStatus::Invalid);
+    }
+
+    #[test]
+    fn test_verify_shard_file_blocks() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        let path = tmp.path().join("s.ec00").to_str().unwrap().to_string();
+        let block_size: i64 = 4;
+        let data = b"0123456789"; // 10 bytes, 3 blocks
+        std::fs::write(&path, data).unwrap();
+
+        let mut b = ShardChecksumBuilder::new(block_size);
+        b.write(data);
+        let (covered, packed) = b.finalize();
+        let entry = EcShardChecksums {
+            shard_id: 0,
+            covered_size: covered,
+            block_crc32c: packed,
+        };
+
+        // Clean file => no mismatches.
+        let mm = verify_shard_file_blocks(&path, &entry, block_size).unwrap();
+        assert!(mm.is_empty());
+
+        // Corrupt block index 1 (bytes 4..8).
+        let mut corrupt = data.to_vec();
+        corrupt[5] ^= 0xff;
+        std::fs::write(&path, &corrupt).unwrap();
+        let mm = verify_shard_file_blocks(&path, &entry, block_size).unwrap();
+        assert_eq!(mm, vec![1]);
+
+        // Truncation => all blocks mismatched.
+        std::fs::write(&path, b"012").unwrap();
+        let mm = verify_shard_file_blocks(&path, &entry, block_size).unwrap();
+        assert_eq!(mm, vec![0, 1, 2]);
+    }
+}

seaweed-volume/src/storage/erasure_coding/ec_decoder.rs

@@ -0,0 +1,697 @@
+//! EC decoding: reconstruct a .dat file from EC shards.
+//!
+//! Rebuilds the original .dat + .idx files from data shards (.ec00-.ec09)
+//! and the sorted index (.ecx) + deletion journal (.ecj).
+
+use std::fs::File;
+use std::io::{self, Read, Write};
+
+use crate::storage::erasure_coding::ec_shard::*;
+use crate::storage::idx;
+use crate::storage::needle::needle::get_actual_size;
+use crate::storage::super_block::SUPER_BLOCK_SIZE;
+use crate::storage::types::*;
+use crate::storage::volume::{fsync_dir, volume_file_name};
+
+/// Calculate .dat file size from the max offset entry in .ecx.
+/// Reads the volume version from the first EC shard (.ec00) superblock,
+/// then scans .ecx entries to find the largest (offset + needle_actual_size).
+///
+/// `dir` is used both for reading `.ec00` and `.ecx`. For split-disk
+/// reconciled volumes call [`find_dat_file_size_with_dirs`] instead.
+pub fn find_dat_file_size(dir: &str, collection: &str, volume_id: VolumeId) -> io::Result<i64> {
+    find_dat_file_size_with_dirs(dir, dir, collection, volume_id)
+}
+
+/// Like [`find_dat_file_size`] but lets the caller pass separate dirs
+/// for `.ec00` (the data shard) and `.ecx` (the sealed index). This
+/// is the form needed when shards are split across data dirs and the
+/// `.ecx` lives on a sibling disk's idx dir (#9252).
+pub fn find_dat_file_size_with_dirs(
+    ec00_dir: &str,
+    ecx_dir: &str,
+    collection: &str,
+    volume_id: VolumeId,
+) -> io::Result<i64> {
+    let ec00_base = volume_file_name(ec00_dir, collection, volume_id);
+    let ecx_base = volume_file_name(ecx_dir, collection, volume_id);
+
+    // Read volume version from .ec00 superblock
+    let ec00_path = format!("{}.ec00", ec00_base);
+    let mut ec00 = File::open(&ec00_path)?;
+    let mut sb_buf = [0u8; SUPER_BLOCK_SIZE];
+    ec00.read_exact(&mut sb_buf)?;
+    let version = Version(sb_buf[0]);
+
+    // Start with at least the superblock size
+    let mut dat_size: i64 = SUPER_BLOCK_SIZE as i64;
+
+    // Scan .ecx entries
+    let ecx_path = format!("{}.ecx", ecx_base);
+    let ecx_data = std::fs::read(&ecx_path)?;
+    let entry_count = ecx_data.len() / NEEDLE_MAP_ENTRY_SIZE;
+
+    for i in 0..entry_count {
+        let start = i * NEEDLE_MAP_ENTRY_SIZE;
+        let (_, offset, size) =
+            idx_entry_from_bytes(&ecx_data[start..start + NEEDLE_MAP_ENTRY_SIZE]);
+        if size.is_deleted() {
+            continue;
+        }
+        let entry_stop = offset.to_actual_offset() + get_actual_size(size, version);
+        if entry_stop > dat_size {
+            dat_size = entry_stop;
+        }
+    }
+
+    Ok(dat_size)
+}
+
+/// What it takes to rebuild a volume's .dat from its EC data shards.
+///
+/// Mirrors Go's `WriteDatFile(baseFileName, datFileSize,
+/// encodedDatFileSize, shardFileNames)` shape — Go passes per-shard
+/// paths so a reconciled volume with shards split across disks of the
+/// same volume server can still be decoded back to a regular .dat
+/// (seaweedfs/seaweedfs#9252).
+#[derive(Clone, Copy, Debug)]
+pub struct DatRebuild<'a> {
+    /// Where the produced `.dat` is written.
+    pub dat_dir: &'a str,
+    pub collection: &'a str,
+    pub volume_id: VolumeId,
+    /// The number of bytes to write, i.e. the live data extent from
+    /// [`find_dat_file_size`].
+    pub dat_file_size: i64,
+    /// The .dat size at encode time, which fixed the shard block layout:
+    /// deletions can move the live extent below the large-block row
+    /// boundary, and deriving the layout from the shrunk extent would read
+    /// the shards in the wrong block order. Zero when the .vif does not
+    /// record the encode-time size; the layout is then inferred from the
+    /// shard size.
+    pub encoded_dat_file_size: i64,
+    pub data_shards: usize,
+    /// `shard_dirs[i]` is the directory holding shard `i`. `None` means every
+    /// data shard sits in `dat_dir`.
+    pub shard_dirs: Option<&'a [String]>,
+    /// The volume's shard block layout, e.g. `EcVolume::large_block_size()`
+    /// / `small_block_size()` from its .vif EC config.
+    pub large_block_size: usize,
+    pub small_block_size: usize,
+}
+
+/// Reconstruct a .dat file from EC data shards.
+///
+/// Reads from .ec00-.ec09 and writes a new .dat file, from one directory or
+/// from the per-shard directories of a cross-disk reconciled volume.
+pub fn write_dat_file_from_shards(spec: &DatRebuild<'_>) -> io::Result<()> {
+    let DatRebuild {
+        dat_dir,
+        collection,
+        volume_id,
+        dat_file_size,
+        encoded_dat_file_size,
+        data_shards,
+        shard_dirs,
+        large_block_size,
+        small_block_size,
+    } = *spec;
+    let same_dir: Vec<String>;
+    let shard_dirs: &[String] = match shard_dirs {
+        Some(dirs) => dirs,
+        None => {
+            same_dir = vec![dat_dir.to_string(); data_shards];
+            &same_dir
+        }
+    };
+    if data_shards == 0 {
+        return Err(io::Error::new(
+            io::ErrorKind::InvalidInput,
+            "no data shards",
+        ));
+    }
+    if shard_dirs.len() < data_shards {
+        return Err(io::Error::new(
+            io::ErrorKind::InvalidInput,
+            format!(
+                "shard_dirs len {} < data_shards {}",
+                shard_dirs.len(),
+                data_shards
+            ),
+        ));
+    }
+    let base = volume_file_name(dat_dir, collection, volume_id);
+    let dat_path = format!("{}.dat", base);
+    // Write to a temp file and atomically rename into place, so a crash
+    // mid-write never leaves a partial .dat at the final name beside the
+    // source shards.
+    let tmp_path = format!("{}.tmp", dat_path);
+
+    let write_result = (|| -> io::Result<()> {
+        // Open data shards from their individual home dirs.
+        let mut shards: Vec<EcVolumeShard> = (0..data_shards as u8)
+            .map(|i| EcVolumeShard::new(&shard_dirs[i as usize], collection, volume_id, i))
+            .collect();
+
+        for shard in &mut shards {
+            shard.open()?;
+        }
+
+        let mut encoded_dat_file_size = encoded_dat_file_size;
+        if encoded_dat_file_size <= 0 {
+            // .vif without the encode-time size: infer the padded layout from
+            // the physical shard size, which reads the shards in the same
+            // block order.
+            let shard_size = std::fs::metadata(shards[0].file_name())?.len() as i64;
+            // A shard size that is an exact multiple of the large block size
+            // is ambiguous: N large rows, or N-1 large rows plus a full
+            // small-block region. The two layouts only agree below the last
+            // large row.
+            let large = large_block_size as i64;
+            if shard_size % large == 0
+                && dat_file_size > (shard_size / large - 1) * large * data_shards as i64
+            {
+                return Err(io::Error::new(
+                    io::ErrorKind::InvalidData,
+                    format!(
+                        "shard size {} does not identify the block layout; re-encode to record the dat size in .vif",
+                        shard_size
+                    ),
+                ));
+            }
+            encoded_dat_file_size = data_shards as i64 * shard_size;
+        }
+        if dat_file_size > encoded_dat_file_size {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!(
+                    "dat file size {} exceeds encoded dat file size {}",
+                    dat_file_size, encoded_dat_file_size
+                ),
+            ));
+        }
+
+        let mut dat_file = File::create(&tmp_path)?;
+        let mut remaining = dat_file_size;
+        let mut encoded_remaining = encoded_dat_file_size;
+        let large_row_size = (large_block_size * data_shards) as i64;
+
+        let mut shard_offset: u64 = 0;
+
+        // Read large blocks
+        while encoded_remaining >= large_row_size && remaining > 0 {
+            for (i, shard) in shards[..data_shards].iter().enumerate() {
+                let to_write = large_block_size.min(remaining as usize);
+                let mut buf = vec![0u8; to_write];
+                let n = shard.read_at(&mut buf, shard_offset)?;
+                if n != to_write {
+                    return Err(io::Error::new(
+                        io::ErrorKind::UnexpectedEof,
+                        format!("short read of large block on shard {}", i),
+                    ));
+                }
+                dat_file.write_all(&buf)?;
+                remaining -= to_write as i64;
+                if remaining <= 0 {
+                    break;
+                }
+            }
+            encoded_remaining -= large_row_size;
+            shard_offset += large_block_size as u64;
+        }
+
+        // Read small blocks
+        while remaining > 0 {
+            for (i, shard) in shards[..data_shards].iter().enumerate() {
+                let to_write = small_block_size.min(remaining as usize);
+                let mut buf = vec![0u8; to_write];
+                let n = shard.read_at(&mut buf, shard_offset)?;
+                if n != to_write {
+                    return Err(io::Error::new(
+                        io::ErrorKind::UnexpectedEof,
+                        format!("short read of small block on shard {}", i),
+                    ));
+                }
+                dat_file.write_all(&buf)?;
+                remaining -= to_write as i64;
+                if remaining <= 0 {
+                    break;
+                }
+            }
+            shard_offset += small_block_size as u64;
+        }
+
+        for shard in &mut shards {
+            shard.close();
+        }
+
+        // fsync, rename, then fsync the dir so the decoded .dat is durable and
+        // atomically published before the caller deletes the source shards.
+        dat_file.sync_all()?;
+        drop(dat_file);
+        // Windows rename does not replace an existing file on every version;
+        // remove the destination first, matching the compaction commit path.
+        #[cfg(windows)]
+        {
+            let _ = std::fs::remove_file(&dat_path);
+        }
+        std::fs::rename(&tmp_path, &dat_path)?;
+        fsync_dir(&dat_path)?;
+        Ok(())
+    })();
+    if write_result.is_err() {
+        let _ = std::fs::remove_file(&tmp_path);
+    }
+    write_result
+}
+
+/// Write .idx file from .ecx index + .ecj deletion journal.
+///
+/// Copies sorted .ecx entries to .idx, then appends tombstones for
+/// deleted needles from .ecj.
+pub fn write_idx_file_from_ec_index(
+    dir: &str,
+    collection: &str,
+    volume_id: VolumeId,
+) -> io::Result<()> {
+    let base = volume_file_name(dir, collection, volume_id);
+    let ecx_path = format!("{}.ecx", base);
+    let ecj_path = format!("{}.ecj", base);
+    let idx_path = format!("{}.idx", base);
+    // Write to a temp file and atomically rename into place, so a crash
+    // mid-write never leaves a partial .idx at the final name beside the
+    // source shards.
+    let tmp_path = format!("{}.tmp", idx_path);
+
+    let write_result = (|| -> io::Result<()> {
+        // Copy .ecx to the temp .idx
+        std::fs::copy(&ecx_path, &tmp_path)?;
+
+        // Append deletions from .ecj as tombstones. Read the journal directly
+        // and treat only NotFound as "no journal": Path::exists would also
+        // swallow a permission/IO error and silently skip deletions, which
+        // would resurrect deleted needles as live.
+        let mut idx_file = std::fs::OpenOptions::new().append(true).open(&tmp_path)?;
+        match std::fs::read(&ecj_path) {
+            Ok(ecj_data) => {
+                let count = ecj_data.len() / NEEDLE_ID_SIZE;
+                for i in 0..count {
+                    let start = i * NEEDLE_ID_SIZE;
+                    let needle_id = NeedleId::from_bytes(&ecj_data[start..start + NEEDLE_ID_SIZE]);
+                    idx::write_index_entry(
+                        &mut idx_file,
+                        needle_id,
+                        Offset::default(),
+                        TOMBSTONE_FILE_SIZE,
+                    )?;
+                }
+            }
+            Err(e) if e.kind() == io::ErrorKind::NotFound => {}
+            Err(e) => return Err(e),
+        }
+
+        // fsync, rename, then fsync the dir so the decoded .idx is durable and
+        // atomically published before the caller deletes the source shards.
+        idx_file.sync_all()?;
+        drop(idx_file);
+        // Windows rename does not replace an existing file on every version;
+        // remove the destination first, matching the compaction commit path.
+        #[cfg(windows)]
+        {
+            let _ = std::fs::remove_file(&idx_path);
+        }
+        std::fs::rename(&tmp_path, &idx_path)?;
+        fsync_dir(&idx_path)?;
+        Ok(())
+    })();
+    if write_result.is_err() {
+        let _ = std::fs::remove_file(&tmp_path);
+    }
+    write_result
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::storage::erasure_coding::ec_encoder;
+    use crate::storage::needle::needle::Needle;
+    use crate::storage::needle_map::NeedleMapKind;
+    use crate::storage::volume::{Volume, VolumeSpec};
+    use tempfile::TempDir;
+
+    #[test]
+    fn test_ec_full_round_trip() {
+        let tmp = TempDir::new().unwrap();
+        let dir = tmp.path().to_str().unwrap();
+
+        // Create volume with data
+        let mut v = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+
+        let test_data: Vec<(NeedleId, Vec<u8>)> = (1..=3)
+            .map(|i| {
+                let data = format!("EC round trip data for needle {}", i);
+                (NeedleId(i), data.into_bytes())
+            })
+            .collect();
+
+        for (id, data) in &test_data {
+            let mut n = Needle {
+                id: *id,
+                cookie: Cookie(id.0 as u32),
+                data: data.clone(),
+                data_size: data.len() as u32,
+                ..Needle::default()
+            };
+            v.write_needle(&mut n, true, false).unwrap();
+        }
+        v.sync_to_disk().unwrap();
+        let original_dat_size = v.dat_file_size().unwrap();
+        v.close();
+
+        // Read original .dat for comparison
+        let original_dat = std::fs::read(format!("{}/1.dat", dir)).unwrap();
+
+        // Encode to EC
+        let data_shards = 10;
+        let parity_shards = 4;
+        let block_size =
+            ec_encoder::write_ec_files(dir, dir, "", VolumeId(1), data_shards, parity_shards)
+                .unwrap();
+
+        // Delete original .dat and .idx
+        std::fs::remove_file(format!("{}/1.dat", dir)).unwrap();
+        std::fs::remove_file(format!("{}/1.idx", dir)).unwrap();
+
+        // Reconstruct from EC shards
+        write_dat_file_from_shards(&DatRebuild {
+            dat_dir: dir,
+            collection: "",
+            volume_id: VolumeId(1),
+            dat_file_size: original_dat_size as i64,
+            encoded_dat_file_size: original_dat_size as i64,
+            data_shards,
+            shard_dirs: None,
+            large_block_size: block_size as usize,
+            small_block_size: block_size as usize,
+        })
+        .unwrap();
+        write_idx_file_from_ec_index(dir, "", VolumeId(1)).unwrap();
+
+        // Atomic publish must rename the temp files away, never leaving them behind.
+        assert!(!std::path::Path::new(&format!("{}/1.dat.tmp", dir)).exists());
+        assert!(!std::path::Path::new(&format!("{}/1.idx.tmp", dir)).exists());
+
+        // Verify reconstructed .dat matches original
+        let reconstructed_dat = std::fs::read(format!("{}/1.dat", dir)).unwrap();
+        assert_eq!(
+            original_dat[..original_dat_size as usize],
+            reconstructed_dat[..original_dat_size as usize],
+            "reconstructed .dat should match original"
+        );
+
+        // Verify we can load and read from reconstructed volume
+        let v2 = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+
+        for (id, expected_data) in &test_data {
+            let mut n = Needle {
+                id: *id,
+                ..Needle::default()
+            };
+            v2.read_needle(&mut n).unwrap();
+            assert_eq!(&n.data, expected_data, "needle {} data should match", id);
+        }
+    }
+
+    #[test]
+    fn test_decode_missing_shard_leaves_no_dat() {
+        let tmp = TempDir::new().unwrap();
+        let dir = tmp.path().to_str().unwrap();
+        // No shard files exist, so de-striping must fail and publish nothing:
+        // neither the final .dat nor a partial .dat.tmp may remain.
+        let res = write_dat_file_from_shards(&DatRebuild {
+            dat_dir: dir,
+            collection: "",
+            volume_id: VolumeId(7),
+            dat_file_size: 100,
+            encoded_dat_file_size: 100,
+            data_shards: 10,
+            shard_dirs: None,
+            large_block_size: ERASURE_CODING_LARGE_BLOCK_SIZE,
+            small_block_size: ERASURE_CODING_SMALL_BLOCK_SIZE,
+        });
+        assert!(res.is_err());
+        assert!(!std::path::Path::new(&format!("{}/7.dat", dir)).exists());
+        assert!(!std::path::Path::new(&format!("{}/7.dat.tmp", dir)).exists());
+    }
+
+    // Decoding when .vif does not record the encode-time size: the layout is
+    // inferred from the shard size, except when that is an exact large-block
+    // multiple and the live extent reaches the ambiguous region.
+    #[test]
+    fn test_write_dat_file_fallback_layout() {
+        use crate::storage::erasure_coding::ec_bitrot::{
+            DEFAULT_BITROT_BLOCK_SIZE, ShardChecksumBuilder,
+        };
+        use reed_solomon_erasure::galois_8::ReedSolomon;
+
+        const LARGE: usize = 10000;
+        const SMALL: usize = 100;
+        let data_shards = 10usize;
+        let parity_shards = 4usize;
+        let large_row_size = (LARGE * data_shards) as i64;
+
+        let tmp = TempDir::new().unwrap();
+
+        let encode = |name: &str, dat_size: i64| -> (String, Vec<String>, Vec<u8>) {
+            let dir = tmp.path().join(name);
+            std::fs::create_dir(&dir).unwrap();
+            let dir = dir.to_str().unwrap().to_string();
+            let original: Vec<u8> = (0..dat_size as usize)
+                .map(|i| ((i as u64).wrapping_mul(2654435761) >> 8) as u8)
+                .collect();
+            std::fs::write(format!("{}/1.dat", dir), &original).unwrap();
+            let dat_file = File::open(format!("{}/1.dat", dir)).unwrap();
+            let rs = ReedSolomon::new(data_shards, parity_shards).unwrap();
+            let total = data_shards + parity_shards;
+            let mut shards: Vec<EcVolumeShard> = (0..total as u8)
+                .map(|i| EcVolumeShard::new(&dir, "", VolumeId(1), i))
+                .collect();
+            for shard in &mut shards {
+                shard.create().unwrap();
+            }
+            let mut builders: Vec<ShardChecksumBuilder> = (0..total)
+                .map(|_| ShardChecksumBuilder::new(DEFAULT_BITROT_BLOCK_SIZE as i64))
+                .collect();
+            ec_encoder::encode_dat_file(
+                &dat_file,
+                dat_size,
+                &rs,
+                &mut shards,
+                &mut builders,
+                ec_encoder::EcEncodeLayout {
+                    data_shards,
+                    parity_shards,
+                    buffer_size: SMALL,
+                    large_block_size: LARGE,
+                    small_block_size: SMALL,
+                },
+            )
+            .unwrap();
+            for shard in &mut shards {
+                shard.close();
+            }
+            let shard_dirs: Vec<String> = (0..data_shards).map(|_| dir.clone()).collect();
+            (dir, shard_dirs, original)
+        };
+
+        let decode_to = |dir: &str,
+                         sub: &str,
+                         live: i64,
+                         encoded: i64,
+                         shard_dirs: &[String]|
+         -> io::Result<Vec<u8>> {
+            let out = format!("{}/{}", dir, sub);
+            std::fs::create_dir_all(&out).unwrap();
+            write_dat_file_from_shards(&DatRebuild {
+                dat_dir: &out,
+                collection: "",
+                volume_id: VolumeId(1),
+                dat_file_size: live,
+                encoded_dat_file_size: encoded,
+                data_shards: 10,
+                shard_dirs: Some(shard_dirs),
+                large_block_size: LARGE,
+                small_block_size: SMALL,
+            })?;
+            Ok(std::fs::read(format!("{}/1.dat", out)).unwrap())
+        };
+
+        // a small-block tail that is not a large-block multiple is unambiguous
+        let (dir, shard_dirs, original) = encode("plain", large_row_size + 2530);
+        let live = large_row_size / 2;
+        let decoded = decode_to(&dir, "out", live, 0, &shard_dirs).unwrap();
+        assert_eq!(&original[..live as usize], &decoded[..]);
+
+        // datSize just under one large row: the full small-block region makes
+        // each shard exactly one large block, indistinguishable from one large row
+        let (dir, shard_dirs, _) = encode("ambig1", large_row_size - 1);
+        let err = decode_to(&dir, "out", large_row_size / 2, 0, &shard_dirs).unwrap_err();
+        assert!(
+            err.to_string()
+                .contains("does not identify the block layout")
+        );
+
+        // two-row equivalent: decoding within the agreed prefix still works
+        let (dir, shard_dirs, original) = encode("ambig2", 2 * large_row_size - 1);
+        let decoded = decode_to(&dir, "outa", large_row_size, 0, &shard_dirs).unwrap();
+        assert_eq!(&original[..large_row_size as usize], &decoded[..]);
+        let err = decode_to(&dir, "outb", large_row_size + 1, 0, &shard_dirs).unwrap_err();
+        assert!(
+            err.to_string()
+                .contains("does not identify the block layout")
+        );
+    }
+
+    // Decoding after deletions moved the live extent below the large-block row
+    // boundary: the shard block layout is fixed by the encode-time .dat size,
+    // so de-striping must not derive it from the shrunk live extent.
+    #[test]
+    fn test_write_dat_file_after_tail_deletion() {
+        use crate::storage::erasure_coding::ec_bitrot::{
+            DEFAULT_BITROT_BLOCK_SIZE, ShardChecksumBuilder,
+        };
+        use reed_solomon_erasure::galois_8::ReedSolomon;
+
+        const LARGE: usize = 10000;
+        const SMALL: usize = 100;
+        let data_shards = 10usize;
+        let parity_shards = 4usize;
+        let large_row_size = (LARGE * data_shards) as i64;
+        let small_row_size = (SMALL * data_shards) as i64;
+
+        // one full large-block row plus a small-block tail
+        let dat_size = large_row_size + 2 * small_row_size + 530;
+
+        let tmp = TempDir::new().unwrap();
+        let dir = tmp.path().to_str().unwrap();
+
+        let original: Vec<u8> = (0..dat_size as usize)
+            .map(|i| ((i as u64).wrapping_mul(2654435761) >> 8) as u8)
+            .collect();
+        std::fs::write(format!("{}/1.dat", dir), &original).unwrap();
+
+        let dat_file = File::open(format!("{}/1.dat", dir)).unwrap();
+        let rs = ReedSolomon::new(data_shards, parity_shards).unwrap();
+        let total = data_shards + parity_shards;
+        let mut shards: Vec<EcVolumeShard> = (0..total as u8)
+            .map(|i| EcVolumeShard::new(dir, "", VolumeId(1), i))
+            .collect();
+        for shard in &mut shards {
+            shard.create().unwrap();
+        }
+        let mut builders: Vec<ShardChecksumBuilder> = (0..total)
+            .map(|_| ShardChecksumBuilder::new(DEFAULT_BITROT_BLOCK_SIZE as i64))
+            .collect();
+        ec_encoder::encode_dat_file(
+            &dat_file,
+            dat_size,
+            &rs,
+            &mut shards,
+            &mut builders,
+            ec_encoder::EcEncodeLayout {
+                data_shards,
+                parity_shards,
+                buffer_size: SMALL,
+                large_block_size: LARGE,
+                small_block_size: SMALL,
+            },
+        )
+        .unwrap();
+        for shard in &mut shards {
+            shard.close();
+        }
+
+        let shard_size = std::fs::metadata(format!("{}/1.ec00", dir)).unwrap().len() as i64;
+        let padded_size = data_shards as i64 * shard_size;
+        let shard_dirs: Vec<String> = (0..data_shards).map(|_| dir.to_string()).collect();
+
+        // decode into a separate dir so the output does not collide with the source .dat
+        let out_dir = tmp.path().join("out");
+        std::fs::create_dir(&out_dir).unwrap();
+        let out = out_dir.to_str().unwrap();
+        let decode = |live_size: i64, encoded_size: i64| -> Vec<u8> {
+            write_dat_file_from_shards(&DatRebuild {
+                dat_dir: out,
+                collection: "",
+                volume_id: VolumeId(1),
+                dat_file_size: live_size,
+                encoded_dat_file_size: encoded_size,
+                data_shards,
+                shard_dirs: Some(&shard_dirs),
+                large_block_size: LARGE,
+                small_block_size: SMALL,
+            })
+            .unwrap();
+            let path = format!("{}/1.dat", out);
+            let decoded = std::fs::read(&path).unwrap();
+            std::fs::remove_file(&path).unwrap();
+            decoded
+        };
+
+        for live_size in [
+            LARGE as i64 - 1,                  // within the first large block
+            LARGE as i64 + 42,                 // partial second large block
+            large_row_size / 2,                // mid large row
+            large_row_size,                    // exactly the large row
+            large_row_size + 5 * SMALL as i64, // into the small-block tail
+            dat_size,                          // nothing deleted
+        ] {
+            assert_eq!(
+                &original[..live_size as usize],
+                &decode(live_size, dat_size)[..],
+                "live size {} with encode-time layout",
+                live_size
+            );
+            assert_eq!(
+                &original[..live_size as usize],
+                &decode(live_size, padded_size)[..],
+                "live size {} with padded layout",
+                live_size
+            );
+        }
+
+        // deriving the layout from the shrunk live extent reorders the data
+        let control = decode(large_row_size / 2, large_row_size / 2);
+        assert_ne!(&original[..(large_row_size / 2) as usize], &control[..]);
+
+        // the live extent can never exceed the encode-time size
+        assert!(
+            write_dat_file_from_shards(&DatRebuild {
+                dat_dir: out,
+                collection: "",
+                volume_id: VolumeId(1),
+                dat_file_size: dat_size + 1,
+                encoded_dat_file_size: dat_size,
+                data_shards,
+                shard_dirs: Some(&shard_dirs),
+                large_block_size: LARGE,
+                small_block_size: SMALL,
+            })
+            .is_err()
+        );
+    }
+}

seaweed-volume/src/storage/erasure_coding/ec_locate.rs

@@ -0,0 +1,266 @@
+//! EC data location: maps needle offset/size to shard intervals.
+//!
+//! Determines which shard(s) contain data for a given needle and at what
+//! offsets within those shards. Handles both large (1GB) and small (1MB)
+//! block sections.
+
+use crate::storage::erasure_coding::ec_shard::*;
+use crate::storage::types::*;
+
+/// An interval to read from EC shards.
+#[derive(Debug, Clone)]
+pub struct Interval {
+    pub block_index: usize,
+    pub inner_block_offset: i64,
+    pub size: i64,
+    pub is_large_block: bool,
+    pub large_block_rows_count: usize,
+}
+
+impl Interval {
+    pub fn to_shard_id_and_offset(
+        &self,
+        data_shards: u32,
+        large_block_size: i64,
+        small_block_size: i64,
+    ) -> (ShardId, i64) {
+        let data_shards_usize = data_shards as usize;
+        let shard_id = (self.block_index % data_shards_usize) as ShardId;
+        let row_index = self.block_index / data_shards_usize;
+
+        let block_size = if self.is_large_block {
+            large_block_size
+        } else {
+            small_block_size
+        };
+
+        let mut offset = row_index as i64 * block_size + self.inner_block_offset;
+        if !self.is_large_block {
+            // Small blocks come after large blocks in the shard file
+            offset += self.large_block_rows_count as i64 * large_block_size;
+        }
+
+        (shard_id, offset)
+    }
+}
+
+/// Locate the EC shard intervals needed to read data at the given offset and size.
+///
+/// `shard_size` is the size of a single shard file.
+pub fn locate_data(
+    offset: i64,
+    size: Size,
+    shard_size: i64,
+    data_shards: u32,
+    large_block_size: i64,
+    small_block_size: i64,
+) -> Vec<Interval> {
+    let mut intervals = Vec::new();
+    let data_size = size.0 as i64;
+
+    if data_size <= 0 || shard_size <= 0 {
+        return intervals;
+    }
+
+    let large_row_size = large_block_size * data_shards as i64;
+    let small_row_size = small_block_size * data_shards as i64;
+
+    // Number of large block rows. Mirrors Go's shardDatSize/largeBlockLength:
+    // the caller's ecd-size fallback already subtracts 1 to disambiguate the
+    // exact-multiple case, so no further -1 here — a shard size that IS an
+    // exact multiple (dat_file_size path) means real full large rows.
+    let n_large_block_rows = (shard_size / large_block_size) as usize;
+    let large_section_size = n_large_block_rows as i64 * large_row_size;
+
+    let mut remaining_offset = offset;
+    let mut remaining_size = data_size;
+
+    // In large block section?
+    if remaining_offset < large_section_size {
+        let available_in_large = large_section_size - remaining_offset;
+        let to_read = remaining_size.min(available_in_large);
+
+        add_intervals(
+            &mut intervals,
+            remaining_offset,
+            to_read,
+            large_block_size,
+            large_row_size,
+            true,
+            n_large_block_rows,
+        );
+
+        remaining_offset += to_read;
+        remaining_size -= to_read;
+    }
+
+    // In small block section?
+    if remaining_size > 0 {
+        let small_offset = remaining_offset - large_section_size;
+        add_intervals(
+            &mut intervals,
+            small_offset,
+            remaining_size,
+            small_block_size,
+            small_row_size,
+            false,
+            n_large_block_rows,
+        );
+    }
+
+    intervals
+}
+
+fn add_intervals(
+    intervals: &mut Vec<Interval>,
+    offset: i64,
+    size: i64,
+    block_size: i64,
+    _row_size: i64,
+    is_large_block: bool,
+    large_block_rows_count: usize,
+) {
+    let mut pos = offset;
+    let end = offset + size;
+
+    while pos < end {
+        let block_index = (pos / block_size) as usize;
+        let inner_offset = pos % block_size;
+        let remaining_in_block = block_size - inner_offset;
+        let interval_size = remaining_in_block.min(end - pos);
+
+        intervals.push(Interval {
+            block_index,
+            inner_block_offset: inner_offset,
+            size: interval_size,
+            is_large_block,
+            large_block_rows_count,
+        });
+
+        pos += interval_size;
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_interval_to_shard_id() {
+        let data_shards = 10;
+        let large_block_size = ERASURE_CODING_LARGE_BLOCK_SIZE as i64;
+        let _shard_size = 1024 * 1024; // Example shard size
+
+        // Block index 0 → shard 0
+        let interval = Interval {
+            block_index: 0,
+            inner_block_offset: 100,
+            size: 50,
+            is_large_block: true,
+            large_block_rows_count: 1,
+        };
+        let (shard_id, offset) = interval.to_shard_id_and_offset(
+            data_shards,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        assert_eq!(shard_id, 0);
+        assert_eq!(offset, 100);
+
+        // Block index 5 → shard 5
+        let interval = Interval {
+            block_index: 5,
+            inner_block_offset: 0,
+            size: 1024,
+            is_large_block: true,
+            large_block_rows_count: 1,
+        };
+        let (shard_id, _offset) = interval.to_shard_id_and_offset(
+            data_shards,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        assert_eq!(shard_id, 5);
+
+        // Block index 12 (data_shards=10) → row_index 1, shard_id 2
+        let interval = Interval {
+            block_index: 12,
+            inner_block_offset: 200,
+            size: 50,
+            is_large_block: true,
+            large_block_rows_count: 5,
+        };
+        let (shard_id, offset) = interval.to_shard_id_and_offset(
+            data_shards,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        assert_eq!(shard_id, 2); // 12 % 10 = 2
+        assert_eq!(offset, large_block_size + 200); // row 1 offset + inner_block_offset
+
+        // Block index 10 → shard 0 (second row)
+        let interval = Interval {
+            block_index: 10,
+            inner_block_offset: 0,
+            size: 100,
+            is_large_block: true,
+            large_block_rows_count: 2,
+        };
+        let (shard_id, offset) = interval.to_shard_id_and_offset(
+            data_shards,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        assert_eq!(shard_id, 0);
+        assert_eq!(offset, ERASURE_CODING_LARGE_BLOCK_SIZE as i64); // row 1 offset
+    }
+
+    #[test]
+    fn test_locate_data_small_file() {
+        // Small file: 100 bytes at offset 50, shard size = 1MB
+        let intervals = locate_data(
+            50,
+            Size(100),
+            1024 * 1024,
+            10,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        assert!(!intervals.is_empty());
+
+        // Should be a single small block interval (no large block rows for 1MB shard)
+        assert_eq!(intervals.len(), 1);
+        assert!(!intervals[0].is_large_block);
+    }
+
+    #[test]
+    fn test_locate_data_empty() {
+        let intervals = locate_data(
+            0,
+            Size(0),
+            1024 * 1024,
+            10,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        assert!(intervals.is_empty());
+    }
+
+    #[test]
+    fn test_small_block_after_large() {
+        let interval = Interval {
+            block_index: 0,
+            inner_block_offset: 0,
+            size: 100,
+            is_large_block: false,
+            large_block_rows_count: 2,
+        };
+        let (_shard_id, offset) = interval.to_shard_id_and_offset(
+            10,
+            ERASURE_CODING_LARGE_BLOCK_SIZE as i64,
+            ERASURE_CODING_SMALL_BLOCK_SIZE as i64,
+        );
+        // Should be after 2 large block rows
+        assert_eq!(offset, 2 * ERASURE_CODING_LARGE_BLOCK_SIZE as i64);
+    }
+}

seaweed-volume/src/storage/erasure_coding/ec_shard.rs

@@ -0,0 +1,254 @@
+//! EcVolumeShard: a single shard file (.ec00-.ec13) of an erasure-coded volume.
+
+use std::fs::{self, File, OpenOptions};
+use std::io::{self, Write};
+
+use crate::storage::types::*;
+use crate::storage::volume_open::open_volume_file;
+
+pub const DATA_SHARDS_COUNT: usize = 10;
+pub const PARITY_SHARDS_COUNT: usize = 4;
+pub const TOTAL_SHARDS_COUNT: usize = DATA_SHARDS_COUNT + PARITY_SHARDS_COUNT;
+pub const MAX_SHARD_COUNT: usize = 32;
+pub const MIN_TOTAL_DISKS: usize = TOTAL_SHARDS_COUNT / PARITY_SHARDS_COUNT + 1;
+pub const ERASURE_CODING_LARGE_BLOCK_SIZE: usize = 1024 * 1024 * 1024; // 1GB
+pub const ERASURE_CODING_SMALL_BLOCK_SIZE: usize = 1024 * 1024; // 1MB
+
+pub type ShardId = u8;
+
+/// A single erasure-coded shard file.
+pub struct EcVolumeShard {
+    pub volume_id: VolumeId,
+    pub shard_id: ShardId,
+    pub collection: String,
+    pub dir: String,
+    pub disk_type: DiskType,
+    ecd_file: Option<File>,
+    ecd_file_size: i64,
+}
+
+impl EcVolumeShard {
+    /// Create a new shard reference (does not open the file).
+    pub fn new(dir: &str, collection: &str, volume_id: VolumeId, shard_id: ShardId) -> Self {
+        EcVolumeShard {
+            volume_id,
+            shard_id,
+            collection: collection.to_string(),
+            dir: dir.to_string(),
+            disk_type: DiskType::default(),
+            ecd_file: None,
+            ecd_file_size: 0,
+        }
+    }
+
+    /// Shard file name, e.g. "dir/collection_42.ec03"
+    pub fn file_name(&self) -> String {
+        let base =
+            crate::storage::volume::volume_file_name(&self.dir, &self.collection, self.volume_id);
+        format!("{}.ec{:02}", base, self.shard_id)
+    }
+
+    /// Open the shard file for reading.
+    pub fn open(&mut self) -> io::Result<()> {
+        let path = self.file_name();
+        let file = open_volume_file(OpenOptions::new().read(true), &path)?;
+        self.ecd_file_size = file.metadata()?.len() as i64;
+        self.ecd_file = Some(file);
+        Ok(())
+    }
+
+    /// Create the shard file for writing.
+    pub fn create(&mut self) -> io::Result<()> {
+        let path = self.file_name();
+        let file = open_volume_file(
+            OpenOptions::new()
+                .read(true)
+                .write(true)
+                .create(true)
+                .truncate(true),
+            &path,
+        )?;
+        self.ecd_file = Some(file);
+        self.ecd_file_size = 0;
+        Ok(())
+    }
+
+    /// Read data at a specific offset.
+    pub fn read_at(&self, buf: &mut [u8], offset: u64) -> io::Result<usize> {
+        let file = self
+            .ecd_file
+            .as_ref()
+            .ok_or_else(|| io::Error::other("shard file not open"))?;
+
+        #[cfg(unix)]
+        {
+            use std::os::unix::fs::FileExt;
+            file.read_at(buf, offset)
+        }
+
+        #[cfg(not(unix))]
+        {
+            use std::io::{Read, Seek, SeekFrom};
+            // File::read_at is unix-only; fall back to seek + read.
+            // We need a mutable reference for seek/read, so clone the handle.
+            let mut f = file.try_clone()?;
+            f.seek(SeekFrom::Start(offset))?;
+            f.read(buf)
+        }
+    }
+
+    /// Write data to the shard file (appends).
+    pub fn write_all(&mut self, data: &[u8]) -> io::Result<()> {
+        let file = self
+            .ecd_file
+            .as_mut()
+            .ok_or_else(|| io::Error::other("shard file not open"))?;
+        file.write_all(data)?;
+        self.ecd_file_size += data.len() as i64;
+        Ok(())
+    }
+
+    pub fn file_size(&self) -> i64 {
+        self.ecd_file_size
+    }
+
+    /// A duplicate of the mounted shard handle, for a reader that has to
+    /// outlive the store guard.
+    ///
+    /// This is the same descriptor `read_at` serves from, so it carries the
+    /// `O_NOATIME` from `open_volume_file` and keeps pointing at the shard
+    /// that was mounted, whatever later happens to the path. `dup` shares the
+    /// kernel file offset, which is why every read through it must be
+    /// positional (`read_at`), never seek-based.
+    pub fn try_clone_file(&self) -> io::Result<File> {
+        self.ecd_file
+            .as_ref()
+            .ok_or_else(|| io::Error::other("shard file not open"))?
+            .try_clone()
+    }
+
+    /// Protobuf descriptor for this shard. Mirrors Go's ToEcShardInfo.
+    pub fn to_ec_shard_info(&self) -> crate::pb::volume_server_pb::EcShardInfo {
+        crate::pb::volume_server_pb::EcShardInfo {
+            shard_id: self.shard_id as u32,
+            size: self.file_size(),
+            collection: self.collection.clone(),
+            volume_id: self.volume_id.0,
+            ..Default::default()
+        }
+    }
+
+    /// Close the shard file.
+    pub fn close(&mut self) {
+        if let Some(ref file) = self.ecd_file {
+            let _ = file.sync_all();
+        }
+        self.ecd_file = None;
+    }
+
+    /// Delete the shard file from disk.
+    pub fn destroy(&mut self) {
+        self.close();
+        let _ = fs::remove_file(self.file_name());
+    }
+}
+
+/// ShardBits: bitmap tracking which shards are present.
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
+pub struct ShardBits(pub u32);
+
+impl ShardBits {
+    pub fn add_shard_id(&mut self, id: ShardId) {
+        assert!((id as usize) < 32, "shard id {} out of bounds (max 31)", id,);
+        self.0 |= 1 << id;
+    }
+
+    pub fn remove_shard_id(&mut self, id: ShardId) {
+        assert!((id as usize) < 32, "shard id {} out of bounds (max 31)", id,);
+        self.0 &= !(1 << id);
+    }
+
+    pub fn has_shard_id(&self, id: ShardId) -> bool {
+        if (id as usize) >= 32 {
+            return false;
+        }
+        self.0 & (1 << id) != 0
+    }
+
+    pub fn shard_id_count(&self) -> usize {
+        self.0.count_ones() as usize
+    }
+
+    /// Iterator over present shard IDs.
+    pub fn shard_ids(&self) -> Vec<ShardId> {
+        let mut ids = Vec::with_capacity(self.shard_id_count());
+        for i in 0..32 {
+            if self.has_shard_id(i) {
+                ids.push(i);
+            }
+        }
+        ids
+    }
+
+    pub fn minus(&self, other: ShardBits) -> ShardBits {
+        ShardBits(self.0 & !other.0)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_shard_bits() {
+        let mut bits = ShardBits::default();
+        assert_eq!(bits.shard_id_count(), 0);
+
+        bits.add_shard_id(0);
+        bits.add_shard_id(3);
+        bits.add_shard_id(13);
+        assert_eq!(bits.shard_id_count(), 3);
+        assert!(bits.has_shard_id(0));
+        assert!(bits.has_shard_id(3));
+        assert!(!bits.has_shard_id(1));
+
+        bits.remove_shard_id(3);
+        assert!(!bits.has_shard_id(3));
+        assert_eq!(bits.shard_id_count(), 2);
+    }
+
+    #[test]
+    fn test_shard_bits_ids() {
+        let mut bits = ShardBits::default();
+        bits.add_shard_id(1);
+        bits.add_shard_id(5);
+        bits.add_shard_id(9);
+        assert_eq!(bits.shard_ids(), vec![1, 5, 9]);
+    }
+
+    #[test]
+    fn test_shard_bits_minus() {
+        let mut a = ShardBits::default();
+        a.add_shard_id(0);
+        a.add_shard_id(1);
+        a.add_shard_id(2);
+
+        let mut b = ShardBits::default();
+        b.add_shard_id(1);
+
+        let c = a.minus(b);
+        assert_eq!(c.shard_ids(), vec![0, 2]);
+    }
+
+    #[test]
+    fn test_shard_file_name() {
+        let shard = EcVolumeShard::new("/data", "pics", VolumeId(42), 3);
+        assert_eq!(shard.file_name(), "/data/pics_42.ec03");
+    }
+
+    #[test]
+    fn test_shard_file_name_no_collection() {
+        let shard = EcVolumeShard::new("/data", "", VolumeId(7), 13);
+        assert_eq!(shard.file_name(), "/data/7.ec13");
+    }
+}

seaweed-volume/src/storage/erasure_coding/mod.rs

@@ -0,0 +1,17 @@
+//! Erasure coding module for volume data protection.
+//!
+//! Encodes a volume's .dat file into 10 data + 4 parity shards using
+//! Reed-Solomon erasure coding. Can reconstruct from any 10 of 14 shards.
+
+pub mod ec_bitrot;
+pub mod ec_decoder;
+pub mod ec_encoder;
+pub mod ec_locate;
+pub mod ec_shard;
+pub mod ec_volume;
+
+pub use ec_shard::{
+    DATA_SHARDS_COUNT, EcVolumeShard, MAX_SHARD_COUNT, MIN_TOTAL_DISKS, PARITY_SHARDS_COUNT,
+    ShardId, TOTAL_SHARDS_COUNT,
+};
+pub use ec_volume::EcVolume;

seaweed-volume/src/storage/idx/mod.rs

@@ -0,0 +1,252 @@
+//! Index file (.idx) format: sequential 17-byte entries.
+//!
+//! Each entry: NeedleId(8) + Offset(5) + Size(4) = 17 bytes.
+
+use crate::storage::needle::needle::get_actual_size;
+use crate::storage::types::*;
+use std::io::{self, Read, Seek, SeekFrom};
+
+pub(crate) const ROWS_TO_READ: usize = 1024;
+
+/// Walk all entries in an .idx file, calling `f` for each.
+/// Mirrors Go's `WalkIndexFile()`.
+pub fn walk_index_file<R, F>(reader: &mut R, start_from: u64, mut f: F) -> io::Result<()>
+where
+    R: Read + Seek,
+    F: FnMut(NeedleId, Offset, Size) -> io::Result<()>,
+{
+    let reader_offset = start_from * NEEDLE_MAP_ENTRY_SIZE as u64;
+    reader.seek(SeekFrom::Start(reader_offset))?;
+
+    let mut buf = vec![0u8; NEEDLE_MAP_ENTRY_SIZE * ROWS_TO_READ];
+
+    loop {
+        let count = match reader.read(&mut buf) {
+            Ok(0) => return Ok(()),
+            Ok(n) => n,
+            Err(ref e) if e.kind() == io::ErrorKind::UnexpectedEof => return Ok(()),
+            Err(e) => return Err(e),
+        };
+
+        let mut i = 0;
+        while i + NEEDLE_MAP_ENTRY_SIZE <= count {
+            let (key, offset, size) = idx_entry_from_bytes(&buf[i..i + NEEDLE_MAP_ENTRY_SIZE]);
+            f(key, offset, size)?;
+            i += NEEDLE_MAP_ENTRY_SIZE;
+        }
+    }
+}
+
+/// Verify the integrity of an .idx/.ecx index file: walk every entry, sort by
+/// (offset, size), flag overlapping needles, and check the file is a whole
+/// number of entries. Mirrors Go's `idx.CheckIndexFile`. No data-file reads.
+pub fn check_index_file<R: Read + Seek>(
+    reader: &mut R,
+    idx_file_size: i64,
+    version: Version,
+) -> (u64, Vec<String>) {
+    let mut errs = Vec::new();
+    // (walk index, id, actual offset, size)
+    let mut entries: Vec<(usize, NeedleId, i64, Size)> = Vec::new();
+    let mut i = 0usize;
+    if let Err(e) = walk_index_file(reader, 0, |id, offset, size| {
+        entries.push((i, id, offset.to_actual_offset(), size));
+        i += 1;
+        Ok(())
+    }) {
+        errs.push(format!("walk index file: {}", e));
+    }
+
+    entries.sort_by(|a, b| a.2.cmp(&b.2).then(a.3.0.cmp(&b.3.0)));
+
+    // Offset-0 logical tombstones (remote-tier deletes) occupy no physical extent,
+    // so they cannot overlap anything — exclude them from the overlap check. They
+    // are still counted below for the index-size check.
+    let physical: Vec<(usize, NeedleId, i64, Size)> = entries
+        .iter()
+        .copied()
+        .filter(|e| !(e.2 == 0 && e.3.is_deleted()))
+        .collect();
+
+    for j in 1..physical.len() {
+        let (index, id, offset, size) = physical[j];
+        let (_, last_id, last_offset, last_size) = physical[j - 1];
+
+        let end = match get_actual_size(size, version) {
+            0 => offset,
+            s => offset + s - 1,
+        };
+        let last_end = match get_actual_size(last_size, version) {
+            0 => last_offset,
+            s => last_offset + s - 1,
+        };
+
+        if offset <= last_end {
+            errs.push(format!(
+                "needle {} (#{}) at [{}-{}] overlaps needle {} at [{}-{}]",
+                id.0,
+                index + 1,
+                offset,
+                end,
+                last_id.0,
+                last_offset,
+                last_end
+            ));
+        }
+    }
+
+    let count = entries.len() as u64;
+    let expected = count as i64 * NEEDLE_MAP_ENTRY_SIZE as i64;
+    if expected != idx_file_size {
+        errs.push(format!(
+            "expected an index file of size {}, got {}",
+            idx_file_size, expected
+        ));
+    }
+
+    (count, errs)
+}
+
+/// Write a single index entry to a writer.
+pub fn write_index_entry<W: io::Write>(
+    writer: &mut W,
+    key: NeedleId,
+    offset: Offset,
+    size: Size,
+) -> io::Result<()> {
+    let mut buf = [0u8; NEEDLE_MAP_ENTRY_SIZE];
+    idx_entry_to_bytes(&mut buf, key, offset, size);
+    writer.write_all(&buf)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::io::Cursor;
+
+    #[test]
+    fn test_walk_index_file() {
+        // Create a small index with 3 entries
+        let mut data = Vec::new();
+        let entries = vec![
+            (NeedleId(1), Offset::from_actual_offset(0), Size(100)),
+            (NeedleId(2), Offset::from_actual_offset(128), Size(200)),
+            (NeedleId(3), Offset::from_actual_offset(384), Size(300)),
+        ];
+        for (key, offset, size) in &entries {
+            let mut buf = [0u8; NEEDLE_MAP_ENTRY_SIZE];
+            idx_entry_to_bytes(&mut buf, *key, *offset, *size);
+            data.extend_from_slice(&buf);
+        }
+
+        let mut cursor = Cursor::new(data);
+        let mut collected = Vec::new();
+        walk_index_file(&mut cursor, 0, |key, offset, size| {
+            collected.push((key, offset.to_actual_offset(), size));
+            Ok(())
+        })
+        .unwrap();
+
+        assert_eq!(collected.len(), 3);
+        assert_eq!(collected[0].0, NeedleId(1));
+        assert_eq!(collected[0].1, 0);
+        assert_eq!(collected[0].2, Size(100));
+        assert_eq!(collected[1].0, NeedleId(2));
+        assert_eq!(collected[2].0, NeedleId(3));
+    }
+
+    #[test]
+    fn test_walk_empty() {
+        let mut cursor = Cursor::new(Vec::new());
+        let mut count = 0;
+        walk_index_file(&mut cursor, 0, |_, _, _| {
+            count += 1;
+            Ok(())
+        })
+        .unwrap();
+        assert_eq!(count, 0);
+    }
+
+    fn idx_bytes(entries: &[(NeedleId, Offset, Size)]) -> Vec<u8> {
+        let mut data = Vec::new();
+        for (key, offset, size) in entries {
+            let mut buf = [0u8; NEEDLE_MAP_ENTRY_SIZE];
+            idx_entry_to_bytes(&mut buf, *key, *offset, *size);
+            data.extend_from_slice(&buf);
+        }
+        data
+    }
+
+    #[test]
+    fn test_check_index_file_clean() {
+        let data = idx_bytes(&[
+            (NeedleId(1), Offset::from_actual_offset(0), Size(50)),
+            (NeedleId(2), Offset::from_actual_offset(100_000), Size(50)),
+        ]);
+        let size = data.len() as i64;
+        let (count, errs) = check_index_file(&mut Cursor::new(data), size, Version(3));
+        assert_eq!(count, 2);
+        assert!(errs.is_empty(), "{:?}", errs);
+    }
+
+    #[test]
+    fn test_check_index_file_detects_overlap() {
+        let data = idx_bytes(&[
+            (NeedleId(1), Offset::from_actual_offset(0), Size(100)),
+            (NeedleId(2), Offset::from_actual_offset(8), Size(100)),
+        ]);
+        let size = data.len() as i64;
+        let (count, errs) = check_index_file(&mut Cursor::new(data), size, Version(3));
+        assert_eq!(count, 2);
+        assert_eq!(errs.len(), 1, "{:?}", errs);
+        assert!(errs[0].contains("overlaps"), "{:?}", errs);
+    }
+
+    #[test]
+    fn test_check_index_file_ignores_offset0_tombstone() {
+        // A live needle near the start plus an offset-0 logical tombstone (remote-tier
+        // delete) must NOT be flagged as overlapping; the tombstone row still counts.
+        let data = idx_bytes(&[
+            (NeedleId(1), Offset::from_actual_offset(8), Size(100)),
+            (NeedleId(2), Offset::from_actual_offset(0), Size(-1)),
+        ]);
+        let size = data.len() as i64;
+        let (count, errs) = check_index_file(&mut Cursor::new(data), size, Version(3));
+        assert_eq!(count, 2, "tombstone row is still counted: {:?}", errs);
+        assert!(
+            errs.is_empty(),
+            "offset-0 tombstone must not overlap: {:?}",
+            errs
+        );
+    }
+
+    #[test]
+    fn test_check_index_file_detects_size_mismatch() {
+        let data = idx_bytes(&[(NeedleId(1), Offset::from_actual_offset(0), Size(50))]);
+        let claimed = data.len() as i64 + 5;
+        let (_count, errs) = check_index_file(&mut Cursor::new(data), claimed, Version(3));
+        assert!(
+            errs.iter().any(|e| e.contains("expected an index file")),
+            "{:?}",
+            errs
+        );
+    }
+
+    #[test]
+    fn test_write_index_entry() {
+        let mut buf = Vec::new();
+        write_index_entry(
+            &mut buf,
+            NeedleId(42),
+            Offset::from_actual_offset(8 * 10),
+            Size(512),
+        )
+        .unwrap();
+        assert_eq!(buf.len(), NEEDLE_MAP_ENTRY_SIZE);
+
+        let (key, offset, size) = idx_entry_from_bytes(&buf);
+        assert_eq!(key, NeedleId(42));
+        assert_eq!(offset.to_actual_offset(), 80);
+        assert_eq!(size, Size(512));
+    }
+}

seaweed-volume/src/storage/mod.rs

@@ -0,0 +1,16 @@
+pub mod disk_location;
+pub mod erasure_coding;
+pub mod idx;
+pub mod needle;
+pub mod needle_map;
+pub mod store;
+pub mod store_ec_mirror;
+pub mod store_ec_reconcile;
+pub mod super_block;
+pub mod types;
+pub mod volume;
+pub mod volume_idx_rebuild;
+pub mod volume_idx_repair;
+pub mod volume_open;
+pub mod volume_report;
+pub mod volume_report_hash;

seaweed-volume/src/storage/needle/crc.rs

@@ -0,0 +1,75 @@
+//! CRC32-Castagnoli checksum for needle data integrity.
+//!
+//! Matches Go's `crc32.MakeTable(crc32.Castagnoli)` exactly.
+//! The CRC is stored as raw u32 (not the `.Value()` legacy transform).
+
+/// CRC32-Castagnoli checksum wrapper.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub struct CRC(pub u32);
+
+impl CRC {
+    /// Compute CRC from a byte slice (starting from 0).
+    pub fn new(data: &[u8]) -> Self {
+        CRC(0).update(data)
+    }
+
+    /// Update the CRC with additional bytes.
+    pub fn update(self, data: &[u8]) -> Self {
+        CRC(crc32c::crc32c_append(self.0, data))
+    }
+
+    /// Legacy `.Value()` function — deprecated in Go but needed for backward compat check.
+    /// Formula: (crc >> 15 | crc << 17) + 0xa282ead8
+    pub fn legacy_value(&self) -> u32 {
+        self.0.rotate_right(15).wrapping_add(0xa282ead8)
+    }
+}
+
+impl From<u32> for CRC {
+    fn from(v: u32) -> Self {
+        CRC(v)
+    }
+}
+
+impl From<CRC> for u32 {
+    fn from(c: CRC) -> Self {
+        c.0
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_crc_empty() {
+        let crc = CRC::new(&[]);
+        assert_eq!(crc.0, 0);
+    }
+
+    #[test]
+    fn test_crc_known_value() {
+        // CRC32-C of "hello" — verify it produces a non-zero deterministic value
+        let crc = CRC::new(b"hello");
+        assert_ne!(crc.0, 0);
+        // Same input produces same output
+        assert_eq!(crc, CRC::new(b"hello"));
+    }
+
+    #[test]
+    fn test_crc_incremental() {
+        let crc1 = CRC::new(b"hello world");
+        let crc2 = CRC::new(b"hello").update(b" world");
+        assert_eq!(crc1, crc2);
+    }
+
+    #[test]
+    fn test_crc_legacy_value() {
+        let crc = CRC(0x12345678);
+        let v = crc.legacy_value();
+        // (0x12345678 >> 15 | 0x12345678 << 17) + 0xa282ead8, worked out by hand so
+        // the test checks the rotate rather than restating it.
+        let expected = 0x4f730f40_u32;
+        assert_eq!(v, expected);
+    }
+}

seaweed-volume/src/storage/needle/mod.rs

@@ -0,0 +1,11 @@
+pub mod crc;
+#[expect(
+    clippy::module_inception,
+    reason = "needle/needle.rs mirrors the Go package layout"
+)]
+pub mod needle;
+pub mod ttl;
+
+pub use crc::CRC;
+pub use needle::Needle;
+pub use ttl::TTL;

seaweed-volume/src/storage/needle/needle.rs

@@ -0,0 +1,972 @@
+//! Needle: the individual file object stored in a volume.
+//!
+//! Binary format (Version 2/3):
+//!   Header (16 bytes): Cookie(4) + NeedleId(8) + Size(4)
+//!   Body (Size bytes):
+//!     DataSize(4) + Data(DataSize) + Flags(1)
+//!     [if HasName]:    NameSize(1) + Name(NameSize)
+//!     [if HasMime]:    MimeSize(1) + Mime(MimeSize)
+//!     [if HasLastMod]: LastModified(5)
+//!     [if HasTtl]:     TTL(2)
+//!     [if HasPairs]:   PairsSize(2) + Pairs(PairsSize)
+//!   Tail:
+//!     Checksum(4) + [if V3: AppendAtNs(8)] + Padding(0-7)
+
+use super::crc::CRC;
+use super::ttl::TTL;
+use crate::storage::types::*;
+
+// Flag bits (matching Go constants)
+pub const FLAG_IS_COMPRESSED: u8 = 0x01;
+pub const FLAG_HAS_NAME: u8 = 0x02;
+pub const FLAG_HAS_MIME: u8 = 0x04;
+pub const FLAG_HAS_LAST_MODIFIED_DATE: u8 = 0x08;
+pub const FLAG_HAS_TTL: u8 = 0x10;
+pub const FLAG_HAS_PAIRS: u8 = 0x20;
+pub const FLAG_IS_CHUNK_MANIFEST: u8 = 0x80;
+
+pub const LAST_MODIFIED_BYTES_LENGTH: usize = 5;
+pub const TTL_BYTES_LENGTH: usize = 2;
+
+#[derive(Debug, Clone, Default)]
+pub struct Needle {
+    pub cookie: Cookie,
+    pub id: NeedleId,
+    pub size: Size, // sum of body content fields
+
+    // Version 2+ fields
+    pub data_size: u32,
+    pub data: Vec<u8>,
+    pub flags: u8,
+    pub name_size: u8,
+    pub name: Vec<u8>, // max 255 bytes
+    pub mime_size: u8,
+    pub mime: Vec<u8>, // max 255 bytes
+    pub pairs_size: u16,
+    pub pairs: Vec<u8>,     // max 64KB, JSON
+    pub last_modified: u64, // stored as 5 bytes on disk
+    pub ttl: Option<TTL>,
+
+    // Tail fields
+    pub checksum: CRC,
+    pub append_at_ns: u64, // Version 3 only
+    pub padding: Vec<u8>,
+}
+
+impl Needle {
+    // ---- Flag accessors (matching Go) ----
+
+    pub fn is_compressed(&self) -> bool {
+        self.flags & FLAG_IS_COMPRESSED != 0
+    }
+    pub fn set_is_compressed(&mut self) {
+        self.flags |= FLAG_IS_COMPRESSED;
+    }
+
+    pub fn has_name(&self) -> bool {
+        self.flags & FLAG_HAS_NAME != 0
+    }
+    pub fn set_has_name(&mut self) {
+        self.flags |= FLAG_HAS_NAME;
+    }
+
+    pub fn has_mime(&self) -> bool {
+        self.flags & FLAG_HAS_MIME != 0
+    }
+    pub fn set_has_mime(&mut self) {
+        self.flags |= FLAG_HAS_MIME;
+    }
+
+    pub fn has_last_modified_date(&self) -> bool {
+        self.flags & FLAG_HAS_LAST_MODIFIED_DATE != 0
+    }
+    pub fn set_has_last_modified_date(&mut self) {
+        self.flags |= FLAG_HAS_LAST_MODIFIED_DATE;
+    }
+
+    pub fn has_ttl(&self) -> bool {
+        self.flags & FLAG_HAS_TTL != 0
+    }
+    pub fn set_has_ttl(&mut self) {
+        self.flags |= FLAG_HAS_TTL;
+    }
+
+    pub fn has_pairs(&self) -> bool {
+        self.flags & FLAG_HAS_PAIRS != 0
+    }
+    pub fn set_has_pairs(&mut self) {
+        self.flags |= FLAG_HAS_PAIRS;
+    }
+
+    pub fn is_chunk_manifest(&self) -> bool {
+        self.flags & FLAG_IS_CHUNK_MANIFEST != 0
+    }
+    pub fn set_is_chunk_manifest(&mut self) {
+        self.flags |= FLAG_IS_CHUNK_MANIFEST;
+    }
+
+    // ---- Header parsing ----
+
+    /// Parse the 16-byte needle header.
+    pub fn parse_header(bytes: &[u8]) -> (Cookie, NeedleId, Size) {
+        assert!(bytes.len() >= NEEDLE_HEADER_SIZE);
+        let cookie = Cookie::from_bytes(&bytes[0..COOKIE_SIZE]);
+        let id = NeedleId::from_bytes(&bytes[COOKIE_SIZE..COOKIE_SIZE + NEEDLE_ID_SIZE]);
+        let size = Size::from_bytes(&bytes[COOKIE_SIZE + NEEDLE_ID_SIZE..NEEDLE_HEADER_SIZE]);
+        (cookie, id, size)
+    }
+
+    /// Parse needle header into self.
+    pub fn read_header(&mut self, bytes: &[u8]) {
+        let (cookie, id, size) = Self::parse_header(bytes);
+        self.cookie = cookie;
+        self.id = id;
+        self.size = size;
+    }
+
+    // ---- Body reading (Version 2/3) ----
+
+    /// Read version 2/3 body metadata only — skips copying the data payload.
+    /// Sets `data_size` and all metadata fields but leaves `data` empty.
+    pub fn read_body_v2_meta_only(&mut self, bytes: &[u8]) -> Result<(), NeedleError> {
+        let len_bytes = bytes.len();
+        let mut index = 0;
+
+        // DataSize (4 bytes)
+        if index + 4 > len_bytes {
+            return Err(NeedleError::IndexOutOfRange(1));
+        }
+        self.data_size = u32::from_be_bytes([
+            bytes[index],
+            bytes[index + 1],
+            bytes[index + 2],
+            bytes[index + 3],
+        ]);
+        index += 4;
+
+        // Skip data bytes (do NOT copy them)
+        if index + self.data_size as usize > len_bytes {
+            return Err(NeedleError::IndexOutOfRange(1));
+        }
+        index += self.data_size as usize;
+
+        // Read non-data metadata
+        self.read_body_v2_non_data(&bytes[index..])?;
+        Ok(())
+    }
+
+    /// Read full needle from bytes but skip copying the data payload.
+    /// Sets all metadata fields, checksum, etc. but leaves `data` empty.
+    pub fn read_bytes_meta_only(
+        &mut self,
+        bytes: &[u8],
+        offset: i64,
+        expected_size: Size,
+        version: Version,
+    ) -> Result<(), NeedleError> {
+        self.read_header(bytes);
+
+        if self.size != expected_size {
+            return Err(NeedleError::SizeMismatch {
+                offset,
+                id: self.id,
+                found: self.size,
+                expected: expected_size,
+            });
+        }
+
+        let body_start = NEEDLE_HEADER_SIZE;
+        let body_end = body_start + self.size.0 as usize;
+
+        if version == VERSION_1 {
+            // V1 has no metadata — data is the entire body
+            self.data_size = self.size.0 as u32;
+        } else if self.size.0 == 0 {
+            // Tombstones have no DataSize/body section; metadata starts at the tail.
+            self.data_size = 0;
+        } else {
+            self.read_body_v2_meta_only(&bytes[body_start..body_end])?;
+        }
+
+        // Read tail but skip CRC validation (no data to check against)
+        self.read_tail_meta_only(&bytes[body_end..], version)?;
+        Ok(())
+    }
+
+    /// Paged meta-only parse: accepts the 20-byte header+DataSize prefix and the
+    /// meta tail bytes (everything after the data payload). This avoids reading
+    /// the data payload from disk at all, matching Go's `ReadNeedleMeta`.
+    pub fn read_paged_meta(
+        &mut self,
+        header_bytes: &[u8], // first 20 bytes: NEEDLE_HEADER_SIZE + DATA_SIZE_SIZE
+        meta_bytes: &[u8],   // tail: non-data body metadata + checksum + timestamp + padding
+        offset: i64,
+        expected_size: Size,
+        version: Version,
+    ) -> Result<(), NeedleError> {
+        // Parse the 16-byte header
+        self.read_header(header_bytes);
+
+        if self.size != expected_size {
+            return Err(NeedleError::SizeMismatch {
+                offset,
+                id: self.id,
+                found: self.size,
+                expected: expected_size,
+            });
+        }
+
+        if version == VERSION_1 {
+            self.data_size = self.size.0 as u32;
+        } else if self.size.0 == 0 {
+            // Tombstone
+            self.data_size = 0;
+        } else {
+            // Extract DataSize from bytes 16..20
+            self.data_size = u32::from_be_bytes([
+                header_bytes[NEEDLE_HEADER_SIZE],
+                header_bytes[NEEDLE_HEADER_SIZE + 1],
+                header_bytes[NEEDLE_HEADER_SIZE + 2],
+                header_bytes[NEEDLE_HEADER_SIZE + 3],
+            ]);
+
+            // meta_bytes starts with the non-data body metadata (flags, name, mime, etc.)
+            // followed by the tail (checksum + timestamp + padding).
+            // readNeedleDataVersion2NonData returns the index where it stopped.
+            let index = self.read_body_v2_non_data(meta_bytes)?;
+            self.read_tail_meta_only(&meta_bytes[index..], version)?;
+            return Ok(());
+        }
+
+        // For VERSION_1 or tombstones, meta_bytes IS the tail
+        self.read_tail_meta_only(meta_bytes, version)?;
+        Ok(())
+    }
+
+    /// Read tail without CRC validation (used when data was not read).
+    fn read_tail_meta_only(
+        &mut self,
+        tail_bytes: &[u8],
+        version: Version,
+    ) -> Result<(), NeedleError> {
+        if tail_bytes.len() < NEEDLE_CHECKSUM_SIZE {
+            return Err(NeedleError::TailTooShort);
+        }
+
+        self.checksum = CRC(u32::from_be_bytes([
+            tail_bytes[0],
+            tail_bytes[1],
+            tail_bytes[2],
+            tail_bytes[3],
+        ]));
+
+        if version == VERSION_3 {
+            let ts_offset = NEEDLE_CHECKSUM_SIZE;
+            if tail_bytes.len() < ts_offset + TIMESTAMP_SIZE {
+                return Err(NeedleError::TailTooShort);
+            }
+            self.append_at_ns = u64::from_be_bytes([
+                tail_bytes[ts_offset],
+                tail_bytes[ts_offset + 1],
+                tail_bytes[ts_offset + 2],
+                tail_bytes[ts_offset + 3],
+                tail_bytes[ts_offset + 4],
+                tail_bytes[ts_offset + 5],
+                tail_bytes[ts_offset + 6],
+                tail_bytes[ts_offset + 7],
+            ]);
+        }
+
+        Ok(())
+    }
+
+    /// Read the version 2/3 body data from bytes (size bytes starting after header).
+    /// Returns IndexOutOfRange errors for truncated data (matching Go's readNeedleDataVersion2).
+    pub fn read_body_v2(&mut self, bytes: &[u8]) -> Result<(), NeedleError> {
+        let len_bytes = bytes.len();
+        let mut index = 0;
+
+        // DataSize (4 bytes)
+        if index + 4 > len_bytes {
+            return Ok(()); // tolerate EOF
+        }
+        self.data_size = u32::from_be_bytes([
+            bytes[index],
+            bytes[index + 1],
+            bytes[index + 2],
+            bytes[index + 3],
+        ]);
+        index += 4;
+
+        // Data
+        if index + self.data_size as usize > len_bytes {
+            return Err(NeedleError::IndexOutOfRange(1));
+        }
+        self.data = bytes[index..index + self.data_size as usize].to_vec();
+        index += self.data_size as usize;
+
+        // Read non-data metadata
+        self.read_body_v2_non_data(&bytes[index..])?;
+        Ok(())
+    }
+
+    /// Read version 2/3 metadata fields (everything after Data).
+    /// Returns IndexOutOfRange errors for truncated data (matching Go's readNeedleDataVersion2).
+    fn read_body_v2_non_data(&mut self, bytes: &[u8]) -> Result<usize, NeedleError> {
+        let len_bytes = bytes.len();
+        let mut index = 0;
+
+        // Flags (1 byte)
+        if index < len_bytes {
+            self.flags = bytes[index];
+            index += 1;
+        } else {
+            return Ok(index);
+        }
+
+        // Name
+        if index < len_bytes && self.has_name() {
+            self.name_size = bytes[index];
+            index += 1;
+            if index + self.name_size as usize > len_bytes {
+                return Err(NeedleError::IndexOutOfRange(2));
+            }
+            self.name = bytes[index..index + self.name_size as usize].to_vec();
+            index += self.name_size as usize;
+        }
+
+        // Mime
+        if index < len_bytes && self.has_mime() {
+            self.mime_size = bytes[index];
+            index += 1;
+            if index + self.mime_size as usize > len_bytes {
+                return Err(NeedleError::IndexOutOfRange(3));
+            }
+            self.mime = bytes[index..index + self.mime_size as usize].to_vec();
+            index += self.mime_size as usize;
+        }
+
+        // LastModified (5 bytes)
+        if index < len_bytes && self.has_last_modified_date() {
+            if index + LAST_MODIFIED_BYTES_LENGTH > len_bytes {
+                return Err(NeedleError::IndexOutOfRange(4));
+            }
+            self.last_modified = bytes_to_u64_5(&bytes[index..index + LAST_MODIFIED_BYTES_LENGTH]);
+            index += LAST_MODIFIED_BYTES_LENGTH;
+        }
+
+        // TTL (2 bytes)
+        if index < len_bytes && self.has_ttl() {
+            if index + TTL_BYTES_LENGTH > len_bytes {
+                return Err(NeedleError::IndexOutOfRange(5));
+            }
+            self.ttl = Some(TTL::from_bytes(&bytes[index..index + TTL_BYTES_LENGTH]));
+            index += TTL_BYTES_LENGTH;
+        }
+
+        // Pairs
+        if index < len_bytes && self.has_pairs() {
+            if index + 2 > len_bytes {
+                return Err(NeedleError::IndexOutOfRange(6));
+            }
+            self.pairs_size = u16::from_be_bytes([bytes[index], bytes[index + 1]]);
+            index += 2;
+            if index + self.pairs_size as usize > len_bytes {
+                return Err(NeedleError::IndexOutOfRange(7));
+            }
+            self.pairs = bytes[index..index + self.pairs_size as usize].to_vec();
+            index += self.pairs_size as usize;
+        }
+
+        Ok(index)
+    }
+
+    // ---- Tail reading ----
+
+    /// Read the needle tail (checksum + optional timestamp + padding).
+    pub fn read_tail(&mut self, tail_bytes: &[u8], version: Version) -> Result<(), NeedleError> {
+        if tail_bytes.len() < NEEDLE_CHECKSUM_SIZE {
+            return Err(NeedleError::TailTooShort);
+        }
+
+        let expected_checksum = CRC(u32::from_be_bytes([
+            tail_bytes[0],
+            tail_bytes[1],
+            tail_bytes[2],
+            tail_bytes[3],
+        ]));
+
+        if !self.data.is_empty() {
+            let data_checksum = CRC::new(&self.data);
+            // Go double-checks: n.Checksum != crc && uint32(n.Checksum) != crc.Value()
+            // The crc.Value() path is a deprecated legacy transform for backward compat
+            // with seaweed versions prior to commit 056c480eb.
+            if expected_checksum != data_checksum
+                && expected_checksum.0 != data_checksum.legacy_value()
+            {
+                return Err(NeedleError::CrcMismatch {
+                    needle_id: self.id,
+                    got: data_checksum.0,
+                    want: expected_checksum.0,
+                });
+            }
+            self.checksum = data_checksum;
+        } else {
+            self.checksum = expected_checksum;
+        }
+
+        if version == VERSION_3 {
+            let ts_offset = NEEDLE_CHECKSUM_SIZE;
+            if tail_bytes.len() < ts_offset + TIMESTAMP_SIZE {
+                return Err(NeedleError::TailTooShort);
+            }
+            self.append_at_ns = u64::from_be_bytes([
+                tail_bytes[ts_offset],
+                tail_bytes[ts_offset + 1],
+                tail_bytes[ts_offset + 2],
+                tail_bytes[ts_offset + 3],
+                tail_bytes[ts_offset + 4],
+                tail_bytes[ts_offset + 5],
+                tail_bytes[ts_offset + 6],
+                tail_bytes[ts_offset + 7],
+            ]);
+        }
+
+        Ok(())
+    }
+
+    // ---- Full read from bytes ----
+
+    /// Read a complete needle from its raw bytes (header + body + tail).
+    pub fn read_bytes(
+        &mut self,
+        bytes: &[u8],
+        offset: i64,
+        expected_size: Size,
+        version: Version,
+    ) -> Result<(), NeedleError> {
+        self.read_header(bytes);
+
+        if self.size != expected_size {
+            return Err(NeedleError::SizeMismatch {
+                offset,
+                id: self.id,
+                found: self.size,
+                expected: expected_size,
+            });
+        }
+
+        let body_start = NEEDLE_HEADER_SIZE;
+        let body_end = body_start + self.size.0 as usize;
+
+        if version == VERSION_1 {
+            self.data = bytes[body_start..body_end].to_vec();
+        } else {
+            self.read_body_v2(&bytes[body_start..body_end])?;
+        }
+
+        self.read_tail(&bytes[body_end..], version)?;
+        Ok(())
+    }
+
+    // ---- Write (serialize) ----
+
+    /// Serialize the needle to bytes for writing to a .dat file (Version 2/3).
+    pub fn write_bytes(&mut self, version: Version) -> Vec<u8> {
+        let mut buf = Vec::with_capacity(256);
+
+        // Compute sizes (matching Go writeNeedleCommon)
+        if self.name.len() >= 255 {
+            self.name_size = 255;
+        } else {
+            self.name_size = self.name.len() as u8;
+        }
+        self.data_size = self.data.len() as u32;
+        self.mime_size = self.mime.len() as u8;
+
+        // Compute n.Size (body size, excluding header)
+        if self.data_size > 0 {
+            let mut s: i32 = 4 + self.data_size as i32 + 1; // DataSize + Data + Flags
+            if self.has_name() {
+                s += 1 + self.name_size as i32;
+            }
+            if self.has_mime() {
+                s += 1 + self.mime_size as i32;
+            }
+            if self.has_last_modified_date() {
+                s += LAST_MODIFIED_BYTES_LENGTH as i32;
+            }
+            if self.has_ttl() {
+                s += TTL_BYTES_LENGTH as i32;
+            }
+            if self.has_pairs() {
+                s += 2 + self.pairs_size as i32;
+            }
+            self.size = Size(s);
+        } else {
+            self.size = Size(0);
+        }
+
+        // Header: Cookie(4) + NeedleId(8) + Size(4) = 16 bytes
+        let mut header = [0u8; NEEDLE_HEADER_SIZE];
+        self.cookie.to_bytes(&mut header[0..COOKIE_SIZE]);
+        self.id
+            .to_bytes(&mut header[COOKIE_SIZE..COOKIE_SIZE + NEEDLE_ID_SIZE]);
+        self.size
+            .to_bytes(&mut header[COOKIE_SIZE + NEEDLE_ID_SIZE..NEEDLE_HEADER_SIZE]);
+        buf.extend_from_slice(&header);
+
+        // Body
+        if self.data_size > 0 {
+            buf.extend_from_slice(&self.data_size.to_be_bytes());
+            buf.extend_from_slice(&self.data);
+            buf.push(self.flags);
+            if self.has_name() {
+                buf.push(self.name_size);
+                buf.extend_from_slice(&self.name[..self.name_size as usize]);
+            }
+            if self.has_mime() {
+                buf.push(self.mime_size);
+                buf.extend_from_slice(&self.mime);
+            }
+            if self.has_last_modified_date() {
+                // Write 5 bytes of last_modified (lower 5 bytes of u64 big-endian)
+                let lm_bytes = self.last_modified.to_be_bytes();
+                buf.extend_from_slice(&lm_bytes[8 - LAST_MODIFIED_BYTES_LENGTH..8]);
+            }
+            if self.has_ttl() {
+                if let Some(ref ttl) = self.ttl {
+                    let mut ttl_buf = [0u8; 2];
+                    ttl.to_bytes(&mut ttl_buf);
+                    buf.extend_from_slice(&ttl_buf);
+                } else {
+                    buf.extend_from_slice(&[0u8; 2]);
+                }
+            }
+            if self.has_pairs() {
+                buf.extend_from_slice(&self.pairs_size.to_be_bytes());
+                buf.extend_from_slice(&self.pairs);
+            }
+        }
+
+        // Compute checksum
+        self.checksum = CRC::new(&self.data);
+
+        // Tail: Checksum + [V3: AppendAtNs] + Padding
+        buf.extend_from_slice(&self.checksum.0.to_be_bytes());
+        if version == VERSION_3 {
+            buf.extend_from_slice(&self.append_at_ns.to_be_bytes());
+        }
+
+        // Padding to 8-byte alignment
+        let padding = padding_length(self.size, version).0 as usize;
+        buf.extend(std::iter::repeat_n(0u8, padding));
+
+        buf
+    }
+
+    /// Total disk size of this needle including header, body, checksum, timestamp, and padding.
+    pub fn disk_size(&self, version: Version) -> i64 {
+        get_actual_size(self.size, version)
+    }
+
+    /// Compute ETag string from checksum (matching Go).
+    pub fn etag(&self) -> String {
+        etag_from_checksum(self.checksum.0)
+    }
+}
+
+// ============================================================================
+// Helper functions (matching Go)
+// ============================================================================
+
+/// Compute padding to align needle to NEEDLE_PADDING_SIZE (8 bytes).
+pub fn padding_length(needle_size: Size, version: Version) -> Size {
+    if version == VERSION_3 {
+        Size(
+            NEEDLE_PADDING_SIZE as i32
+                - ((NEEDLE_HEADER_SIZE as i32
+                    + needle_size.0
+                    + NEEDLE_CHECKSUM_SIZE as i32
+                    + TIMESTAMP_SIZE as i32)
+                    % NEEDLE_PADDING_SIZE as i32),
+        )
+    } else {
+        Size(
+            NEEDLE_PADDING_SIZE as i32
+                - ((NEEDLE_HEADER_SIZE as i32 + needle_size.0 + NEEDLE_CHECKSUM_SIZE as i32)
+                    % NEEDLE_PADDING_SIZE as i32),
+        )
+    }
+}
+
+/// Body length = Size + Checksum + [Timestamp] + Padding.
+pub fn needle_body_length(needle_size: Size, version: Version) -> i64 {
+    if version == VERSION_3 {
+        needle_size.0 as i64
+            + NEEDLE_CHECKSUM_SIZE as i64
+            + TIMESTAMP_SIZE as i64
+            + padding_length(needle_size, version).0 as i64
+    } else {
+        needle_size.0 as i64
+            + NEEDLE_CHECKSUM_SIZE as i64
+            + padding_length(needle_size, version).0 as i64
+    }
+}
+
+/// Total actual size on disk: Header + Body.
+pub fn get_actual_size(size: Size, version: Version) -> i64 {
+    NEEDLE_HEADER_SIZE as i64 + needle_body_length(size, version)
+}
+
+/// Read 5 bytes as a u64 (big-endian, zero-padded high bytes).
+fn bytes_to_u64_5(bytes: &[u8]) -> u64 {
+    assert!(bytes.len() >= 5);
+    // The 5 bytes are the LOWER 5 bytes of a big-endian u64.
+    // In Go: util.BytesToUint64(bytes[index : index+5]) reads into a uint64
+    // Go's BytesToUint64 copies into the LAST 5 bytes of an 8-byte array (big-endian).
+    let mut buf = [0u8; 8];
+    buf[3..8].copy_from_slice(&bytes[..5]);
+    u64::from_be_bytes(buf)
+}
+
+/// ETag formatted as Go: hex of big-endian u32 bytes.
+pub fn etag_from_checksum(checksum: u32) -> String {
+    let bits = checksum.to_be_bytes();
+    format!(
+        "{:02x}{:02x}{:02x}{:02x}",
+        bits[0], bits[1], bits[2], bits[3]
+    )
+}
+
+// ============================================================================
+// FileId
+// ============================================================================
+
+/// FileId = VolumeId + NeedleId + Cookie.
+/// String format: "<volume_id>,<needle_id_hex><cookie_hex>"
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct FileId {
+    pub volume_id: VolumeId,
+    pub key: NeedleId,
+    pub cookie: Cookie,
+}
+
+impl FileId {
+    pub fn new(volume_id: VolumeId, key: NeedleId, cookie: Cookie) -> Self {
+        FileId {
+            volume_id,
+            key,
+            cookie,
+        }
+    }
+
+    /// Parse "volume_id,needle_id_cookie" or "volume_id/needle_id_cookie".
+    pub fn parse(s: &str) -> Result<Self, String> {
+        let (vid_str, rest) = if let Some(pos) = s.find(',') {
+            (&s[..pos], &s[pos + 1..])
+        } else if let Some(pos) = s.find('/') {
+            (&s[..pos], &s[pos + 1..])
+        } else {
+            return Err(format!("invalid file id: {}", s));
+        };
+
+        let volume_id =
+            VolumeId::parse(vid_str).map_err(|e| format!("invalid volume id: {}", e))?;
+        let (key, cookie) = parse_needle_id_cookie(rest)?;
+        Ok(FileId {
+            volume_id,
+            key,
+            cookie,
+        })
+    }
+
+    /// Format the needle_id + cookie part as a hex string (stripping leading zeros).
+    pub fn needle_id_cookie_string(&self) -> String {
+        format_needle_id_cookie(self.key, self.cookie)
+    }
+}
+
+impl std::fmt::Display for FileId {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{},{}", self.volume_id, self.needle_id_cookie_string())
+    }
+}
+
+/// Format NeedleId + Cookie as hex, stripping leading zero bytes from NeedleId only.
+/// Matches Go: strips leading zero bytes up to NeedleIdSize (8), so cookie is always present.
+fn format_needle_id_cookie(key: NeedleId, cookie: Cookie) -> String {
+    // Encode 12 bytes: 8 for NeedleId + 4 for Cookie
+    let mut bytes = [0u8; 12];
+    key.to_bytes(&mut bytes[0..8]);
+    cookie.to_bytes(&mut bytes[8..12]);
+
+    // Strip leading zero bytes, but only within NeedleId portion (first 8 bytes)
+    let mut nonzero_index = 0;
+    while nonzero_index < NEEDLE_ID_SIZE && bytes[nonzero_index] == 0 {
+        nonzero_index += 1;
+    }
+    hex::encode(&bytes[nonzero_index..])
+}
+
+/// Parse "needle_id_cookie_hex" or "needle_id_cookie_hex_delta" into (NeedleId, Cookie).
+/// Matches Go's ParsePath + ParseNeedleIdCookie: supports an optional `_delta` suffix
+/// where delta is a decimal number added to the NeedleId (used for sub-file addressing).
+/// Rejects strings that are too short or too long, and deltas that overflow the needle id.
+pub fn parse_needle_id_cookie(s: &str) -> Result<(NeedleId, Cookie), String> {
+    // Go ParsePath: check for "_" suffix containing a decimal delta
+    let (hex_part, delta) = if let Some(underscore_pos) = s.rfind('_') {
+        if underscore_pos > 0 {
+            let delta_str = &s[underscore_pos + 1..];
+            let d: u64 = delta_str
+                .parse()
+                .map_err(|e| format!("Parse delta error: {}", e))?;
+            (&s[..underscore_pos], Some(d))
+        } else {
+            (s, None)
+        }
+    } else {
+        (s, None)
+    };
+
+    // Go: len(key_hash_string) <= CookieSize*2 => error (must be > 8 hex chars)
+    if hex_part.len() <= COOKIE_SIZE * 2 {
+        return Err("KeyHash is too short.".to_string());
+    }
+    // Go: len(key_hash_string) > (NeedleIdSize+CookieSize)*2 => error (must be <= 24 hex chars)
+    if hex_part.len() > (NEEDLE_ID_SIZE + COOKIE_SIZE) * 2 {
+        return Err("KeyHash is too long.".to_string());
+    }
+
+    // Split: last CookieSize*2 hex chars are cookie, rest is needle id
+    let split = hex_part.len() - COOKIE_SIZE * 2;
+    let needle_id_hex = &hex_part[..split];
+    let cookie_hex = &hex_part[split..];
+
+    // Go parses both with strconv.ParseUint, which accepts odd-length hex
+    let mut key = NeedleId(
+        u64::from_str_radix(needle_id_hex, 16)
+            .map_err(|e| format!("Parse needleId error: {}", e))?,
+    );
+    let cookie = Cookie(
+        u32::from_str_radix(cookie_hex, 16).map_err(|e| format!("Parse cookie error: {}", e))?,
+    );
+
+    // Apply delta if present (Go: n.Id += Uint64ToNeedleId(d))
+    if let Some(d) = delta {
+        key = NeedleId(
+            key.0
+                .checked_add(d)
+                .ok_or_else(|| format!("delta {} overflows needle id {:x}", d, key.0))?,
+        );
+    }
+
+    Ok((key, cookie))
+}
+
+// ============================================================================
+// Errors
+// ============================================================================
+
+#[derive(Debug, thiserror::Error)]
+pub enum NeedleError {
+    #[error(
+        "size mismatch at offset {offset}: found id={id} size={found:?}, expected size={expected:?}"
+    )]
+    SizeMismatch {
+        offset: i64,
+        id: NeedleId,
+        found: Size,
+        expected: Size,
+    },
+
+    #[error("CRC mismatch for needle {needle_id}: got {got:08x}, want {want:08x}")]
+    CrcMismatch {
+        needle_id: NeedleId,
+        got: u32,
+        want: u32,
+    },
+
+    #[error("index out of range ({0})")]
+    IndexOutOfRange(u32),
+
+    #[error("needle tail too short")]
+    TailTooShort,
+
+    #[error("unsupported version: {0}")]
+    UnsupportedVersion(u8),
+
+    #[error("IO error: {0}")]
+    Io(#[from] std::io::Error),
+}
+
+// ============================================================================
+// Tests
+// ============================================================================
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_parse_header() {
+        let mut buf = [0u8; NEEDLE_HEADER_SIZE];
+        let cookie = Cookie(0xdeadbeef);
+        let id = NeedleId(0x123456789abcdef0);
+        let size = Size(1024);
+        cookie.to_bytes(&mut buf[0..4]);
+        id.to_bytes(&mut buf[4..12]);
+        size.to_bytes(&mut buf[12..16]);
+
+        let (c, i, s) = Needle::parse_header(&buf);
+        assert_eq!(c, cookie);
+        assert_eq!(i, id);
+        assert_eq!(s, size);
+    }
+
+    #[test]
+    fn test_needle_write_read_round_trip_v3() {
+        let mut n = Needle {
+            cookie: Cookie(42),
+            id: NeedleId(100),
+            data: b"hello world".to_vec(),
+            flags: 0,
+            ..Needle::default()
+        };
+        n.set_has_name();
+        n.name = b"test.txt".to_vec();
+        n.name_size = 8;
+        n.set_has_mime();
+        n.mime = b"text/plain".to_vec();
+        n.mime_size = 10;
+        n.set_has_last_modified_date();
+        n.last_modified = 1234567890;
+        n.set_has_ttl();
+        n.ttl = Some(TTL {
+            count: 5,
+            unit: super::super::ttl::TTL_UNIT_DAY,
+        });
+        n.append_at_ns = 999999999;
+
+        let bytes = n.write_bytes(VERSION_3);
+
+        // Verify total size matches expected
+        let expected_size = get_actual_size(n.size, VERSION_3);
+        assert_eq!(bytes.len() as i64, expected_size);
+
+        // Read it back
+        let mut n2 = Needle::default();
+        n2.read_bytes(&bytes, 0, n.size, VERSION_3).unwrap();
+
+        assert_eq!(n2.cookie, n.cookie);
+        assert_eq!(n2.id, n.id);
+        assert_eq!(n2.data, n.data);
+        assert_eq!(n2.name, n.name);
+        assert_eq!(n2.mime, n.mime);
+        assert_eq!(n2.last_modified, n.last_modified);
+        assert_eq!(n2.ttl, n.ttl);
+        assert_eq!(n2.checksum, n.checksum);
+        assert_eq!(n2.append_at_ns, n.append_at_ns);
+    }
+
+    #[test]
+    fn test_needle_write_read_round_trip_v2() {
+        let mut n = Needle {
+            cookie: Cookie(77),
+            id: NeedleId(200),
+            data: b"data v2".to_vec(),
+            flags: 0,
+            ..Needle::default()
+        };
+
+        let bytes = n.write_bytes(VERSION_2);
+        let expected_size = get_actual_size(n.size, VERSION_2);
+        assert_eq!(bytes.len() as i64, expected_size);
+
+        let mut n2 = Needle::default();
+        n2.read_bytes(&bytes, 0, n.size, VERSION_2).unwrap();
+
+        assert_eq!(n2.data, n.data);
+        assert_eq!(n2.checksum, n.checksum);
+    }
+
+    #[test]
+    fn test_read_bytes_meta_only_handles_tombstone_v3() {
+        let mut tombstone = Needle {
+            cookie: Cookie(0x1234abcd),
+            id: NeedleId(300),
+            append_at_ns: 999_999,
+            ..Needle::default()
+        };
+
+        let bytes = tombstone.write_bytes(VERSION_3);
+
+        let mut meta = Needle::default();
+        meta.read_bytes_meta_only(&bytes, 0, Size(0), VERSION_3)
+            .unwrap();
+
+        assert_eq!(meta.cookie, tombstone.cookie);
+        assert_eq!(meta.id, tombstone.id);
+        assert_eq!(meta.size, Size(0));
+        assert_eq!(meta.data_size, 0);
+        assert_eq!(meta.append_at_ns, tombstone.append_at_ns);
+        assert_eq!(meta.checksum, tombstone.checksum);
+    }
+
+    #[test]
+    fn test_padding_alignment() {
+        // All actual sizes should be multiples of 8
+        for size_val in 0..50 {
+            let s = Size(size_val);
+            let actual_v2 = get_actual_size(s, VERSION_2);
+            let actual_v3 = get_actual_size(s, VERSION_3);
+            assert_eq!(actual_v2 % 8, 0, "V2 size {} not aligned", size_val);
+            assert_eq!(actual_v3 % 8, 0, "V3 size {} not aligned", size_val);
+        }
+    }
+
+    #[test]
+    fn test_file_id_parse() {
+        let fid = FileId::parse("3,01637037d6").unwrap();
+        assert_eq!(fid.volume_id, VolumeId(3));
+        // The hex "01637037d6" is 5 bytes = 0x0163..., padded to 12 bytes
+        assert!(!fid.key.is_empty() || !fid.cookie.0 == 0);
+    }
+
+    #[test]
+    fn test_file_id_round_trip() {
+        let fid = FileId::new(VolumeId(5), NeedleId(0x123456), Cookie(0xabcd));
+        let s = fid.to_string();
+        let fid2 = FileId::parse(&s).unwrap();
+        assert_eq!(fid, fid2);
+    }
+
+    #[test]
+    fn test_needle_id_cookie_format() {
+        let s = format_needle_id_cookie(NeedleId(1), Cookie(0x12345678));
+        let (key, cookie) = parse_needle_id_cookie(&s).unwrap();
+        assert_eq!(key, NeedleId(1));
+        assert_eq!(cookie, Cookie(0x12345678));
+    }
+
+    #[test]
+    fn test_parse_rejects_delta_overflow() {
+        assert!(parse_needle_id_cookie("ffffffffffffffff00000000_1").is_err());
+        let (key, _) = parse_needle_id_cookie("fffffffffffffffe00000000_1").unwrap();
+        assert_eq!(key, NeedleId(u64::MAX));
+    }
+
+    #[test]
+    fn test_parse_odd_length_needle_id() {
+        // Go's shell formats purge fids with strconv.FormatUint, which does not
+        // pad the needle id hex to an even length.
+        let (key, cookie) = parse_needle_id_cookie("100000000").unwrap();
+        assert_eq!(key, NeedleId(1));
+        assert_eq!(cookie, Cookie(0));
+
+        let fid = FileId::parse("3,12300000000").unwrap();
+        assert_eq!(fid.volume_id, VolumeId(3));
+        assert_eq!(fid.key, NeedleId(0x123));
+        assert_eq!(fid.cookie, Cookie(0));
+    }
+}

seaweed-volume/src/storage/needle/ttl.rs

@@ -0,0 +1,383 @@
+//! Time-to-live encoding for needles.
+//!
+//! TTL is stored as 2 bytes: Count(1) + Unit(1).
+//! Supported units: minute(m), hour(h), day(d), week(w), month(M), year(y).
+
+use std::fmt;
+
+/// TTL unit constants (matching Go).
+pub const TTL_UNIT_EMPTY: u8 = 0;
+pub const TTL_UNIT_MINUTE: u8 = 1;
+pub const TTL_UNIT_HOUR: u8 = 2;
+pub const TTL_UNIT_DAY: u8 = 3;
+pub const TTL_UNIT_WEEK: u8 = 4;
+pub const TTL_UNIT_MONTH: u8 = 5;
+pub const TTL_UNIT_YEAR: u8 = 6;
+
+pub const TTL_BYTES_LENGTH: usize = 2;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub struct TTL {
+    pub count: u8,
+    pub unit: u8,
+}
+
+impl TTL {
+    pub const EMPTY: TTL = TTL { count: 0, unit: 0 };
+
+    pub fn is_empty(&self) -> bool {
+        self.count == 0 && self.unit == 0
+    }
+
+    /// Load from 2 bytes.
+    pub fn from_bytes(input: &[u8]) -> Self {
+        if input.len() < 2 {
+            return TTL::EMPTY;
+        }
+        TTL {
+            count: input[0],
+            unit: input[1],
+        }
+    }
+
+    /// Serialize to 2 bytes.
+    pub fn to_bytes(&self, output: &mut [u8]) {
+        assert!(output.len() >= 2);
+        output[0] = self.count;
+        output[1] = self.unit;
+    }
+
+    /// Encode as u32: (count << 8) | unit.
+    pub fn to_u32(&self) -> u32 {
+        if self.count == 0 {
+            return 0;
+        }
+        ((self.count as u32) << 8) + (self.unit as u32)
+    }
+
+    /// Decode from u32.
+    pub fn from_u32(v: u32) -> Self {
+        if v == 0 {
+            return TTL::EMPTY;
+        }
+        TTL {
+            count: (v >> 8) as u8,
+            unit: (v & 0xFF) as u8,
+        }
+    }
+
+    /// Convert to total seconds.
+    pub fn to_seconds(&self) -> u64 {
+        unit_to_seconds(self.count as u64, self.unit)
+    }
+
+    /// Parse from string like "3m", "4h", "5d", "6w", "7M", "8y".
+    /// If the string is all digits (no unit suffix), defaults to minutes.
+    /// Matches Go's ReadTTL which calls fitTtlCount to normalize:
+    /// e.g. "120m" -> 2h, "7d" -> 1w, "24h" -> 1d.
+    pub fn read(s: &str) -> Result<Self, String> {
+        let s = s.trim();
+        if s.is_empty() {
+            return Ok(TTL::EMPTY);
+        }
+        let last_byte = s.as_bytes()[s.len() - 1];
+        let (num_str, unit_byte) = if last_byte.is_ascii_digit() {
+            // All digits — default to minutes (matching Go)
+            (s, b'm')
+        } else {
+            (&s[..s.len() - 1], last_byte)
+        };
+        let count: u32 = num_str
+            .parse()
+            .map_err(|e| format!("invalid TTL count: {}", e))?;
+        let unit = match unit_byte {
+            b'm' => TTL_UNIT_MINUTE,
+            b'h' => TTL_UNIT_HOUR,
+            b'd' => TTL_UNIT_DAY,
+            b'w' => TTL_UNIT_WEEK,
+            b'M' => TTL_UNIT_MONTH,
+            b'y' => TTL_UNIT_YEAR,
+            _ => return Err(format!("unknown TTL unit: {}", unit_byte as char)),
+        };
+        // Match Go's ReadTTL: normalize via fitTtlCount
+        Ok(fit_ttl_count(count, unit))
+    }
+
+    /// Minutes representation.
+    pub fn minutes(&self) -> u32 {
+        (self.to_seconds() / 60) as u32
+    }
+}
+
+fn unit_to_seconds(count: u64, unit: u8) -> u64 {
+    match unit {
+        TTL_UNIT_EMPTY => 0,
+        TTL_UNIT_MINUTE => count * 60,
+        TTL_UNIT_HOUR => count * 60 * 60,
+        TTL_UNIT_DAY => count * 60 * 60 * 24,
+        TTL_UNIT_WEEK => count * 60 * 60 * 24 * 7,
+        TTL_UNIT_MONTH => count * 60 * 60 * 24 * 30,
+        TTL_UNIT_YEAR => count * 60 * 60 * 24 * 365,
+        _ => 0,
+    }
+}
+
+/// Fit a count+unit into a TTL that fits in a single byte count.
+/// Converts to seconds first, then finds the coarsest unit that fits.
+/// Matches Go's fitTtlCount called from ReadTTL.
+fn fit_ttl_count(count: u32, unit: u8) -> TTL {
+    if count == 0 || unit == TTL_UNIT_EMPTY {
+        return TTL::EMPTY;
+    }
+
+    // Always convert to seconds and normalize (matches Go).
+    let seconds = unit_to_seconds(count as u64, unit);
+    if seconds == 0 {
+        return TTL::EMPTY;
+    }
+
+    const YEAR_SECS: u64 = 3600 * 24 * 365;
+    const MONTH_SECS: u64 = 3600 * 24 * 30;
+    const WEEK_SECS: u64 = 3600 * 24 * 7;
+    const DAY_SECS: u64 = 3600 * 24;
+    const HOUR_SECS: u64 = 3600;
+    const MINUTE_SECS: u64 = 60;
+
+    // First pass: try exact fits from largest to smallest
+    if seconds.is_multiple_of(YEAR_SECS) && seconds / YEAR_SECS < 256 {
+        return TTL {
+            count: (seconds / YEAR_SECS) as u8,
+            unit: TTL_UNIT_YEAR,
+        };
+    }
+    if seconds.is_multiple_of(MONTH_SECS) && seconds / MONTH_SECS < 256 {
+        return TTL {
+            count: (seconds / MONTH_SECS) as u8,
+            unit: TTL_UNIT_MONTH,
+        };
+    }
+    if seconds.is_multiple_of(WEEK_SECS) && seconds / WEEK_SECS < 256 {
+        return TTL {
+            count: (seconds / WEEK_SECS) as u8,
+            unit: TTL_UNIT_WEEK,
+        };
+    }
+    if seconds.is_multiple_of(DAY_SECS) && seconds / DAY_SECS < 256 {
+        return TTL {
+            count: (seconds / DAY_SECS) as u8,
+            unit: TTL_UNIT_DAY,
+        };
+    }
+    if seconds.is_multiple_of(HOUR_SECS) && seconds / HOUR_SECS < 256 {
+        return TTL {
+            count: (seconds / HOUR_SECS) as u8,
+            unit: TTL_UNIT_HOUR,
+        };
+    }
+    // Minutes: truncating division
+    if seconds / MINUTE_SECS < 256 {
+        return TTL {
+            count: (seconds / MINUTE_SECS) as u8,
+            unit: TTL_UNIT_MINUTE,
+        };
+    }
+    // Second pass: truncating division from smallest to largest
+    if seconds / HOUR_SECS < 256 {
+        return TTL {
+            count: (seconds / HOUR_SECS) as u8,
+            unit: TTL_UNIT_HOUR,
+        };
+    }
+    if seconds / DAY_SECS < 256 {
+        return TTL {
+            count: (seconds / DAY_SECS) as u8,
+            unit: TTL_UNIT_DAY,
+        };
+    }
+    if seconds / WEEK_SECS < 256 {
+        return TTL {
+            count: (seconds / WEEK_SECS) as u8,
+            unit: TTL_UNIT_WEEK,
+        };
+    }
+    if seconds / MONTH_SECS < 256 {
+        return TTL {
+            count: (seconds / MONTH_SECS) as u8,
+            unit: TTL_UNIT_MONTH,
+        };
+    }
+    if seconds / YEAR_SECS < 256 {
+        return TTL {
+            count: (seconds / YEAR_SECS) as u8,
+            unit: TTL_UNIT_YEAR,
+        };
+    }
+    TTL::EMPTY
+}
+
+fn unit_to_char(unit: u8) -> char {
+    match unit {
+        TTL_UNIT_MINUTE => 'm',
+        TTL_UNIT_HOUR => 'h',
+        TTL_UNIT_DAY => 'd',
+        TTL_UNIT_WEEK => 'w',
+        TTL_UNIT_MONTH => 'M',
+        TTL_UNIT_YEAR => 'y',
+        _ => ' ',
+    }
+}
+
+impl fmt::Display for TTL {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        if self.count == 0 || self.unit == TTL_UNIT_EMPTY {
+            return write!(f, "");
+        }
+        write!(f, "{}{}", self.count, unit_to_char(self.unit))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_ttl_parse() {
+        let ttl = TTL::read("3m").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 3,
+                unit: TTL_UNIT_MINUTE
+            }
+        );
+        assert_eq!(ttl.to_seconds(), 180);
+    }
+
+    #[test]
+    fn test_ttl_parse_hours() {
+        // 24h normalizes to 1d via fitTtlCount
+        let ttl = TTL::read("24h").unwrap();
+        assert_eq!(ttl.to_seconds(), 86400);
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 1,
+                unit: TTL_UNIT_DAY
+            }
+        );
+    }
+
+    #[test]
+    fn test_ttl_display() {
+        let ttl = TTL {
+            count: 5,
+            unit: TTL_UNIT_DAY,
+        };
+        assert_eq!(ttl.to_string(), "5d");
+    }
+
+    #[test]
+    fn test_ttl_bytes_round_trip() {
+        let ttl = TTL {
+            count: 10,
+            unit: TTL_UNIT_WEEK,
+        };
+        let mut buf = [0u8; 2];
+        ttl.to_bytes(&mut buf);
+        let ttl2 = TTL::from_bytes(&buf);
+        assert_eq!(ttl, ttl2);
+    }
+
+    #[test]
+    fn test_ttl_u32_round_trip() {
+        let ttl = TTL {
+            count: 42,
+            unit: TTL_UNIT_HOUR,
+        };
+        let v = ttl.to_u32();
+        let ttl2 = TTL::from_u32(v);
+        assert_eq!(ttl, ttl2);
+    }
+
+    #[test]
+    fn test_ttl_empty() {
+        assert!(TTL::EMPTY.is_empty());
+        assert_eq!(TTL::EMPTY.to_seconds(), 0);
+        assert_eq!(TTL::EMPTY.to_u32(), 0);
+    }
+
+    #[test]
+    fn test_ttl_overflow_normalizes() {
+        // Go's ReadTTL calls fitTtlCount: 300m = 18000s = 5h (exact fit)
+        let ttl = TTL::read("300m").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 5,
+                unit: TTL_UNIT_HOUR
+            }
+        );
+
+        // 256h = 921600s. Doesn't fit in hours (256 >= 256), doesn't fit exact in days.
+        // Second pass: 921600/86400 = 10 (truncated) < 256 -> 10d
+        let ttl = TTL::read("256h").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 10,
+                unit: TTL_UNIT_DAY
+            }
+        );
+    }
+
+    #[test]
+    fn test_ttl_normalizes_unit() {
+        // Go's ReadTTL calls fitTtlCount which normalizes to coarsest unit.
+        // 120m -> 2h, 7d -> 1w, 24h -> 1d.
+        let ttl = TTL::read("120m").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 2,
+                unit: TTL_UNIT_HOUR
+            }
+        );
+
+        let ttl = TTL::read("7d").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 1,
+                unit: TTL_UNIT_WEEK
+            }
+        );
+
+        let ttl = TTL::read("24h").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 1,
+                unit: TTL_UNIT_DAY
+            }
+        );
+
+        // Values that don't simplify stay as-is
+        let ttl = TTL::read("5d").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 5,
+                unit: TTL_UNIT_DAY
+            }
+        );
+
+        let ttl = TTL::read("3m").unwrap();
+        assert_eq!(
+            ttl,
+            TTL {
+                count: 3,
+                unit: TTL_UNIT_MINUTE
+            }
+        );
+    }
+}

seaweed-volume/src/storage/needle_map/compact_map.rs

@@ -0,0 +1,375 @@
+//! CompactMap: memory-efficient in-memory map of NeedleId -> (Offset, Size).
+//!
+//! Port of Go's CompactMap from weed/storage/needle_map/compact_map.go.
+//! Uses segmented sorted arrays with compressed keys (u16 instead of u64)
+//! to achieve ~10 bytes per entry instead of ~40-48 bytes with HashMap.
+//!
+//! NeedleId is split into: chunk = id / SEGMENT_CHUNK_SIZE, compact_key = id % SEGMENT_CHUNK_SIZE.
+//! Each segment stores up to SEGMENT_CHUNK_SIZE entries in a sorted Vec, searched via binary search.
+//! Best case (ordered inserts): O(1). Worst case: O(log n) per segment.
+
+use std::collections::HashMap;
+
+use super::NeedleValue;
+use crate::storage::types::*;
+
+/// Maximum entries per segment. Must be <= u16::MAX (65535).
+const SEGMENT_CHUNK_SIZE: u64 = 50_000;
+
+/// Compact key: only the low bits of NeedleId within a segment.
+type CompactKey = u16;
+
+/// Segment chunk identifier: NeedleId / SEGMENT_CHUNK_SIZE.
+type Chunk = u64;
+
+/// Compact entry: 10 bytes (2 + 4 + 4) vs 16 bytes for full NeedleId + NeedleValue.
+#[derive(Clone, Copy)]
+struct CompactEntry {
+    key: CompactKey,           // 2 bytes
+    offset: [u8; OFFSET_SIZE], // 4 bytes
+    size: Size,                // 4 bytes
+}
+
+impl CompactEntry {
+    fn to_needle_value(self) -> NeedleValue {
+        NeedleValue {
+            offset: Offset::from_bytes(&self.offset),
+            size: self.size,
+        }
+    }
+}
+
+/// A sorted segment of compact entries for a given chunk.
+struct Segment {
+    list: Vec<CompactEntry>,
+    chunk: Chunk,
+    first_key: CompactKey,
+    last_key: CompactKey,
+}
+
+impl Segment {
+    fn new(chunk: Chunk) -> Self {
+        Segment {
+            list: Vec::new(),
+            chunk,
+            first_key: u16::MAX,
+            last_key: 0,
+        }
+    }
+
+    fn compact_key(&self, id: NeedleId) -> CompactKey {
+        (id.0 - SEGMENT_CHUNK_SIZE * self.chunk) as CompactKey
+    }
+
+    /// Binary search for a compact key. Returns (index, found).
+    /// If not found, index is the insertion point.
+    fn bsearch(&self, id: NeedleId) -> (usize, bool) {
+        let ck = self.compact_key(id);
+
+        if self.list.is_empty() {
+            return (0, false);
+        }
+        if ck == self.first_key {
+            return (0, true);
+        }
+        if ck < self.first_key {
+            return (0, false);
+        }
+        if ck == self.last_key {
+            return (self.list.len() - 1, true);
+        }
+        if ck > self.last_key {
+            return (self.list.len(), false);
+        }
+
+        let i = self.list.partition_point(|e| e.key < ck);
+        if i < self.list.len() && self.list[i].key == ck {
+            (i, true)
+        } else {
+            (i, false)
+        }
+    }
+
+    /// Insert or update. Returns old NeedleValue if updating.
+    fn set(&mut self, id: NeedleId, offset: Offset, size: Size) -> Option<NeedleValue> {
+        let (i, found) = self.bsearch(id);
+
+        if found {
+            let old = self.list[i].to_needle_value();
+            let mut offset_bytes = [0u8; OFFSET_SIZE];
+            offset.to_bytes(&mut offset_bytes);
+            self.list[i].offset = offset_bytes;
+            self.list[i].size = size;
+            return Some(old);
+        }
+
+        // Insert at sorted position
+        let ck = self.compact_key(id);
+        let mut offset_bytes = [0u8; OFFSET_SIZE];
+        offset.to_bytes(&mut offset_bytes);
+
+        let entry = CompactEntry {
+            key: ck,
+            offset: offset_bytes,
+            size,
+        };
+
+        // Match Go panic: don't exceed segment capacity
+        if self.list.len() >= SEGMENT_CHUNK_SIZE as usize {
+            panic!(
+                "attempted to write more than {} entries on CompactMapSegment",
+                SEGMENT_CHUNK_SIZE
+            );
+        }
+
+        if self.list.len() == SEGMENT_CHUNK_SIZE as usize - 1 {
+            // Pin capacity to exact size when maxing out
+            let mut new_list = Vec::with_capacity(SEGMENT_CHUNK_SIZE as usize);
+            new_list.extend_from_slice(&self.list[..i]);
+            new_list.push(entry);
+            new_list.extend_from_slice(&self.list[i..]);
+            self.list = new_list;
+        } else {
+            self.list.insert(i, entry);
+        }
+
+        if ck < self.first_key {
+            self.first_key = ck;
+        }
+        if ck > self.last_key {
+            self.last_key = ck;
+        }
+
+        None
+    }
+
+    fn get(&self, id: NeedleId) -> Option<NeedleValue> {
+        let (i, found) = self.bsearch(id);
+        if found {
+            Some(self.list[i].to_needle_value())
+        } else {
+            None
+        }
+    }
+
+    /// Mark as deleted by negating size. Returns previous size if not already deleted.
+    /// Matches Go behavior: checks !IsDeleted() (i.e., size >= 0).
+    fn delete(&mut self, id: NeedleId) -> Option<Size> {
+        let (i, found) = self.bsearch(id);
+        if found && !self.list[i].size.is_deleted() {
+            let old_size = self.list[i].size;
+            if self.list[i].size.0 == 0 {
+                self.list[i].size = TOMBSTONE_FILE_SIZE;
+            } else {
+                self.list[i].size = Size(-self.list[i].size.0);
+            }
+            Some(old_size)
+        } else {
+            None
+        }
+    }
+}
+
+/// Memory-efficient map of NeedleId -> (Offset, Size).
+/// Segments NeedleIds into chunks of 50,000 and stores compact 10-byte entries
+/// in sorted arrays, using only 2 bytes for the key within each segment.
+pub struct CompactMap {
+    segments: HashMap<Chunk, Segment>,
+}
+
+impl CompactMap {
+    pub fn new() -> Self {
+        CompactMap {
+            segments: HashMap::new(),
+        }
+    }
+
+    fn _segment_for_key(&mut self, id: NeedleId) -> &mut Segment {
+        let chunk = id.0 / SEGMENT_CHUNK_SIZE;
+        self.segments
+            .entry(chunk)
+            .or_insert_with(|| Segment::new(chunk))
+    }
+
+    /// Insert or update. Returns old NeedleValue if updating.
+    pub fn set(&mut self, id: NeedleId, offset: Offset, size: Size) -> Option<NeedleValue> {
+        let chunk = id.0 / SEGMENT_CHUNK_SIZE;
+        let segment = self
+            .segments
+            .entry(chunk)
+            .or_insert_with(|| Segment::new(chunk));
+        segment.set(id, offset, size)
+    }
+
+    pub fn get(&self, id: NeedleId) -> Option<NeedleValue> {
+        let chunk = id.0 / SEGMENT_CHUNK_SIZE;
+        self.segments.get(&chunk)?.get(id)
+    }
+
+    /// Mark as deleted. Returns previous size if was valid.
+    pub fn delete(&mut self, id: NeedleId) -> Option<Size> {
+        let chunk = id.0 / SEGMENT_CHUNK_SIZE;
+        self.segments.get_mut(&chunk)?.delete(id)
+    }
+
+    /// Remove entry entirely (used during idx loading).
+    pub fn remove(&mut self, id: NeedleId) -> Option<NeedleValue> {
+        let chunk = id.0 / SEGMENT_CHUNK_SIZE;
+        let segment = self.segments.get_mut(&chunk)?;
+        let (i, found) = segment.bsearch(id);
+        if found {
+            let entry = segment.list.remove(i);
+            // Update first/last keys
+            if segment.list.is_empty() {
+                segment.first_key = u16::MAX;
+                segment.last_key = 0;
+            } else {
+                segment.first_key = segment.list[0].key;
+                segment.last_key = segment.list[segment.list.len() - 1].key;
+            }
+            Some(entry.to_needle_value())
+        } else {
+            None
+        }
+    }
+
+    /// Iterate all entries in ascending NeedleId order.
+    pub fn ascending_visit<F, E>(&self, mut f: F) -> Result<(), E>
+    where
+        F: FnMut(NeedleId, &NeedleValue) -> Result<(), E>,
+    {
+        let mut chunks: Vec<Chunk> = self.segments.keys().copied().collect();
+        chunks.sort_unstable();
+
+        for chunk in chunks {
+            let segment = &self.segments[&chunk];
+            for entry in &segment.list {
+                let id = NeedleId(SEGMENT_CHUNK_SIZE * segment.chunk + entry.key as u64);
+                let nv = entry.to_needle_value();
+                f(id, &nv)?;
+            }
+        }
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn offset(v: u32) -> Offset {
+        Offset::from_actual_offset(v as i64 * NEEDLE_PADDING_SIZE as i64)
+    }
+
+    #[test]
+    fn test_compact_map_basic() {
+        let mut m = CompactMap::new();
+
+        // Insert
+        assert!(m.set(NeedleId(1), offset(100), Size(50)).is_none());
+        assert!(m.set(NeedleId(2), offset(200), Size(60)).is_none());
+
+        // Get
+        let nv = m.get(NeedleId(1)).unwrap();
+        assert_eq!(nv.size, Size(50));
+
+        // Update returns old value
+        let old = m.set(NeedleId(1), offset(300), Size(70)).unwrap();
+        assert_eq!(old.size, Size(50));
+
+        // Get updated value
+        let nv = m.get(NeedleId(1)).unwrap();
+        assert_eq!(nv.size, Size(70));
+
+        // Miss
+        assert!(m.get(NeedleId(999)).is_none());
+    }
+
+    #[test]
+    fn test_compact_map_delete() {
+        let mut m = CompactMap::new();
+        m.set(NeedleId(1), offset(100), Size(50));
+
+        // Delete returns old size
+        let old = m.delete(NeedleId(1)).unwrap();
+        assert_eq!(old, Size(50));
+
+        // Get returns deleted (negative size)
+        let nv = m.get(NeedleId(1)).unwrap();
+        assert!(nv.size.is_deleted());
+
+        // Delete again returns None (already deleted)
+        assert!(m.delete(NeedleId(1)).is_none());
+    }
+
+    #[test]
+    fn test_compact_map_zero_size_delete() {
+        let mut m = CompactMap::new();
+        m.set(NeedleId(1), offset(100), Size(0));
+
+        let old = m.delete(NeedleId(1)).unwrap();
+        assert_eq!(old, Size(0));
+
+        let nv = m.get(NeedleId(1)).unwrap();
+        assert_eq!(nv.size, TOMBSTONE_FILE_SIZE);
+    }
+
+    #[test]
+    fn test_compact_map_cross_segment() {
+        let mut m = CompactMap::new();
+
+        // Insert across multiple segments
+        m.set(NeedleId(1), offset(1), Size(1));
+        m.set(NeedleId(50_000), offset(2), Size(2));
+        m.set(NeedleId(100_000), offset(3), Size(3));
+
+        assert_eq!(m.get(NeedleId(1)).unwrap().size, Size(1));
+        assert_eq!(m.get(NeedleId(50_000)).unwrap().size, Size(2));
+        assert_eq!(m.get(NeedleId(100_000)).unwrap().size, Size(3));
+    }
+
+    #[test]
+    fn test_compact_map_ascending_visit() {
+        let mut m = CompactMap::new();
+        m.set(NeedleId(100_005), offset(3), Size(3));
+        m.set(NeedleId(5), offset(1), Size(1));
+        m.set(NeedleId(50_005), offset(2), Size(2));
+
+        let mut visited = Vec::new();
+        m.ascending_visit(|id, nv| {
+            visited.push((id, nv.size));
+            Ok::<_, String>(())
+        })
+        .unwrap();
+
+        assert_eq!(visited.len(), 3);
+        assert_eq!(visited[0].0, NeedleId(5));
+        assert_eq!(visited[1].0, NeedleId(50_005));
+        assert_eq!(visited[2].0, NeedleId(100_005));
+    }
+
+    #[test]
+    fn test_compact_map_remove() {
+        let mut m = CompactMap::new();
+        m.set(NeedleId(1), offset(100), Size(50));
+        m.set(NeedleId(2), offset(200), Size(60));
+
+        let removed = m.remove(NeedleId(1)).unwrap();
+        assert_eq!(removed.size, Size(50));
+
+        assert!(m.get(NeedleId(1)).is_none());
+        assert_eq!(m.get(NeedleId(2)).unwrap().size, Size(60));
+    }
+
+    #[test]
+    fn test_compact_map_reverse_insert_order() {
+        let mut m = CompactMap::new();
+        // Insert in reverse order to test sorted insert
+        for i in (0..100).rev() {
+            m.set(NeedleId(i), offset(i as u32), Size(i as i32));
+        }
+        for i in 0..100 {
+            assert_eq!(m.get(NeedleId(i)).unwrap().size, Size(i as i32));
+        }
+    }
+}

seaweed-volume/src/storage/needle_map/file_pool.rs

@@ -0,0 +1,252 @@
+//! Bounded pool of open index-file descriptors.
+//!
+//! Read-only volumes — cloud-tiered ones above all — outnumber writable ones by
+//! orders of magnitude on a large server, and a volume that pins its `.idx` and
+//! `.sdx` for the life of the process costs two descriptors whether or not
+//! anybody reads it. At ~600K volumes per server that alone exhausts any fd
+//! limit. Neither file is needed except while a lookup is in flight, so
+//! [`SortedFileNeedleMap`](super::sorted_file::SortedFileNeedleMap) borrows them
+//! from this pool: an idle volume holds nothing, a busy one keeps its handles
+//! hot rather than paying an `open()` per needle.
+//!
+//! Mirrors Go's `weed/storage/needle_map_file_pool.go`. Handles are handed out
+//! as `Arc<File>`, so an eviction cannot close a descriptor a reader still
+//! holds — the file closes when the last borrower drops its `Arc`.
+
+use std::collections::{BTreeMap, HashMap};
+use std::fs::{File, OpenOptions};
+use std::io;
+use std::sync::{Arc, Mutex, OnceLock};
+
+use crate::storage::volume_open::open_volume_file;
+
+/// Descriptors the pool keeps open. Matches Go's `maxPooledIndexFiles`.
+pub const MAX_POOLED_INDEX_FILES: usize = 1024;
+
+struct Entry {
+    file: Arc<File>,
+    tick: u64,
+}
+
+#[derive(Default)]
+struct Inner {
+    entries: HashMap<String, Entry>,
+    /// Recency order, oldest tick first, so eviction is a `pop_first`.
+    order: BTreeMap<u64, String>,
+    next_tick: u64,
+}
+
+pub struct IndexFilePool {
+    capacity: usize,
+    inner: Mutex<Inner>,
+}
+
+/// Writable and read-only handles for the same path are pooled separately so a
+/// read never depends on the file being openable for write — a volume served
+/// off a read-only mount still answers lookups.
+fn pool_key(path: &str, writable: bool) -> String {
+    if writable {
+        format!("{path}\0rw")
+    } else {
+        path.to_string()
+    }
+}
+
+impl IndexFilePool {
+    pub fn new(capacity: usize) -> Self {
+        IndexFilePool {
+            capacity: capacity.max(1),
+            inner: Mutex::new(Inner::default()),
+        }
+    }
+
+    /// Hand out an open handle for `path`, reusing the pooled one when there is
+    /// one. The descriptor lives as long as the returned `Arc`.
+    pub fn borrow(&self, path: &str, writable: bool) -> io::Result<Arc<File>> {
+        let key = pool_key(path, writable);
+        if let Some(file) = self.touch(&key) {
+            return Ok(file);
+        }
+
+        // Opened outside the lock: a cold open blocks on disk, and holding a
+        // process-wide mutex across it would serialize every volume's lookups.
+        let file = Arc::new(open_volume_file(
+            OpenOptions::new().read(true).write(writable),
+            path,
+        )?);
+        Ok(self.insert(key, file))
+    }
+
+    /// Forget the pooled handles for `path`, so a later rename or delete of that
+    /// path cannot be served from a descriptor on the old inode.
+    pub fn discard(&self, path: &str) {
+        let mut inner = self.inner.lock().unwrap();
+        for key in [pool_key(path, false), pool_key(path, true)] {
+            if let Some(entry) = inner.entries.remove(&key) {
+                inner.order.remove(&entry.tick);
+            }
+        }
+    }
+
+    /// Descriptors currently pooled. Test-only visibility into the bound.
+    #[cfg(test)]
+    pub fn pooled_count(&self) -> usize {
+        self.inner.lock().unwrap().entries.len()
+    }
+
+    fn touch(&self, key: &str) -> Option<Arc<File>> {
+        let mut inner = self.inner.lock().unwrap();
+        let tick = inner.next_tick;
+        let entry = inner.entries.get_mut(key)?;
+        let file = entry.file.clone();
+        let old_tick = std::mem::replace(&mut entry.tick, tick);
+        inner.order.remove(&old_tick);
+        inner.order.insert(tick, key.to_string());
+        inner.next_tick += 1;
+        Some(file)
+    }
+
+    fn insert(&self, key: String, file: Arc<File>) -> Arc<File> {
+        let mut inner = self.inner.lock().unwrap();
+        if let Some(entry) = inner.entries.get(&key) {
+            // Another borrower opened the same path first; keep one descriptor.
+            return entry.file.clone();
+        }
+        let tick = inner.next_tick;
+        inner.next_tick += 1;
+        inner.order.insert(tick, key.clone());
+        inner.entries.insert(
+            key,
+            Entry {
+                file: file.clone(),
+                tick,
+            },
+        );
+        while inner.entries.len() > self.capacity {
+            let Some((_, oldest)) = inner.order.pop_first() else {
+                break;
+            };
+            inner.entries.remove(&oldest);
+        }
+        file
+    }
+}
+
+/// Process-wide pool shared by every read-only volume on this server.
+pub fn pooled_index_files() -> &'static IndexFilePool {
+    static POOL: OnceLock<IndexFilePool> = OnceLock::new();
+    POOL.get_or_init(|| IndexFilePool::new(MAX_POOLED_INDEX_FILES))
+}
+
+/// Descriptors this process holds on `.idx`/`.sdx` files under `dir`, read from
+/// `/proc/self/fd` where it exists and from `lsof` otherwise. `None` when
+/// neither is available, so a caller can skip rather than assert vacuously.
+#[cfg(test)]
+pub(crate) fn open_index_fds(dir: &std::path::Path) -> Option<usize> {
+    let prefix = std::fs::canonicalize(dir).unwrap_or_else(|_| dir.to_path_buf());
+    let is_index = |target: &std::path::Path| {
+        target.starts_with(&prefix)
+            && matches!(
+                target.extension().and_then(|e| e.to_str()),
+                Some("idx") | Some("sdx")
+            )
+    };
+
+    if let Ok(entries) = std::fs::read_dir("/proc/self/fd") {
+        return Some(
+            entries
+                .filter_map(|e| std::fs::read_link(e.ok()?.path()).ok())
+                .filter(|target| is_index(target))
+                .count(),
+        );
+    }
+
+    let out = std::process::Command::new("lsof")
+        .args(["-p", &std::process::id().to_string(), "-F", "n"])
+        .output()
+        .ok()?;
+    Some(
+        String::from_utf8_lossy(&out.stdout)
+            .lines()
+            .filter_map(|line| line.strip_prefix('n'))
+            .filter(|line| is_index(std::path::Path::new(line)))
+            .count(),
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::io::Write;
+
+    fn write_file(dir: &std::path::Path, name: &str, contents: &[u8]) -> String {
+        let path = dir.join(name);
+        let mut f = File::create(&path).unwrap();
+        f.write_all(contents).unwrap();
+        path.to_str().unwrap().to_string()
+    }
+
+    #[test]
+    fn evicted_handle_stays_usable_for_its_borrower() {
+        let dir = tempfile::tempdir().unwrap();
+        let first = write_file(dir.path(), "first", b"first");
+        let second = write_file(dir.path(), "second", b"second");
+
+        let pool = IndexFilePool::new(1);
+        let borrowed = pool.borrow(&first, false).unwrap();
+
+        // Pushes the single slot over, evicting the entry still in use.
+        let _other = pool.borrow(&second, false).unwrap();
+        assert_eq!(pool.pooled_count(), 1);
+
+        let mut buf = [0u8; 5];
+        #[cfg(unix)]
+        {
+            use std::os::unix::fs::FileExt;
+            borrowed.read_exact_at(&mut buf, 0).unwrap();
+        }
+        assert_eq!(&buf, b"first");
+    }
+
+    #[test]
+    fn borrow_reuses_the_pooled_handle() {
+        let dir = tempfile::tempdir().unwrap();
+        let path = write_file(dir.path(), "idx", b"x");
+
+        let pool = IndexFilePool::new(4);
+        let a = pool.borrow(&path, false).unwrap();
+        let b = pool.borrow(&path, false).unwrap();
+        assert!(Arc::ptr_eq(&a, &b));
+        assert_eq!(pool.pooled_count(), 1);
+
+        // A writable handle is pooled separately from the read-only one.
+        let w = pool.borrow(&path, true).unwrap();
+        assert!(!Arc::ptr_eq(&a, &w));
+        assert_eq!(pool.pooled_count(), 2);
+    }
+
+    #[test]
+    fn discard_drops_both_handles() {
+        let dir = tempfile::tempdir().unwrap();
+        let path = write_file(dir.path(), "idx", b"x");
+
+        let pool = IndexFilePool::new(4);
+        let _r = pool.borrow(&path, false).unwrap();
+        let _w = pool.borrow(&path, true).unwrap();
+        assert_eq!(pool.pooled_count(), 2);
+
+        pool.discard(&path);
+        assert_eq!(pool.pooled_count(), 0);
+    }
+
+    #[test]
+    fn pool_stays_within_capacity() {
+        let dir = tempfile::tempdir().unwrap();
+        let pool = IndexFilePool::new(3);
+        for i in 0..10 {
+            let path = write_file(dir.path(), &format!("f{i}"), b"x");
+            let _ = pool.borrow(&path, false).unwrap();
+        }
+        assert_eq!(pool.pooled_count(), 3);
+    }
+}

seaweed-volume/src/storage/needle_map/idx_metric.rs

@@ -0,0 +1,159 @@
+//! Rebuild a needle map's counters from the whole `.idx` history.
+//!
+//! The volume's garbage ratio is `deleted_size / content_size`, and both are
+//! additive over the life of the volume: an overwritten needle keeps its
+//! bytes in `content_size` and adds them to `deleted_size`. A backend whose
+//! table only holds the final value per key (redb) cannot recover that from
+//! the table, so on load the counters come from the `.idx` file instead.
+//!
+//! This mirrors Go's `needleMapMetricFromIndexFile`: walk the index newest
+//! entry first with a bloom filter of the keys already seen, so the memory
+//! cost is a few bits per entry instead of a map of every key. The counting
+//! rule reproduces what the live `on_put`/`on_delete` path accumulates:
+//!
+//! - every live entry is one put: `file_count`, `file_byte_count`;
+//! - a live entry with a newer entry for the same key was overwritten or
+//!   deleted later, so it is also one deletion: `deletion_count`,
+//!   `deletion_byte_count`;
+//! - a tombstone only marks its key as seen.
+//!
+//! A bloom false positive (0.1%) can only add a spurious deletion, which
+//! over-reports garbage slightly; it never hides any.
+
+use std::io::{self, Read, Seek, SeekFrom};
+use std::sync::atomic::Ordering;
+
+use xxhash_rust::xxh64::xxh64;
+
+use super::NeedleMapMetric;
+use crate::storage::types::*;
+
+/// Entries read per batch while walking backwards (64 KiB of index).
+const BATCH_ENTRIES: usize = 4096;
+/// Same target false-positive rate as the Go server's filter.
+const FALSE_POSITIVE_RATE: f64 = 0.001;
+
+/// Minimal bloom filter over needle ids, double hashing with xxh64.
+struct SeenKeys {
+    bits: Vec<u64>,
+    bit_count: u64,
+    hashes: u64,
+}
+
+impl SeenKeys {
+    fn new(expected: u64, false_positive_rate: f64) -> Self {
+        let n = expected.max(1) as f64;
+        let ln2 = std::f64::consts::LN_2;
+        let bit_count = (-(n * false_positive_rate.ln()) / (ln2 * ln2))
+            .ceil()
+            .max(64.0) as u64;
+        let hashes = ((bit_count as f64 / n) * ln2).round().clamp(1.0, 16.0) as u64;
+        SeenKeys {
+            bits: vec![0u64; bit_count.div_ceil(64) as usize],
+            bit_count,
+            hashes,
+        }
+    }
+
+    /// Whether `key` was (probably) seen before; marks it seen either way.
+    fn test_and_add(&mut self, key: u64) -> bool {
+        let bytes = key.to_le_bytes();
+        let h1 = xxh64(&bytes, 0);
+        let h2 = xxh64(&bytes, 0x9E37_79B9_7F4A_7C15) | 1;
+        let mut seen = true;
+        for i in 0..self.hashes {
+            let bit = h1.wrapping_add(i.wrapping_mul(h2)) % self.bit_count;
+            let word = (bit / 64) as usize;
+            let mask = 1u64 << (bit % 64);
+            if self.bits[word] & mask == 0 {
+                seen = false;
+                self.bits[word] |= mask;
+            }
+        }
+        seen
+    }
+}
+
+/// Walk `reader` (an `.idx` file) newest entry first and return the counters
+/// a live volume would hold after applying the same history. A torn partial
+/// entry at the tail is ignored, as `walk_index_file` does. The reader is
+/// left positioned at the start of the file.
+pub(super) fn metrics_from_idx<R: Read + Seek>(
+    reader: &mut R,
+    version: Version,
+) -> io::Result<NeedleMapMetric> {
+    let metric = NeedleMapMetric::default();
+    let file_size = reader.seek(SeekFrom::End(0))?;
+    let entry_count = file_size / NEEDLE_MAP_ENTRY_SIZE as u64;
+    let mut seen = SeenKeys::new(entry_count, FALSE_POSITIVE_RATE);
+    let mut buf = vec![0u8; NEEDLE_MAP_ENTRY_SIZE * BATCH_ENTRIES];
+
+    let mut remaining = entry_count;
+    while remaining > 0 {
+        let batch = remaining.min(BATCH_ENTRIES as u64) as usize;
+        let first_entry = remaining - batch as u64;
+        let len = batch * NEEDLE_MAP_ENTRY_SIZE;
+        reader.seek(SeekFrom::Start(first_entry * NEEDLE_MAP_ENTRY_SIZE as u64))?;
+        reader.read_exact(&mut buf[..len])?;
+        for i in (0..batch).rev() {
+            let entry = &buf[i * NEEDLE_MAP_ENTRY_SIZE..(i + 1) * NEEDLE_MAP_ENTRY_SIZE];
+            let (key, offset, size) = idx_entry_from_bytes(entry);
+            metric.maybe_set_max_needle_end(offset, size, version);
+            let superseded = seen.test_and_add(key.into());
+            if offset.is_zero() || size.is_deleted() {
+                // Tombstone: reserves no bytes, only marks the key as seen.
+                continue;
+            }
+            metric.maybe_set_max_file_key(key);
+            metric.file_count.fetch_add(1, Ordering::Relaxed);
+            metric
+                .file_byte_count
+                .fetch_add(size.0 as u64, Ordering::Relaxed);
+            if superseded && size.0 > 0 {
+                metric.deletion_count.fetch_add(1, Ordering::Relaxed);
+                metric
+                    .deletion_byte_count
+                    .fetch_add(size.0 as u64, Ordering::Relaxed);
+            }
+        }
+        remaining = first_entry;
+    }
+
+    reader.seek(SeekFrom::Start(0))?;
+    Ok(metric)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_seen_keys_reports_repeats_and_not_fresh_keys() {
+        let mut seen = SeenKeys::new(10_000, FALSE_POSITIVE_RATE);
+        // Fresh keys may occasionally collide (that is the false-positive
+        // rate), but only rarely.
+        let fresh_reported_seen = (0..10_000u64).filter(|&key| seen.test_and_add(key)).count();
+        assert!(
+            fresh_reported_seen < 50,
+            "fresh keys reported seen: {fresh_reported_seen}"
+        );
+        // A repeated key is never reported fresh: no false negatives.
+        for key in 0..10_000u64 {
+            assert!(seen.test_and_add(key), "repeated key {key} reported fresh");
+        }
+        // Over 100k never-inserted keys the false-positive rate stays near
+        // the 0.1% target; allow a generous margin.
+        let false_positives = (1_000_000..1_100_000u64)
+            .filter(|k| {
+                let bytes = k.to_le_bytes();
+                let h1 = xxh64(&bytes, 0);
+                let h2 = xxh64(&bytes, 0x9E37_79B9_7F4A_7C15) | 1;
+                (0..seen.hashes).all(|i| {
+                    let bit = h1.wrapping_add(i.wrapping_mul(h2)) % seen.bit_count;
+                    seen.bits[(bit / 64) as usize] & (1u64 << (bit % 64)) != 0
+                })
+            })
+            .count();
+        assert!(false_positives < 500, "false positives: {false_positives}");
+    }
+}

seaweed-volume/src/storage/store_ec_mirror.rs

@@ -0,0 +1,420 @@
+//! Physical EC sidecar mirroring across disks of the same volume
+//! server. Mirrors `weed/storage/store_ec_mirror.go`.
+
+use std::collections::HashMap;
+use std::fs;
+use std::io::{self, Read, Write};
+use std::path::Path;
+
+use tracing::{info, warn};
+
+use crate::storage::disk_location::{DiskLocation, parse_collection_volume_id_pub};
+use crate::storage::store::Store;
+use crate::storage::types::VolumeId;
+
+// Listed in `EcVolume::new`'s open order.
+const EC_MIRRORED_SIDECARS: &[&str] = &[".ecx", ".ecj", ".vif"];
+
+#[derive(Clone, Debug, PartialEq, Eq, Hash)]
+struct EcKey {
+    collection: String,
+    vid: VolumeId,
+}
+
+#[derive(Clone, Debug)]
+struct EcxOwner {
+    location: usize,
+    idx_dir: String,
+    data_dir: String,
+}
+
+impl Store {
+    /// Mirror EC sidecars onto every shard-bearing disk that lacks
+    /// them, so each disk mounts self-contained. Runs before
+    /// `reconcile_ec_shards_across_disks` so the orphan pass can
+    /// prefer the local idx_directory.
+    pub fn mirror_ec_metadata_to_shard_disks(&mut self) {
+        if self.locations.len() < 2 {
+            return;
+        }
+        let owners = self.index_ecx_owners_for_mirror();
+        if owners.is_empty() {
+            return;
+        }
+
+        // Two-pass: gather work under an immutable borrow, then apply
+        // copies under independent mutable borrows.
+        struct Mirror<'a> {
+            target_idx: usize,
+            owner: &'a EcxOwner,
+            collection: String,
+            vid: VolumeId,
+        }
+        let mut mirrors: Vec<Mirror> = Vec::new();
+        for (loc_idx, loc) in self.locations.iter().enumerate() {
+            let orphans = collect_shard_disk_volumes(loc);
+            for (key, _shards) in orphans {
+                let Some(owner) = owners.get(&key) else {
+                    continue;
+                };
+                if owner.location == loc_idx {
+                    continue;
+                }
+                if disk_has_all_sidecars(loc, &key.collection, key.vid) {
+                    continue;
+                }
+                mirrors.push(Mirror {
+                    target_idx: loc_idx,
+                    owner,
+                    collection: key.collection,
+                    vid: key.vid,
+                });
+            }
+        }
+
+        for m in mirrors {
+            let loc_dir = self.locations[m.target_idx].directory.clone();
+            let loc_idx_dir = self.locations[m.target_idx].idx_directory.clone();
+            match mirror_sidecars_for_volume(
+                &m.owner.idx_dir,
+                &m.owner.data_dir,
+                &loc_dir,
+                &loc_idx_dir,
+                &m.collection,
+                m.vid,
+            ) {
+                Ok(0) => {}
+                Ok(copied) => {
+                    info!(
+                        volume_id = m.vid.0,
+                        collection = %m.collection,
+                        from = %m.owner.data_dir,
+                        to = %loc_dir,
+                        copied,
+                        "mirrored EC sidecar(s) for same-disk invariant",
+                    );
+                }
+                Err(e) => {
+                    warn!(
+                        volume_id = m.vid.0,
+                        collection = %m.collection,
+                        from = %m.owner.data_dir,
+                        to = %loc_dir,
+                        error = %e,
+                        "mirror EC sidecars failed; cross-disk fallback will handle this volume",
+                    );
+                }
+            }
+        }
+    }
+
+    // Records both idx_dir (where .ecx was found) and data_dir, so
+    // the mirror can resolve .vif from data_dir even when .ecx lives
+    // in idx_directory.
+    fn index_ecx_owners_for_mirror(&self) -> HashMap<EcKey, EcxOwner> {
+        let mut owners: HashMap<EcKey, EcxOwner> = HashMap::new();
+        for (loc_idx, loc) in self.locations.iter().enumerate() {
+            let mut seen: Vec<&str> = Vec::with_capacity(2);
+            for scan in [loc.idx_directory.as_str(), loc.directory.as_str()] {
+                if scan.is_empty() || seen.contains(&scan) {
+                    continue;
+                }
+                seen.push(scan);
+                let Ok(read) = fs::read_dir(scan) else {
+                    continue;
+                };
+                for ent in read.flatten() {
+                    if ent.file_type().map(|ft| ft.is_dir()).unwrap_or(false) {
+                        continue;
+                    }
+                    let name = ent.file_name().to_string_lossy().into_owned();
+                    let Some(base) = name.strip_suffix(".ecx") else {
+                        continue;
+                    };
+                    let Some((collection, vid)) = parse_collection_volume_id_pub(base) else {
+                        continue;
+                    };
+                    owners
+                        .entry(EcKey { collection, vid })
+                        .or_insert_with(|| EcxOwner {
+                            location: loc_idx,
+                            idx_dir: scan.to_string(),
+                            data_dir: loc.directory.clone(),
+                        });
+                }
+            }
+        }
+        owners
+    }
+}
+
+// Checks the modern routing and the opposite directory — without
+// that fallback, a destination with a legacy pre-`-dir.idx` .ecx in
+// its data dir would be re-mirrored into idx_directory.
+fn disk_has_all_sidecars(loc: &DiskLocation, collection: &str, vid: VolumeId) -> bool {
+    for ext in EC_MIRRORED_SIDECARS {
+        let primary = sidecar_dest_path(&loc.directory, &loc.idx_directory, collection, vid, ext);
+        if path_is_regular_file(&primary) {
+            continue;
+        }
+        if loc.idx_directory != loc.directory {
+            let (fallback_data, fallback_idx) = if *ext == ".vif" {
+                (loc.idx_directory.as_str(), loc.directory.as_str())
+            } else {
+                (loc.directory.as_str(), loc.directory.as_str())
+            };
+            let fallback = sidecar_dest_path(fallback_data, fallback_idx, collection, vid, ext);
+            if path_is_regular_file(&fallback) {
+                continue;
+            }
+        }
+        return false;
+    }
+    true
+}
+
+fn path_is_regular_file(path: &str) -> bool {
+    fs::metadata(path).map(|m| !m.is_dir()).unwrap_or(false)
+}
+
+// `.ecx`/`.ecj` route to idx_directory, `.vif` to directory.
+fn sidecar_dest_path(
+    data_dir: &str,
+    idx_dir: &str,
+    collection: &str,
+    vid: VolumeId,
+    ext: &str,
+) -> String {
+    let dir = if ext == ".vif" { data_dir } else { idx_dir };
+    if collection.is_empty() {
+        format!("{}/{}{}", dir, vid.0, ext)
+    } else {
+        format!("{}/{}_{}{}", dir, collection, vid.0, ext)
+    }
+}
+
+fn mirror_sidecars_for_volume(
+    src_idx_dir: &str,
+    src_data_dir: &str,
+    dst_data_dir: &str,
+    dst_idx_dir: &str,
+    collection: &str,
+    vid: VolumeId,
+) -> io::Result<usize> {
+    let mut copied = 0usize;
+    for ext in EC_MIRRORED_SIDECARS {
+        let dst = sidecar_dest_path(dst_data_dir, dst_idx_dir, collection, vid, ext);
+        // An existing local copy is authoritative — it may be newer
+        // than the owner's after a delete journal append.
+        if fs::metadata(&dst).is_ok() {
+            continue;
+        }
+
+        let candidates = [
+            sidecar_dest_path(src_data_dir, src_idx_dir, collection, vid, ext),
+            sidecar_dest_path(src_idx_dir, src_data_dir, collection, vid, ext),
+        ];
+        let mut src_path: Option<String> = None;
+        for c in candidates.iter() {
+            if fs::metadata(c).map(|m| !m.is_dir()).unwrap_or(false) {
+                src_path = Some(c.clone());
+                break;
+            }
+        }
+        let Some(src) = src_path else {
+            continue;
+        };
+        copy_sidecar_atomic(Path::new(&src), Path::new(&dst))?;
+        copied += 1;
+    }
+    Ok(copied)
+}
+
+fn copy_sidecar_atomic(src: &Path, dst: &Path) -> io::Result<()> {
+    if let Some(parent) = dst.parent() {
+        fs::create_dir_all(parent)?;
+    }
+    let mut src_file = fs::File::open(src)?;
+    let tmp = {
+        let mut s = dst.as_os_str().to_owned();
+        s.push(".mirror.tmp");
+        std::path::PathBuf::from(s)
+    };
+    let _ = fs::remove_file(&tmp);
+    let mut dst_file = fs::OpenOptions::new()
+        .write(true)
+        .create_new(true)
+        .open(&tmp)?;
+    let mut buf = [0u8; 64 * 1024];
+    loop {
+        let n = src_file.read(&mut buf)?;
+        if n == 0 {
+            break;
+        }
+        dst_file.write_all(&buf[..n])?;
+    }
+    dst_file.sync_all()?;
+    drop(dst_file);
+    if let Err(e) = fs::rename(&tmp, dst) {
+        let _ = fs::remove_file(&tmp);
+        return Err(e);
+    }
+    Ok(())
+}
+
+fn collect_shard_disk_volumes(loc: &DiskLocation) -> HashMap<EcKey, Vec<String>> {
+    let mut out: HashMap<EcKey, Vec<String>> = HashMap::new();
+    let Ok(read) = fs::read_dir(&loc.directory) else {
+        return out;
+    };
+    for ent in read.flatten() {
+        if ent.file_type().map(|ft| ft.is_dir()).unwrap_or(false) {
+            continue;
+        }
+        let name = ent.file_name().to_string_lossy().into_owned();
+        let Some(dot) = name.rfind('.') else {
+            continue;
+        };
+        let (base, ext) = name.split_at(dot);
+        if crate::storage::disk_location::is_ec_shard_extension(ext).is_none() {
+            continue;
+        }
+        match ent.metadata() {
+            Ok(meta) if meta.len() > 0 => {}
+            _ => continue,
+        }
+        let Some((collection, vid)) = parse_collection_volume_id_pub(base) else {
+            continue;
+        };
+        out.entry(EcKey { collection, vid }).or_default().push(name);
+    }
+    out
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::config::MinFreeSpace;
+    use crate::storage::needle_map::NeedleMapKind;
+    use crate::storage::types::DiskType;
+    use crate::storage::volume::{VifEcShardConfig, VifVolumeInfo};
+    use tempfile::TempDir;
+
+    fn plant_shard(dir: &Path, collection: &str, vid: u32, shard_id: u8) {
+        let path = if collection.is_empty() {
+            dir.join(format!("{}.ec{:02}", vid, shard_id))
+        } else {
+            dir.join(format!("{}_{}.ec{:02}", collection, vid, shard_id))
+        };
+        fs::write(&path, b"shard data nonempty").unwrap();
+    }
+
+    fn plant_ecx(dir: &Path, collection: &str, vid: u32, bytes: &[u8]) {
+        let path = dir.join(format!("{}_{}.ecx", collection, vid));
+        fs::write(&path, bytes).unwrap();
+    }
+
+    fn plant_ecj(dir: &Path, collection: &str, vid: u32, bytes: &[u8]) {
+        let path = dir.join(format!("{}_{}.ecj", collection, vid));
+        fs::write(&path, bytes).unwrap();
+    }
+
+    fn plant_vif(dir: &Path, collection: &str, vid: u32, data_shards: u32, parity_shards: u32) {
+        let vif = VifVolumeInfo {
+            version: 3,
+            ec_shard_config: Some(VifEcShardConfig {
+                data_shards,
+                parity_shards,
+                ..Default::default()
+            }),
+            ..Default::default()
+        };
+        let path = dir.join(format!("{}_{}.vif", collection, vid));
+        fs::write(&path, serde_json::to_string(&vif).unwrap()).unwrap();
+    }
+
+    fn add_loc(store: &mut Store, dir: &Path) {
+        store
+            .add_location(
+                dir.to_str().unwrap(),
+                dir.to_str().unwrap(),
+                100,
+                DiskType::HardDrive,
+                MinFreeSpace::Percent(0.0),
+                Vec::new(),
+            )
+            .unwrap();
+    }
+
+    #[test]
+    fn mirror_copies_sidecars_to_shard_only_disk() {
+        let tmp = TempDir::new().unwrap();
+        let dir0 = tmp.path().join("data0");
+        let dir1 = tmp.path().join("data1");
+        fs::create_dir_all(&dir0).unwrap();
+        fs::create_dir_all(&dir1).unwrap();
+
+        let collection = "video-recordings";
+        let vid = 4121u32;
+
+        plant_shard(&dir0, collection, vid, 0);
+        plant_shard(&dir0, collection, vid, 12);
+        plant_shard(&dir1, collection, vid, 1);
+
+        let ecx = vec![0xA1u8; 20];
+        let ecj = vec![0xB2u8; 16];
+        plant_ecx(&dir1, collection, vid, &ecx);
+        plant_ecj(&dir1, collection, vid, &ecj);
+        plant_vif(&dir1, collection, vid, 10, 4);
+
+        let mut store = Store::new(NeedleMapKind::InMemory);
+        add_loc(&mut store, &dir0);
+        add_loc(&mut store, &dir1);
+
+        for ext in [".ecx", ".ecj", ".vif"] {
+            let dst = dir0.join(format!("{}_{}{}", collection, vid, ext));
+            assert!(
+                dst.exists(),
+                "mirror did not install sidecar {} on dir0",
+                ext
+            );
+        }
+        let ecx_dst = fs::read(dir0.join(format!("{}_{}.ecx", collection, vid))).unwrap();
+        assert_eq!(ecx_dst, ecx, ".ecx mirrored bytes differ from source");
+        let ecj_dst = fs::read(dir0.join(format!("{}_{}.ecj", collection, vid))).unwrap();
+        assert_eq!(ecj_dst, ecj, ".ecj mirrored bytes differ from source");
+    }
+
+    #[test]
+    fn mirror_preserves_existing_destination_sidecars() {
+        let tmp = TempDir::new().unwrap();
+        let dir0 = tmp.path().join("data0");
+        let dir1 = tmp.path().join("data1");
+        fs::create_dir_all(&dir0).unwrap();
+        fs::create_dir_all(&dir1).unwrap();
+
+        let collection = "video-recordings";
+        let vid = 7777u32;
+
+        plant_shard(&dir0, collection, vid, 0);
+        plant_shard(&dir0, collection, vid, 12);
+        plant_shard(&dir1, collection, vid, 1);
+
+        let ecx_owner = vec![0xC3u8; 20];
+        let ecx_local = vec![0x5Au8; 20];
+        let ecj_bytes = vec![0xD4u8; 16];
+        plant_ecx(&dir1, collection, vid, &ecx_owner);
+        plant_ecj(&dir1, collection, vid, &ecj_bytes);
+        plant_vif(&dir1, collection, vid, 10, 4);
+        plant_ecx(&dir0, collection, vid, &ecx_local);
+        plant_ecj(&dir0, collection, vid, &ecj_bytes);
+        plant_vif(&dir0, collection, vid, 10, 4);
+
+        let mut store = Store::new(NeedleMapKind::InMemory);
+        add_loc(&mut store, &dir0);
+        add_loc(&mut store, &dir1);
+
+        let post = fs::read(dir0.join(format!("{}_{}.ecx", collection, vid))).unwrap();
+        assert_eq!(post, ecx_local, "mirror overwrote dir0's existing .ecx");
+    }
+}

seaweed-volume/src/storage/super_block.rs

@@ -0,0 +1,289 @@
+//! SuperBlock: the 8-byte (+ optional extra) header at the start of every .dat file.
+//!
+//! Byte layout:
+//!   [0]     Version
+//!   [1]     ReplicaPlacement byte
+//!   [2..4]  TTL (2 bytes)
+//!   [4..6]  CompactionRevision (u16 big-endian)
+//!   [6..8]  ExtraSize (u16 big-endian)
+//!   [8..]   Extra data (protobuf, ExtraSize bytes) — only for Version 2/3
+
+use crate::storage::needle::ttl::TTL;
+use crate::storage::types::Version;
+
+pub const SUPER_BLOCK_SIZE: usize = 8;
+
+/// SuperBlock metadata at the start of a volume .dat file.
+#[derive(Debug, Clone)]
+pub struct SuperBlock {
+    pub version: Version,
+    pub replica_placement: ReplicaPlacement,
+    pub ttl: TTL,
+    pub compaction_revision: u16,
+    pub extra_size: u16,
+    pub extra_data: Vec<u8>, // raw protobuf bytes (SuperBlockExtra)
+}
+
+impl SuperBlock {
+    /// Total block size on disk (base 8 + extra).
+    pub fn block_size(&self) -> usize {
+        match self.version.0 {
+            2 | 3 => SUPER_BLOCK_SIZE + self.extra_size as usize,
+            _ => SUPER_BLOCK_SIZE,
+        }
+    }
+
+    /// Serialize to bytes.
+    pub fn to_bytes(&self) -> Vec<u8> {
+        let mut header = vec![0u8; SUPER_BLOCK_SIZE];
+        header[0] = self.version.0;
+        header[1] = self.replica_placement.to_byte();
+        self.ttl.to_bytes(&mut header[2..4]);
+        header[4..6].copy_from_slice(&self.compaction_revision.to_be_bytes());
+
+        if !self.extra_data.is_empty() {
+            // Go checks extraSize > 256*256-2 and calls glog.Fatalf; guard against u16 overflow.
+            assert!(
+                self.extra_data.len() <= 65534,
+                "super block extra data too large: {} > 65534",
+                self.extra_data.len()
+            );
+            let extra_size = self.extra_data.len() as u16;
+            header[6..8].copy_from_slice(&extra_size.to_be_bytes());
+            header.extend_from_slice(&self.extra_data);
+        }
+
+        header
+    }
+
+    /// Parse from bytes (must be at least SUPER_BLOCK_SIZE bytes).
+    pub fn from_bytes(bytes: &[u8]) -> Result<Self, SuperBlockError> {
+        if bytes.len() < SUPER_BLOCK_SIZE {
+            return Err(SuperBlockError::TooShort(bytes.len()));
+        }
+
+        let version = Version(bytes[0]);
+        let replica_placement = ReplicaPlacement::from_byte(bytes[1])?;
+        let ttl = TTL::from_bytes(&bytes[2..4]);
+        let compaction_revision = u16::from_be_bytes([bytes[4], bytes[5]]);
+        let extra_size = u16::from_be_bytes([bytes[6], bytes[7]]);
+
+        let extra_data = if extra_size > 0 && bytes.len() >= SUPER_BLOCK_SIZE + extra_size as usize
+        {
+            bytes[SUPER_BLOCK_SIZE..SUPER_BLOCK_SIZE + extra_size as usize].to_vec()
+        } else {
+            vec![]
+        };
+
+        Ok(SuperBlock {
+            version,
+            replica_placement,
+            ttl,
+            compaction_revision,
+            extra_size,
+            extra_data,
+        })
+    }
+
+    pub fn initialized(&self) -> bool {
+        true // ReplicaPlacement and TTL are always valid after construction
+    }
+}
+
+impl Default for SuperBlock {
+    fn default() -> Self {
+        SuperBlock {
+            version: Version::current(),
+            replica_placement: ReplicaPlacement::default(),
+            ttl: TTL::EMPTY,
+            compaction_revision: 0,
+            extra_size: 0,
+            extra_data: vec![],
+        }
+    }
+}
+
+// ============================================================================
+// ReplicaPlacement
+// ============================================================================
+
+/// Replication strategy encoded as a single byte.
+///
+/// Byte value = DiffDataCenterCount * 100 + DiffRackCount * 10 + SameRackCount
+///
+/// Examples:
+///   "000" → no replication (1 copy total)
+///   "010" → 1 copy in different rack (2 copies total)
+///   "100" → 1 copy in different datacenter
+///   "200" → 2 copies in different datacenters
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub struct ReplicaPlacement {
+    pub same_rack_count: u8,
+    pub diff_rack_count: u8,
+    pub diff_data_center_count: u8,
+}
+
+impl ReplicaPlacement {
+    /// Parse from a string like "000", "010", "100".
+    /// Accepts 0-3 character strings, padding with leading zeros to match Go behavior.
+    /// E.g. "" -> "000", "1" -> "001", "01" -> "001", "010" -> "010"
+    pub fn from_string(s: &str) -> Result<Self, SuperBlockError> {
+        let s = s.trim();
+        if s.is_empty() {
+            return Ok(ReplicaPlacement::default());
+        }
+        // Pad with leading zeros to 3 chars, matching Go's NewReplicaPlacementFromString
+        let padded = match s.len() {
+            1 => format!("00{}", s),
+            2 => format!("0{}", s),
+            3 => s.to_string(),
+            _ => return Err(SuperBlockError::InvalidReplicaPlacement(s.to_string())),
+        };
+        let chars: Vec<char> = padded.chars().collect();
+        let dc = chars[0]
+            .to_digit(10)
+            .ok_or_else(|| SuperBlockError::InvalidReplicaPlacement(s.to_string()))?
+            as u8;
+        let rack = chars[1]
+            .to_digit(10)
+            .ok_or_else(|| SuperBlockError::InvalidReplicaPlacement(s.to_string()))?
+            as u8;
+        let same = chars[2]
+            .to_digit(10)
+            .ok_or_else(|| SuperBlockError::InvalidReplicaPlacement(s.to_string()))?
+            as u8;
+        // Go validates: value = dc*100 + rack*10 + same must fit in a byte
+        let value = dc as u16 * 100 + rack as u16 * 10 + same as u16;
+        if value > 255 {
+            return Err(SuperBlockError::InvalidReplicaPlacement(s.to_string()));
+        }
+        Ok(ReplicaPlacement {
+            diff_data_center_count: dc,
+            diff_rack_count: rack,
+            same_rack_count: same,
+        })
+    }
+
+    /// Parse from a single byte.
+    pub fn from_byte(b: u8) -> Result<Self, SuperBlockError> {
+        Ok(ReplicaPlacement {
+            diff_data_center_count: b / 100,
+            diff_rack_count: (b % 100) / 10,
+            same_rack_count: b % 10,
+        })
+    }
+
+    /// Encode as a single byte.
+    pub fn to_byte(&self) -> u8 {
+        self.diff_data_center_count * 100 + self.diff_rack_count * 10 + self.same_rack_count
+    }
+
+    /// Total number of copies (including the original).
+    pub fn get_copy_count(&self) -> u8 {
+        self.diff_data_center_count + self.diff_rack_count + self.same_rack_count + 1
+    }
+
+    /// Whether this placement requires replication (more than 1 copy).
+    pub fn has_replication(&self) -> bool {
+        self.get_copy_count() > 1
+    }
+}
+
+impl std::fmt::Display for ReplicaPlacement {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}{}{}",
+            self.diff_data_center_count, self.diff_rack_count, self.same_rack_count
+        )
+    }
+}
+
+// ============================================================================
+// Errors
+// ============================================================================
+
+#[derive(Debug, thiserror::Error)]
+pub enum SuperBlockError {
+    #[error("super block too short: {0} bytes")]
+    TooShort(usize),
+
+    #[error("invalid replica placement: {0}")]
+    InvalidReplicaPlacement(String),
+}
+
+// ============================================================================
+// Tests
+// ============================================================================
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::storage::types::*;
+
+    #[test]
+    fn test_super_block_round_trip() {
+        let sb = SuperBlock {
+            version: VERSION_3,
+            replica_placement: ReplicaPlacement::from_string("010").unwrap(),
+            ttl: TTL { count: 5, unit: 3 },
+            compaction_revision: 42,
+            extra_size: 0,
+            extra_data: vec![],
+        };
+
+        let bytes = sb.to_bytes();
+        assert_eq!(bytes.len(), SUPER_BLOCK_SIZE);
+
+        let sb2 = SuperBlock::from_bytes(&bytes).unwrap();
+        assert_eq!(sb2.version, sb.version);
+        assert_eq!(sb2.replica_placement, sb.replica_placement);
+        assert_eq!(sb2.ttl, sb.ttl);
+        assert_eq!(sb2.compaction_revision, sb.compaction_revision);
+    }
+
+    #[test]
+    fn test_super_block_with_extra() {
+        let sb = SuperBlock {
+            version: VERSION_3,
+            replica_placement: ReplicaPlacement::default(),
+            ttl: TTL::EMPTY,
+            compaction_revision: 0,
+            extra_size: 3,
+            extra_data: vec![1, 2, 3],
+        };
+
+        let bytes = sb.to_bytes();
+        assert_eq!(bytes.len(), SUPER_BLOCK_SIZE + 3);
+
+        let sb2 = SuperBlock::from_bytes(&bytes).unwrap();
+        assert_eq!(sb2.extra_data, vec![1, 2, 3]);
+    }
+
+    #[test]
+    fn test_replica_placement_byte_round_trip() {
+        let rp = ReplicaPlacement::from_string("123").unwrap();
+        assert_eq!(rp.diff_data_center_count, 1);
+        assert_eq!(rp.diff_rack_count, 2);
+        assert_eq!(rp.same_rack_count, 3);
+        assert_eq!(rp.to_byte(), 123);
+        assert_eq!(rp.get_copy_count(), 7); // 1+2+3+1
+
+        let rp2 = ReplicaPlacement::from_byte(123).unwrap();
+        assert_eq!(rp, rp2);
+    }
+
+    #[test]
+    fn test_replica_placement_no_replication() {
+        let rp = ReplicaPlacement::from_string("000").unwrap();
+        assert!(!rp.has_replication());
+        assert_eq!(rp.get_copy_count(), 1);
+    }
+
+    #[test]
+    fn test_replica_placement_display() {
+        let rp = ReplicaPlacement::from_string("010").unwrap();
+        assert_eq!(rp.to_string(), "010");
+        assert!(rp.has_replication());
+    }
+}

seaweed-volume/src/storage/types.rs

@@ -0,0 +1,659 @@
+//! Core storage types: NeedleId, Offset, Size, Cookie, DiskType.
+//!
+//! These types define the binary-compatible on-disk format matching the Go implementation.
+//! CRITICAL: Byte layout must match exactly for cross-compatibility.
+
+use std::fmt;
+
+// ============================================================================
+// Constants
+// ============================================================================
+
+pub const NEEDLE_ID_SIZE: usize = 8;
+pub const NEEDLE_ID_EMPTY: u64 = 0;
+pub const COOKIE_SIZE: usize = 4;
+pub const SIZE_SIZE: usize = 4;
+pub const NEEDLE_HEADER_SIZE: usize = COOKIE_SIZE + NEEDLE_ID_SIZE + SIZE_SIZE; // 16
+pub const DATA_SIZE_SIZE: usize = 4;
+pub const TIMESTAMP_SIZE: usize = 8;
+pub const NEEDLE_PADDING_SIZE: usize = 8;
+pub const NEEDLE_CHECKSUM_SIZE: usize = 4;
+
+/// 5-byte offset mode (matching Go production builds with `-tags 5BytesOffset`).
+/// Max volume size: 8TB. Index entry: 17 bytes (8 + 5 + 4).
+#[cfg(feature = "5bytes")]
+pub const OFFSET_SIZE: usize = 5;
+#[cfg(feature = "5bytes")]
+pub const MAX_POSSIBLE_VOLUME_SIZE: u64 = 4 * 1024 * 1024 * 1024 * 8 * 256; // 8TB
+
+/// 4-byte offset mode (matching Go default build without `5BytesOffset`).
+/// Max volume size: 32GB. Index entry: 16 bytes (8 + 4 + 4).
+#[cfg(not(feature = "5bytes"))]
+pub const OFFSET_SIZE: usize = 4;
+#[cfg(not(feature = "5bytes"))]
+pub const MAX_POSSIBLE_VOLUME_SIZE: u64 = 4 * 1024 * 1024 * 1024 * 8; // 32GB
+
+pub const NEEDLE_MAP_ENTRY_SIZE: usize = NEEDLE_ID_SIZE + OFFSET_SIZE + SIZE_SIZE;
+
+// ============================================================================
+// NeedleId
+// ============================================================================
+
+/// 64-bit unique identifier for a needle within a volume.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord, Default)]
+pub struct NeedleId(pub u64);
+
+impl NeedleId {
+    pub fn to_bytes(&self, bytes: &mut [u8]) {
+        assert!(bytes.len() >= NEEDLE_ID_SIZE);
+        bytes[0..8].copy_from_slice(&self.0.to_be_bytes());
+    }
+
+    pub fn from_bytes(bytes: &[u8]) -> Self {
+        assert!(bytes.len() >= NEEDLE_ID_SIZE);
+        NeedleId(u64::from_be_bytes([
+            bytes[0], bytes[1], bytes[2], bytes[3], bytes[4], bytes[5], bytes[6], bytes[7],
+        ]))
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.0 == 0
+    }
+
+    /// Parse a hex string into a NeedleId.
+    pub fn parse(s: &str) -> Result<Self, std::num::ParseIntError> {
+        u64::from_str_radix(s, 16).map(NeedleId)
+    }
+}
+
+impl fmt::Display for NeedleId {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{:x}", self.0)
+    }
+}
+
+impl From<u64> for NeedleId {
+    fn from(v: u64) -> Self {
+        NeedleId(v)
+    }
+}
+
+impl From<NeedleId> for u64 {
+    fn from(v: NeedleId) -> Self {
+        v.0
+    }
+}
+
+// ============================================================================
+// Cookie
+// ============================================================================
+
+/// Random 32-bit value to mitigate brute-force lookups.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Default)]
+pub struct Cookie(pub u32);
+
+impl Cookie {
+    pub fn to_bytes(&self, bytes: &mut [u8]) {
+        assert!(bytes.len() >= COOKIE_SIZE);
+        bytes[0..4].copy_from_slice(&self.0.to_be_bytes());
+    }
+
+    pub fn from_bytes(bytes: &[u8]) -> Self {
+        assert!(bytes.len() >= COOKIE_SIZE);
+        Cookie(u32::from_be_bytes([bytes[0], bytes[1], bytes[2], bytes[3]]))
+    }
+
+    /// Parse a hex string into a Cookie.
+    pub fn parse(s: &str) -> Result<Self, std::num::ParseIntError> {
+        u32::from_str_radix(s, 16).map(Cookie)
+    }
+}
+
+impl fmt::Display for Cookie {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{:x}", self.0)
+    }
+}
+
+impl From<u32> for Cookie {
+    fn from(v: u32) -> Self {
+        Cookie(v)
+    }
+}
+
+// ============================================================================
+// Size
+// ============================================================================
+
+/// Needle size as stored in the index. Negative = deleted.
+///
+/// - Positive: valid needle with that many bytes of body content
+/// - TombstoneFileSize (-1): tombstone marker
+/// - Other negative: deleted, absolute value was the original size
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Default)]
+pub struct Size(pub i32);
+
+/// Special marker for a tombstone (deletion marker) entry.
+pub const TOMBSTONE_FILE_SIZE: Size = Size(-1);
+
+impl Size {
+    pub fn is_tombstone(&self) -> bool {
+        self.0 == TOMBSTONE_FILE_SIZE.0
+    }
+
+    pub fn is_deleted(&self) -> bool {
+        self.0 < 0 || self.0 == TOMBSTONE_FILE_SIZE.0
+    }
+
+    pub fn is_valid(&self) -> bool {
+        self.0 > 0 && !self.is_tombstone()
+    }
+
+    /// Raw storage size. For tombstones returns 0; for negative returns abs value.
+    pub fn raw(&self) -> u32 {
+        if self.is_tombstone() {
+            return 0;
+        }
+        if self.0 < 0 {
+            return -self.0 as u32;
+        }
+        self.0 as u32
+    }
+
+    pub fn to_bytes(&self, bytes: &mut [u8]) {
+        assert!(bytes.len() >= SIZE_SIZE);
+        bytes[0..4].copy_from_slice(&(self.0 as u32).to_be_bytes());
+    }
+
+    pub fn from_bytes(bytes: &[u8]) -> Self {
+        assert!(bytes.len() >= SIZE_SIZE);
+        let v = u32::from_be_bytes([bytes[0], bytes[1], bytes[2], bytes[3]]);
+        Size(v as i32)
+    }
+}
+
+impl From<i32> for Size {
+    fn from(v: i32) -> Self {
+        Size(v)
+    }
+}
+
+impl From<Size> for i32 {
+    fn from(v: Size) -> Self {
+        v.0
+    }
+}
+
+// ============================================================================
+// Offset
+// ============================================================================
+
+/// Offset encoding for needle positions in .dat files.
+///
+/// The offset is stored divided by NEEDLE_PADDING_SIZE (8).
+///
+/// With `5bytes` feature (default, matching Go production builds):
+///   5 bytes can address up to 8TB.
+///   On-disk layout: [b3][b2][b1][b0][b4] (big-endian 4 bytes + 1 high byte)
+///
+/// Without `5bytes` feature (matching Go default build):
+///   4 bytes can address up to 32GB.
+///   On-disk layout: [b3][b2][b1][b0] (big-endian 4 bytes)
+#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Default)]
+pub struct Offset {
+    pub b0: u8,
+    pub b1: u8,
+    pub b2: u8,
+    pub b3: u8,
+    #[cfg(feature = "5bytes")]
+    pub b4: u8,
+}
+
+impl Offset {
+    /// Convert to the actual byte offset in the .dat file.
+    pub fn to_actual_offset(&self) -> i64 {
+        let stored = self.b0 as i64
+            + (self.b1 as i64) * 256
+            + (self.b2 as i64) * 65536
+            + (self.b3 as i64) * 16777216;
+        #[cfg(feature = "5bytes")]
+        let stored = stored + (self.b4 as i64) * 4294967296; // 1 << 32
+        stored * NEEDLE_PADDING_SIZE as i64
+    }
+
+    /// Create an Offset from an actual byte offset.
+    pub fn from_actual_offset(offset: i64) -> Self {
+        let smaller = offset / NEEDLE_PADDING_SIZE as i64;
+        Offset {
+            b0: smaller as u8,
+            b1: (smaller >> 8) as u8,
+            b2: (smaller >> 16) as u8,
+            b3: (smaller >> 24) as u8,
+            #[cfg(feature = "5bytes")]
+            b4: (smaller >> 32) as u8,
+        }
+    }
+
+    /// Serialize to bytes in the .idx file format.
+    /// 5-byte layout: [b3][b2][b1][b0][b4]
+    /// 4-byte layout: [b3][b2][b1][b0]
+    pub fn to_bytes(&self, bytes: &mut [u8]) {
+        assert!(bytes.len() >= OFFSET_SIZE);
+        bytes[0] = self.b3;
+        bytes[1] = self.b2;
+        bytes[2] = self.b1;
+        bytes[3] = self.b0;
+        #[cfg(feature = "5bytes")]
+        {
+            bytes[4] = self.b4;
+        }
+    }
+
+    /// Deserialize from bytes in the .idx file format.
+    pub fn from_bytes(bytes: &[u8]) -> Self {
+        assert!(bytes.len() >= OFFSET_SIZE);
+        Offset {
+            b3: bytes[0],
+            b2: bytes[1],
+            b1: bytes[2],
+            b0: bytes[3],
+            #[cfg(feature = "5bytes")]
+            b4: bytes[4],
+        }
+    }
+
+    pub fn is_zero(&self) -> bool {
+        #[cfg(feature = "5bytes")]
+        {
+            self.b0 == 0 && self.b1 == 0 && self.b2 == 0 && self.b3 == 0 && self.b4 == 0
+        }
+        #[cfg(not(feature = "5bytes"))]
+        {
+            self.b0 == 0 && self.b1 == 0 && self.b2 == 0 && self.b3 == 0
+        }
+    }
+}
+
+impl fmt::Display for Offset {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{}", self.to_actual_offset())
+    }
+}
+
+// ============================================================================
+// DiskType
+// ============================================================================
+
+#[derive(Debug, Clone, PartialEq, Eq, Hash, Default)]
+pub enum DiskType {
+    #[default]
+    HardDrive,
+    Ssd,
+    Custom(String),
+}
+
+impl DiskType {
+    pub fn from_string(s: &str) -> Self {
+        match s.to_lowercase().as_str() {
+            "" | "hdd" => DiskType::HardDrive,
+            "ssd" => DiskType::Ssd,
+            other => DiskType::Custom(other.to_string()),
+        }
+    }
+
+    pub fn readable_string(&self) -> &str {
+        match self {
+            DiskType::HardDrive => "hdd",
+            DiskType::Ssd => "ssd",
+            DiskType::Custom(s) => s,
+        }
+    }
+}
+
+impl fmt::Display for DiskType {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            DiskType::HardDrive => write!(f, ""),
+            DiskType::Ssd => write!(f, "ssd"),
+            DiskType::Custom(s) => write!(f, "{}", s),
+        }
+    }
+}
+
+// ============================================================================
+// VolumeId
+// ============================================================================
+
+/// Volume identifier, stored as u32.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord, Default)]
+pub struct VolumeId(pub u32);
+
+impl VolumeId {
+    pub fn parse(s: &str) -> Result<Self, std::num::ParseIntError> {
+        s.parse::<u32>().map(VolumeId)
+    }
+
+    pub fn next(&self) -> VolumeId {
+        VolumeId(self.0 + 1)
+    }
+}
+
+impl fmt::Display for VolumeId {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{}", self.0)
+    }
+}
+
+impl From<u32> for VolumeId {
+    fn from(v: u32) -> Self {
+        VolumeId(v)
+    }
+}
+
+// ============================================================================
+// Version
+// ============================================================================
+
+/// Needle storage format version.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct Version(pub u8);
+
+pub const VERSION_1: Version = Version(1);
+pub const VERSION_2: Version = Version(2);
+pub const VERSION_3: Version = Version(3);
+
+impl Version {
+    pub fn current() -> Self {
+        VERSION_3
+    }
+
+    pub fn is_supported(&self) -> bool {
+        self.0 >= 1 && self.0 <= 3
+    }
+}
+
+impl Default for Version {
+    fn default() -> Self {
+        VERSION_3
+    }
+}
+
+impl From<u8> for Version {
+    fn from(v: u8) -> Self {
+        Version(v)
+    }
+}
+
+// ============================================================================
+// ReadOption
+// ============================================================================
+
+/// Options controlling needle read behavior, matching Go's `ReadOption` in store.go.
+///
+/// Fields are split into request-side options (set by the caller) and response-side
+/// flags (set during the read to communicate status back).
+#[derive(Debug, Clone, Default)]
+pub struct ReadOption {
+    // -- request --
+    /// If true, allow reading needles that have been soft-deleted.
+    pub read_deleted: bool,
+    /// If true, attempt to read only metadata for large needles (> PagedReadLimit).
+    pub attempt_meta_only: bool,
+    /// If true, the caller requires metadata only (no data payload).
+    pub must_meta_only: bool,
+
+    // -- response --
+    /// Set to true when the read actually returned metadata only.
+    pub is_meta_only: bool,
+    /// Compaction revision at the time of the read (for consistency during streaming).
+    pub volume_revision: u16,
+    /// Set to true when the offset exceeded MaxPossibleVolumeSize (4-byte offset wrap).
+    pub is_out_of_range: bool,
+
+    // -- slow-read / streaming --
+    /// When true, the read lock is acquired and released per chunk instead of held
+    /// for the entire read, reducing write latency at the cost of higher read P99.
+    pub has_slow_read: bool,
+    /// Buffer size for chunked streaming reads (used with `has_slow_read`).
+    pub read_buffer_size: i32,
+}
+
+// ============================================================================
+// NeedleMapEntry helpers (for .idx file)
+// ============================================================================
+
+/// Parse a single .idx file entry (17 bytes) into (NeedleId, Offset, Size).
+pub fn idx_entry_from_bytes(bytes: &[u8]) -> (NeedleId, Offset, Size) {
+    assert!(bytes.len() >= NEEDLE_MAP_ENTRY_SIZE);
+    let key = NeedleId::from_bytes(&bytes[..NEEDLE_ID_SIZE]);
+    let offset = Offset::from_bytes(&bytes[NEEDLE_ID_SIZE..NEEDLE_ID_SIZE + OFFSET_SIZE]);
+    let size = Size::from_bytes(
+        &bytes[NEEDLE_ID_SIZE + OFFSET_SIZE..NEEDLE_ID_SIZE + OFFSET_SIZE + SIZE_SIZE],
+    );
+    (key, offset, size)
+}
+
+/// Write a single .idx file entry (17 bytes).
+pub fn idx_entry_to_bytes(bytes: &mut [u8], key: NeedleId, offset: Offset, size: Size) {
+    assert!(bytes.len() >= NEEDLE_MAP_ENTRY_SIZE);
+    key.to_bytes(&mut bytes[..NEEDLE_ID_SIZE]);
+    offset.to_bytes(&mut bytes[NEEDLE_ID_SIZE..NEEDLE_ID_SIZE + OFFSET_SIZE]);
+    size.to_bytes(
+        &mut bytes[NEEDLE_ID_SIZE + OFFSET_SIZE..NEEDLE_ID_SIZE + OFFSET_SIZE + SIZE_SIZE],
+    );
+}
+
+// ============================================================================
+// Tests
+// ============================================================================
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_needle_id_round_trip() {
+        let id = NeedleId(0x123456789abcdef0);
+        let mut buf = [0u8; 8];
+        id.to_bytes(&mut buf);
+        let id2 = NeedleId::from_bytes(&buf);
+        assert_eq!(id, id2);
+    }
+
+    #[test]
+    fn test_needle_id_display() {
+        let id = NeedleId(255);
+        assert_eq!(id.to_string(), "ff");
+    }
+
+    #[test]
+    fn test_needle_id_parse() {
+        let id = NeedleId::parse("ff").unwrap();
+        assert_eq!(id, NeedleId(255));
+    }
+
+    #[test]
+    fn test_cookie_round_trip() {
+        let cookie = Cookie(0xdeadbeef);
+        let mut buf = [0u8; 4];
+        cookie.to_bytes(&mut buf);
+        let cookie2 = Cookie::from_bytes(&buf);
+        assert_eq!(cookie, cookie2);
+    }
+
+    #[test]
+    fn test_size_semantics() {
+        assert!(Size(100).is_valid());
+        assert!(!Size(100).is_deleted());
+        assert!(!Size(100).is_tombstone());
+        assert_eq!(Size(100).raw(), 100);
+
+        assert!(Size(-50).is_deleted());
+        assert!(!Size(-50).is_tombstone());
+        assert_eq!(Size(-50).raw(), 50);
+
+        assert!(TOMBSTONE_FILE_SIZE.is_deleted());
+        assert!(TOMBSTONE_FILE_SIZE.is_tombstone());
+        assert_eq!(TOMBSTONE_FILE_SIZE.raw(), 0);
+
+        assert!(!Size(0).is_valid());
+        assert!(!Size(0).is_deleted());
+    }
+
+    #[test]
+    fn test_size_round_trip() {
+        let size = Size(12345);
+        let mut buf = [0u8; 4];
+        size.to_bytes(&mut buf);
+        let size2 = Size::from_bytes(&buf);
+        assert_eq!(size, size2);
+    }
+
+    #[test]
+    fn test_size_negative_round_trip() {
+        // Negative sizes round-trip through u32 bit pattern
+        let size = Size(-50);
+        let mut buf = [0u8; 4];
+        size.to_bytes(&mut buf);
+        let size2 = Size::from_bytes(&buf);
+        assert_eq!(size, size2);
+    }
+
+    #[test]
+    fn test_offset_round_trip() {
+        // Test with a known actual offset
+        let actual_offset: i64 = 8 * 1000000; // must be multiple of 8
+        let offset = Offset::from_actual_offset(actual_offset);
+        assert_eq!(offset.to_actual_offset(), actual_offset);
+
+        // Test byte serialization
+        let mut buf = [0u8; 5];
+        offset.to_bytes(&mut buf);
+        let offset2 = Offset::from_bytes(&buf);
+        assert_eq!(offset.to_actual_offset(), offset2.to_actual_offset());
+    }
+
+    #[test]
+    fn test_offset_zero() {
+        let offset = Offset::default();
+        assert!(offset.is_zero());
+        assert_eq!(offset.to_actual_offset(), 0);
+    }
+
+    #[test]
+    fn test_offset_max() {
+        // Max stored value depends on offset size
+        #[cfg(feature = "5bytes")]
+        let max_stored: i64 = (1i64 << 40) - 1; // 5-byte max
+        #[cfg(not(feature = "5bytes"))]
+        let max_stored: i64 = (1i64 << 32) - 1; // 4-byte max
+        let max_actual = max_stored * NEEDLE_PADDING_SIZE as i64;
+        let offset = Offset::from_actual_offset(max_actual);
+        assert_eq!(offset.to_actual_offset(), max_actual);
+    }
+
+    #[test]
+    fn test_offset_size_constants() {
+        #[cfg(feature = "5bytes")]
+        {
+            assert_eq!(OFFSET_SIZE, 5);
+            assert_eq!(NEEDLE_MAP_ENTRY_SIZE, 17); // 8 + 5 + 4
+            assert_eq!(MAX_POSSIBLE_VOLUME_SIZE, 4 * 1024 * 1024 * 1024 * 8 * 256);
+            // 8TB
+        }
+        #[cfg(not(feature = "5bytes"))]
+        {
+            assert_eq!(OFFSET_SIZE, 4);
+            assert_eq!(NEEDLE_MAP_ENTRY_SIZE, 16); // 8 + 4 + 4
+            assert_eq!(MAX_POSSIBLE_VOLUME_SIZE, 4 * 1024 * 1024 * 1024 * 8); // 32GB
+        }
+    }
+
+    #[test]
+    fn test_idx_entry_round_trip() {
+        let key = NeedleId(0xdeadbeef12345678);
+        let offset = Offset::from_actual_offset(8 * 999);
+        let size = Size(4096);
+
+        let mut buf = [0u8; NEEDLE_MAP_ENTRY_SIZE];
+        idx_entry_to_bytes(&mut buf, key, offset, size);
+
+        let (key2, offset2, size2) = idx_entry_from_bytes(&buf);
+        assert_eq!(key, key2);
+        assert_eq!(offset.to_actual_offset(), offset2.to_actual_offset());
+        assert_eq!(size, size2);
+    }
+
+    #[test]
+    fn test_volume_id() {
+        let vid = VolumeId::parse("42").unwrap();
+        assert_eq!(vid, VolumeId(42));
+        assert_eq!(vid.to_string(), "42");
+        assert_eq!(vid.next(), VolumeId(43));
+    }
+
+    #[test]
+    fn test_version() {
+        assert!(VERSION_1.is_supported());
+        assert!(VERSION_2.is_supported());
+        assert!(VERSION_3.is_supported());
+        assert!(!Version(0).is_supported());
+        assert!(!Version(4).is_supported());
+        assert_eq!(Version::current(), VERSION_3);
+    }
+
+    #[test]
+    fn test_disk_type() {
+        assert_eq!(DiskType::from_string(""), DiskType::HardDrive);
+        assert_eq!(DiskType::from_string("hdd"), DiskType::HardDrive);
+        assert_eq!(DiskType::from_string("SSD"), DiskType::Ssd);
+        assert_eq!(
+            DiskType::from_string("nvme"),
+            DiskType::Custom("nvme".to_string())
+        );
+        assert_eq!(DiskType::HardDrive.readable_string(), "hdd");
+        assert_eq!(DiskType::Ssd.readable_string(), "ssd");
+    }
+
+    #[test]
+    fn test_read_option_default() {
+        let ro = ReadOption::default();
+        assert!(!ro.read_deleted);
+        assert!(!ro.attempt_meta_only);
+        assert!(!ro.must_meta_only);
+        assert!(!ro.is_meta_only);
+        assert_eq!(ro.volume_revision, 0);
+        assert!(!ro.is_out_of_range);
+        assert!(!ro.has_slow_read);
+        assert_eq!(ro.read_buffer_size, 0);
+    }
+
+    #[test]
+    fn test_read_option_custom() {
+        let ro = ReadOption {
+            read_deleted: true,
+            attempt_meta_only: true,
+            has_slow_read: true,
+            read_buffer_size: 1024 * 1024,
+            ..ReadOption::default()
+        };
+        assert!(ro.read_deleted);
+        assert!(ro.attempt_meta_only);
+        assert!(!ro.must_meta_only);
+        assert!(!ro.is_meta_only);
+        assert!(ro.has_slow_read);
+        assert_eq!(ro.read_buffer_size, 1024 * 1024);
+    }
+
+    #[test]
+    fn test_read_option_clone() {
+        let ro = ReadOption {
+            is_out_of_range: true,
+            volume_revision: 42,
+            ..ReadOption::default()
+        };
+        let ro2 = ro.clone();
+        assert!(ro2.is_out_of_range);
+        assert_eq!(ro2.volume_revision, 42);
+    }
+}

seaweed-volume/src/storage/volume_idx_rebuild.rs

@@ -0,0 +1,344 @@
+//! Rebuild a missing .idx from the .dat it indexes. Mirrors
+//! `weed/storage/volume_idx_rebuild.go`.
+
+use std::fs::{self, OpenOptions};
+use std::io::{BufWriter, Write};
+use std::path::Path;
+
+use crate::storage::idx;
+use crate::storage::needle::Needle;
+use crate::storage::super_block::SuperBlock;
+use crate::storage::types::*;
+use crate::storage::volume::{
+    Volume, VolumeError, VolumeFileVisitor, fsync_dir, needle_disk_end, scan_volume_file,
+};
+
+/// Writes one .idx row per .dat record, in .dat append order, which is the
+/// shape the volume server's own writes leave behind.
+struct VolumeFileScanner4RebuildIdx<W: Write> {
+    writer: W,
+    dat_size: i64,
+    version: Version,
+    stopped: bool,
+}
+
+impl<W: Write> VolumeFileVisitor for VolumeFileScanner4RebuildIdx<W> {
+    fn visit_super_block(&mut self, _sb: &SuperBlock) -> Result<(), VolumeError> {
+        Ok(())
+    }
+
+    fn read_needle_body(&self) -> bool {
+        false
+    }
+
+    fn visit_needle(&mut self, n: &Needle, offset: i64) -> Result<(), VolumeError> {
+        // A record reaching past the end of .dat is a torn append or a corrupt
+        // header: nothing beyond it is indexable, and a row pointing past EOF
+        // would fail every read of that needle. The all-zero header case ends
+        // the walk upstream.
+        if self.stopped {
+            return Ok(());
+        }
+        if needle_disk_end(Offset::from_actual_offset(offset), n.size, self.version) > self.dat_size
+        {
+            self.stopped = true;
+            return Ok(());
+        }
+        let size = if n.size.is_valid() {
+            n.size
+        } else {
+            TOMBSTONE_FILE_SIZE
+        };
+        idx::write_index_entry(
+            &mut self.writer,
+            n.id,
+            Offset::from_actual_offset(offset),
+            size,
+        )?;
+        Ok(())
+    }
+}
+
+impl Volume {
+    /// Regenerate the volume's .idx from its .dat. The whole index is derivable
+    /// from the data file, so a volume whose index directory has no .idx -- a
+    /// --dir.idx pointed at an empty directory, or a lost index -- comes back on
+    /// its own instead of mounting with every needle invisible. The rows go to a
+    /// temp file that is renamed in, so an interrupted rebuild leaves no partial
+    /// index behind.
+    pub(crate) fn rebuild_idx_file(&self) -> Result<(), VolumeError> {
+        let idx_path = self.file_name(".idx");
+        let dat_path = self.file_name(".dat");
+        let tmp_path = format!("{idx_path}.tmp");
+
+        let rebuild = || -> Result<(), VolumeError> {
+            if let Some(parent) = Path::new(&idx_path).parent() {
+                fs::create_dir_all(parent)?;
+            }
+            let tmp_file = OpenOptions::new()
+                .write(true)
+                .create(true)
+                .truncate(true)
+                .open(&tmp_path)?;
+            let mut scanner = VolumeFileScanner4RebuildIdx {
+                writer: BufWriter::new(&tmp_file),
+                dat_size: fs::metadata(&dat_path)?.len() as i64,
+                version: self.version(),
+                stopped: false,
+            };
+            scan_volume_file(&dat_path, &mut scanner)?;
+            scanner.writer.flush()?;
+            drop(scanner);
+            tmp_file.sync_all()?;
+            fs::rename(&tmp_path, &idx_path)?;
+            fsync_dir(&idx_path)?;
+            Ok(())
+        };
+
+        let result = rebuild();
+        if result.is_err() {
+            let _ = fs::remove_file(&tmp_path);
+        }
+        result
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::storage::needle::Needle;
+    use crate::storage::needle::crc::CRC;
+    use crate::storage::needle_map::NeedleMapKind;
+    use crate::storage::types::*;
+    use crate::storage::volume::{Volume, VolumeSpec};
+    use std::fs;
+    use std::path::Path;
+    use tempfile::TempDir;
+
+    fn needle(id: u64) -> Needle {
+        let data = format!("payload-{id}").into_bytes();
+        Needle {
+            id: NeedleId(id),
+            cookie: Cookie(0x55),
+            data_size: data.len() as u32,
+            checksum: CRC::new(&data),
+            data,
+            ..Needle::default()
+        }
+    }
+
+    // Pointing --dir.idx at a directory with no .idx used to mount the volume on
+    // an empty index; the index is derivable from the .dat, so it must be
+    // rebuilt in place instead.
+    #[test]
+    fn test_load_moved_idx_directory_rebuilds_idx() {
+        let root = TempDir::new().unwrap();
+        let data_dir = root.path().join("data");
+        let old_idx_dir = root.path().join("idxA");
+        let new_idx_dir = root.path().join("idxB");
+        for dir in [&data_dir, &old_idx_dir, &new_idx_dir] {
+            fs::create_dir_all(dir).unwrap();
+        }
+        let data = data_dir.to_str().unwrap();
+        let old_idx = old_idx_dir.to_str().unwrap();
+        let new_idx = new_idx_dir.to_str().unwrap();
+
+        let mut v = Volume::new(
+            data,
+            old_idx,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        for id in 1..=3 {
+            v.write_needle(&mut needle(id), true, false).unwrap();
+        }
+        v.delete_needle(&mut needle(2)).unwrap();
+        v.sync_to_disk().unwrap();
+        let (want_count, want_deleted) = (v.file_count(), v.deleted_count());
+        drop(v);
+
+        let seeded = fs::read(format!("{old_idx}/1.idx")).unwrap();
+
+        let reopened = Volume::new(
+            data,
+            new_idx,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+
+        assert!(
+            Path::new(&format!("{new_idx}/1.idx")).exists(),
+            "idx not rebuilt in the new idx dir"
+        );
+        assert_eq!(reopened.file_count(), want_count);
+        assert_eq!(reopened.deleted_count(), want_deleted);
+        assert_eq!(
+            fs::read(format!("{new_idx}/1.idx")).unwrap(),
+            seeded,
+            "rebuilt idx differs from the one the server wrote"
+        );
+
+        for id in [1, 3] {
+            let mut got = needle(id);
+            got.data.clear();
+            reopened.read_needle(&mut got).unwrap();
+            assert_eq!(got.data, format!("payload-{id}").into_bytes());
+        }
+    }
+
+    // A .dat padded with zeros must not be indexed as needle 0 rows: the walk
+    // stops where the records do.
+    #[test]
+    fn test_rebuild_idx_stops_at_zero_padded_dat_tail() {
+        let root = TempDir::new().unwrap();
+        let dir = root.path().to_str().unwrap();
+
+        let mut v = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        v.write_needle(&mut needle(1), true, false).unwrap();
+        v.sync_to_disk().unwrap();
+        drop(v);
+
+        let seeded = fs::read(format!("{dir}/1.idx")).unwrap();
+        let dat = fs::OpenOptions::new()
+            .write(true)
+            .open(format!("{dir}/1.dat"))
+            .unwrap();
+        let dat_size = dat.metadata().unwrap().len();
+        dat.set_len(dat_size + 4096).unwrap();
+        drop(dat);
+        fs::remove_file(format!("{dir}/1.idx")).unwrap();
+
+        let reopened = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        drop(reopened);
+
+        assert_eq!(
+            fs::read(format!("{dir}/1.idx")).unwrap(),
+            seeded,
+            "the zero-padded tail leaked into the rebuilt idx"
+        );
+    }
+
+    // A .dat whose last append was torn mid-body must not gain an index row
+    // that points past the end of the file.
+    #[test]
+    fn test_rebuild_idx_skips_truncated_dat_tail() {
+        let root = TempDir::new().unwrap();
+        let dir = root.path().to_str().unwrap();
+
+        let mut v = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        v.write_needle(&mut needle(1), true, false).unwrap();
+        v.sync_to_disk().unwrap();
+        let kept = fs::read(format!("{dir}/1.idx")).unwrap();
+        v.write_needle(&mut needle(2), true, false).unwrap();
+        v.sync_to_disk().unwrap();
+        drop(v);
+
+        // Chop the second needle's body, leaving its header intact.
+        let dat = fs::OpenOptions::new()
+            .write(true)
+            .open(format!("{dir}/1.dat"))
+            .unwrap();
+        let torn_size = dat.metadata().unwrap().len() - 8;
+        dat.set_len(torn_size).unwrap();
+        drop(dat);
+        fs::remove_file(format!("{dir}/1.idx")).unwrap();
+
+        let reopened = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        drop(reopened);
+
+        assert_eq!(
+            fs::read(format!("{dir}/1.idx")).unwrap(),
+            kept,
+            "the torn record leaked into the rebuilt idx"
+        );
+    }
+
+    // A corrupt header carrying a negative size advances the .dat walk
+    // backwards, which cycles forever between it and the record before it.
+    #[test]
+    fn test_rebuild_idx_stops_at_negative_size_header() {
+        let root = TempDir::new().unwrap();
+        let dir = root.path().to_str().unwrap();
+
+        let mut v = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        v.write_needle(&mut needle(1), true, false).unwrap();
+        v.sync_to_disk().unwrap();
+        let kept = fs::read(format!("{dir}/1.idx")).unwrap();
+        v.write_needle(&mut needle(2), true, false).unwrap();
+        v.sync_to_disk().unwrap();
+        drop(v);
+
+        let rows = fs::read(format!("{dir}/1.idx")).unwrap();
+        let (_, offset, _) = idx_entry_from_bytes(&rows[NEEDLE_MAP_ENTRY_SIZE..]);
+        let corrupt_at = offset.to_actual_offset();
+
+        // Overwrite the second needle's size field with a negative i32.
+        use std::io::{Seek, SeekFrom, Write};
+        let mut dat = fs::OpenOptions::new()
+            .write(true)
+            .open(format!("{dir}/1.dat"))
+            .unwrap();
+        dat.seek(SeekFrom::Start(
+            corrupt_at as u64 + COOKIE_SIZE as u64 + NEEDLE_ID_SIZE as u64,
+        ))
+        .unwrap();
+        dat.write_all(&[0xff, 0xff, 0xf0, 0x00]).unwrap();
+        drop(dat);
+        fs::remove_file(format!("{dir}/1.idx")).unwrap();
+
+        // Pre-fix the rebuild walked backwards from here and never terminated.
+        let reopened = Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap();
+        drop(reopened);
+
+        assert_eq!(
+            fs::read(format!("{dir}/1.idx")).unwrap(),
+            kept,
+            "the corrupt record leaked into the rebuilt idx"
+        );
+    }
+}

seaweed-volume/src/storage/volume_idx_repair.rs

@@ -0,0 +1,418 @@
+//! Recovery for .idx rows that deletes on a tiered read-only volume
+//! overwrote. Mirrors `weed/storage/volume_idx_repair.go`.
+
+use std::collections::HashMap;
+use std::fs::{self, File, OpenOptions};
+use std::io::{self, BufReader, Write};
+use std::path::Path;
+
+use tracing::info;
+
+use crate::storage::idx;
+use crate::storage::needle::Needle;
+use crate::storage::needle::needle::needle_body_length;
+use crate::storage::types::*;
+use crate::storage::volume::{Volume, VolumeError, fsync_dir};
+
+/// Needles found in the head of .dat, keyed by id, plus the ids in .dat order.
+type DatHeadNeedles = (HashMap<NeedleId, (Offset, Size)>, Vec<NeedleId>);
+
+impl Volume {
+    /// Restore the .idx rows that deletes on a tiered read-only volume used to
+    /// overwrite.
+    ///
+    /// The sorted-file needle map opened .idx read-write but never seeded its
+    /// write position, so a delete wrote its (key, offset 0, tombstone) row at
+    /// .idx offset 0 and advanced one row at a time instead of appending. Every
+    /// delete therefore replaced one more row at the front of .idx, and the rows
+    /// it replaced -- the Put rows indexing the first needles in .dat -- were
+    /// lost. Reads for those needles return not-found even though .dat still
+    /// holds them intact.
+    ///
+    /// The damage leaves a fingerprint: .idx begins with a run of offset-0
+    /// tombstones. A healthy .idx never does. Its first row is the Put for the
+    /// first needle in .dat, and an offset-0 tombstone -- a delete against a
+    /// tiered volume, which appends no .dat record and so has no extent to point
+    /// at -- can only ever land at the tail.
+    ///
+    /// .idx and .dat grow in lockstep, so the clobbered rows indexed exactly the
+    /// first N records of .dat. Re-deriving them is a header-only walk over the
+    /// head of .dat, which stays cheap even when .dat is served from a remote
+    /// tier.
+    ///
+    /// The recovered rows go back in front, where the rows they replace used to
+    /// sit, and every existing row keeps its relative order behind them. That
+    /// restores .idx to .dat append order, which several readers lean on: the
+    /// fingerprint is gone so a later load stops after one row, and the last row
+    /// is the .dat-tail needle again.
+    pub(crate) fn repair_idx_head_tombstones(&self) -> Result<usize, VolumeError> {
+        if !self.has_data_backend() {
+            return Ok(0);
+        }
+        let version = self.version();
+        if !version.is_supported() {
+            return Ok(0);
+        }
+        let first_needle_offset = self.super_block.block_size() as i64;
+        let idx_path = self.file_name(".idx");
+
+        let clobbered = idx_head_tombstone_count(&idx_path)?;
+        if clobbered == 0 {
+            return Ok(0);
+        }
+
+        info!(
+            volume_id = self.id.0,
+            idx = %idx_path,
+            clobbered,
+            "idx starts with offset-0 tombstones, recovering the rows they overwrote from .dat"
+        );
+
+        let (mut lost, order) = self.scan_dat_head(version, first_needle_offset, clobbered)?;
+        drop_indexed_keys(&idx_path, &mut lost)?;
+        if lost.is_empty() {
+            return Ok(0);
+        }
+
+        let mut rows = Vec::with_capacity(lost.len() * NEEDLE_MAP_ENTRY_SIZE);
+        let mut restored = 0;
+        for key in order {
+            let Some((offset, size)) = lost.get(&key).copied() else {
+                continue;
+            };
+            idx::write_index_entry(&mut rows, key, offset, size)?;
+            restored += 1;
+        }
+        prepend_idx_rows(&idx_path, &rows)?;
+        Ok(restored)
+    }
+
+    /// Read the headers of the first `limit` records of .dat and return the
+    /// needles they hold, in .dat order. A record with an invalid size is a
+    /// delete marker, so the key it names drops out of the result rather than
+    /// being resurrected.
+    fn scan_dat_head(
+        &self,
+        version: Version,
+        first_needle_offset: i64,
+        limit: usize,
+    ) -> Result<DatHeadNeedles, VolumeError> {
+        let mut found: HashMap<NeedleId, (Offset, Size)> = HashMap::new();
+        let mut order: Vec<NeedleId> = Vec::new();
+        let mut offset = first_needle_offset;
+
+        for _ in 0..limit {
+            let mut header = [0u8; NEEDLE_HEADER_SIZE];
+            match self.read_exact_at_backend(&mut header, offset as u64) {
+                Ok(()) => {}
+                Err(VolumeError::Io(ref e)) if e.kind() == io::ErrorKind::UnexpectedEof => break,
+                Err(e) => return Err(e),
+            }
+            let (_cookie, id, size) = Needle::parse_header(&header);
+            if size.0 == 0 && id.is_empty() {
+                break;
+            }
+            if size.is_valid() {
+                if found
+                    .insert(id, (Offset::from_actual_offset(offset), size))
+                    .is_none()
+                {
+                    order.push(id);
+                }
+            } else {
+                found.remove(&id);
+            }
+            offset += NEEDLE_HEADER_SIZE as i64 + needle_body_length(size, version);
+        }
+
+        Ok((found, order))
+    }
+}
+
+/// Report how many rows at the front of .idx are offset-0 tombstones. A healthy
+/// .idx answers 0 on its first row.
+fn idx_head_tombstone_count(idx_path: &str) -> Result<usize, VolumeError> {
+    if !Path::new(idx_path).exists() {
+        return Ok(0);
+    }
+    let mut reader = BufReader::new(File::open(idx_path)?);
+
+    let mut clobbered = 0usize;
+    let walk = idx::walk_index_file(&mut reader, 0, |_key, offset, size| {
+        if !offset.is_zero() || !size.is_tombstone() {
+            return Err(stop_walk());
+        }
+        clobbered += 1;
+        Ok(())
+    });
+    finish_walk(walk)?;
+    Ok(clobbered)
+}
+
+/// Remove every candidate the .idx already names, whether by a Put row or a
+/// tombstone. What is left is only what the clobbered rows held.
+fn drop_indexed_keys(
+    idx_path: &str,
+    candidates: &mut HashMap<NeedleId, (Offset, Size)>,
+) -> Result<(), VolumeError> {
+    if candidates.is_empty() {
+        return Ok(());
+    }
+    let mut reader = BufReader::new(File::open(idx_path)?);
+    let walk = idx::walk_index_file(&mut reader, 0, |key, _offset, _size| {
+        candidates.remove(&key);
+        if candidates.is_empty() {
+            return Err(stop_walk());
+        }
+        Ok(())
+    });
+    finish_walk(walk)
+}
+
+/// Rewrite .idx as `rows` followed by its current contents, through a temp file
+/// and a rename so a crash mid-write never leaves a partial index at the live
+/// name.
+fn prepend_idx_rows(idx_path: &str, rows: &[u8]) -> Result<(), VolumeError> {
+    let tmp_path = format!("{}.tmp", idx_path);
+    let mut src = File::open(idx_path)?;
+    let commit = (|| -> io::Result<()> {
+        let src_perm = src.metadata()?.permissions();
+        let mut dst = OpenOptions::new()
+            .write(true)
+            .create(true)
+            .truncate(true)
+            .open(&tmp_path)?;
+        // The rename replaces .idx with this file, so it has to carry the mode
+        // the index already had rather than whatever the umask allows.
+        dst.set_permissions(src_perm)?;
+        dst.write_all(rows)?;
+        io::copy(&mut src, &mut dst)?;
+        dst.sync_all()?;
+        drop(dst);
+        fs::rename(&tmp_path, idx_path)?;
+        fsync_dir(idx_path)
+    })();
+    if commit.is_err() {
+        let _ = fs::remove_file(&tmp_path);
+    }
+    Ok(commit?)
+}
+
+/// Sentinel that ends an `idx::walk_index_file` early once the answer is known.
+fn stop_walk() -> io::Error {
+    io::Error::new(io::ErrorKind::Interrupted, "stop idx walk")
+}
+
+fn finish_walk(walk: io::Result<()>) -> Result<(), VolumeError> {
+    match walk {
+        Err(ref e) if e.kind() == io::ErrorKind::Interrupted => Ok(()),
+        other => Ok(other?),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::storage::needle::crc::CRC;
+    use crate::storage::needle_map::NeedleMapKind;
+    use crate::storage::volume::VolumeSpec;
+    use std::os::unix::fs::{FileExt, PermissionsExt};
+    use tempfile::TempDir;
+
+    fn open_volume(dir: &str) -> Volume {
+        Volume::new(
+            dir,
+            dir,
+            VolumeId(1),
+            NeedleMapKind::InMemory,
+            &VolumeSpec::default(),
+        )
+        .unwrap()
+    }
+
+    fn write_test_volume(dir: &str, needle_count: u64) -> Vec<Vec<u8>> {
+        let mut v = open_volume(dir);
+        let mut written = Vec::new();
+        for i in 1..=needle_count {
+            let data = format!("needle-{}-payload", i).into_bytes();
+            let mut n = Needle {
+                id: NeedleId(i),
+                cookie: Cookie(0x1234_0000 + i as u32),
+                data_size: data.len() as u32,
+                checksum: CRC::new(&data),
+                data: data.clone(),
+                ..Needle::default()
+            };
+            v.write_needle(&mut n, true, false).unwrap();
+            written.push(data);
+        }
+        written
+    }
+
+    /// Reproduce the damage a delete on a tiered read-only volume used to do: it
+    /// writes (key, offset 0, tombstone) rows over the front of .idx instead of
+    /// appending them.
+    fn clobber_idx_head(idx_path: &str, keys: &[u64]) {
+        let file = OpenOptions::new().write(true).open(idx_path).unwrap();
+        for (i, key) in keys.iter().enumerate() {
+            let mut row = Vec::new();
+            idx::write_index_entry(
+                &mut row,
+                NeedleId(*key),
+                Offset::from_actual_offset(0),
+                TOMBSTONE_FILE_SIZE,
+            )
+            .unwrap();
+            file.write_at(&row, (i * NEEDLE_MAP_ENTRY_SIZE) as u64)
+                .unwrap();
+        }
+    }
+
+    fn idx_size(idx_path: &str) -> u64 {
+        std::fs::metadata(idx_path).unwrap().len()
+    }
+
+    fn idx_rows(idx_path: &str) -> Vec<(NeedleId, i64, Size)> {
+        let mut reader = BufReader::new(File::open(idx_path).unwrap());
+        let mut rows = Vec::new();
+        idx::walk_index_file(&mut reader, 0, |key, offset, size| {
+            rows.push((key, offset.to_actual_offset(), size));
+            Ok(())
+        })
+        .unwrap();
+        rows
+    }
+
+    fn read_needle_data(v: &Volume, id: u64) -> Result<Vec<u8>, VolumeError> {
+        let mut n = Needle {
+            id: NeedleId(id),
+            ..Needle::default()
+        };
+        v.read_needle(&mut n)?;
+        Ok(n.data)
+    }
+
+    #[test]
+    fn test_repair_idx_head_tombstones_restores_clobbered_rows() {
+        let tmp = TempDir::new().unwrap();
+        let dir = tmp.path().to_str().unwrap();
+        let idx_path = format!("{}/1.idx", dir);
+
+        let written = write_test_volume(dir, 12);
+        let size_before = idx_size(&idx_path);
+
+        // The rewrite replaces .idx wholesale, so it must not widen the mode.
+        fs::set_permissions(&idx_path, fs::Permissions::from_mode(0o600)).unwrap();
+
+        // Deletes against needles 9..12 land on the front of .idx and take the
+        // rows indexing needles 1..4 with them.
+        clobber_idx_head(&idx_path, &[9, 10, 11, 12]);
+
+        let v = open_volume(dir);
+        for i in 1..=8u64 {
+            assert_eq!(
+                read_needle_data(&v, i).unwrap(),
+                written[(i - 1) as usize],
+                "needle {} not recovered",
+                i
+            );
+        }
+
+        let want = size_before + 4 * NEEDLE_MAP_ENTRY_SIZE as u64;
+        assert_eq!(idx_size(&idx_path), want, "idx size after recovery");
+
+        assert_eq!(
+            fs::metadata(&idx_path).unwrap().permissions().mode() & 0o777,
+            0o600,
+            "idx mode after recovery"
+        );
+
+        // The recovered rows go back in front, so .idx is in .dat append order
+        // again: the fingerprint is gone and the last row is still the .dat tail.
+        let rows = idx_rows(&idx_path);
+        assert_eq!(
+            rows[0].1,
+            v.super_block.block_size() as i64,
+            "first row does not index the first needle in .dat"
+        );
+        for i in 1..4 {
+            assert!(
+                rows[i - 1].1 < rows[i].1,
+                "recovered rows are not in .dat order: {:?} then {:?}",
+                rows[i - 1],
+                rows[i]
+            );
+        }
+
+        // The recovery is idempotent: a second load finds nothing left to restore.
+        drop(v);
+        let _v = open_volume(dir);
+        assert_eq!(idx_size(&idx_path), want, "idx grew on second load");
+    }
+
+    #[test]
+    fn test_repair_idx_head_tombstones_leaves_healthy_idx_alone() {
+        let tmp = TempDir::new().unwrap();
+        let dir = tmp.path().to_str().unwrap();
+        let idx_path = format!("{}/1.idx", dir);
+
+        write_test_volume(dir, 6);
+        let mut v = open_volume(dir);
+        v.delete_needle(&mut Needle {
+            id: NeedleId(3),
+            ..Needle::default()
+        })
+        .unwrap();
+        drop(v);
+
+        let size_before = idx_size(&idx_path);
+        let v = open_volume(dir);
+
+        assert_eq!(
+            idx_size(&idx_path),
+            size_before,
+            "healthy idx was rewritten"
+        );
+        assert!(
+            matches!(
+                read_needle_data(&v, 3),
+                Err(VolumeError::Deleted) | Err(VolumeError::NotFound)
+            ),
+            "needle 3 should stay deleted"
+        );
+    }
+
+    /// A needle deleted before the damage must not be resurrected: its tombstone
+    /// row survives at the tail of .idx, so the key is still indexed and stays
+    /// out of the recovery.
+    #[test]
+    fn test_repair_idx_head_tombstones_keeps_deleted_needles_deleted() {
+        let tmp = TempDir::new().unwrap();
+        let dir = tmp.path().to_str().unwrap();
+        let idx_path = format!("{}/1.idx", dir);
+
+        write_test_volume(dir, 8);
+        let mut v = open_volume(dir);
+        v.delete_needle(&mut Needle {
+            id: NeedleId(2),
+            ..Needle::default()
+        })
+        .unwrap();
+        drop(v);
+
+        clobber_idx_head(&idx_path, &[7, 8]);
+
+        let v = open_volume(dir);
+        assert!(
+            matches!(
+                read_needle_data(&v, 2),
+                Err(VolumeError::Deleted) | Err(VolumeError::NotFound)
+            ),
+            "needle 2 was resurrected"
+        );
+        assert!(
+            read_needle_data(&v, 1).is_ok(),
+            "needle 1 should have been recovered"
+        );
+    }
+}

seaweed-volume/src/storage/volume_open.rs

@@ -0,0 +1,27 @@
+//! Opening volume data and index files without access-time updates.
+
+use std::fs::{File, OpenOptions};
+use std::io;
+use std::path::Path;
+
+/// Open a volume data or index file with `O_NOATIME`. Nothing reads these
+/// files' atime, but without the flag every needle read dirties the inode —
+/// even relatime writes atime on the first read after each write, so an
+/// actively written volume pays a metadata write per read/write cycle.
+/// Matches Go's `backend.OpenVolumeFile`.
+pub fn open_volume_file(opts: &OpenOptions, path: impl AsRef<Path>) -> io::Result<File> {
+    let path = path.as_ref();
+    #[cfg(target_os = "linux")]
+    {
+        use std::os::unix::fs::OpenOptionsExt;
+        let mut noatime = opts.clone();
+        noatime.custom_flags(libc::O_NOATIME);
+        match noatime.open(path) {
+            // O_NOATIME is refused unless we own the file or hold CAP_FOWNER.
+            Err(e) if e.raw_os_error() == Some(libc::EPERM) => opts.open(path),
+            result => result,
+        }
+    }
+    #[cfg(not(target_os = "linux"))]
+    opts.open(path)
+}

seaweed-volume/src/storage/volume_report.rs

@@ -0,0 +1,130 @@
+//! Mirror of `weed/storage/store_volume_report.go`.
+
+use std::collections::HashMap;
+use std::sync::Mutex;
+
+/// Identifies one reported copy. Keyed by disk as well as id because a volume
+/// id can be mounted on two disks, and reporting one of them would leave the
+/// other's changes untold.
+pub type VolumeReportKey = (u32, u32);
+
+/// What the master was told about one volume copy: the hash that detects
+/// change, and the heartbeat pass that last found the copy held.
+#[derive(Clone, Copy)]
+struct ReportedVolume {
+    hash: u64,
+    pass: u64,
+}
+
+/// Everything a heartbeat reads and writes about what the master was told,
+/// under one lock. The full-list flag and the generation that answers it have
+/// to move together: split across atomics, a request landing between two of
+/// them is answered by a heartbeat that never carried a list. Go holds a single
+/// mutex over the same fields.
+#[derive(Default)]
+struct ReportState {
+    /// Set once the master says it compares digests. Until then the whole list
+    /// goes every time, which is what an older master needs.
+    deltas_accepted: bool,
+    full_list_needed: bool,
+    /// Counts requests for the whole list, so one arriving while a heartbeat is
+    /// being built is not marked satisfied by it.
+    full_list_generation: u64,
+    /// Numbers heartbeats, so one can mark the copies it finds held without
+    /// building a second map of them.
+    pass: u64,
+    last_reported: HashMap<VolumeReportKey, ReportedVolume>,
+}
+
+/// Remembers what the master was last told about each volume, so a heartbeat
+/// can carry only what moved since.
+///
+/// Per-connection: a server that reconnects, or reaches a different master,
+/// knows nothing about what that master holds and starts again from the full
+/// list. The default has told no master anything, so it sends the whole list
+/// until one accepts changes.
+#[derive(Default)]
+pub struct VolumeReportState {
+    state: Mutex<ReportState>,
+}
+
+impl VolumeReportState {
+    /// Drops everything known about the master's view.
+    pub fn reset(&self) {
+        let mut state = self.state.lock().unwrap();
+        state.deltas_accepted = false;
+        state.full_list_needed = true;
+        state.full_list_generation += 1;
+        state.last_reported.clear();
+    }
+
+    pub fn accept_deltas(&self) {
+        self.state.lock().unwrap().deltas_accepted = true;
+    }
+
+    pub fn request_full_list(&self) {
+        let mut state = self.state.lock().unwrap();
+        state.full_list_needed = true;
+        state.full_list_generation += 1;
+    }
+
+    /// Opens a heartbeat: whether it must carry the whole list, the request it
+    /// answers, and the pass number that marks the copies it finds still held.
+    pub fn begin(&self) -> (bool, u64, u64) {
+        let mut state = self.state.lock().unwrap();
+        state.pass += 1;
+        (
+            state.full_list_needed || !state.deltas_accepted,
+            state.full_list_generation,
+            state.pass,
+        )
+    }
+
+    /// Reports whether the master needs telling about this volume, given what
+    /// it was last told. For a caller that is only taking a snapshot and so
+    /// must leave the reporting state alone; a heartbeat calls `record`.
+    pub fn changed(&self, key: VolumeReportKey, hash: u64) -> bool {
+        self.state
+            .lock()
+            .unwrap()
+            .last_reported
+            .get(&key)
+            .is_none_or(|previous| previous.hash != hash)
+    }
+
+    /// Marks one volume copy as held by the heartbeat being built, and reports
+    /// whether the master needs telling about it. It updates the entry already
+    /// held rather than build a second map beside it, so a server whose volumes
+    /// are quiet allocates nothing per volume per heartbeat.
+    pub fn record(&self, key: VolumeReportKey, hash: u64, pass: u64) -> bool {
+        let mut state = self.state.lock().unwrap();
+        match state.last_reported.get_mut(&key) {
+            Some(previous) => {
+                let changed = previous.hash != hash;
+                previous.hash = hash;
+                previous.pass = pass;
+                changed
+            }
+            None => {
+                state
+                    .last_reported
+                    .insert(key, ReportedVolume { hash, pass });
+                true
+            }
+        }
+    }
+
+    /// Closes the heartbeat. Copies this pass did not find are forgotten, so one
+    /// that comes back is reported again.
+    pub fn commit(&self, pass: u64, generation: u64) {
+        let mut state = self.state.lock().unwrap();
+        state
+            .last_reported
+            .retain(|_, reported| reported.pass == pass);
+        // A request that arrived while this heartbeat was being built asked
+        // about a later state than it carries, so it stands.
+        if state.full_list_generation == generation {
+            state.full_list_needed = false;
+        }
+    }
+}

seaweed-volume/src/storage/volume_report_hash.rs

@@ -0,0 +1,111 @@
+//! Mirror of `weed/storage/volume_report_hash.go`.
+//!
+//! The master compares the digest a volume server reports against one it
+//! computes itself, so this has to agree with the Go implementation
+//! byte-for-byte. `report_hash_vectors` pins that against values produced by
+//! the Go side; do not change the layout without regenerating them there.
+
+use xxhash_rust::xxh64::xxh64;
+
+use crate::pb::master_pb;
+
+/// Digests everything a volume server reports about a volume.
+///
+/// It must cover every field of `VolumeInformationMessage`: a change the hash
+/// misses is a change the master would never be told about.
+pub fn report_hash(m: &master_pb::VolumeInformationMessage) -> u64 {
+    let mut buf = [0u8; 57];
+    buf[0..4].copy_from_slice(&m.id.to_le_bytes());
+    buf[4..12].copy_from_slice(&m.size.to_le_bytes());
+    buf[12..20].copy_from_slice(&m.file_count.to_le_bytes());
+    buf[20..28].copy_from_slice(&m.delete_count.to_le_bytes());
+    buf[28..36].copy_from_slice(&m.deleted_byte_count.to_le_bytes());
+    // The master stores these narrowed, so hash what it will hold, not what the
+    // wire type could carry.
+    buf[36..40].copy_from_slice(&((m.replica_placement as u8) as u32).to_le_bytes());
+    buf[40..44].copy_from_slice(&((m.version as u8) as u32).to_le_bytes());
+    buf[44..48].copy_from_slice(&normalize_ttl(m.ttl).to_le_bytes());
+    buf[48..52].copy_from_slice(&m.compact_revision.to_le_bytes());
+    buf[52..56].copy_from_slice(&m.disk_id.to_le_bytes());
+    if m.read_only {
+        buf[56] = 1;
+    }
+    if m.read_only_can_delete {
+        buf[56] |= 2;
+    }
+    let mut h = xxh64(&buf, 0);
+
+    h = fold(h, xxh64(&(m.modified_at_second as u64).to_le_bytes(), 0));
+    h = fold(h, xxh64(m.collection.as_bytes(), 0));
+    h = fold(h, xxh64(m.disk_type.as_bytes(), 0));
+    // Not the remote storage key: the master does not keep it, so a change to
+    // it alters nothing its copy holds.
+    h = fold(h, xxh64(m.remote_storage_name.as_bytes(), 0));
+    h
+}
+
+/// A ttl whose count is zero encodes as zero however the unit is set, matching
+/// what the master stores after decoding it.
+fn normalize_ttl(ttl: u32) -> u32 {
+    let count = (ttl >> 8) & 0xff;
+    if count == 0 {
+        return 0;
+    }
+    (count << 8) | (ttl & 0xff)
+}
+
+/// Combines two hashes order-dependently, so swapping two string fields is not
+/// invisible.
+fn fold(h: u64, x: u64) -> u64 {
+    let h = (h ^ x).wrapping_mul(0x9E37_79B9_7F4A_7C15);
+    h ^ (h >> 29)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    // Produced by the Go implementation. If these drift, every volume server
+    // running this build reports a digest the master can never match, and falls
+    // back to sending its whole volume list forever.
+    #[test]
+    fn report_hash_vectors() {
+        let empty = master_pb::VolumeInformationMessage::default();
+        assert_eq!(report_hash(&empty), 10988706248825469653);
+
+        let one = master_pb::VolumeInformationMessage {
+            id: 1,
+            ..Default::default()
+        };
+        assert_eq!(report_hash(&one), 2035849960016744285);
+
+        let full = master_pb::VolumeInformationMessage {
+            id: 42,
+            size: 1 << 30,
+            collection: "c".to_string(),
+            file_count: 7,
+            delete_count: 2,
+            deleted_byte_count: 99,
+            read_only: true,
+            read_only_can_delete: false,
+            replica_placement: 10,
+            version: 3,
+            ttl: 3 << 8,
+            compact_revision: 5,
+            modified_at_second: 1700000000,
+            remote_storage_name: "s3".to_string(),
+            remote_storage_key: "k/1.dat".to_string(),
+            disk_type: "ssd".to_string(),
+            disk_id: 2,
+        };
+        assert_eq!(report_hash(&full), 2748844479819636032);
+    }
+
+    #[test]
+    fn ttl_with_no_count_is_dropped() {
+        assert_eq!(normalize_ttl(0), 0);
+        assert_eq!(normalize_ttl(3), 0);
+        assert_eq!(normalize_ttl(3 << 8), 3 << 8);
+        assert_eq!(normalize_ttl((3 << 8) | 4), (3 << 8) | 4);
+    }
+}

seaweed-volume/src/version.rs

@@ -0,0 +1,79 @@
+//! Version helpers aligned with Go's util/version package.
+
+use std::sync::OnceLock;
+
+#[cfg(feature = "5bytes")]
+const SIZE_LIMIT: &str = "8000GB"; // Matches Go production builds (5BytesOffset)
+#[cfg(not(feature = "5bytes"))]
+const SIZE_LIMIT: &str = "30GB"; // Matches Go default build (!5BytesOffset)
+
+pub fn size_limit() -> &'static str {
+    SIZE_LIMIT
+}
+
+pub fn commit() -> &'static str {
+    option_env!("SEAWEEDFS_COMMIT")
+        .or(option_env!("GIT_COMMIT"))
+        .or(option_env!("GIT_SHA"))
+        .unwrap_or("")
+}
+
+pub fn version_number() -> &'static str {
+    static VERSION_NUMBER: OnceLock<String> = OnceLock::new();
+    VERSION_NUMBER
+        .get_or_init(|| {
+            parse_go_version_number().unwrap_or_else(|| env!("CARGO_PKG_VERSION").to_string())
+        })
+        .as_str()
+}
+
+pub fn version() -> &'static str {
+    static VERSION: OnceLock<String> = OnceLock::new();
+    VERSION
+        .get_or_init(|| format!("{} {}", size_limit(), version_number()))
+        .as_str()
+}
+
+pub fn full_version() -> &'static str {
+    static FULL: OnceLock<String> = OnceLock::new();
+    FULL.get_or_init(|| format!("{} {}", version(), commit()))
+        .as_str()
+}
+
+pub fn server_header() -> &'static str {
+    static HEADER: OnceLock<String> = OnceLock::new();
+    HEADER
+        .get_or_init(|| format!("SeaweedFS Volume {}", version()))
+        .as_str()
+}
+
+fn parse_go_version_number() -> Option<String> {
+    let src = include_str!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/../weed/util/version/constants.go"
+    ));
+    let mut major: Option<u32> = None;
+    let mut minor: Option<u32> = None;
+    for line in src.lines() {
+        let l = line.trim();
+        if l.starts_with("MAJOR_VERSION") {
+            major = parse_int32_line(l);
+        } else if l.starts_with("MINOR_VERSION") {
+            minor = parse_int32_line(l);
+        }
+        if major.is_some() && minor.is_some() {
+            break;
+        }
+    }
+    match (major, minor) {
+        (Some(maj), Some(min)) => Some(format!("{}.{:02}", maj, min)),
+        _ => None,
+    }
+}
+
+fn parse_int32_line(line: &str) -> Option<u32> {
+    let start = line.find("int32(")? + "int32(".len();
+    let rest = &line[start..];
+    let end = rest.find(')')?;
+    rest[..end].trim().parse::<u32>().ok()
+}

seaweed-volume/tests/admin_auth_coverage.rs

@@ -0,0 +1,182 @@
+//! Parity guard for the gRPC admin-auth gate.
+//!
+//! Every `VolumeServer` handler must either call `check_grpc_admin_auth` or be
+//! listed here as intentionally ungated, with the reason it stays open. Mirrors
+//! the Go side's `TestVolumeServerAdminAuthCoverage`
+//! (`weed/server/volume_grpc_admin_auth_coverage_test.go`). The gate had
+//! silently dropped off 15 handlers on the Rust side; this keeps the two
+//! implementations from drifting apart again.
+
+use std::collections::{HashMap, HashSet};
+
+/// Handlers that intentionally run without `check_grpc_admin_auth`, each with
+/// the reason. Kept in sync with `ungatedVolumeServerRPCs` on the Go side.
+///
+/// The split is by caller, not by how destructive the call is: the guard checks
+/// the peer IP against `-whiteList`, which holds masters, shell hosts and
+/// workers -- not every peer volume server. Gating a volume-server ->
+/// volume-server call therefore breaks replication, EC and tiering, so those
+/// calls stay open and need a cluster-peer identity before they can be closed.
+fn ungated_handlers() -> HashMap<&'static str, &'static str> {
+    [
+        // Cluster-internal: issued volume server -> volume server.
+        (
+            "copy_file",
+            "replica sync and EC task pull whole files from a peer",
+        ),
+        (
+            "read_needle_blob",
+            "replica sync, vacuum and EC rebuild read needles from a peer",
+        ),
+        (
+            "read_needle_meta",
+            "replica sync compares needle metadata across peers",
+        ),
+        ("write_needle_blob", "replica sync repairs a peer's needle"),
+        (
+            "receive_file",
+            "EC shard distribution pushes shards to a peer",
+        ),
+        (
+            "read_volume_file_status",
+            "the copy path queries the source volume server",
+        ),
+        (
+            "volume_ec_shard_read",
+            "a volume server reads EC shards held by a peer",
+        ),
+        (
+            "volume_ec_blob_delete",
+            "EC delete is fanned out to the shard holders",
+        ),
+        (
+            "volume_ec_shards_info",
+            "EC verification polls shard holders",
+        ),
+        (
+            "volume_ec_shards_mount",
+            "EC shard distribution mounts on the receiving peer",
+        ),
+        (
+            "volume_incremental_copy",
+            "volume backup pulls increments from a peer",
+        ),
+        (
+            "volume_sync_status",
+            "sync compares volume state across peers",
+        ),
+        (
+            "volume_tail_sender",
+            "the tail source streams to the receiving peer",
+        ),
+        (
+            "volume_status",
+            "replica sync and the master's vacuum loop poll volume status",
+        ),
+        // Read-only or liveness: no state change.
+        ("ping", "liveness probe"),
+        ("get_state", "read-only volume server state"),
+        ("query", "read-only data query"),
+        (
+            "vacuum_volume_check",
+            "read-only garbage ratio; the vacuum steps that act on it are gated",
+        ),
+        (
+            "volume_server_status",
+            "read-only status, the gRPC counterpart of the /status page",
+        ),
+    ]
+    .into_iter()
+    .collect()
+}
+
+/// Whether a handler body actually invokes the gate. We can't run a real AST
+/// pass like the Go side without pulling in a parser, so approximate it: drop
+/// `//` line comments (the realistic way a stray mention would sneak in) and
+/// look for the call form `check_grpc_admin_auth(`, not the bare identifier, so
+/// a comment or doc reference can't make an ungated handler pass.
+fn calls_gate(body: &str) -> bool {
+    body.lines()
+        .map(|line| line.split_once("//").map_or(line, |(code, _)| code))
+        .any(|code| code.contains("check_grpc_admin_auth("))
+}
+
+/// Enumerate the `VolumeServer` trait handlers and whether each gates on
+/// `check_grpc_admin_auth`, by scanning the source of the trait impl block.
+fn handler_gating(src: &str) -> Vec<(String, bool)> {
+    let impl_start = src
+        .find("impl VolumeServer for VolumeGrpcService {")
+        .expect("locate the VolumeServer trait impl");
+    let after = &src[impl_start..];
+    // rustfmt keeps every method body indented, so the impl's own closing brace
+    // is the first line that begins at column 0.
+    let impl_end = after
+        .match_indices("\n}")
+        .next()
+        .map(|(i, _)| i + 1)
+        .expect("locate the end of the trait impl");
+    let block = &after[..impl_end];
+
+    let marker = "\n    async fn ";
+    let starts: Vec<usize> = block.match_indices(marker).map(|(i, _)| i).collect();
+    let mut handlers = Vec::with_capacity(starts.len());
+    for (i, &start) in starts.iter().enumerate() {
+        let name_start = start + marker.len();
+        let name_end = name_start
+            + block[name_start..]
+                .find('(')
+                .expect("handler signature has an argument list");
+        let name = block[name_start..name_end].to_string();
+        let body_end = starts.get(i + 1).copied().unwrap_or(block.len());
+        let gated = calls_gate(&block[start..body_end]);
+        handlers.push((name, gated));
+    }
+    handlers
+}
+
+#[test]
+fn volume_server_admin_auth_coverage() {
+    let src = std::fs::read_to_string(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/src/server/grpc_server.rs"
+    ))
+    .expect("read grpc_server.rs");
+
+    let handlers = handler_gating(&src);
+    assert!(
+        handlers.len() >= 40,
+        "parsed only {} handlers from the trait impl, expected the full service",
+        handlers.len()
+    );
+
+    let ungated = ungated_handlers();
+    let mut problems = Vec::new();
+    for (name, gated) in &handlers {
+        match (gated, ungated.contains_key(name.as_str())) {
+            (true, true) => problems.push(format!(
+                "{name} calls check_grpc_admin_auth but is also listed as intentionally ungated; drop it from ungated_handlers"
+            )),
+            (false, false) => problems.push(format!(
+                "{name} does not call check_grpc_admin_auth and is not listed as intentionally ungated; gate it, or add it with the reason it must stay open"
+            )),
+            _ => {}
+        }
+    }
+
+    // Keep the exemption list honest: an entry for a handler that no longer
+    // exists hides the fact that nothing is being exempted.
+    let names: HashSet<&str> = handlers.iter().map(|(n, _)| n.as_str()).collect();
+    for name in ungated.keys() {
+        if !names.contains(name) {
+            problems.push(format!(
+                "ungated_handlers lists \"{name}\", which is not a VolumeServer handler"
+            ));
+        }
+    }
+
+    assert!(
+        problems.is_empty(),
+        "admin-auth coverage gaps:\n{}",
+        problems.join("\n")
+    );
+}