Fix other sentences that carry faults the review named

A search of the four documents, the help text and config.example.yml
for the review's faults found these.

- prune help, and snapshot purge and snapshot list in README.md: a
  snapshot is dropped or reported when its manifest is gone, not its
  metadata or directory.
- README.md, version: the date printed is the commit date.
- README.md, Configuration: names the commands that start without a
  config that loads.
- README.md, snapshot purge: "each such snapshot" named nothing.
- README.md and docs/DATAMODEL.md: a backup can upload unchanged
  content again, so "uploads only what changed" was false.
- docs/REPOSTRUCTURE.md: db.zst.age holds more than one snapshot's
  rows.
- ARCHITECTURE.md: "database records" read as a noun.

Model: fable-5-1

ARCHITECTURE.md

@@ -1,411 +1,401 @@
-# Vaultik Architecture
+# vaultik architecture
 
-This document describes the internal architecture of Vaultik, focusing on the data model, type instantiation, and the relationships between core modules.
+This document describes how the `vaultik` code is organised. It covers three
+things:
 
-## Overview
+- the packages
+- how they are wired together
+- what happens during a backup, a restore, a verify and a prune
 
-Vaultik is a backup system that uses content-defined chunking for deduplication and packs chunks into large, compressed, encrypted blobs for efficient cloud storage. The system is built around dependency injection using [uber-go/fx](https://github.com/uber-go/fx).
+Two other documents cover the stored data:
 
-## Data Flow
+- [docs/DATAMODEL.md](docs/DATAMODEL.md) describes the local index.
+- [docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md) describes the layout of the
+  destination.
+
+## Terms
+
+- **Snapshot**: one backup of the paths listed under one name in the config.
+- **Chunk**: a piece of a file's content. A chunk is named by the SHA-256 hash
+  of its content.
+- **Blob**: one object at the destination. It holds many chunks, compressed and
+  then encrypted.
+- **Local index**: the SQLite database on the backed-up machine.
+- **Destination**: the place the backups are stored.
+- **Manifest**: the unencrypted list of the blobs one snapshot uses.
+- **Remote key**: a hash of the snapshot ID. It names the snapshot's directory
+  at the destination.
+
+## How a backup moves data
 
 ```
-Source Files
+source files
      │
      ▼
-┌─────────────────┐
-│    Scanner      │  Walks directories, detects changed files
-└────────┬────────┘
-         │
-         ▼
-┌─────────────────┐
-│    Chunker      │  Splits files into variable-size chunks (FastCDC)
-└────────┬────────┘
-         │
-         ▼
-┌─────────────────┐
-│    Packer       │  Accumulates chunks, compresses (zstd), encrypts (age)
-└────────┬────────┘
-         │
-         ▼
-┌─────────────────┐
-│   S3 Client     │  Uploads blobs to remote storage
-└─────────────────┘
+  Scanner    walks the paths and finds the files that changed
+     │
+     ▼
+  Chunker    cuts each changed file into chunks
+     │
+     ▼
+  Packer     packs new chunks into a blob, compressed and encrypted
+     │
+     ▼
+  Storer     uploads the blob to the destination
 ```
 
-## Data Model
+## Packages
 
-### Core Entities
+- `cmd/vaultik`: the `main` function. It starts optional profiling and calls
+  `cli.Entry`.
+- `internal/cli`: the commands, their flags and their help text. It builds the
+  application, takes the lock and handles interrupts.
+- `internal/vaultik`: the code behind each command. `Vaultik` holds every
+  dependency a command uses.
+- `internal/snapshot`: the `Scanner`, which backs up one path, and the
+  `SnapshotManager`, which creates snapshot records and uploads their metadata.
+- `internal/chunker`: cuts a stream into chunks with the FastCDC algorithm.
+- `internal/blob`: the `Packer`, which collects chunks into blobs.
+- `internal/blobgen`: the writer that hashes, compresses and encrypts a blob,
+  and the reader that reverses it.
+- `internal/database`: the local index. It holds the schema and the repository
+  types that read and write the tables.
+  [docs/DATAMODEL.md](docs/DATAMODEL.md#code) says which type serves which
+  table.
+- `internal/storage`: the `Storer` interface and its three implementations, for
+  S3, a local directory and rclone.
+- `internal/s3`: the S3 client that the S3 `Storer` uses.
+- `internal/config`: loads and checks the config file.
+- `internal/ui`: formats the status lines, warnings and errors a user sees.
+- `internal/log`: the logger. It writes to stderr.
+- `internal/pidlock`: the lock that keeps two data-changing commands from
+  running at once.
+- `internal/globals`: the version, commit and commit date, set when the binary
+  is built.
+- `internal/types`: distinct types for IDs, hashes and paths.
 
-The database tracks five primary entities and their relationships:
+## Wiring
 
-```
-┌──────────────┐     ┌──────────────┐     ┌──────────────┐
-│   Snapshot   │────▶│     File     │────▶│    Chunk     │
-└──────────────┘     └──────────────┘     └──────────────┘
-       │                                         │
-       │                                         │
-       ▼                                         ▼
-┌──────────────┐                          ┌──────────────┐
-│     Blob     │◀─────────────────────────│  BlobChunk   │
-└──────────────┘                          └──────────────┘
-```
-
-### Entity Descriptions
-
-#### File (`database.File`)
-Represents a file, directory, or symlink in the backup system. Stores metadata needed for restoration:
-- Path, mtime
-- Size, mode, ownership (uid, gid)
-- Symlink target (if applicable)
-
-It also stores `source_path`, the source directory the scan found it under, made absolute and with symlinks resolved. Restore does not read it.
-
-#### Chunk (`database.Chunk`)
-A content-addressed unit of data. Files are split into variable-size chunks using the FastCDC algorithm:
-- `ChunkHash`: SHA256 hash of chunk content (primary key)
-- `Size`: Chunk size in bytes
-
-Chunk sizes vary between `avgChunkSize/4` and `avgChunkSize*4` (2.5MB-40MB for the 10MB default average).
-
-#### FileChunk (`database.FileChunk`)
-Maps files to their constituent chunks:
-- `FileID`: Reference to the file
-- `Idx`: Position of this chunk within the file (0-indexed)
-- `ChunkHash`: Reference to the chunk
-
-#### Blob (`database.Blob`)
-The final storage unit uploaded to S3. Contains many compressed and encrypted chunks:
-- `ID`: UUID assigned at creation
-- `Hash`: `hex(SHA256(SHA256(uncompressed blob contents)))`, computed before compression and encryption (see [docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#blobs-directory-blobs))
-- `UncompressedSize`: Total raw chunk data before compression
-- `CompressedSize`: Size after zstd compression and age encryption
-- `CreatedTS`, `FinishedTS`, `UploadedTS`: Lifecycle timestamps
-
-Blob creation process:
-1. Chunks are accumulated (up to MaxBlobSize, typically 10GB)
-2. As each chunk is added, its uncompressed bytes are fed to a running SHA-256
-3. Concurrently, the same bytes are compressed with zstd, then encrypted with age (recipients configured in config), and written to a temporary file
-4. On finalize, the blob's name is the double SHA-256 of the uncompressed contents — `hex(SHA256(SHA256(...)))` — not a hash of the compressed, encrypted bytes
-5. The finished file is uploaded to `blobs/{hash[0:2]}/{hash[2:4]}/{hash}` and then deleted
-
-A backup needs free temporary space, because each blob is written whole to a temporary file before it is uploaded (up to about `blob_size_limit`; an rclone destination that cannot stream uploads needs about twice that) and the metadata export writes copies of the local index. Temporary files go to `$TMPDIR` (default `/tmp`); with `TMPDIR` unset, SQLite writes one of those copies to `/var/tmp`.
-
-#### BlobChunk (`database.BlobChunk`)
-Maps chunks to their position within blobs:
-- `BlobID`: Reference to the blob
-- `ChunkHash`: Reference to the chunk
-- `Offset`: Byte offset within the uncompressed blob
-- `Length`: Chunk size
-
-#### Snapshot (`database.Snapshot`)
-Represents a point-in-time backup:
-- `ID`: Format is `{hostname}_{snapshot-name}_{RFC3339}` (e.g. `server1_home_2025-06-01T12:00:00Z`)
-- Tracks file count, chunk count, blob count, sizes, compression ratio
-- `CompletedAt`: Null until snapshot finishes successfully
-
-#### SnapshotFile / SnapshotBlob
-Join tables linking snapshots to their files and blobs.
-
-### Relationship Summary
-
-```
-Snapshot 1──────────▶ N SnapshotFile N ◀────────── 1 File
-Snapshot 1──────────▶ N SnapshotBlob N ◀────────── 1 Blob
-File     1──────────▶ N FileChunk    N ◀────────── 1 Chunk
-Blob     1──────────▶ N BlobChunk    N ◀────────── 1 Chunk
-```
-
-## Type Instantiation
-
-### Application Startup
-
-The CLI uses fx for dependency injection. Here's the instantiation order:
+The application is assembled with [uber-go/fx](https://github.com/uber-go/fx), a
+dependency injection library. `cli.NewApp` lists the providers:
 
 ```go
-// cli/app.go: NewApp()
-fx.New(
-    fx.Supply(config.Path(opts.ConfigPath)),         // 1. Config path
-    fx.Supply(opts.LogOptions),                      // 2. Log options
-    fx.Provide(globals.New),                         // 3. Globals
-    fx.Provide(log.New),                             // 4. Logger config
-    config.Module,                                   // 5. Config
-    database.Module,                                 // 6. Database + Repositories
-    log.Module,                                      // 7. Logger initialization
-    storage.Module,                                  // 8. Storage backend (S3/file/rclone)
-    snapshot.Module,                                 // 9. SnapshotManager + ScannerFactory
-    fx.Provide(vaultik.New),                         // 10. Vaultik orchestrator
-)
+fx.Supply(config.Path(opts.ConfigPath)),
+fx.Supply(opts.LogOptions),
+fx.Provide(globals.New),
+fx.Provide(log.New),
+config.Module,
+database.Module,
+log.Module,
+storage.Module,
+snapshot.Module,
+fx.Provide(vaultik.New),
+fx.Invoke(setupGlobals),
+fx.NopLogger,
 ```
 
-### Key Type Instantiation Points
+Each module provides these:
 
-#### 1. Config (`config.Config`)
-- **Created by**: `config.Module` via `config.LoadConfig()`
-- **When**: Application startup (fx DI)
-- **Contains**: All configuration from YAML file (S3 credentials, encryption keys, paths, etc.)
+- `config.Module` provides `*config.Config`, from `config.Load`.
+- `database.Module` provides `*database.DB` and `*database.Repositories`. It
+  creates the index file with mode `0600` when it does not exist.
+- `storage.Module` provides the `storage.Storer` that the config selects.
+- `snapshot.Module` provides the `snapshot.ScannerFactory` and the
+  `*snapshot.SnapshotManager`.
+- `vaultik.New` provides `*vaultik.Vaultik`.
 
-#### 2. Database (`database.DB`)
-- **Created by**: `database.Module` via `database.New()`
-- **When**: Application startup (fx DI)
-- **Contains**: SQLite connection, path reference
+A command runs through `cli.RunOperation`. It works like this:
 
-#### 3. Repositories (`database.Repositories`)
-- **Created by**: `database.Module` via `database.NewRepositories()`
-- **When**: Application startup (fx DI)
-- **Contains**: All repository interfaces (Files, Chunks, Blobs, Snapshots, etc.)
+- `RunWithApp` takes the lock when the command changes data, and then starts the
+  application.
+- The command's function runs in its own goroutine.
+- When the function returns, it asks the application to shut down.
+- On `SIGINT` or `SIGTERM`, the stop hook cancels the `Vaultik` context. It then
+  waits for the function to return, so that its temporary files are removed.
+- The wait ends after 30 seconds at most.
 
-#### 4. Vaultik (`vaultik.Vaultik`)
-- **Created by**: `vaultik.New(VaultikParams)`
-- **When**: Application startup (fx DI)
-- **Contains**: All dependencies for backup operations
+The `config` commands, `database delete`, `completion` and `version` do not
+build the application.
 
-```go
-type Vaultik struct {
-    Globals         *globals.Globals
-    Config          *config.Config
-    DB              *database.DB
-    Repositories    *database.Repositories
-    Storage         storage.Storer
-    ScannerFactory  snapshot.ScannerFactory
-    SnapshotManager *snapshot.SnapshotManager
-    Shutdowner      fx.Shutdowner
-    Fs              afero.Fs
-    ctx             context.Context
-    cancel          context.CancelFunc
-}
-```
+## Backup
 
-#### 5. SnapshotManager (`snapshot.SnapshotManager`)
-- **Created by**: `snapshot.Module` via `snapshot.NewSnapshotManager()`
-- **When**: Application startup (fx DI)
-- **Responsibility**: Creates/completes snapshots, exports metadata to S3
+`Vaultik.CreateSnapshot` runs a backup. It does these steps in order:
 
-#### 6. Scanner (`snapshot.Scanner`)
-- **Created by**: `ScannerFactory(ScannerParams)`
-- **When**: Each `CreateSnapshot()` call
-- **Contains**: Chunker, Packer, progress reporter
+- `EnsureStorageBinding` checks that the local index belongs to the configured
+  destination.
+- `markDestination` checks the marker file and then writes it.
+- `PruneDatabase` removes unfinished snapshots and unused records from the
+  local index.
+- `createNamedSnapshot` runs once for each snapshot name.
+- With `--prune`, `runPostBackupPrune` removes old snapshots and unused data.
 
-```go
-// vaultik/snapshot.go: CreateSnapshot()
-scanner := v.ScannerFactory(snapshot.ScannerParams{
-    EnableProgress: !opts.Cron,
-    Fs:             v.Fs,
-})
-```
+`CreateSnapshot` calls `skipPostBackupPrune` instead of `runPostBackupPrune` in
+these cases:
 
-#### 7. Chunker (`chunker.Chunker`)
-- **Created by**: `chunker.NewChunker(avgChunkSize)`
-- **When**: Inside `snapshot.NewScanner()`
-- **Configuration**:
-  - `avgChunkSize`: From config (default 10MB)
-  - `minChunkSize`: avgChunkSize / 4
-  - `maxChunkSize`: avgChunkSize * 4
+- `--skip-errors` skipped a file or directory that could not be read.
+- A configured path held nothing.
+- An earlier snapshot of the same name found a mount point that this backup did
+  not find.
 
-#### 8. Packer (`blob.Packer`)
-- **Created by**: `blob.NewPacker(PackerConfig)`
-- **When**: Inside `snapshot.NewScanner()`
-- **Configuration**:
-  - `MaxBlobSize`: Maximum blob size before finalization (typically 10GB)
-  - `CompressionLevel`: zstd level (1-19)
-  - `Recipients`: age public keys for encryption
+`PruneDatabase` must succeed before anything is scanned. An unfinished snapshot
+left in the index would make the backup skip files that were never fully
+stored.
 
-```go
-// snapshot/scanner.go: NewScanner()
-packerCfg := blob.PackerConfig{
-    MaxBlobSize:      cfg.MaxBlobSize,
-    CompressionLevel: cfg.CompressionLevel,
-    Recipients:       cfg.AgeRecipients,
-    Repositories:     cfg.Repositories,
-    Fs:               cfg.FS,
-}
-packer, err := blob.NewPacker(packerCfg)
-```
+### One snapshot
 
-## Module Responsibilities
+`createNamedSnapshot` does these steps:
 
-### `internal/cli`
-Entry point for fx application. Combines all modules and handles signal interrupts.
+- `resolveSnapshotPaths` makes each configured path absolute and resolves
+  symlinks.
+- `SnapshotManager.CreateSnapshotWithName` creates the snapshot record. The
+  record is not yet marked complete.
+- `Scanner.Scan` backs up each path in turn.
+- `finalizeSnapshotMetadata` uploads the snapshot's metadata and marks the
+  snapshot complete.
 
-Key functions:
-- `NewApp(AppOptions)` → Creates fx.App with all modules
-- `RunApp(ctx, app)` → Starts app, handles graceful shutdown
-- `RunWithApp(ctx, opts)` → Convenience wrapper
+A new `Scanner` is created for each snapshot, from `ScannerFactory`. It owns a
+`Chunker` and a `Packer`.
 
-### `internal/vaultik`
-Main orchestrator containing all dependencies and command implementations.
+### Scanning one path
 
-Key methods:
-- `New(VaultikParams)` → Constructor (fx DI)
-- `CreateSnapshot(opts)` → Main backup operation
-- `ListSnapshots(jsonOutput)` → List available snapshots
-- `VerifySnapshot(id, deep)` → Verify snapshot integrity
-- `PurgeSnapshots(...)` → Remove old snapshots
+`Scanner.Scan` works in these steps:
 
-### `internal/chunker`
-Content-defined chunking using FastCDC algorithm.
+- `repairInterruptedBlobs` deletes the records of blobs whose upload never
+  finished. Their data is then stored again.
+- `loadDatabaseState` loads the known files and chunks into memory.
+- `scanPhase` walks the path. It sorts each entry into changed or unchanged.
+- `detectDeletedFiles` counts the files that the previous snapshot of this name
+  had and that are now gone.
+- `batchAddFilesToSnapshot` records the unchanged files in the new snapshot.
+- `recordDirectories` records every directory, and the mount points found.
+- `processPhase` reads each changed file and stores it.
 
-Key types:
-- `Chunk` → Hash, Data, Offset, Size
-- `Chunker` → avgChunkSize, minChunkSize, maxChunkSize
+The walk handles each kind of entry differently:
 
-Key methods:
-- `NewChunker(avgChunkSize)` → Constructor
-- `ChunkReaderStreaming(reader, callback)` → Stream chunks with callback (preferred)
-- `ChunkReader(reader)` → Return all chunks at once (memory-intensive)
+- A regular file or a symlink is compared with its record in the index.
+- A directory is recorded with its metadata. It has no content to store.
+- A device node, a named pipe or a socket is skipped.
+- An entry that matches an exclude pattern is skipped. For a directory, so is
+  everything in it.
 
-### `internal/blob`
-Blob packing: accumulates chunks, compresses, encrypts, tracks metadata.
+### Deduplication
 
-Key types:
-- `Packer` → Thread-safe blob accumulator
-- `ChunkRef` → Hash + Data for adding to packer
-- `FinishedBlob` → Completed blob ready for upload
-- `BlobWithReader` → FinishedBlob + io.Reader for streaming upload
+Data is stored once at two levels.
 
-Key methods:
-- `NewPacker(PackerConfig)` → Constructor
-- `AddChunk(ChunkRef)` → Add chunk to current blob
-- `FinalizeBlob()` → Compress, encrypt, hash current blob
-- `Flush()` → Finalize any in-progress blob
-- `SetBlobHandler(func)` → Set callback for upload
+- **Files.** A file is not read again when its record in the index still matches
+  it.
+- **Chunks.** A chunk is named by the SHA-256 hash of its content. A chunk whose
+  hash is already in the index is not stored again.
 
-### `internal/snapshot`
+A file's record matches when all of these are unchanged:
 
-#### Scanner
-Orchestrates the backup process for a directory.
+- the size
+- the modification time
+- the mode
+- the owner and group
+- the link target
 
-Key methods:
-- `NewScanner(ScannerConfig)` → Constructor (creates Chunker + Packer)
-- `Scan(ctx, path, snapshotID)` → Main scan operation
+One case overrides the file check. A file is read again when it lists a chunk
+that no uploaded blob holds. This happens when the only snapshot that used the
+blob holding that chunk is removed. `loadFilesToRechunk` finds these files.
 
-Scan phases:
-1. **Phase 1**: Walk directory, identify files needing processing
-2. **Phase 1b**: Count as deleted the files that the previous snapshot of the same name lists, the walk did not find, and are gone from disk; a file that is excluded, or inside an excluded directory, is not counted
-3. **Phase 2**: Process files (chunk → pack → upload)
+### Chunks
 
-#### SnapshotManager
-Manages snapshot lifecycle and metadata export.
+`chunker.Chunker` cuts a file with the FastCDC algorithm. The cut points depend
+on the content, so an insertion early in a file moves only nearby chunks.
 
-Key methods:
-- `CreateSnapshot(ctx, hostname, version, commit)` → Create snapshot record
-- `PopulateSnapshotBlobs(ctx, snapshotID)` → Record every blob the snapshot references
-- `ExportSnapshotMetadata(ctx, dbPath, snapshotID)` → Export to S3
-- `MarkSnapshotComplete(ctx, snapshotID)` → Record completion, only after a successful export
-- `CompleteSnapshot(ctx, snapshotID)` → Convenience: populate blobs, then mark complete (no export between)
+`chunk_size` sets the average chunk size. No chunk is larger than four times
+this size. Only a file's last chunk can be smaller than a quarter of it. With
+the default of 10 MB, those limits are 40 MB and 2.5 MB.
 
-### `internal/database`
-SQLite database for local index. Single-writer mode for thread safety.
+`ChunkReaderStreaming` passes each chunk to a callback as it is found. The whole
+file is never held in memory.
 
-Key types:
-- `DB` → Database connection wrapper
-- `Repositories` → Collection of all repository interfaces
+### Blobs
 
-Repository interfaces:
-- `FilesRepository` → CRUD for File records
-- `ChunksRepository` → CRUD for Chunk records
-- `BlobsRepository` → CRUD for Blob records
-- `SnapshotsRepository` → CRUD for Snapshot records
-- Plus join table repositories (FileChunks, BlobChunks, etc.)
+`blob.Packer` collects new chunks into a blob. For each blob it does this:
 
-## Snapshot Creation Flow
+- It gives the blob a UUID and creates a temporary file.
+- It adds each chunk's bytes to a running SHA-256 hash.
+- In the same pass it compresses the bytes with zstd and encrypts them with
+  age. The result goes to the temporary file.
+- It records each chunk's offset and length within the blob.
+- It closes the blob when the next chunk would pass `blob_size_limit`, and at
+  the end of each path.
 
-```
-CreateSnapshot(opts)
-    │
-    ├─► markDestination()   // Check the marker (see Storage Layout in S3), then write it
-    │
-    ├─► PruneDatabase()   // Critical: avoid dedup errors
-    │
-    ├─► SnapshotManager.CreateSnapshot()   // Create DB record
-    │
-    ├─► For each source directory:
-    │       │
-    │       ├─► scanner.Scan(ctx, path, snapshotID)
-    │       │       │
-    │       │       ├─► Phase 1: scanPhase()
-    │       │       │       Walk directory
-    │       │       │       Check file metadata changes
-    │       │       │       Build list of files to process
-    │       │       │
-    │       │       ├─► Phase 1b: detectDeletedFiles()
-    │       │       │
-    │       │       └─► Phase 2: processPhase()
-    │       │               For each file:
-    │       │                   chunker.ChunkReaderStreaming()
-    │       │                   For each chunk:
-    │       │                       packer.AddChunk()
-    │       │                       If blob full → FinalizeBlob()
-    │       │                           → handleBlobReady()
-    │       │                           → s3Client.PutObjectWithProgress()
-    │       │               packer.Flush()  // Final blob
-    │       │
-    │       └─► Accumulate statistics
-    │
-    ├─► SnapshotManager.PopulateSnapshotBlobs()   // record referenced blobs
-    │
-    ├─► SnapshotManager.UpdateSnapshotStatsExtended()
-    │
-    ├─► SnapshotManager.ExportSnapshotMetadata()
-    │       │
-    │       ├─► Copy database to temp file
-    │       ├─► Clean to only current snapshot data (VACUUM)
-    │       ├─► Compress binary SQLite with zstd
-    │       ├─► Encrypt with age
-    │       ├─► Upload db.zst.age to storage
-    │       └─► Upload manifest.json.zst to storage
-    │
-    └─► SnapshotManager.MarkSnapshotComplete()   // only after the export succeeds
-```
+The blob's name is `hex(SHA256(SHA256(uncompressed contents)))`. It is computed
+from the chunk data before compression and encryption.
+[docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#blobs) says what follows from
+that.
 
-## Deduplication Strategy
+When a blob is closed, `Scanner.handleBlobReady` stores it:
 
-1. **File-level**: Files and symlinks unchanged since last backup are skipped (metadata comparison: size, mtime, mode, uid, gid, and a symlink's target), unless the file lists a chunk that no uploaded blob holds; such a file is re-chunked
+- It looks the blob up at the destination. A blob that is already there is not
+  uploaded again.
+- Otherwise it uploads the temporary file to
+  `blobs/<hash[0:2]>/<hash[2:4]>/<hash>`.
+- It records the upload in the index and links the blob to the snapshot.
+- It deletes the temporary file.
 
-2. **Chunk-level**: Chunks are content-addressed by SHA256 hash. If a chunk hash already exists in the database, the chunk data is not re-uploaded.
+A file's record is written to the index only after every blob that holds its
+chunks is uploaded.
 
-3. **Blob-level**: Blobs contain only unique chunks. Duplicate chunks within a blob are skipped.
+### Snapshot metadata
 
-## Storage Layout in S3
+`finalizeSnapshotMetadata` does these steps in order:
 
-```
-bucket/
-├── vaultik-destination.txt          # Marker written by every backup
-│
-├── blobs/
-│   └── {hash[0:2]}/
-│       └── {hash[2:4]}/
-│           └── {full-hash}          # Compressed+encrypted blob
-│
-└── metadata/
-    └── {remote-key}/
-        ├── db.zst.age               # Encrypted binary SQLite database
-        └── manifest.json.zst        # Blob list (for pruning/verification)
-```
+- `PopulateSnapshotBlobs` records every blob the snapshot uses.
+- `UpdateSnapshotStatsExtended` stores the snapshot's totals.
+- `ExportSnapshotMetadata` uploads the metadata.
+- `MarkSnapshotComplete` marks the snapshot complete.
 
-A destination that answers a look-up of `vaultik-destination.txt` with not
-found is treated as not present while the local index holds snapshots stored
-there (`checkDestinationMarker` in `internal/vaultik/destination_marker.go`):
-`snapshot create` refuses to write to it, `prune`, `snapshot purge` and
-`remote nuke` fail, and `snapshot list` and `snapshot remove` behave as for a
-destination that cannot be listed. This keeps those commands from taking an
-unplugged volume's empty mount point for a destination whose snapshots are
-all gone. When the look-up fails for any reason other than not found, for
-example when S3 refuses it to credentials that allow only PUT, the check is
-skipped, so a host with such credentials does not notice an emptied
-destination; see
-[docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#marker-file-vaultik-destinationtxt).
-The `file://` backend never creates its destination directory, so a write to a
-missing one fails.
+Completion is recorded last on purpose. A backup that stops during the upload
+leaves the snapshot unfinished, and the next backup removes it and stores its
+data again.
 
-The `{remote-key}` directory name is a one-way double SHA-256 hash of the human
-snapshot ID, so the human ID (hostname, snapshot name, timestamp) is never
-written to the store as a directory name. See
-[docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#remote-key-derivation) for the
-derivation and a worked example.
+`ExportSnapshotMetadata` works on a copy of the index, never on the index
+itself:
 
-## Thread Safety
+- It copies the index into a temporary directory with `VACUUM INTO`.
+- It deletes the rows of every other snapshot from the copy. It also deletes the
+  files and blobs that this snapshot does not use. It then deletes each chunk
+  that no remaining file and no remaining blob uses.
+- It keeps the `local_meta` and `schema_migrations` tables as they are. The
+  uploaded database therefore names the destination, and lists the migrations
+  that have run.
+- It runs `VACUUM`, so that no deleted row is left in the file.
+- It compresses the copy with zstd and encrypts it with age.
+- It builds the manifest from the copy.
+- It uploads `metadata/<remote-key>/db.zst.age`, and then
+  `metadata/<remote-key>/manifest.json.zst`.
 
-- `Packer`: Thread-safe via mutex. Multiple goroutines can call `AddChunk()`.
-- `Scanner`: Uses `packerMu` mutex to coordinate blob finalization.
-- `Database`: Single-writer mode (`MaxOpenConns=1`) ensures SQLite thread safety.
-- `Repositories.WithTx()`: Handles transaction lifecycle automatically.
+The manifest is uploaded last. A directory under `metadata/` with no manifest
+therefore marks a backup that did not finish.
+
+## Restore
+
+`Vaultik.Restore` does these steps:
+
+- `restoreIdentities` parses the age private key. A missing or malformed key
+  stops the restore before anything is downloaded.
+- `downloadSnapshotDB` downloads and decrypts the snapshot's database into a
+  temporary directory.
+- `getFilesToRestore` selects the files, from the paths given.
+- `restoreAllFiles` downloads blobs and writes the files.
+- With `--verify`, `verifyRestoredFiles` reads each restored file and checks its
+  chunk hashes.
+
+The restore does not read the local index or the manifest. It finds each file's
+chunks and blobs through the downloaded database.
+
+### Checks on the downloaded database
+
+Decryption shows that a database is readable. It does not show that the
+database is honest. The restore therefore checks it:
+
+- `database.OpenReadOnly` opens the file read-only. It refuses a file that has a
+  trigger, a view or a virtual table, or that lacks an expected table.
+- `verifySnapshotDBIdentity` checks that the snapshot in the database hashes to
+  the remote key that was asked for.
+- `containedRestorePath` refuses a stored path that would leave the target
+  directory. It also refuses to write through a symlink below the target.
+- A blob is not decompressed beyond the size that the database gives for it.
+- Each downloaded blob is hashed. The hash must match the blob's name.
+
+### Order of work
+
+`restorePlan` orders the files so that few blobs are on disk at once:
+
+- Downloaded blobs are kept decrypted in a cache directory.
+- A file is restored once every blob it needs is in the cache.
+- When no file is ready, the restore downloads the blobs for the file that
+  needs the fewest.
+- `restoreSweeper` deletes a cached blob once no remaining file needs it.
+
+A regular file is created with mode `0600` and gets its stored mode after its
+content is written. A directory gets its stored mode, owner and modification
+time after the whole restore.
+
+## Verify
+
+`Vaultik.VerifySnapshotWithOptions` checks a snapshot without the private key:
+
+- It downloads the manifest.
+- It checks that the snapshot's encrypted database is present.
+- It checks that each listed blob is present with the listed size.
+
+`Vaultik.RunDeepVerify` needs the private key. It does these steps:
+
+- It downloads the manifest and the database. It checks the database as a
+  restore does.
+- `verifyManifestAgainstDatabase` checks that both list the same blobs with the
+  same sizes.
+- `verifyBlobExistenceFromDB` checks that every blob is present with the
+  recorded size.
+- `verifyBlob` downloads each blob, checks the hash of each chunk in it, and
+  checks the hash that names the blob.
+
+## Prune
+
+`Vaultik.Prune` does three steps:
+
+- `CleanupLocalSnapshots` drops from the local index each snapshot whose
+  manifest is gone from the destination.
+- `PruneDatabase` removes unfinished snapshots and unused records from the
+  local index.
+- `PruneBlobs` deletes unused data at the destination.
+
+`PruneBlobs` reads every manifest at the destination. It then deletes two kinds
+of object, once they are older than `prune_grace_period`:
+
+- a blob that no manifest lists
+- a directory under `metadata/` that has no manifest, and whose snapshot is not
+  in the local index
+
+It deletes nothing when any manifest cannot be read.
+[docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#pruning) explains why this is
+safe, and where it is not.
+
+## The destination
+
+`storage.Storer` is the only way the code reaches the destination. It has these
+methods:
+
+- `Put` and `PutWithProgress` store an object.
+- `Get` reads an object.
+- `Stat` reports an object's size.
+- `Delete` removes an object.
+- `List` and `ListStream` list the objects under a prefix.
+- `DeletePartialUploads` removes what an interrupted upload left behind.
+- `Info` describes the destination.
+
+`storage.NewStorer` picks the implementation from `storage_url`:
+
+- `s3://` gives the S3 implementation.
+- `file://` gives the local directory implementation.
+- `rclone://` gives the rclone implementation.
+
+With no `storage_url`, it builds the S3 implementation from the `s3` settings.
+
+## The marker file and the destination binding
+
+Two checks keep the local index and the destination in step.
+
+- `checkDestinationMarker` looks for `vaultik-destination.txt`. It fails when
+  the file is reported missing while the local index holds snapshots.
+- `EnsureStorageBinding` records the configured destination in the local index
+  on first use. Later it fails when the config names a different destination.
+
+Every command that changes data calls `EnsureStorageBinding` first.
+
+## Concurrency
+
+- The local index is opened with one connection, so its queries run one at a
+  time.
+- The index runs in SQLite's WAL mode. A second process can therefore read it
+  while a backup writes to it.
+- `Packer` guards its state with a mutex.
+- `Scanner` holds `packerMu` while a chunk is added to the packer and while a
+  blob is closed.
+- `Repositories.WithTx` runs a function in one transaction. It commits when the
+  function succeeds and rolls back when it fails.

TODO.md

@@ -24,6 +24,15 @@ owner's; a `v*` tag runs the release build in
 
 # Completed Steps
 
+- 2026-10-15: Stopped `snapshot create --prune` pruning after a backup
+  in which a volume mounted below a configured path in an earlier
+  snapshot of the same name was not mounted
+  ([issue #395](https://git.eeqj.de/sneak/vaultik/issues/395)). Such a
+  prune could delete every snapshot still holding the volume's files.
+  Each snapshot now records in the local index the directories it found
+  on a different filesystem from their parent and found an entry in that
+  the backup does not exclude.
+
 - 2026-10-14: Stopped `snapshot create --prune` pruning after a backup
   that found nothing in a configured path
   ([issue #388](https://git.eeqj.de/sneak/vaultik/issues/388)), as in
@@ -1307,5 +1316,64 @@ owner's; a `v*` tag runs the release build in
 
 # Future Steps
 
-Work planned after 1.0 is listed in the README
-[roadmap](README.md#roadmap).
+Work planned after 1.0, loosely ordered by priority.
+
+## Correctness and operability
+
+- **Outside security audit.** Before 1.0 the encryption and
+  blob-generation code was reviewed. Every bug the review found was
+  fixed. The risks it accepted are listed in
+  [Accepted risks](docs/REPOSTRUCTURE.md#accepted-risks). No outside
+  audit has been done. age, zstd and content-defined chunking are mostly
+  off-the-shelf pieces. The code that joins them needs an outside read:
+    - key handling
+    - recipient parsing
+    - manifest trust boundary
+    - restore-time identity validation
+- **Verify restored content end-to-end in CI.** The current integration
+  test does this for a small synthetic snapshot but not at scale. A
+  nightly job against a multi-GB representative snapshot would catch
+  silent regressions in the chunker, packer, or restore planner.
+
+## Performance
+
+- **Parallel blob downloads during restore.** A restore downloads one
+  blob at a time now. With a fast S3 endpoint and a multi-core machine,
+  restore speed is bound by the fetch, decryption and decompression of
+  each blob. Running several of those in parallel against the disk
+  cache would close most of the remaining gap. This needs to interact
+  correctly with the locality planner and the sweeper.
+- **Bandwidth limiting (`--bwlimit`).** For both upload and download.
+  It is useful for backing up over a shared link. It is tricky to make
+  work correctly with parallel downloads.
+- **Restart of interrupted restore.** Today a restore can be started
+  again, and it then overwrites the partial output. It does not resume
+  from where it stopped, and it does not skip files that are already
+  present. A `--resume` mode that checks targets before fetching blobs
+  would matter for very large restores.
+- **Daemon mode.** A long-running mode that watches for file changes,
+  so that frequent backups, such as hourly ones, skip the full scan. It
+  adds little for the usual backups from cron every 12 to 36 hours. See
+  [issue #204](https://git.eeqj.de/sneak/vaultik/issues/204).
+
+## Usability
+
+- **Man pages and richer `--help` examples.** Cobra generates basic
+  help. Man pages would be a separate target.
+- **`vaultik snapshot diff <a> <b>`.** Show which files changed between
+  two snapshots, without restoring either.
+- **Status reporting hook for `--cron`.** When a backup fails silently
+  in cron, the user has no idea. A configurable hook on completion, for
+  both success and failure, would close the loop. It could be a
+  webhook, an email or `notify-send`.
+
+## Infrastructure
+
+- **Schema migrations after 1.0.** Migrations are supported, but none
+  are added before 1.0 because there is no installed base yet. After
+  1.0, each schema change is a new migration, so an existing local
+  index is migrated when `vaultik` is updated. See
+  [`docs/DATAMODEL.md`](docs/DATAMODEL.md#schema-migrations).
+- **Storage backend coverage tests.** S3, `file://` and `rclone://` all
+  share the `Storer` interface, but the rclone path is the least
+  exercised in CI.

config.example.yml

@@ -1,16 +1,31 @@
 # vaultik configuration file example
-# This file shows all available configuration options with their default values
-# Copy this file and uncomment/modify the values you need
+# This file shows every configuration option. The values are examples, not
+# defaults; where an option has a default, its comment gives it.
+# Copy this file and change the values you need
 
 # Age recipient public keys for encryption
 # Backups are encrypted to these public keys. snapshot create needs at least
 # one; listing, verifying and restoring do not
-# Generate with: age-keygen | grep "public key"
+# Generate a key pair, and show its public key, with:
+#   age-keygen -o vaultik_backup_private_key.txt
+#   grep 'public key' vaultik_backup_private_key.txt
+# Keep the private key file: restoring needs it. Do not pipe age-keygen
+# straight into grep, which would discard the private key
 age_recipients:
   - age1cj2k2addawy294f6k2gr2mf9gps9r3syplryxca3nvxj3daqm96qfp84tz
 
+# Age private key for decryption
+# Only snapshot restore and snapshot verify --deep need it. The
+# VAULTIK_AGE_SECRET_KEY environment variable overrides this setting
+# Do not set it on a machine you back up: prefer the environment variable,
+# supplied only on the machine you restore on
+# Default: not set
+#age_secret_key: YOUR_AGE_PRIVATE_KEY
+
 # Named snapshots - each snapshot can contain multiple paths
 # Each snapshot gets its own ID and can have snapshot-specific excludes
+# Required: at least one snapshot with at least one path. Only the config
+# commands, version and completion run without one
 snapshots:
   testing:
     paths:
@@ -258,6 +273,11 @@ exclude:
 
 # Storage URL - use either this OR the s3 section below
 # Supports: s3://bucket/prefix, file:///path, rclone://remote/path
+# An s3:// URL also needs s3.access_key_id and s3.secret_access_key below
+# A file:// path must be absolute, and the directory must already exist:
+# vaultik never creates it
+# An rclone:// URL needs the remote to be defined in rclone's own config
+# (rclone config)
 storage_url: "rclone://myremote/path/to/backups"
 
 # S3-compatible storage configuration
@@ -283,6 +303,8 @@ storage_url: "rclone://myremote/path/to/backups"
 #  #region: us-east-1
 #
 #  # Use SSL/TLS for S3 connections
+#  # Only read when storage_url is not set and the endpoint has no scheme;
+#  # with an s3:// storage_url, use ?ssl=false in the URL instead
 #  # Default: true
 #  #use_ssl: true
 #
@@ -295,6 +317,7 @@ storage_url: "rclone://myremote/path/to/backups"
 
 # Path to local SQLite index database
 # This database tracks file state for incremental backups
+# The VAULTIK_INDEX_PATH environment variable overrides this setting
 # Default: the platform data directory, e.g.
 #   macOS: ~/Library/Application Support/vaultik/index.sqlite
 #   Linux: ~/.local/share/vaultik/index.sqlite

docs/DATAMODEL.md

@@ -1,297 +1,418 @@
-# Vaultik Data Model
+# vaultik data model
 
-## Overview
+`vaultik` keeps a SQLite database on the backed-up machine, called the local
+index. This document describes its tables and how each command uses them.
 
-Vaultik uses a local SQLite database to track file metadata, chunk mappings, and blob associations during the backup process. This database serves as an index for incremental backups and enables efficient deduplication.
+The local index has two jobs:
 
-**Important Notes:**
+- It records which files, chunks and blobs are already stored. The next backup
+  then does not store them again.
+- It is the source of each snapshot's own database. A backup uploads a trimmed,
+  encrypted copy of the index with every snapshot.
 
-- **Version Compatibility**: In rare cases, you may need to use the same version
-  of Vaultik to restore a backup as was used to create it. This ensures
-  compatibility with the metadata format stored in S3.
+A restore does not need the local index. It reads the copy that was uploaded
+with the snapshot.
 
-## Schema Migrations
+## Terms
 
-Vaultik supports schema migrations. They are the numbered SQL files in
-`internal/database/schema/`, embedded in the binary: `000.sql` creates the
-`schema_migrations` table, which records each migration that has run, and
-`001.sql` creates the application tables. `database.New` opens a database and
-applies, in order, every migration that database has not yet recorded.
+- **Snapshot**: one backup of the paths listed under one name in the config.
+- **Chunk**: a piece of a file's content, named by the SHA-256 hash of that
+  content.
+- **Blob**: one object at the destination. It holds many chunks, compressed and
+  then encrypted.
+- **Destination**: the place the backups are stored.
+- **Manifest**: the unencrypted list of the blobs one snapshot uses.
+- **Remote key**: a hash of the snapshot ID. It names the snapshot's directory
+  at the destination.
 
-**Before 1.0** no migrations are added, because nothing is installed anywhere
-yet. A schema change edits `001.sql` (and the code that uses the affected
-tables) directly. A local database created before the change has already
-recorded `001.sql` as run, so it keeps the old schema and can become unusable;
-`vaultik database delete` followed by a full backup rebuilds it.
-
-**After 1.0** each schema change is a new numbered file, so an existing local
-database is migrated the first time an updated vaultik opens it. A file that
-has shipped in a release is never edited.
-
-## Database Tables
-
-### 1. `files`
-Stores metadata about files in the filesystem being backed up.
-
-**Columns:**
-- `id` (TEXT PRIMARY KEY) - UUID for the file record
-- `path` (TEXT NOT NULL UNIQUE) - Absolute file path
-- `mtime` (INTEGER NOT NULL) - Modification time, whole seconds since the Unix epoch
-- `mtime_nsec` (INTEGER NOT NULL) - Nanoseconds within that second, 0 to 999999999
-- `size` (INTEGER NOT NULL) - File size in bytes
-- `mode` (INTEGER NOT NULL) - Unix file permissions and type
-- `uid` (INTEGER NOT NULL) - User ID of file owner
-- `gid` (INTEGER NOT NULL) - Group ID of file owner
-- `link_target` (TEXT) - Symlink target path (NULL for regular files)
-
-**Indexes:**
-- `idx_files_path` on `path` for efficient lookups
-
-**Purpose:** Tracks file metadata to detect changes between backup runs. Used for incremental backup decisions. The UUID primary key provides stable references that don't change if files are moved.
-
-### 2. `chunks`
-Stores information about content-defined chunks created from files.
-
-**Columns:**
-- `chunk_hash` (TEXT PRIMARY KEY) - SHA256 hash of chunk content
-- `size` (INTEGER NOT NULL) - Chunk size in bytes
+## Where the index is
 
-**Purpose:** Enables deduplication by tracking unique chunks across all files.
-
-### 3. `file_chunks`
-Maps files to their constituent chunks in order.
-
-**Columns:**
-- `file_id` (TEXT) - File ID (FK to files.id)
-- `idx` (INTEGER) - Chunk index within file (0-based)
-- `chunk_hash` (TEXT) - Chunk hash (FK to chunks.chunk_hash)
-- PRIMARY KEY (`file_id`, `idx`)
+The index is the file `index.sqlite` in the platform data directory. That is
+`~/.local/share/vaultik/` on Linux and `~/Library/Application Support/vaultik/`
+on macOS. The `index_path` setting and the `VAULTIK_INDEX_PATH` environment
+variable change the path.
 
-**Purpose:** Allows reconstruction of files from chunks during restore.
+The file is readable only by its owner, because it lists every backed-up path.
+`vaultik` sets that mode each time it opens the index.
 
-### 4. `chunk_files`
-Reverse mapping showing which files contain each chunk.
+## Schema migrations
 
-**Columns:**
-- `chunk_hash` (TEXT) - Chunk hash (FK to chunks.chunk_hash)
-- `file_id` (TEXT) - File ID (FK to files.id)
-- `file_offset` (INTEGER) - Byte offset of chunk within file
-- `length` (INTEGER) - Length of chunk in bytes
-- PRIMARY KEY (`chunk_hash`, `file_id`)
+`vaultik` supports schema migrations. They are the numbered SQL files in
+`internal/database/schema/`, which are built into the binary.
 
-**Purpose:** Supports efficient queries for chunk usage and deduplication statistics.
+- `000.sql` creates the `schema_migrations` table. That table records each
+  migration that has run.
+- `001.sql` creates every other table.
 
-### 5. `blobs`
-Stores information about packed, compressed, and encrypted blob files.
+`database.New` opens the index. It then applies, in order, every migration that
+the index has not recorded.
 
-**Columns:**
-- `id` (TEXT PRIMARY KEY) - UUID assigned when blob creation starts
-- `blob_hash` (TEXT UNIQUE) - `hex(SHA256(SHA256(uncompressed blob contents)))`, computed before compression and encryption (NULL until finalized); see [REPOSTRUCTURE.md](REPOSTRUCTURE.md#blobs-directory-blobs)
-- `created_ts` (INTEGER NOT NULL) - Creation timestamp
-- `finished_ts` (INTEGER) - Finalization timestamp (NULL if in progress)
-- `uncompressed_size` (INTEGER NOT NULL DEFAULT 0) - Total size of chunks before compression
-- `compressed_size` (INTEGER NOT NULL DEFAULT 0) - Size after compression and encryption. Once uploaded, the size of the object in storage: when the blob was already stored, under another compression level or other recipients, this can differ from the size of the local copy
-- `uploaded_ts` (INTEGER) - Upload completion timestamp (NULL if not uploaded)
+Before version 1.0, no migration is added, because nothing is installed anywhere
+yet. A schema change edits `001.sql` directly, with the code that uses the
+changed tables.
 
-**Purpose:** Tracks blob lifecycle from creation through upload. The UUID primary key allows immediate association of chunks with blobs.
+This has one consequence. An index created before such a change has already
+recorded `001.sql` as run, so it keeps the old schema and can become unusable.
+To rebuild it, run `vaultik database delete` and then a full backup.
 
-### 6. `blob_chunks`
-Maps chunks to the blobs that contain them.
-
-**Columns:**
-- `blob_id` (TEXT) - Blob ID (FK to blobs.id)
-- `chunk_hash` (TEXT) - Chunk hash (FK to chunks.chunk_hash)
-- `offset` (INTEGER) - Byte offset of chunk within blob (before compression)
-- `length` (INTEGER) - Length of chunk in bytes
-- PRIMARY KEY (`blob_id`, `chunk_hash`)
-
-**Purpose:** Enables chunk retrieval from blobs during restore operations.
-
-### 7. `snapshots`
-Tracks backup snapshots.
-
-**Columns:**
-- `id` (TEXT PRIMARY KEY) - Snapshot ID (format: `hostname_name_timestamp`, e.g. `server1_home_2025-06-01T12:00:00Z`: the hostname up to its first `.`, the snapshot name, and an RFC 3339 UTC timestamp)
-- `hostname` (TEXT) - Hostname where backup was created
-- `vaultik_version` (TEXT) - Version of Vaultik used
-- `vaultik_git_revision` (TEXT) - Git revision of Vaultik used
-- `started_at` (INTEGER) - Start timestamp
-- `completed_at` (INTEGER) - Completion timestamp (NULL if in progress)
-- `file_count` (INTEGER) - Number of files in snapshot
-- `chunk_count` (INTEGER) - Number of chunks this snapshot stored that were not stored before
-- `new_chunk_size` (INTEGER) - Total size of those chunks before compression; `snapshot list` shows it as NEW CHUNK SIZE
-- `blob_count` (INTEGER) - Number of blobs this snapshot created
-- `total_size` (INTEGER) - Total size of all files
-- `uncompressed_chunk_size` (INTEGER) - Total size before compression of the distinct chunks the snapshot's files reference, recorded when its backup ran; `snapshot list` shows it as UNCOMPRESSED SIZE
-- `blob_size` (INTEGER) - Total compressed size of all referenced blobs
-- `blob_uncompressed_size` (INTEGER) - Total uncompressed size of all referenced blobs
-- `compression_ratio` (REAL) - Compression ratio achieved
-- `compression_level` (INTEGER) - Compression level used for this snapshot
-- `upload_bytes` (INTEGER) - Total bytes uploaded during this snapshot
-- `upload_duration_ms` (INTEGER) - Total milliseconds spent uploading to S3
-
-**Purpose:** Provides snapshot metadata and statistics including version tracking for compatibility.
-
-### 8. `snapshot_files`
-Maps snapshots to the files they contain.
-
-**Columns:**
-- `snapshot_id` (TEXT) - Snapshot ID (FK to snapshots.id)
-- `file_id` (TEXT) - File ID (FK to files.id)
-- PRIMARY KEY (`snapshot_id`, `file_id`)
-
-**Purpose:** Records which files are included in each snapshot.
-
-### 9. `snapshot_blobs`
-Maps snapshots to the blobs they reference.
-
-**Columns:**
-- `snapshot_id` (TEXT) - Snapshot ID (FK to snapshots.id)
-- `blob_id` (TEXT) - Blob ID (FK to blobs.id)
-- `blob_hash` (TEXT) - Denormalized blob hash for manifest generation
-- PRIMARY KEY (`snapshot_id`, `blob_id`)
-
-**Purpose:** Tracks blob dependencies for snapshots and enables manifest generation.
-
-### 10. `uploads`
-Tracks blob upload metrics.
-
-**Columns:**
-- `blob_hash` (TEXT PRIMARY KEY) - Hash of uploaded blob
-- `snapshot_id` (TEXT NOT NULL) - The snapshot that triggered this upload (FK to snapshots.id)
-- `uploaded_at` (INTEGER) - Upload timestamp
-- `size` (INTEGER) - Size of uploaded blob
-- `duration_ms` (INTEGER) - Upload duration in milliseconds
-
-**Purpose:** Performance monitoring and tracking which blobs were newly created (uploaded) during each snapshot.
-
-## Data Flow and Operations
-
-### 1. Backup Process
-
-1. **File Scanning**
-   - `INSERT OR REPLACE INTO files` - Update file metadata
-   - `SELECT * FROM files WHERE path = ?` - Check if file has changed
-   - `INSERT INTO snapshot_files` - Add file to current snapshot
-
-2. **Chunking** (for changed files)
-   - `INSERT OR IGNORE INTO chunks` - Store new chunks
-   - `INSERT INTO file_chunks` - Map chunks to file
-   - `INSERT INTO chunk_files` - Create reverse mapping
-
-3. **Blob Packing**
-   - `INSERT INTO blobs` - Create blob record with UUID (blob_hash NULL)
-   - `INSERT INTO blob_chunks` - Associate chunks with blob immediately
-   - `UPDATE blobs SET blob_hash = ?, finished_ts = ?` - Finalize blob after packing
-
-4. **Upload**
-   - `UPDATE blobs SET uploaded_ts = ?, compressed_size = ?` - Mark blob as uploaded, with the size of the object in storage (the upload is skipped when the blob is already stored)
-   - `INSERT INTO uploads` - Record upload metrics with snapshot_id
-   - `INSERT INTO snapshot_blobs` - Associate blob with snapshot
-
-5. **Snapshot Completion**
-   - `UPDATE snapshots SET completed_at = ?, stats...` - Finalize snapshot
-   - Generate and upload blob manifest from `snapshot_blobs`
-
-### 2. Incremental Backup
-
-1. **Change Detection**
-   - `SELECT * FROM files WHERE path = ?` - Get previous file metadata
-   - Compare mtime, size, mode to detect changes
-   - Re-chunk a file that lists a chunk no uploaded blob holds, even when its metadata is unchanged
-   - Skip unchanged files but still add to `snapshot_files`
-
-2. **Chunk Reuse**
-   - `SELECT * FROM blob_chunks WHERE chunk_hash = ?` - Find existing chunks
-   - `INSERT INTO snapshot_blobs` - Reference existing blobs for unchanged files
-
-### 3. Snapshot Metadata Export
-
-After a snapshot is completed:
-1. Copy database to temporary file
-2. Clean temporary database to contain only current snapshot data
-3. VACUUM the trimmed database so deleted rows leave no pages behind
-4. Compress with zstd and encrypt with age
-5. Upload to S3 as `metadata/{remote-key}/db.zst.age`
-6. Generate blob manifest and upload as `metadata/{remote-key}/manifest.json.zst`
-
-The `{remote-key}` directory name is a one-way hash of the human snapshot ID, so the ID is never written to the store in plaintext. The hash uses no secret, so a guessed hostname and snapshot name can still be confirmed against a listing; see [REPOSTRUCTURE.md](REPOSTRUCTURE.md#remote-key-derivation) and its [Accepted Risks](REPOSTRUCTURE.md#accepted-risks).
-
-### 4. Restore Process
-
-The restore process doesn't use the local database. Instead:
-1. Downloads and decrypts the snapshot's metadata database (`db.zst.age`) from S3
-2. Downloads the blobs holding the chunks of the files being restored, found through that database's `blob_chunks` table; the manifest is not read
-3. Reconstructs files from decrypted and decompressed chunks
+After version 1.0, each schema change is a new numbered file. An existing index
+is then migrated the first time an updated `vaultik` opens it. A file that was
+part of a release is never edited.
 
-### 5. Pruning
+Migrations apply to the local index only. A snapshot's database at the
+destination is read as it was written. Each snapshot records the `vaultik`
+version that wrote it.
 
-1. **Identify Unreferenced Blobs**
-   - Query blobs not referenced by any remaining snapshot
-   - Delete from S3, once older than `prune_grace_period`, and from the local
-     database
-2. **Delete Metadata Directories With No Manifest**
-   - Delete each directory under `metadata/` that has no `manifest.json.zst`
-     and no local snapshot record, once every object in it is older than
-     `prune_grace_period`
+## Tables
 
-### 6. Incomplete Snapshot Cleanup
+Times are whole seconds since the Unix epoch, in UTC. The two exceptions are
+noted at their columns.
 
-Before each backup:
-1. Query incomplete snapshots (where `completed_at IS NULL`)
-2. Delete each one and all its associations, without checking S3 for its metadata
-3. Clean up orphaned files, chunks, and blobs
+### `files`
 
-## Repository Pattern
+One row for each backed-up path: a regular file, a symlink or a directory.
 
-Vaultik uses a repository pattern for database access:
-
-- `FileRepository` - CRUD operations for files and file metadata
-- `ChunkRepository` - CRUD operations for content chunks
-- `FileChunkRepository` - Manage file-to-chunk mappings
-- `ChunkFileRepository` - Manage chunk-to-file reverse mappings  
-- `BlobRepository` - Manage blob lifecycle (creation, finalization, upload)
-- `BlobChunkRepository` - Manage blob-to-chunk associations
-- `SnapshotRepository` - Manage snapshots and their relationships
-- `UploadRepository` - Track blob upload metrics
+- `id`: a UUID. Primary key.
+- `path`: the absolute path. Unique.
+- `source_path`: the configured path that the scan found this entry under. A
+  restore does not read it.
+- `mtime`: the modification time.
+- `mtime_nsec`: the nanoseconds within that second.
+- `size`: the size in bytes.
+- `mode`: the permission bits and the file type.
+- `uid`: the owner's user ID.
+- `gid`: the owner's group ID.
+- `link_target`: the target of a symlink. It is `NULL` for other entries.
 
-Each repository provides methods like:
-- `Create()` - Insert new record
-- `GetByID()` / `GetByPath()` / `GetByHash()` - Retrieve records
-- `Update()` - Update existing records
-- `Delete()` - Remove records
-- Specialized queries for each entity type (e.g., `DeleteOrphaned()`, `GetIncompleteByHostname()`)
+`idx_files_path` indexes `path`.
 
-## Transaction Management
+There is one row for each path, and it is updated when the file changes. The
+row is shared by every snapshot in the local index that lists the path. An
+older snapshot's own view of the file is in that snapshot's database at the
+destination.
 
-All database operations that modify multiple tables are wrapped in transactions:
+### `chunks`
+
+One row for each distinct chunk.
+
+- `chunk_hash`: the SHA-256 hash of the chunk's content, in hex. Primary key.
+- `size`: the size in bytes.
+
+### `file_chunks`
+
+The chunks that make up each file, in order.
+
+- `file_id`: the file. References `files.id`.
+- `idx`: the position of the chunk within the file, counted from 0.
+- `chunk_hash`: the chunk. References `chunks.chunk_hash`.
+
+The primary key is (`file_id`, `idx`). `idx_file_chunks_chunk_hash` indexes
+`chunk_hash`.
+
+### `chunk_files`
+
+The reverse of `file_chunks`: the files that contain each chunk.
+
+- `chunk_hash`: the chunk. References `chunks.chunk_hash`.
+- `file_id`: the file. References `files.id`.
+- `file_offset`: the byte offset of the chunk within the file.
+- `length`: the length of the chunk in bytes.
+
+The primary key is (`chunk_hash`, `file_id`). `idx_chunk_files_file_id` indexes
+`file_id`.
+
+### `blobs`
+
+One row for each blob.
+
+- `id`: a UUID, given when packing of the blob starts. Primary key.
+- `blob_hash`: the name of the blob at the destination. Unique. Until the blob
+  is closed, it holds a placeholder that starts with `temp-placeholder-`.
+- `created_ts`: when packing started.
+- `finished_ts`: when the blob was closed. It is `NULL` before that.
+- `uncompressed_size`: the total size of the blob's chunks.
+- `compressed_size`: the size after compression and encryption.
+- `uploaded_ts`: when the upload finished. It is `NULL` before that.
+
+`blob_hash` is `hex(SHA256(SHA256(uncompressed contents)))`. See
+[REPOSTRUCTURE.md](REPOSTRUCTURE.md#blobs).
+
+Once a blob is uploaded, `compressed_size` is the size of the object at the
+destination. A blob that was already there can have a different size from the
+local copy. That happens when it was written with another compression level or
+other recipients.
+
+### `blob_chunks`
+
+The chunks that each blob holds.
+
+- `blob_id`: the blob. References `blobs.id`.
+- `chunk_hash`: the chunk. References `chunks.chunk_hash`.
+- `offset`: the byte offset of the chunk within the uncompressed blob.
+- `length`: the length of the chunk in bytes.
+
+The primary key is (`blob_id`, `chunk_hash`). `idx_blob_chunks_chunk_hash`
+indexes `chunk_hash`.
+
+A restore finds the blob for each chunk through this table.
+
+### `snapshots`
+
+One row for each snapshot.
+
+- `id`: the snapshot ID. Primary key.
+- `hostname`: the hostname of the machine that made the backup.
+- `vaultik_version`: the `vaultik` version that made it.
+- `vaultik_git_revision`: the git commit of that version.
+- `started_at`: when the backup started.
+- `completed_at`: when the backup finished. It is `NULL` until then.
+- `file_count`: the number of files and symlinks in the snapshot. Directories
+  are not counted.
+- `chunk_count`: the number of chunks this snapshot stored that were not stored
+  before.
+- `new_chunk_size`: the total size of those chunks before compression.
+- `blob_count`: the number of blobs this snapshot created.
+- `total_size`: the total size of all files in the snapshot.
+- `uncompressed_chunk_size`: the total size, before compression, of the
+  distinct chunks the snapshot's files use.
+- `blob_size`: the total compressed size of the blobs the snapshot uses.
+- `blob_uncompressed_size`: the total uncompressed size of those blobs.
+- `compression_ratio`: the compression ratio of those blobs.
+- `compression_level`: the compression level the backup used.
+- `upload_bytes`: the bytes uploaded during the backup.
+- `upload_duration_ms`: the time spent uploading, in milliseconds.
+
+The snapshot ID has the form `<hostname>_<snapshot-name>_<timestamp>`. An
+example is `server1_home_2025-06-01T12:00:00Z`. It has three parts:
+
+- the hostname, up to its first `.`
+- the snapshot name
+- the start time, in UTC
+
+`snapshot list` shows `uncompressed_chunk_size` as `UNCOMPRESSED SIZE` and
+`new_chunk_size` as `NEW CHUNK SIZE`.
+
+A snapshot with `completed_at` set to `NULL` is unfinished. `snapshot list`
+does not show it. The next backup deletes it, and so does `prune`.
+
+### `snapshot_files`
+
+The files in each snapshot.
+
+- `snapshot_id`: the snapshot. References `snapshots.id`.
+- `file_id`: the file. References `files.id`.
+
+The primary key is (`snapshot_id`, `file_id`). `idx_snapshot_files_file_id`
+indexes `file_id`.
+
+### `snapshot_blobs`
+
+The blobs that each snapshot uses.
+
+- `snapshot_id`: the snapshot. References `snapshots.id`.
+- `blob_id`: the blob. References `blobs.id`.
+- `blob_hash`: a copy of the blob's hash.
+
+The primary key is (`snapshot_id`, `blob_id`). `idx_snapshot_blobs_blob_id`
+indexes `blob_id`.
+
+The manifest is built from this table.
+
+### `snapshot_mount_points`
+
+The mount points that a snapshot's backup found.
+
+- `snapshot_id`: the snapshot. References `snapshots.id`.
+- `path`: the absolute path of the directory.
+
+The primary key is (`snapshot_id`, `path`).
+
+A directory is recorded here when all of these hold:
+
+- It is at or below a configured path.
+- The backup found an entry in it that is not excluded.
+- It is on a different filesystem from its parent directory.
+
+The device number is read after the directory has been listed. Listing an
+automount point mounts its volume, so the number is that of the volume. An
+automount point whose volume cannot be mounted is not recorded, because the
+backup finds no entry in it.
+
+`snapshot create --prune` reads this table. It skips the prune when an earlier
+snapshot of the same name recorded a mount point that the new backup did not
+find. That happens when the volume is not mounted.
+
+### `uploads`
+
+One row for each blob that a backup uploaded.
+
+- `blob_hash`: the blob. References `blobs.blob_hash`. Primary key.
+- `snapshot_id`: the snapshot whose backup uploaded it. References
+  `snapshots.id`.
+- `uploaded_at`: when the upload started. It is written as a date and time, not
+  as seconds.
+- `size`: the size of the uploaded blob.
+- `duration_ms`: the duration of the upload, in milliseconds.
+
+`idx_uploads_snapshot_id` indexes `snapshot_id`.
+
+A blob that was already at the destination gets no row here.
+
+### `local_meta`
+
+Settings that belong to this copy of the index.
+
+- `key`: the name of the setting. Primary key.
+- `value`: its value.
+
+The only key is `storage_url`. It holds the destination that the index was
+first used with. A command that changes data refuses to run when the config
+names a different destination.
+
+### `schema_migrations`
+
+The migrations that have run.
+
+- `version`: the number of the migration. Primary key.
+- `applied_at`: when it ran. It is written as a date and time, not as seconds.
+
+## How commands use the tables
+
+### Backup
+
+A backup works on the tables in this order:
+
+- It deletes unfinished snapshots, and the rows that no snapshot uses.
+- It creates a row in `snapshots`, with `completed_at` set to `NULL`.
+- It deletes the `blobs` rows that were never uploaded.
+- It walks each path and compares every entry with its row in `files`.
+- It links each unchanged file to the snapshot in `snapshot_files`.
+- It records each mount point in `snapshot_mount_points`.
+- It reads each changed file and stores it. See below.
+- It fills `snapshot_blobs` with every blob the snapshot uses.
+- It stores the snapshot's totals in `snapshots`.
+- It uploads the snapshot's metadata.
+- It sets `completed_at`.
+
+A file counts as changed when its row in `files` differs from the file on disk
+in any of these:
+
+- size
+- modification time
+- mode
+- owner or group
+- link target
+
+A file is also read again when it lists a chunk that no uploaded blob holds.
+
+Storing a changed file writes these rows:
+
+- A row in `blobs`, when packing of a blob starts.
+- When the blob is closed, in one transaction:
+    - a row in `chunks` for each new chunk in the blob
+    - the blob's rows in `blob_chunks`
+    - the blob's `blob_hash`, sizes and `finished_ts`
+- When the blob is stored at the destination, in one transaction:
+    - `uploaded_ts` and `compressed_size` in `blobs`
+    - a row in `snapshot_blobs`
+    - a row in `uploads`, unless the blob was already at the destination
+- The row in `files`, with its `file_chunks`, `chunk_files` and
+  `snapshot_files` rows.
+
+The row in `files` is written last. It is written only after every blob that
+holds the file's new chunks is stored.
+
+### Snapshot metadata
+
+The metadata of a snapshot is made from a copy of the index:
+
+- The index is copied into a temporary directory with `VACUUM INTO`.
+- The rows of every other snapshot are deleted from the copy. So are the files
+  and blobs that this snapshot does not use. A chunk is then deleted when no
+  remaining file and no remaining blob uses it.
+- The `local_meta` and `schema_migrations` tables are kept as they are. The
+  uploaded database therefore names the destination, and lists the migrations
+  that have run.
+- `VACUUM` rewrites the copy, so that no deleted row is left in the file.
+- The copy is compressed with zstd and encrypted with age.
+- The manifest is built from the copy's `snapshot_blobs` table.
+- The copy is uploaded as `metadata/<remote-key>/db.zst.age`.
+- The manifest is uploaded as `metadata/<remote-key>/manifest.json.zst`.
+
+The copy is made through SQLite, not by copying the file. The index runs in WAL
+mode, so rows that are already committed can still be in the `-wal` file beside
+it.
+
+### Restore
+
+A restore does not use the local index.
+
+- It downloads and decrypts `db.zst.age` into a temporary directory.
+- It opens that file read-only. It applies no migration to it.
+- It reads the files to restore from `files`.
+- It finds each file's chunks in `file_chunks`.
+- It finds the blob and offset of each chunk in `blob_chunks`.
+- It finds the name of each blob in `blobs`.
+
+It does not read the manifest.
+
+### Prune
+
+`prune` changes the local index in two ways:
+
+- It deletes each snapshot whose manifest is gone from the destination.
+- It deletes unfinished snapshots, and the rows that no snapshot uses.
+
+The second step removes unused rows in this order:
+
+- files that are in no snapshot
+- blobs that are in no snapshot
+- `blob_chunks` rows whose chunk is gone
+- chunks that no file and no blob uses
+
+`prune` then deletes unused data at the destination. That step reads the
+manifests, not the local index. It reads the local index only to keep a
+directory under `metadata/` whose snapshot is still recorded there.
+
+## Code
+
+Each repository type in `internal/database` serves one or more tables:
+
+- `FileRepository`: `files`
+- `ChunkRepository`: `chunks`
+- `FileChunkRepository`: `file_chunks`
+- `ChunkFileRepository`: `chunk_files`
+- `BlobRepository`: `blobs`
+- `BlobChunkRepository`: `blob_chunks`
+- `SnapshotRepository`:
+    - `snapshots`
+    - `snapshot_files`
+    - `snapshot_blobs`
+    - `snapshot_mount_points`
+- `UploadRepository`: `uploads`
+- `LocalMetaRepository`: `local_meta`
+
+No repository type serves `schema_migrations`. The code that applies the
+migrations reads and writes it.
+
+`Repositories` holds one of each type.
+
+`Repositories.WithTx` runs a function in one transaction:
 
 ```go
 err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
-    // Multiple repository operations using tx
+    // several repository calls that use tx
 })
 ```
 
-This ensures consistency, especially important for operations like:
-- Creating file-chunk mappings
-- Associating chunks with blobs
-- Updating snapshot statistics
+The transaction commits when the function returns no error. Otherwise it is
+rolled back.
 
-## Performance Considerations
+## SQLite settings
 
-1. **Indexes**: 
-   - Primary keys are automatically indexed
-   - `idx_files_path` on `files(path)` for efficient file lookups
-
-2. **Prepared Statements**: All queries use prepared statements for performance and security
-
-3. **Batch Operations**: Where possible, operations are batched within transactions
-
-4. **Write-Ahead Logging**: The local index runs in SQLite WAL mode with a 10-second busy timeout, so a read-only command such as `snapshot list` can read it while a backup writes to it. Committed rows can sit in the `-wal` file beside the index until a checkpoint, so the metadata export copies the index through SQLite (`VACUUM INTO`), not as a file
-
-## Data Integrity
-
-1. **Foreign Keys**: Enforced through CASCADE DELETE and application-level repository methods
-2. **Unique Constraints**: Chunk hashes, file paths, and blob hashes are unique
-3. **Null Handling**: Nullable fields clearly indicate in-progress operations
-4. **Timestamp Tracking**: All major operations record timestamps for auditing
+- The index runs in WAL mode. A command that only reads, such as
+  `snapshot list`, can read the index while a backup writes to it.
+- A connection waits up to 10 seconds for a lock before it fails.
+- Foreign keys are enforced.
+- Each open database uses one connection, so its queries run one at a time.
+- Queries pass their values as parameters.

docs/REPOSTRUCTURE.md

@@ -1,184 +1,416 @@
-# Vaultik S3 Repository Structure
+# vaultik destination layout
 
-This document describes the structure and organization of data stored in the S3 bucket by Vaultik.
+This document describes what `vaultik` stores at the destination. It covers the
+layout, what the layout protects, and the risks it accepts.
 
-## Overview
+The destination is the place the backups are stored. It is an S3 bucket and
+prefix, a local directory or an rclone remote. The layout is the same on all
+three.
 
-Vaultik stores all backup data in an S3-compatible object store. The repository consists of two main components, plus a marker file:
-1. **Blobs** - The actual backup data (content-addressed, encrypted)
-2. **Metadata** - Snapshot information and manifests (partially encrypted)
+## Terms
 
-## Directory Structure
+- **Snapshot**: one backup of the paths listed under one name in the config.
+- **Snapshot ID**: the name of a snapshot, in the form
+  `<hostname>_<snapshot-name>_<timestamp>`.
+- **Chunk**: a piece of a file's content, named by the SHA-256 hash of that
+  content.
+- **Blob**: one object under `blobs/`. It holds many chunks.
+- **Local index**: the SQLite database on the backed-up machine.
+- **Recipient**: an age public key that backups are encrypted to.
+
+## Layout
 
 ```
-<bucket>/<prefix>/
+<destination>/
 ├── vaultik-destination.txt
 ├── blobs/
 │   └── <hash[0:2]>/
 │       └── <hash[2:4]>/
-│           └── <full-hash>
+│           └── <hash>
 └── metadata/
     └── <remote-key>/
         ├── db.zst.age
         └── manifest.json.zst
 ```
 
-The metadata subdirectory is named with the **remote key**, a one-way hash of the snapshot ID, not with the human-readable snapshot ID itself. See [Remote Key Derivation](#remote-key-derivation).
+The destination holds three things:
 
-## Marker File (`vaultik-destination.txt`)
+- `vaultik-destination.txt`, the marker file
+- `blobs/`, which holds the file contents
+- `metadata/`, which holds one directory for each snapshot
 
-Every backup writes this file at the root before it uploads anything. It holds one fixed line of plain text and names no host, path or time. While the local index holds snapshots stored there, a destination that answers a look-up of the marker with not found is treated as not present: `snapshot create` refuses to write to it, and `prune`, `snapshot purge`, `snapshot remove`, `snapshot list` and `remote nuke` behave as for a destination that cannot be listed. An unplugged volume leaves an empty directory at its mount point, and without the marker that directory would read as a destination whose snapshots are all gone. `remote nuke` deletes the marker last.
+## The marker file
 
-When the look-up of the marker fails for any reason other than not found, the check is skipped and the command goes on. S3 refuses it on every run to credentials that allow only PUT, so a host with such credentials does not notice a destination emptied since its last backup, and its next backup would leave out the data its local index records as already stored. After the destination is emptied, run `vaultik database delete` on such a host before it backs up again.
+Every backup writes `vaultik-destination.txt` before it uploads anything else.
+The file holds one fixed line of text. It names no host, path or time.
 
-## Blobs Directory (`blobs/`)
+The marker tells an emptied or missing destination apart from a new one. An
+unplugged volume leaves an empty directory at its mount point. Without the
+marker, that directory would look like a destination whose snapshots are all
+gone.
 
-### Structure
-- **Path format**: `blobs/<first-2-chars>/<next-2-chars>/<full-hash>`
-- **Example**: `blobs/ca/fe/cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678`
-- **Sharding**: The two-level directory structure (using the first 4 characters of the hash) prevents any single directory from containing too many objects
+The check applies while the local index holds snapshots. When the destination
+then answers that the marker does not exist, commands treat the destination as
+not present:
 
-### Content
-- **What it contains**: Packed collections of content-defined chunks from files
-- **Format**: Zstandard compressed, then Age encrypted
-- **Encryption**: Always encrypted with Age using the configured recipients
-- **Naming**: Content-addressed. The blob's name is `hex(SHA256(SHA256(uncompressed blob contents)))` — the double SHA-256 of the concatenated chunk data, computed before compression and encryption, not a hash of the stored (compressed, encrypted) bytes. One consequence: only a holder of the age private key can check a stored blob's integrity, because matching a blob to its name means decrypting and decompressing it first — which is what `restore` and `verify --deep` do. Implemented in `internal/blobgen` (`DoubleSHA256`). This is the canonical description of blob naming; other documents and comments point here.
+- `snapshot create` refuses to write to it.
+- `prune`, `snapshot purge` and `remote nuke` fail.
+- `snapshot list` shows the local index alone.
+- `snapshot remove` deletes nothing at the destination.
 
-### Why Encrypted
-Blobs contain the actual file data from backups and must be encrypted for security. The content-addressing ensures deduplication while the encryption ensures privacy.
+The check is skipped when the look-up fails for any other reason. S3 refuses
+the look-up to credentials that can only write. A machine with such credentials
+therefore does not notice an emptied destination. Its next backup would not
+upload data that its local index says is stored. After a destination is emptied,
+run `vaultik database delete` on such a machine before it backs up again.
 
-## Metadata Directory (`metadata/`)
+`remote nuke` deletes the marker last.
 
-Each snapshot has its own subdirectory. The directory is **not** named with the human-readable snapshot ID; it is named with the remote key — a one-way hash of that ID. The human ID is never written to the destination store as a directory name (see [Remote Key Derivation](#remote-key-derivation)).
+## Blobs
 
-### Snapshot ID Format
+A blob holds the content of many chunks, one after another.
 
-The human-readable snapshot ID is used in CLI arguments, log lines, and the local database. It is not written to the destination store.
-- **Format**: `<hostname>_<snapshot-name>_<RFC3339>` (or `<hostname>_<RFC3339>` if no
-  name was specified)
-- **Example**: `laptop_home_2024-01-15T14:30:52Z`
-- **Components**:
-  - Short hostname (everything before the first dot is stripped from the FQDN)
-  - Snapshot name from the configured `snapshots:` map (optional)
-  - RFC3339 UTC timestamp
+- **Path**: `blobs/<first 2 characters>/<next 2 characters>/<full hash>`.
+- **Example**:
+  `blobs/ca/fe/cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678`.
+- **Format**: the chunk data, compressed with zstd and then encrypted with age.
+- **Recipients**: the public keys in `age_recipients` when the blob was written.
 
-This ID reveals the hostname, the configured snapshot name, and the backup time, so it is never used as the on-disk directory name — the remote key is used instead.
+The two directory levels keep any one directory from holding too many objects.
 
-### Remote Key Derivation
+### Blob names
 
-The remote key is `hex(SHA256(SHA256("vaultik|" + snapshot-id)))`: a double SHA-256 over the snapshot ID, with a `vaultik|` domain-separation prefix. The result is a 64-character hex string. The hash is not reversible, but it uses no secret: an observer who guesses a candidate hostname and snapshot name can hash it the same way and confirm whether that snapshot is present. The remote key keeps names out of a plain listing; it does not hide them from a guess. Implemented in `internal/snapshot/remotekey.go`.
+A blob's name is `hex(SHA256(SHA256(uncompressed contents)))`. The contents are
+the chunk data, joined in order, before compression and encryption. The name is
+not a hash of the stored bytes.
+
+This has two consequences:
+
+- Checking a blob against its name needs the private key. The blob must be
+  decrypted and decompressed first. `snapshot restore` and
+  `snapshot verify --deep` do this.
+- Without the key, only a blob's presence and size can be checked.
+  `snapshot verify` does this.
+
+`blobgen.DoubleSHA256` in `internal/blobgen` computes the name.
+
+### Uploads
+
+A backup looks a blob up by name before it uploads it. It does not upload a
+blob that is already there.
+
+When the destination refuses the look-up, the backup uploads the blob. An
+object with that name is then replaced by one that holds the same chunk data.
+
+## Metadata
+
+Each snapshot has one directory under `metadata/`. The directory is named with
+the snapshot's remote key, not with its snapshot ID.
+
+### Snapshot IDs
+
+A snapshot ID has the form `<hostname>_<snapshot-name>_<timestamp>`. An example
+is `laptop_home_2024-01-15T14:30:52Z`. It has three parts:
+
+- the hostname, up to its first `.`
+- the snapshot name from the config
+- the time the backup started, in UTC
+
+The snapshot ID is used on the command line, in log records and in the local
+index. It reveals the hostname, the snapshot name and the backup time. It is
+therefore never written to the destination without encryption.
+
+### Remote keys
+
+The remote key is `hex(SHA256(SHA256("vaultik|" + snapshot ID)))`. It is 64 hex
+characters long. `internal/snapshot/remotekey.go` computes it.
+
+For example:
 
-Worked example:
 - Snapshot ID: `server1_home_2025-06-01T12:00:00Z`
 - Remote key: `17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa`
-- Directory: `metadata/17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa/`
+- Directory:
+  `metadata/17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa/`
 
-A plain listing of the destination store therefore shows only these hashes, not the hostname or snapshot name of any backup — but because the hash uses no secret, a guessed hostname and snapshot name can be hashed and confirmed against the listing. The same remote key is stored in the manifest's `snapshot_id` field.
+A listing of the destination therefore shows no hostname and no snapshot name.
 
-### Files in Each Snapshot Directory
+The hash cannot be reversed, but it uses no secret. Someone who guesses a
+snapshot ID can hash it the same way and see whether that snapshot exists. The
+remote key keeps names out of a listing. It does not hide them from a guess.
 
-#### `db.zst.age` - Encrypted Database
-- **What it contains**: Pruned binary SQLite database for this snapshot
-- **Format**: Binary SQLite → Zstandard compressed → Age encrypted
-- **Encryption**: Encrypted with Age
-- **Purpose**: Contains full file metadata, chunk mappings, and all relationships
-- **Why encrypted**: Contains sensitive metadata like file paths, permissions, and ownership
+### `db.zst.age`
+
+The snapshot's database. A restore needs it.
+
+- **Contents**: a SQLite database. It is a copy of the local index without the
+  rows of other snapshots. It lists every file in the snapshot with its
+  metadata, its chunks, and the blob that holds each chunk.
+- **Format**: the database file, compressed with zstd and then encrypted with
+  age.
+
+It is encrypted because it holds file paths, permissions and owners.
+[DATAMODEL.md](DATAMODEL.md) describes its tables.
+
+### `manifest.json.zst`
+
+The list of the blobs the snapshot uses.
+
+- **Format**: JSON, compressed with zstd. It is not encrypted.
 
-#### `manifest.json.zst` - Unencrypted Blob Manifest
-- **What it contains**: JSON list of all blob hashes referenced by this snapshot
-- **Format**: JSON → Zstandard compressed (NOT encrypted)
-- **Encryption**: NOT encrypted
-- **Purpose**: Enables pruning operations without requiring decryption keys
-- **Structure**:
 ```json
 {
-  "snapshot_id": "17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa",
-  "timestamp": "2025-06-01T12:00:00Z",
-  "blob_count": 42,
-  "total_compressed_size": 1048576,
-  "blobs": [
-    { "hash": "cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678", "compressed_size": 24576 },
-    { "hash": "deadbeef1234567890abcdef1234567890abcdef1234567890abcdef12345678", "compressed_size": 32768 }
-  ]
+    "snapshot_id": "17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa",
+    "timestamp": "2025-06-01T12:04:31Z",
+    "blob_count": 2,
+    "total_compressed_size": 57344,
+    "blobs": [
+        {
+            "hash": "cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678",
+            "compressed_size": 24576
+        },
+        {
+            "hash": "deadbeef1234567890abcdef1234567890abcdef1234567890abcdef12345678",
+            "compressed_size": 32768
+        }
+    ]
 }
 ```
-`snapshot_id` is the remote key (a hash), not the human ID; `timestamp` is written in the clear.
 
-### Why Manifest is Unencrypted
-The manifest must be readable without the private key to enable:
-1. **Pruning operations** - Identifying unreferenced blobs for deletion
-2. **Storage analysis** - Understanding space usage without decryption
-3. **Verification** - Checking blob existence without decryption
-4. **Cross-snapshot deduplication analysis** - Finding shared blobs between snapshots
+- `snapshot_id`: the remote key, not the snapshot ID.
+- `timestamp`: when the manifest was written, at the end of the backup.
+- `blob_count`: the number of blobs.
+- `total_compressed_size`: the total size of those blobs, in bytes.
+- `blobs`: the hash and the size of each blob.
 
-The manifest contains the remote key, the backup timestamp, the blob count and total compressed size, and each blob's hash and compressed size. It contains no file names, paths, or other decrypted metadata.
+The manifest holds no file name and no path.
 
-## Security Considerations
+The manifest is not encrypted so that these commands work without the private
+key:
 
-### What's Encrypted
-- **All file content** (in blobs)
-- **All file metadata** (paths, permissions, timestamps, ownership in db.zst.age)
-- **File-to-chunk mappings** (in db.zst.age)
+- `prune`, which must know which blobs are still in use
+- `snapshot verify`, which checks that each blob is present
+- `snapshot list` and `remote info`, which report what the destination holds
 
-### What's Not Encrypted
-- **The remote key** — directory names and the manifest `snapshot_id`, a one-way hash of the snapshot ID (see [Remote Key Derivation](#remote-key-derivation))
-- **The backup timestamp** (in manifest.json.zst)
-- **Blob hashes and their compressed sizes** (in manifest.json.zst)
-- **Blob count and total compressed size per snapshot** (in manifest.json.zst)
+## What is protected
 
-### Privacy Implications
-From the unencrypted data, an observer of the destination store can determine:
-- **When each backup was taken** — not from the directory name, which is a one-way hash, but from the plaintext `timestamp` field in manifest.json.zst, which is published in the clear
-- How many blobs each snapshot references, and the total compressed size
-- The compressed size of each blob, and which blobs are shared between snapshots (deduplication patterns)
-- **Whether a guessed hostname and snapshot name are present** — the remote key is an unkeyed hash, so an observer holding candidate names can hash each one and match it against the directory listing. The human ID is never published, so it cannot be read off directly, but it can be confirmed by guessing.
+These are encrypted:
 
-Together these give an observer a timing-and-size profile of every snapshot. This is an accepted, documented property of the format, not a defect: the manifest is unencrypted so that pruning can run without the private key, and the timing channel could not be closed by encrypting it anyway — object creation times and per-object sizes stay visible at the storage layer on both `s3://` and `file://` destinations regardless.
+- file contents, in the blobs
+- in `db.zst.age`:
+    - file names and paths
+    - permissions and owners
+    - modification times
+    - which chunks make up which file
 
-An observer cannot determine:
-- The hostname or snapshot name of any backup by reading it off the store — the directory name and the manifest `snapshot_id` are unkeyed hashes of the human ID, so the text is never published (though a guessed name can be confirmed, as above)
-- File names or paths
-- File contents
-- File permissions or ownership
-- Directory structure
-- Which chunks belong to which files
+These are not encrypted:
 
-### Accepted Risks
+- the remote key of each snapshot
+- the time each backup finished, in the manifest
+- the hash and the compressed size of each blob, in the manifest
+- the number of blobs and their total size for each snapshot, in the manifest
 
-These are known, deliberate properties of the format and the tooling, recorded so an operator can weigh them rather than discover them.
+Someone who can read the destination therefore learns these:
 
-1. **No proof of authorship.** Restore and `verify --deep` prove that data decrypts with the age private key and matches its unkeyed content hashes. They do not prove who wrote it: anyone who knows a recipient public key and can replace objects on the destination can substitute a snapshot they built. The recipient string is not stored at the destination, but a compromised backed-up host has it. Defences live on the destination side — bucket versioning or object lock, credentials for the source host that cannot delete or overwrite existing versions, and pruning from a trusted host. Note that S3 `PutObject` overwrites an existing key, so PUT permission alone is not append-only.
-2. **Compression reveals sizes.** Blobs and `db.zst.age` are zstd-compressed then age-encrypted; the manifest is compressed only. age does not pad, so an object's size is the exact compressed length of its contents. All new chunks packed into one blob share a single zstd stream (8 MiB window, 4 MiB at compression levels 1-2), and a blob is closed at `blob_size_limit` and at the end of each configured path. Because a stored chunk is never packed again, someone who can write into a backed-up file and watch blob sizes learns something only when their controlled data and a secret land in the same chunk of a file that keeps changing. Advice: back up any outsider-writable directory as its own snapshot.
-3. **Chunking uses no secret.** The FastCDC parameters are fixed and public. The default 10 MB average yields chunks between 2.5 MB and 40 MB, and any file of 2.5 MB or less is a single chunk. At the default 10 GB `blob_size_limit` a blob holds hundreds of chunks, so individual chunk lengths are not visible in the blob's size; lowering the limit toward the chunk size begins to expose them.
-4. **Decrypted data on local disk.** Several commands stage plaintext under `$TMPDIR`: `snapshot restore` writes decrypted blobs under `vaultik-blobcache-*/` (no size cap) and the decrypted metadata database at `vaultik-restore-*/snapshot.db`; `verify --deep` writes that database at `vaultik-verify-*/snapshot.db`; `snapshot create` keeps a plaintext copy of the index at `vaultik-snapshot-*/snapshot.db`. These files are created `0600` and removed on success, but a `kill -9` or a power loss leaves them behind — delete any leftover `vaultik-*` directory under `$TMPDIR` by hand. `$TMPDIR` should be trusted to the same degree as the restore target.
-5. **Store permissions differ per command.** The backed-up host needs only PUT to run `snapshot create`: it writes blobs and metadata and neither reads nor deletes them. Other commands need more — `snapshot verify`, `snapshot restore`, and `prune` list and read; `prune`, `snapshot purge`, `snapshot remove`, and `remote nuke` also delete. The recommended cron line uses `--prune`, which runs `prune` on the backed-up host, so granting that host `--prune` gives it credentials that can delete its own backups. To keep the source host to PUT only, prune from a separate trusted host instead.
-6. **Changing recipients does not re-encrypt existing data.** Deduplicated chunks and same-named blobs already on the destination stay encrypted to the recipients in force when they were written. A new snapshot that reuses them cannot be restored with a newly added recipient's key alone, because those reused objects were never encrypted to it. To make everything readable by a new key, run `vaultik database delete` and take a full backup to a fresh destination or prefix.
-7. **X25519 recipients only.** vaultik rejects age ssh and plugin recipients. Long-lived ciphertext held by a third party (the destination operator) has no fallback if X25519 is ever broken: there is no second recipient type and no post-quantum option.
+- when each backup finished
+- how many blobs each snapshot uses, and how large they are
+- which blobs two snapshots share
+- whether a snapshot ID they guess exists
 
-## Consistency Guarantees
+They do not learn these:
 
-1. **Blobs are immutable** - Once written, a blob is never modified
-2. **Blobs are written before metadata** - A snapshot's metadata is only written after all its blobs are successfully uploaded
-3. **Metadata is written atomically** - Both db.zst.age and manifest.json.zst are written as complete files
-4. **A snapshot is marked complete in the local DB only after its metadata is uploaded** - `finalizeSnapshotMetadata` runs `ExportSnapshotMetadata` first and records completion (`MarkSnapshotComplete`) only once the export succeeds (see the backup data flow in [ARCHITECTURE.md](../ARCHITECTURE.md)). A crash during the export therefore leaves the snapshot incomplete, so the next backup's `PruneDatabase` drops it and re-backs-up its data, rather than leaving a completed-looking row in the local index with no matching metadata on the destination store. (A crash in the brief moment after the export succeeds but before completion is recorded leaves a fully-restorable snapshot on the destination that the local index drops as incomplete on the next run; `snapshot list` then reports it honestly as remote-only, which is the safe direction: the destination copy stays restorable.)
+- any hostname or snapshot name, unless they guess it
+- file names or paths
+- file contents
+- permissions or owners
+- the directory structure
+- which chunks belong to which file
 
-## Pruning Safety
+The times and sizes could not be hidden by encrypting the manifest. The
+destination itself shows when each object was written and how large it is.
 
-The prune operation is safe because:
-1. It keeps every blob listed in any snapshot's manifest and deletes only blobs that no manifest references. Under `metadata/` it deletes only a directory that has no `manifest.json.zst`, whose name is a remote key, and that no local snapshot record claims
-2. Manifests are unencrypted and can be read without keys
-3. If any manifest cannot be downloaded or decoded, prune deletes nothing and exits with an error, rather than treating that snapshot's blobs as unreferenced
-4. It deletes an unreferenced blob only once the blob is older than `prune_grace_period` (default `24h`), and a directory under `metadata/` with no manifest only once every object in it is. A backup uploads its blobs and its `db.zst.age` before its manifest, so a prune running meanwhile, on this host or another sharing the destination, leaves them as long as no backup run takes longer than the grace period. The grace period does not protect a blob a backup reuses from an earlier run that stopped before its manifest: the backup skips the upload of a blob already on the destination, so that blob keeps the time of its first upload, and a prune before the backup's manifest is written can delete it; the new snapshot then cannot be restored. `remote nuke` has no such check
+## Accepted risks
 
-## Restoration Requirements
+These are known properties of the format. They are written down so that you can
+weigh them.
 
-To restore from a backup, you need:
-1. **The Age private key** - To decrypt blobs and database
-2. **The snapshot metadata** - Both files from the snapshot's metadata directory
-3. **All referenced blobs** - As listed in the manifest
+### No proof of who wrote the data
 
-The restoration process:
-1. Download and decrypt the database dump to understand file structure
-2. Download and decrypt the required blobs
-3. Reconstruct files from their chunks
-4. Restore file metadata (permissions, timestamps, etc.)
+A restore and `snapshot verify --deep` prove two things. The data decrypts with
+your private key, and it matches its hashes. They do not prove who wrote it.
+
+The hashes use no secret. Someone who knows a recipient's public key and can
+replace objects at the destination can put a snapshot of their own there. The
+public key is not stored at the destination, but a backed-up machine that is
+broken into has it.
+
+The defences are on the destination's side:
+
+- object versioning or object lock on the bucket
+- credentials for the backed-up machine that cannot delete or overwrite
+- running `prune` from a trusted machine
+
+An S3 `PutObject` replaces an existing object. Permission to write alone
+therefore does not stop overwrites.
+
+### Sizes are visible
+
+An object's size follows exactly from the size of its compressed contents.
+
+All the new chunks in one blob share one zstd stream. The stream looks back 8
+MiB, or 4 MiB at compression levels 1 and 2. A blob is closed at
+`blob_size_limit` and at the end of each configured path. A chunk that is
+already stored is never packed again.
+
+Data that an outsider writes compresses smaller when it matches a secret in the
+same chunk, so the blob gets smaller. The blob's size therefore shows the
+outsider whether their data matched the secret.
+
+This works only when all of these hold:
+
+- The outsider can write into a backed-up file.
+- The outsider can see the blob sizes at the destination.
+- The outsider's data and the secret land in the same chunk.
+- The file keeps changing.
+
+Back up a directory that outsiders can write to as a snapshot of its own.
+
+### Chunk boundaries use no secret
+
+The FastCDC parameters are fixed and public. `chunk_size` sets the average chunk
+size. No chunk is larger than four times this size. Only a file's last chunk can
+be smaller than a quarter of it. With the default of 10 MB, those limits are 40
+MB and 2.5 MB. A file of 2.5 MB or less is one chunk.
+
+At the default `blob_size_limit` of 10 GB, a blob holds hundreds of chunks. The
+size of the blob then does not show the size of any one chunk. A limit close to
+`chunk_size` begins to show chunk sizes.
+
+### Decrypted data on the local disk
+
+These commands write unencrypted data to the temporary directory:
+
+- `snapshot restore` writes decrypted blobs under `vaultik-blobcache-*/`. There
+  is no limit on their total size.
+- `snapshot restore` writes the decrypted database to
+  `vaultik-restore-*/snapshot.db`.
+- `snapshot verify --deep` writes the decrypted database to
+  `vaultik-verify-*/snapshot.db`.
+- `snapshot create` writes a copy of the local index to
+  `vaultik-snapshot-*/snapshot.db`.
+
+These files are readable only by their owner, and are removed when the command
+ends. A `kill -9` or a power loss leaves them behind. Delete any leftover
+`vaultik-*` entry in the temporary directory yourself.
+
+Trust the temporary directory as much as the directory you restore into.
+
+### Commands need different permissions
+
+`snapshot create` needs only permission to write objects. Other commands need
+more:
+
+- `snapshot list`, `snapshot verify`, `snapshot restore` and `remote info` list
+  and read.
+- `snapshot purge`, `snapshot remove` and `remote nuke` list and delete.
+- `prune` lists, reads and deletes.
+
+`snapshot create --prune` runs a prune on the backed-up machine. That machine
+then holds credentials that can delete its own backups. To keep it to
+write-only credentials, run `prune` from a separate, trusted machine.
+
+### New recipients cannot read old data
+
+Changing `age_recipients` does not re-encrypt what is already stored. A chunk
+that is already stored is not written again, so it stays encrypted to the
+earlier recipients.
+
+A new snapshot that uses such chunks cannot be restored with a new recipient's
+key alone. To make everything readable by a new key, run
+`vaultik database delete` and take a full backup to a new destination.
+
+### X25519 recipients only
+
+`vaultik` accepts age X25519 public keys only. It rejects age SSH and plugin
+recipients.
+
+The encrypted data is held by a third party for a long time. There is no second
+kind of recipient to fall back on if X25519 is broken, and no post-quantum
+option.
+
+## Order of writes
+
+A backup writes in a fixed order:
+
+- the marker file
+- the blobs
+- `db.zst.age`
+- `manifest.json.zst`
+
+A snapshot's metadata is written only after all its blobs are uploaded. The
+manifest is written last. A directory under `metadata/` that has no manifest
+therefore belongs to a backup that did not finish.
+
+The local index marks a snapshot complete only after both metadata files are
+uploaded. A backup that stops during that upload leaves the snapshot
+unfinished. The next backup removes it from the local index and stores its data
+again.
+
+One short gap remains. A backup can stop after the upload and before the
+snapshot is marked complete. The destination then holds a snapshot that can be
+restored, and the local index does not list it. `snapshot list` shows it as
+remote only.
+
+An upload that is cut off leaves no partial object under the final name on S3
+or on a local directory. The same holds for an rclone remote that can move
+objects. An rclone remote that cannot move objects can keep a partial object.
+
+## Pruning
+
+`prune` deletes two kinds of object at the destination:
+
+- a blob that no manifest lists
+- a directory under `metadata/` that has no manifest
+
+It reads the manifests, which are not encrypted, so it needs no private key.
+
+These rules keep it from deleting data that is in use:
+
+- It keeps every blob that any manifest lists.
+- It deletes nothing when any manifest cannot be downloaded or decoded.
+- It deletes a blob only once the blob is older than `prune_grace_period`.
+- It deletes a directory only once every object in it is that old.
+- It keeps a directory whose snapshot is in the local index.
+- It leaves alone a directory whose name is not a remote key.
+
+`prune_grace_period` defaults to `24h`. It protects a backup that is still
+running. Such a backup has uploaded blobs and perhaps `db.zst.age`, but not yet
+its manifest. A prune from this machine or another keeps those objects, as long
+as no backup takes longer than the grace period.
+
+The grace period does not protect one case. A backup does not upload a blob
+that is already at the destination, so that blob keeps the time of its first
+upload. The blob can be left over from an earlier backup that stopped before
+its manifest. A prune that runs before the new manifest is written can then
+delete that blob. The new snapshot cannot be restored afterwards.
+
+To avoid this case, give each machine and each user its own destination.
+
+`remote nuke` applies none of these rules.
+
+## What a restore needs
+
+- the age private key
+- the snapshot's `db.zst.age`
+- the blobs that hold the files to restore
+
+A restore does not read the manifest and does not need the local index.
+
+It works in these steps:
+
+- It downloads and decrypts `db.zst.age`.
+- It checks that the database is the one for the snapshot that was asked for.
+- It downloads and decrypts the blobs it needs. It checks each blob against its
+  name.
+- It writes each file from its chunks.
+- It sets each file's permissions and times, and its owner when it runs as
+  root.

internal/cli/config.go

@@ -289,12 +289,14 @@ func newConfigInitCommand() *cobra.Command {
 		Use:   "init",
 		Short: "Write a default config file",
 		Long: `Creates a default configuration file with commented explanations
-for every setting. If a config file already exists at the target path,
-the command refuses to overwrite it.
+for the settings it lists. If a config file already exists at the target
+path, the command refuses to overwrite it. The file is created with mode
+0600, because it can hold credentials.
 
 The config is written to the path from --config, $VAULTIK_CONFIG, or
-the platform default config directory (e.g. ~/Library/Application Support/
-on macOS, ~/.config/ on Linux, /etc/vaultik/ as root).`,
+config.yml in the platform default config directory
+(~/Library/Application Support/vaultik/ on macOS, ~/.config/vaultik/ on
+Linux, /etc/vaultik/ as root).`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			path := configPathForInit()
@@ -332,7 +334,7 @@ on macOS, ~/.config/ on Linux, /etc/vaultik/ as root).`,
 func newConfigEditCommand() *cobra.Command {
 	return &cobra.Command{
 		Use:   "edit",
-		Short: "Open the config file in $EDITOR",
+		Short: "Open the config file in $EDITOR (vi when it is not set)",
 		Args:  cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			path, err := ResolveConfigPath()

internal/cli/config_test.go

@@ -61,8 +61,8 @@ func TestDefaultConfigTemplateParses(t *testing.T) {
 // testRecipient is a valid age public key.
 const testRecipient = "age1278m9q7dp3chsh2dcy82qk27v047zywyvtxwnj4cvt0z65jw6a7q5dqhfj"
 
-// TestConfigSetRecipientOnFreshConfig follows the README quickstart: on the
-// file `config init` writes, `config set age_recipients.0` and
+// TestConfigSetRecipientOnFreshConfig follows the README's getting started
+// steps: on the file `config init` writes, `config set age_recipients.0` and
 // `config set storage_url` give a config that loads with that recipient.
 // Neither is refused, although the config does not load until both are set;
 // once it loads, a recipient that is not a valid key is.

internal/cli/entry_snapshot_create_prune_test.go

@@ -1,6 +1,7 @@
 package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
 
 import (
+	"context"
 	"os"
 	"path/filepath"
 	"strings"
@@ -8,6 +9,7 @@ import (
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
 )
 
 // Flags of `snapshot create` used to build argument vectors below.
@@ -117,6 +119,51 @@ func TestEntrySnapshotCreatePruneExitsNonZeroWhenSourceIsEmpty(t *testing.T) {
 	assert.Len(t, snapshots, 2, "the older snapshot is kept")
 }
 
+// TestEntrySnapshotCreatePruneExitsNonZeroWhenMountPointIsMissing backs
+// up, records in the local index that this backup found the source
+// directory as a mount point, and runs `snapshot create --prune`, which
+// does not find it as one. The command exits non-zero, reports the
+// backup as completed, not as a failed snapshot creation, and keeps the
+// older snapshot. See https://git.eeqj.de/sneak/vaultik/issues/395.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreatePruneExitsNonZeroWhenMountPointIsMissing(
+	t *testing.T,
+) {
+	configPath, storeDir := writeBackupConfig(t)
+
+	code, _, _ := runEntry(t, flagConfig, configPath, cmdSnapshot, cmdCreate)
+	require.Equal(t, 0, code)
+
+	// writeBackupConfig puts the source directory and the local index
+	// beside the config file.
+	dir := filepath.Dir(configPath)
+	ctx := context.Background()
+
+	db, err := database.New(ctx, filepath.Join(dir, "index.sqlite"))
+	require.NoError(t, err)
+
+	repos := database.NewRepositories(db)
+
+	backedUp, err := repos.Snapshots.GetOnlySnapshot(ctx)
+	require.NoError(t, err)
+	require.NoError(t, repos.Snapshots.AddMountPoint(ctx, nil,
+		backedUp.ID.String(), filepath.Join(dir, "source")))
+	require.NoError(t, db.Close())
+
+	code, _, stderr := runEntry(t,
+		flagConfig, configPath, cmdSnapshot, cmdCreate, flagPrune)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr, "Backup completed, but 1 mount point(s) "+
+		"an earlier snapshot recorded were not found")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+
+	snapshots, err := os.ReadDir(filepath.Join(storeDir, "metadata"))
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 2, "the older snapshot is kept")
+}
+
 // writeBackupConfig builds a hermetic config with one small file in its
 // source directory and an empty destination store directory. Returns the
 // config path and the store directory.

internal/cli/prune.go

@@ -17,7 +17,11 @@ func NewPruneCommand() *cobra.Command {
 unreferenced blobs and directories under metadata/ with no manifest
 from the backup destination store.
 
-Local cleanup drops incomplete snapshots and any files, chunks, or
+Needs permission to list, read and delete objects on the destination
+store. It does not need the age private key.
+
+Local cleanup drops snapshots whose manifest is gone from the
+destination store, incomplete snapshots, and any files, chunks, or
 blobs no longer referenced by a completed snapshot. Remote cleanup
 scans every snapshot manifest in the destination store, builds the
 set of still-referenced blob hashes, and deletes any blob not in that

internal/cli/remote.go

@@ -40,11 +40,13 @@ func newRemoteNukeCommand() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "nuke",
 		Short: "Delete ALL snapshot metadata and blobs from the backup destination store",
-		Long: `Removes every snapshot's metadata and every blob from remote
-storage. After this command completes successfully the bucket prefix is
-empty and the next backup starts from scratch.
+		Long: `Removes every snapshot's metadata, every blob and the destination
+marker from the backup destination store, and each removed snapshot from
+the local index. After this command completes successfully the bucket
+prefix is empty and the next backup starts from scratch.
 
-This is destructive and irreversible. Requires --force.`,
+This is destructive and irreversible. Requires --force, and permission
+to list and delete objects on the destination store.`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			if !force {
@@ -75,7 +77,10 @@ func newRemoteInfoCommand() *cobra.Command {
 - Size of all snapshot metadata (per snapshot and total)
 - Count and total size of all blobs
 - Count and size of referenced blobs (from all manifests)
-- Count and size of orphaned blobs (not referenced by any manifest)`,
+- Count and size of orphaned blobs (not referenced by any manifest)
+
+Needs permission to list and read the destination store. It does not
+need the age private key.`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			// Use unified config resolution

internal/cli/root.go

@@ -35,9 +35,10 @@ func NewRootCommand() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "vaultik",
 		Short: "Secure incremental backup tool with asymmetric encryption",
-		Long: `vaultik is a secure incremental backup tool that encrypts data using age
-public keys and uploads to S3-compatible storage. No private keys are needed
-on the source system.`,
+		Long: `vaultik is an incremental backup tool that encrypts data to age public
+keys and uploads it to an S3-compatible object store, a local directory or
+an rclone remote. The backed-up machine needs no private key: only
+'snapshot restore' and 'snapshot verify --deep' need one.`,
 		SilenceUsage: true,
 		// Bare 'vaultik' (no subcommand): print help. The banner is
 		// printed once at process startup by Entry, before cobra
@@ -50,13 +51,15 @@ on the source system.`,
 
 	// Add global flags
 	cmd.PersistentFlags().StringVar(&rootFlags.ConfigPath, "config", "",
-		"Path to config file (default: $VAULTIK_CONFIG or platform config dir)")
+		"Path to config file (default: $VAULTIK_CONFIG, then the platform "+
+			"config directory, then /etc/vaultik/config.yml)")
 	cmd.PersistentFlags().BoolVarP(&rootFlags.Verbose, "verbose", "v", false,
 		"Enable verbose output")
 	cmd.PersistentFlags().BoolVar(&rootFlags.Debug, "debug", false,
 		"Enable debug output")
 	cmd.PersistentFlags().BoolVarP(&rootFlags.Quiet, "quiet", "q", false,
-		"Suppress non-error output")
+		"Suppress status output (warnings, errors, JSON documents, "+
+			"'config get' values and confirmation prompts are still printed)")
 	cmd.PersistentFlags().BoolVar(&rootFlags.SkipErrors, "skip-errors", false,
 		"Skip files that cannot be read when creating a snapshot, or "+
 			"that cannot be restored when restoring, instead of aborting "+

internal/cli/snapshot.go

@@ -60,6 +60,21 @@ func NewSnapshotCommand() *cobra.Command {
 	return cmd
 }
 
+// snapshotCreateLong is the long help text of 'snapshot create'.
+const snapshotCreateLong = `Creates new snapshots of the configured directories.
+
+If snapshot names are provided, only those snapshots are created.
+If no names are provided, all configured snapshots are created.
+
+Needs at least one public key in age_recipients, and permission to write
+objects at the backup destination store. It does not need the age private
+key. With --prune it also needs permission to list, read and delete objects
+there.
+
+The config is read from the path given by --config or VAULTIK_CONFIG;
+otherwise from the platform config directory (~/.config/vaultik/config.yml
+on Linux), then /etc/vaultik/config.yml.`
+
 // newSnapshotCreateCommand creates the 'snapshot create' subcommand
 func newSnapshotCreateCommand() *cobra.Command {
 	opts := &vaultik.SnapshotCreateOptions{}
@@ -67,15 +82,8 @@ func newSnapshotCreateCommand() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "create [snapshot-names...]",
 		Short: "Create new snapshots",
-		Long: `Creates new snapshots of the configured directories.
-
-If snapshot names are provided, only those snapshots are created.
-If no names are provided, all configured snapshots are created.
-
-The config is read from the path given by --config or VAULTIK_CONFIG;
-otherwise from the platform config directory (~/.config/vaultik/config.yml
-on Linux), then /etc/vaultik/config.yml.`,
-		Args: cobra.ArbitraryArgs,
+		Long:  snapshotCreateLong,
+		Args:  cobra.ArbitraryArgs,
 		RunE: func(cmd *cobra.Command, args []string) error {
 			err := opts.Validate()
 			if err != nil {
@@ -116,7 +124,8 @@ on Linux), then /etc/vaultik/config.yml.`,
 				}
 
 				if errors.Is(err, vaultik.ErrFilesUnreadable) ||
-					errors.Is(err, vaultik.ErrSourcePathEmpty) {
+					errors.Is(err, vaultik.ErrSourcePathEmpty) ||
+					errors.Is(err, vaultik.ErrVolumeNotMounted) {
 					log.Error("Backup completed with an error", "error", err)
 					ReportErrorf("Backup completed, but %v", err)
 
@@ -137,7 +146,8 @@ on Linux), then /etc/vaultik/config.yml.`,
 			"directory under metadata/ with no manifest once all its "+
 			"objects are older than prune_grace_period and its snapshot is "+
 			"not in the local index (not after a backup that skipped "+
-			"unreadable files or found a configured path empty)")
+			"unreadable files, found a configured path empty, or did not "+
+			"find a mount point an earlier snapshot found)")
 	cmd.Flags().StringVar(&opts.KeepNewerThan, "keep-newer-than", "",
 		"With --prune: keep snapshots newer than this duration "+
 			"(e.g. 30d, 4w, 6mo, 1y; m is minutes, mo is months) "+
@@ -154,8 +164,15 @@ func newSnapshotListCommand() *cobra.Command {
 		Use:     "list",
 		Aliases: []string{"ls"},
 		Short:   "List all snapshots",
-		Long:    "Lists all snapshots with their ID, timestamp, and compressed size",
-		Args:    cobra.NoArgs,
+		Long: `Lists the snapshots in the local index and on the backup destination
+store, with their ID, timestamp, and sizes.
+
+Needs permission to list and read the destination store. It does not need
+the age private key. A snapshot found only on the destination store is
+listed by its remote key, because its ID is stored only in the local index
+and in the snapshot's encrypted database. If the destination store cannot
+be listed, the command warns and lists the local index alone.`,
+		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			return runVaultikApp(cmd, readOnly, false,
 				"Failed to list snapshots",
@@ -177,11 +194,19 @@ func newSnapshotPurgeCommand() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "purge",
 		Short: "Purge old snapshots",
-		Long: `Removes snapshots based on age or count criteria.
+		Long: `Removes snapshots based on age or count criteria. Exactly one of
+--keep-latest and --older-than is required.
 
-Retention is per-snapshot-name: --keep-latest keeps the latest of each
-configured snapshot name, not the latest globally. Use --snapshot to
-restrict the operation to specific snapshot names.`,
+Retention is per-snapshot-name: --keep-latest keeps the latest snapshot of
+each name, not the latest globally. Use --snapshot to restrict the
+operation to specific snapshot names.
+
+Only snapshots in the local index are removed. Each one is removed from
+the local index and its metadata is deleted from the backup destination
+store. Blobs are not touched: run 'vaultik prune' afterwards.
+
+Needs permission to list and delete objects on the destination store. If
+the destination store cannot be listed, nothing is removed.`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			// Validate flags
@@ -220,14 +245,29 @@ func newSnapshotVerifyCommand() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "verify <snapshot-id>",
 		Short: "Check a snapshot's blobs are present with the listed size",
-		Long: "Checks that every blob the snapshot's manifest lists is present\n" +
-			"in storage with the size the manifest records, and that the\n" +
-			"snapshot's encrypted database is present. It does not read blob\n" +
-			"contents; use --deep to download, decrypt, and re-hash every blob\n" +
-			"to detect corruption -- integrity, not who wrote it.\n\n" +
-			"The snapshot may be named by its ID or, on a host with no local\n" +
-			"index, by the remote key that 'snapshot list' prints for a\n" +
-			"remote-only snapshot (an unambiguous leading part is enough).",
+		Long: `Checks that every blob the snapshot's manifest lists is present
+in storage with the size the manifest records, and that the
+snapshot's encrypted database is present. It does not read blob
+contents and does not need the age private key.
+
+Needs permission to read the backup destination store, and to list it
+when a remote key is given.
+
+Use --deep to download, decrypt, and re-hash every blob to detect
+corruption -- integrity, not who wrote it.
+
+--deep requires the age private key, in the VAULTIK_AGE_SECRET_KEY
+environment variable or as age_secret_key in the config file; the
+environment variable wins. The variable may hold the whole age-keygen
+file, with its comments and all of its keys. Read the key from the
+file rather than typing it, so that it does not land in your shell
+history:
+
+  export VAULTIK_AGE_SECRET_KEY="$(cat vaultik_backup_private_key.txt)"
+
+The snapshot may be named by its ID or, on a host with no local
+index, by the remote key that 'snapshot list' prints for a
+remote-only snapshot (an unambiguous leading part is enough).`,
 		Args: requireSnapshotIDArg,
 		RunE: func(cmd *cobra.Command, args []string) error {
 			snapshotID := args[0]
@@ -262,7 +302,10 @@ func newSnapshotVerifyCommand() *cobra.Command {
 		},
 	}
 
-	cmd.Flags().BoolVar(&opts.Deep, "deep", false, "Download and verify blob hashes")
+	cmd.Flags().BoolVar(&opts.Deep, "deep", false,
+		"Download, decrypt and re-hash every blob; requires the age private "+
+			"key in VAULTIK_AGE_SECRET_KEY, or as age_secret_key in the config "+
+			"file")
 	cmd.Flags().BoolVar(&opts.JSON, "json", false, "Output verification results as JSON")
 
 	return cmd
@@ -286,8 +329,10 @@ know about), which is what 'vaultik prune' does. The command prints the
 'vaultik prune' invocation to run as a follow-up only when it removed the
 snapshot's metadata from the destination store, and never under --json.
 
-An ID that names no snapshot in the local index or on the destination
-store is an error.
+The argument must be a snapshot ID: a remote key is not accepted. An ID
+that names no snapshot in the local index or on the destination store is
+an error. Needs permission to list and delete objects on the destination
+store, unless --local-only is given.
 
 Use --local-only to skip the remote half (e.g. when you want to forget a
 snapshot locally without touching the destination store); the snapshot

internal/cli/snapshot_restore.go

@@ -37,13 +37,21 @@ The snapshot may be named by its ID or, when restoring on a host with no
 local index, by the remote key that 'snapshot list' prints for a
 remote-only snapshot (an unambiguous leading part is enough).
 
-Requires the age private key in the VAULTIK_AGE_SECRET_KEY environment
-variable. The variable may hold the whole age-keygen file (comments and
-all of its identities are accepted); read it from the file rather than
-typing the key, so it does not land in your shell history:
+Each file is restored under its original absolute path inside the target
+directory. File owners are restored only when the command runs as root.
+
+Requires the age private key, in the VAULTIK_AGE_SECRET_KEY environment
+variable or as age_secret_key in the config file; the environment
+variable wins. The variable may hold the whole age-keygen file, with its
+comments and all of its keys. Read the key from the file rather than
+typing it, so that it does not land in your shell history:
 
   export VAULTIK_AGE_SECRET_KEY="$(cat vaultik_backup_private_key.txt)"
 
+Also needs permission to read the backup destination store (and to list
+it when a remote key is given), and free space in the temporary directory
+for the decrypted blobs.
+
 Examples:
   # Restore entire snapshot
   vaultik snapshot restore myhost_docs_2025-01-01T12:00:00Z /restore

internal/database/schema/001.sql

@@ -123,6 +123,16 @@ CREATE TABLE IF NOT EXISTS snapshot_blobs (
 -- Index for efficient blob lookups (used in orphan detection)
 CREATE INDEX IF NOT EXISTS idx_snapshot_blobs_blob_id ON snapshot_blobs(blob_id);
 
+-- Snapshot mount points table: the directories at or below a configured
+-- path that the snapshot's backup found on a different filesystem from
+-- their parent directory, as a mount point is, and found an entry in
+CREATE TABLE IF NOT EXISTS snapshot_mount_points (
+    snapshot_id TEXT NOT NULL,
+    path TEXT NOT NULL,
+    PRIMARY KEY (snapshot_id, path),
+    FOREIGN KEY (snapshot_id) REFERENCES snapshots(id) ON DELETE CASCADE
+);
+
 -- Uploads table: tracks blob upload metrics
 CREATE TABLE IF NOT EXISTS uploads (
     blob_hash TEXT PRIMARY KEY,

internal/database/snapshots.go

@@ -24,7 +24,8 @@ var (
 )
 
 // SnapshotRepository provides access to the snapshots table and its
-// snapshot_files / snapshot_blobs association tables.
+// snapshot_files / snapshot_blobs association tables, and to the mount
+// points each snapshot's backup found (snapshot_mount_points).
 type SnapshotRepository struct {
 	db *DB
 }
@@ -500,27 +501,57 @@ func (r *SnapshotRepository) GetBlobHashes(
 		ORDER BY sb.blob_hash
 	`
 
-	rows, err := r.db.conn.QueryContext(ctx, query, snapshotID)
+	blobs, err := r.queryStrings(ctx, query, snapshotID)
 	if err != nil {
 		return nil, fmt.Errorf("querying blob hashes: %w", err)
 	}
 
-	defer func() { _ = rows.Close() }()
+	return blobs, nil
+}
 
-	var blobs []string
+// AddMountPoint records dir as a directory that the backup of a snapshot
+// found on a different filesystem from its parent directory, as a mount
+// point is, using tx when non-nil.
+func (r *SnapshotRepository) AddMountPoint(
+	ctx context.Context, tx *sql.Tx, snapshotID string, dir string,
+) error {
+	query := `
+		INSERT OR IGNORE INTO snapshot_mount_points (snapshot_id, path)
+		VALUES (?, ?)
+	`
 
-	for rows.Next() {
-		var blobHash string
-
-		err := rows.Scan(&blobHash)
-		if err != nil {
-			return nil, fmt.Errorf("scanning blob hash: %w", err)
-		}
-
-		blobs = append(blobs, blobHash)
+	var err error
+	if tx != nil {
+		_, err = tx.ExecContext(ctx, query, snapshotID, dir)
+	} else {
+		_, err = r.db.ExecWithLog(ctx, query, snapshotID, dir)
 	}
 
-	return blobs, rows.Err()
+	if err != nil {
+		return fmt.Errorf("adding mount point to snapshot: %w", err)
+	}
+
+	return nil
+}
+
+// GetMountPoints returns the directories AddMountPoint recorded for a
+// snapshot, sorted.
+func (r *SnapshotRepository) GetMountPoints(
+	ctx context.Context, snapshotID string,
+) ([]string, error) {
+	query := `
+		SELECT path
+		FROM snapshot_mount_points
+		WHERE snapshot_id = ?
+		ORDER BY path
+	`
+
+	dirs, err := r.queryStrings(ctx, query, snapshotID)
+	if err != nil {
+		return nil, fmt.Errorf("querying mount points: %w", err)
+	}
+
+	return dirs, nil
 }
 
 // GetSnapshotTotalCompressedSize returns the total compressed size of all
@@ -709,6 +740,34 @@ func (r *SnapshotRepository) DeleteSnapshotUploads(
 	return nil
 }
 
+// queryStrings runs query, which selects one text column, and returns
+// its values in the order of the rows.
+func (r *SnapshotRepository) queryStrings(
+	ctx context.Context, query string, args ...any,
+) ([]string, error) {
+	rows, err := r.db.conn.QueryContext(ctx, query, args...)
+	if err != nil {
+		return nil, err
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	var values []string
+
+	for rows.Next() {
+		var value string
+
+		err := rows.Scan(&value)
+		if err != nil {
+			return nil, err
+		}
+
+		values = append(values, value)
+	}
+
+	return values, rows.Err()
+}
+
 // extendedCompressionRatio computes the compression ratio for a snapshot
 // from its stored blob_size and the given uncompressed size. Returns 1.0
 // when the uncompressed size is zero.

internal/database/snapshots_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"math"
+	"slices"
 	"testing"
 	"time"
 
@@ -422,3 +423,52 @@ func TestSnapshotRepositoryDuplicate(t *testing.T) {
 		t.Error("expected error for duplicate snapshot")
 	}
 }
+
+// Deleting a snapshot from the local index deletes the mount points its
+// backup recorded, and no other snapshot's. See
+// https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestSnapshotRepositoryDeleteRemovesMountPoints(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewSnapshotRepository(db)
+
+	for _, id := range []types.SnapshotID{"kept", "deleted"} {
+		err := repo.Create(ctx, nil, &database.Snapshot{
+			ID:             id,
+			Hostname:       testHostname,
+			VaultikVersion: testVersion,
+			StartedAt:      time.Now().Truncate(time.Second),
+		})
+		if err != nil {
+			t.Fatalf("failed to create snapshot: %v", err)
+		}
+
+		err = repo.AddMountPoint(ctx, nil, id.String(), "/data/nas")
+		if err != nil {
+			t.Fatalf("failed to add mount point: %v", err)
+		}
+	}
+
+	err := repo.Delete(ctx, "deleted")
+	if err != nil {
+		t.Fatalf("failed to delete snapshot: %v", err)
+	}
+
+	for id, want := range map[string][]string{
+		"kept":    {"/data/nas"},
+		"deleted": nil,
+	} {
+		got, err := repo.GetMountPoints(ctx, id)
+		if err != nil {
+			t.Fatalf("failed to read mount points: %v", err)
+		}
+
+		if !slices.Equal(got, want) {
+			t.Errorf("mount points of %s: got %v, want %v", id, got, want)
+		}
+	}
+}

internal/snapshot/device_number_test.go

@@ -0,0 +1,40 @@
+//nolint:testpackage // exercises the unexported deviceNumberFromStat
+package snapshot
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// The device number a real backup reads is the same for a directory and
+// a file in it, and differs between the temp directory and /dev.
+// Without it no mount point is ever found. See
+// https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestDeviceNumberFromStat(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	file := filepath.Join(dir, "file.txt")
+	require.NoError(t, os.WriteFile(file, nil, 0o600))
+
+	deviceOf := func(path string) uint64 {
+		t.Helper()
+
+		info, err := os.Stat(path)
+		require.NoError(t, err)
+
+		device, ok := deviceNumberFromStat(info)
+		require.True(t, ok, "no device number read for %s", path)
+
+		return device
+	}
+
+	assert.Equal(t, deviceOf(dir), deviceOf(file))
+
+	// /dev is a different filesystem from the temp dir on Linux, macOS and in Docker.
+	assert.NotEqual(t, deviceOf(dir), deviceOf("/dev"))
+}

internal/snapshot/module.go

@@ -1,6 +1,8 @@
 package snapshot
 
 import (
+	"os"
+
 	"github.com/spf13/afero"
 	"go.uber.org/fx"
 	"sneak.berlin/go/vaultik/internal/config"
@@ -19,6 +21,9 @@ type ScannerParams struct {
 	// PreviousSnapshotID is the most recent completed snapshot of the
 	// same name, "" for none.
 	PreviousSnapshotID string
+	// DeviceNumber is ScannerConfig.DeviceNumber; nil reads it from the
+	// file's stat.
+	DeviceNumber func(info os.FileInfo) (uint64, bool)
 }
 
 // Module exports backup functionality as an fx module.
@@ -59,6 +64,7 @@ func provideScannerFactory(
 			Exclude:            excludes,
 			SkipErrors:         params.SkipErrors,
 			PreviousSnapshotID: params.PreviousSnapshotID,
+			DeviceNumber:       params.DeviceNumber,
 		})
 	}
 }

internal/snapshot/mount_point_test.go

@@ -0,0 +1,245 @@
+package snapshot_test
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+	"syscall"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// A scan records with the snapshot each directory on a different device
+// from its parent directory that the walk found an entry in, as a mount
+// point is, the scanned path included. A filesystem that gives no device
+// numbers, as the in-memory one, records none. See
+// https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestScanRecordsMountPoints(t *testing.T) {
+	t.Parallel()
+
+	const (
+		scannedPath = "/data/source"
+		snapshotID  = "mount-points"
+	)
+
+	tests := []struct {
+		name string
+		// onVolume names the directories on the volume's device; every
+		// other directory is on another one. nil leaves the device
+		// numbers to the in-memory filesystem, which has none.
+		onVolume []string
+		want     []string
+	}{
+		{
+			name:     "volume below the scanned path",
+			onVolume: []string{"nas"},
+			want:     []string{scannedPath + "/nas"},
+		},
+		{
+			name:     "volume at the scanned path",
+			onVolume: []string{"source", "nas"},
+			want:     []string{scannedPath},
+		},
+		{name: "no device numbers", want: nil},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+			fs := afero.NewMemMapFs()
+
+			for _, path := range []string{
+				scannedPath + "/notes.txt", scannedPath + "/nas/a.txt",
+			} {
+				require.NoError(t, fs.MkdirAll(filepath.Dir(path), 0o755))
+				require.NoError(t, afero.WriteFile(fs, path, []byte(path), 0o644))
+			}
+
+			db, err := database.NewTestDB()
+			require.NoError(t, err)
+			t.Cleanup(func() { _ = db.Close() })
+
+			repos := database.NewRepositories(db)
+			createTestSnapshotRecord(ctx, t, repos, snapshotID)
+
+			cfg := snapshot.ScannerConfig{
+				FS:               fs,
+				ChunkSize:        int64(1024 * 16),
+				Repositories:     repos,
+				MaxBlobSize:      int64(1024 * 1024),
+				CompressionLevel: 3,
+				AgeRecipients:    []string{testAgePublicKey},
+			}
+			if tt.onVolume != nil {
+				cfg.DeviceNumber = func(info os.FileInfo) (uint64, bool) {
+					if slices.Contains(tt.onVolume, info.Name()) {
+						return 2, true
+					}
+
+					return 1, true
+				}
+			}
+
+			_, err = snapshot.NewScanner(cfg).Scan(ctx, scannedPath, snapshotID)
+			require.NoError(t, err)
+
+			got, err := repos.Snapshots.GetMountPoints(ctx, snapshotID)
+			require.NoError(t, err)
+			assert.Equal(t, tt.want, got)
+		})
+	}
+}
+
+// automountFs is an in-memory filesystem with a volume at mountPoint
+// that is mounted on first access: opening mountPoint to list it mounts
+// the volume or, if it cannot be mounted, fails with "no such file or
+// directory", as with the autofs daemon. Stat records the device an
+// entry is on at that moment, as a real stat does: unmountedDevice for
+// mountPoint until the volume is mounted; the volume's, 2, for
+// mountPoint once it is and for everything in it; 1 for everything else.
+type automountFs struct {
+	afero.Fs
+
+	mountPoint      string
+	unmountedDevice uint64
+	canMount        bool
+	mounted         bool
+}
+
+//nolint:ireturn // afero.Fs.Open is defined to return the interface.
+func (fs *automountFs) Open(name string) (afero.File, error) {
+	if name == fs.mountPoint && !fs.canMount {
+		return nil, &os.PathError{Op: "open", Path: name, Err: syscall.ENOENT}
+	}
+
+	if name == fs.mountPoint {
+		fs.mounted = true
+	}
+
+	return fs.Fs.Open(name)
+}
+
+func (fs *automountFs) Stat(name string) (os.FileInfo, error) {
+	info, err := fs.Fs.Stat(name)
+	if err != nil {
+		return nil, err
+	}
+
+	device := uint64(1)
+
+	switch {
+	case name == fs.mountPoint && !fs.mounted:
+		device = fs.unmountedDevice
+	case name == fs.mountPoint || strings.HasPrefix(name, fs.mountPoint+"/"):
+		device = 2
+	}
+
+	return statWithDevice{FileInfo: info, device: device}, nil
+}
+
+// statWithDevice is an entry automountFs.Stat looked up, with the device
+// it was on then.
+type statWithDevice struct {
+	os.FileInfo
+
+	device uint64
+}
+
+// deviceAtLookup returns the device automountFs.Stat recorded for info.
+func deviceAtLookup(info os.FileInfo) (uint64, bool) {
+	stat, ok := info.(statWithDevice)
+
+	return stat.device, ok
+}
+
+// A volume mounted on first access is mounted when the walk lists its
+// mount point. Until then the mount point is on the automounter's own
+// device (a systemd automount) or on its parent's (an autofs map that
+// shows its mount points before mounting them). Either way only the
+// mount point is recorded, by the backup that mounts the volume and by a
+// later one that finds it mounted. Listing an automount point whose
+// volume cannot be mounted fails; it is not recorded, because the walk
+// finds no entry in it. See https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestScanRecordsOnlyTheMountPointOfAnAutomountedVolume(t *testing.T) {
+	t.Parallel()
+
+	const scannedPath = "/data/source"
+
+	mountPoint := scannedPath + "/nas"
+
+	tests := []struct {
+		name            string
+		unmountedDevice uint64
+		canMount        bool
+		want            []string
+	}{
+		{
+			name:            "on the automounter's device until listed",
+			unmountedDevice: 3, canMount: true, want: []string{mountPoint},
+		},
+		{
+			name:            "on its parent's device until listed",
+			unmountedDevice: 1, canMount: true, want: []string{mountPoint},
+		},
+		{
+			name:            "volume that cannot be mounted",
+			unmountedDevice: 1, canMount: false, want: nil,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+			fs := &automountFs{
+				Fs: afero.NewMemMapFs(), mountPoint: mountPoint,
+				unmountedDevice: tt.unmountedDevice, canMount: tt.canMount,
+			}
+
+			require.NoError(t, fs.MkdirAll(mountPoint+"/photos", 0o755))
+
+			for _, path := range []string{
+				scannedPath + "/notes.txt", mountPoint + "/photos/a.jpg",
+			} {
+				require.NoError(t, afero.WriteFile(fs, path, []byte(path), 0o644))
+			}
+
+			db, err := database.NewTestDB()
+			require.NoError(t, err)
+			t.Cleanup(func() { _ = db.Close() })
+
+			repos := database.NewRepositories(db)
+
+			for _, snapshotID := range []string{"first", "later"} {
+				createTestSnapshotRecord(ctx, t, repos, snapshotID)
+
+				scanner := snapshot.NewScanner(snapshot.ScannerConfig{
+					FS:               fs,
+					ChunkSize:        int64(1024 * 16),
+					Repositories:     repos,
+					MaxBlobSize:      int64(1024 * 1024),
+					CompressionLevel: 3,
+					AgeRecipients:    []string{testAgePublicKey},
+					DeviceNumber:     deviceAtLookup,
+				})
+
+				_, err = scanner.Scan(ctx, scannedPath, snapshotID)
+				require.NoError(t, err)
+
+				got, err := repos.Snapshots.GetMountPoints(ctx, snapshotID)
+				require.NoError(t, err)
+				assert.Equal(t, tt.want, got, "mount points of the %s backup", snapshotID)
+			}
+		})
+	}
+}

internal/snapshot/scanner.go

@@ -73,6 +73,9 @@ type Scanner struct {
 	skipErrors bool
 	// ui is the user-facing output; never nil (defaults to a discarding writer).
 	ui *ui.Writer
+	// deviceNumber returns the number of the device that holds a file;
+	// never nil (defaults to deviceNumberFromStat).
+	deviceNumber func(info os.FileInfo) (uint64, bool)
 
 	// In-memory cache of known chunk hashes for fast existence checks
 	knownChunks   map[string]struct{}
@@ -136,6 +139,10 @@ type ScannerConfig struct {
 	// PreviousSnapshotID is the most recent completed snapshot of the
 	// same name, "" for none.
 	PreviousSnapshotID string
+	// DeviceNumber returns the number of the device that holds a file,
+	// and false when there is none to read. nil reads it from the file's
+	// stat; tests replace it, since the in-memory test filesystem has none.
+	DeviceNumber func(info os.FileInfo) (uint64, bool)
 }
 
 // ScanResult contains the results of a scan operation. Files and bytes
@@ -214,6 +221,11 @@ func NewScanner(cfg ScannerConfig) *Scanner {
 		uiw = ui.NewWithColor(io.Discard, false)
 	}
 
+	deviceNumber := cfg.DeviceNumber
+	if deviceNumber == nil {
+		deviceNumber = deviceNumberFromStat
+	}
+
 	return &Scanner{
 		fs:                 cfg.FS,
 		chunker:            chunker.NewChunker(cfg.ChunkSize),
@@ -229,6 +241,7 @@ func NewScanner(cfg ScannerConfig) *Scanner {
 		skipErrors:         cfg.SkipErrors,
 		previousSnapshotID: cfg.PreviousSnapshotID,
 		ui:                 uiw,
+		deviceNumber:       deviceNumber,
 		pendingChunkHashes: make(map[string]struct{}),
 	}
 }
@@ -297,7 +310,7 @@ func (s *Scanner) Scan(
 		}
 	}
 
-	err = s.recordDirectories(ctx, scanResult.Directories)
+	err = s.recordDirectories(ctx, scanResult)
 	if err != nil {
 		return nil, err
 	}
@@ -875,6 +888,11 @@ type ScanPhaseResult struct {
 	FilesToProcess   []*FileToProcess
 	UnchangedFileIDs []types.FileID   // IDs of unchanged files to associate with snapshot
 	Directories      []*FileToProcess // Every directory found, to record
+	// MountPoints are the directories, the scanned path included, that
+	// the walk found an entry in and, once it had listed them, on a
+	// different device from their parent directory (see
+	// recordIfMountPoint).
+	MountPoints []string
 }
 
 // scanPhase performs the initial directory scan to identify files to process
@@ -894,7 +912,7 @@ func (s *Scanner) scanPhase(
 	// estimate leaves them out.
 	estimatedTotal := countNonDirectories(knownFiles)
 
-	collector := &scanCollector{}
+	collector := &scanCollector{dirsWithEntries: make(map[string]struct{})}
 
 	// Set up periodic status output
 	startTime := time.Now()
@@ -939,6 +957,8 @@ func (s *Scanner) scanPhase(
 			result.EntriesFound++
 		}
 
+		s.recordIfMountPoint(filePath, path, collector)
+
 		// Handle symlinks and directories
 		handled, err := s.recordSpecialEntry(
 			filePath, info, existingFiles, knownFiles, collector, result)
@@ -973,6 +993,7 @@ func (s *Scanner) scanPhase(
 		FilesToProcess:   collector.filesToProcess,
 		UnchangedFileIDs: collector.unchangedFileIDs,
 		Directories:      collector.directories,
+		MountPoints:      collector.mountPoints,
 	}, nil
 }
 
@@ -1099,6 +1120,29 @@ type scanCollector struct {
 	unchangedFileIDs []types.FileID // Just IDs - no new records needed
 	directories      []*FileToProcess
 	filesScanned     int64
+	// dirsWithEntries holds each directory the walk has found an entry in.
+	dirsWithEntries map[string]struct{}
+	mountPoints     []string
+}
+
+// firstEntryIn notes that the walk found an entry in directory dir and
+// reports whether it is the first one.
+func (c *scanCollector) firstEntryIn(dir string) bool {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	_, seen := c.dirsWithEntries[dir]
+	c.dirsWithEntries[dir] = struct{}{}
+
+	return !seen
+}
+
+// addMountPoint sets a mount point aside to be recorded.
+func (c *scanCollector) addMountPoint(dir string) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	c.mountPoints = append(c.mountPoints, dir)
 }
 
 // addDirectory sets a directory aside to be recorded. It is not counted
@@ -1290,20 +1334,100 @@ func (s *Scanner) buildDirectoryEntry(path string, info os.FileInfo) *database.F
 	}
 }
 
-// recordDirectories records every directory the walk found. A directory
-// has no data to back up, so it is not among the files counted and
-// backed up.
+// recordDirectories records every directory the walk found, and its
+// mount points (see recordMountPoints). A directory has no data to back
+// up, so it is not among the files counted and backed up.
 func (s *Scanner) recordDirectories(
-	ctx context.Context, directories []*FileToProcess,
+	ctx context.Context, scanned *ScanPhaseResult,
 ) error {
-	for _, dir := range directories {
+	for _, dir := range scanned.Directories {
 		err := s.recordNonRegularFile(ctx, dir)
 		if err != nil {
 			return fmt.Errorf("recording directory %s: %w", dir.Path, err)
 		}
 	}
 
-	return nil
+	return s.recordMountPoints(ctx, scanned.MountPoints)
+}
+
+// recordMountPoints records with the snapshot the mount points the walk
+// found (see ScanPhaseResult.MountPoints).
+func (s *Scanner) recordMountPoints(
+	ctx context.Context, mountPoints []string,
+) error {
+	return s.repos.WithTx(ctx, func(txCtx context.Context, tx *sql.Tx) error {
+		for _, dir := range mountPoints {
+			log.Info("Found a mount point", "path", dir)
+
+			err := s.repos.Snapshots.AddMountPoint(txCtx, tx, s.snapshotID, dir)
+			if err != nil {
+				return fmt.Errorf("recording mount point %s: %w", dir, err)
+			}
+		}
+
+		return nil
+	})
+}
+
+// recordIfMountPoint sets the directory that holds entry, a path the
+// walk found, aside to be recorded as a mount point if it is on a
+// different device from its parent directory. The walk calls it for each
+// entry it finds, and it decides on the first entry in a directory, so a
+// directory the walk found no entry in is never recorded. By then the
+// walk has listed the directory, which mounts a volume mounted there on
+// first access, so the directory and its parent are looked up again:
+// before the listing, the directory can be on the automounter's own
+// device (a systemd automount) or on its parent's (an autofs map that
+// shows its mount points before mounting them). The scanned path is in
+// no listing the walk made, so it is skipped as an entry. A directory
+// that cannot be looked up then, or whose parent cannot, is not
+// recorded, so this snapshot does not stop the prune of a later backup
+// that misses its volume.
+func (s *Scanner) recordIfMountPoint(
+	entry, scannedPath string, collector *scanCollector,
+) {
+	dir := filepath.Dir(entry)
+
+	if entry == scannedPath || !collector.firstEntryIn(dir) {
+		return
+	}
+
+	dirDevice, ok := s.deviceNumberAt(dir)
+	if !ok {
+		return
+	}
+
+	parentDevice, ok := s.deviceNumberAt(filepath.Dir(dir))
+	if ok && dirDevice != parentDevice {
+		collector.addMountPoint(dir)
+	}
+}
+
+// deviceNumberAt looks path up now and returns the number of the device
+// that holds it, and false when the lookup fails (logged) or gives no
+// device number.
+func (s *Scanner) deviceNumberAt(path string) (uint64, bool) {
+	info, err := s.fs.Stat(path)
+	if err != nil {
+		log.Warn("Could not look up a directory to check for a mount point",
+			"path", path, "error", err)
+
+		return 0, false
+	}
+
+	return s.deviceNumber(info)
+}
+
+// deviceNumberFromStat returns the number of the device that holds the
+// file info describes, and false when info carries no stat, as from the
+// in-memory test filesystem.
+func deviceNumberFromStat(info os.FileInfo) (uint64, bool) {
+	stat, ok := info.Sys().(*syscall.Stat_t)
+	if !ok {
+		return 0, false
+	}
+
+	return uint64(stat.Dev), true //nolint:unconvert // Dev is an int32 on darwin
 }
 
 // recordNonRegularFile writes a symlink or directory entry to the database

internal/snapshot/snapshot.go

@@ -518,8 +518,8 @@ type CleanupStats struct {
 //
 // The cleanup is performed in a specific order to maintain referential integrity:
 //  1. Delete other snapshots
-//  2. Delete orphaned snapshot associations (snapshot_files, snapshot_blobs)
-//     for deleted snapshots
+//  2. Delete orphaned snapshot associations (snapshot_files, snapshot_blobs,
+//     snapshot_mount_points) for deleted snapshots
 //  3. Delete orphaned files (not in the current snapshot)
 //  4. Delete orphaned chunk-to-file mappings (references to deleted files)
 //  5. Delete orphaned blobs (not in the current snapshot)
@@ -948,8 +948,8 @@ func (sm *SnapshotManager) deleteOtherSnapshots(
 	return nil
 }
 
-// deleteOrphanedSnapshotAssociations deletes snapshot_files and
-// snapshot_blobs for deleted snapshots
+// deleteOrphanedSnapshotAssociations deletes snapshot_files,
+// snapshot_blobs and snapshot_mount_points for deleted snapshots
 func (sm *SnapshotManager) deleteOrphanedSnapshotAssociations(
 	ctx context.Context, tx *sql.Tx, currentSnapshotID string,
 ) error {
@@ -983,6 +983,19 @@ func (sm *SnapshotManager) deleteOrphanedSnapshotAssociations(
 	log.Debug("[Temp DB Cleanup] Deleted snapshot_blobs associations",
 		"count", rowsAffected)
 
+	// Delete the mount points other snapshots recorded
+	query := "DELETE FROM snapshot_mount_points WHERE snapshot_id != ?"
+	database.LogSQL("Execute", query, currentSnapshotID)
+
+	result, err = tx.ExecContext(ctx, query, currentSnapshotID)
+	if err != nil {
+		return fmt.Errorf("deleting orphaned snapshot_mount_points: %w", err)
+	}
+
+	rowsAffected, _ = result.RowsAffected()
+	log.Debug("[Temp DB Cleanup] Deleted snapshot_mount_points rows",
+		"count", rowsAffected)
+
 	return nil
 }
 

internal/snapshot/snapshot_test.go

@@ -7,12 +7,14 @@ import (
 	"database/sql"
 	"io"
 	"path/filepath"
+	"slices"
 	"testing"
 
 	"github.com/spf13/afero"
 	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/types"
 )
 
 const (
@@ -387,6 +389,77 @@ func TestCleanSnapshotDBNonExistentSnapshot(t *testing.T) {
 	}
 }
 
+// The exported metadata database keeps only the mount points that the
+// exported snapshot recorded. See
+// https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestCleanSnapshotDBKeepsOnlyTheSnapshotsMountPoints(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	dbPath := filepath.Join(t.TempDir(), "export.db")
+
+	db, err := database.New(ctx, dbPath)
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+
+	repos := database.NewRepositories(db)
+
+	for _, id := range []types.SnapshotID{"exported", "other"} {
+		err = repos.Snapshots.Create(ctx, nil, &database.Snapshot{ID: id})
+		if err != nil {
+			t.Fatalf("failed to create snapshot: %v", err)
+		}
+
+		err = repos.Snapshots.AddMountPoint(ctx, nil, id.String(), "/data/"+id.String())
+		if err != nil {
+			t.Fatalf("failed to add mount point: %v", err)
+		}
+	}
+
+	err = db.Close()
+	if err != nil {
+		t.Fatalf("failed to close database: %v", err)
+	}
+
+	sm := &SnapshotManager{
+		config: &config.Config{
+			CompressionLevel: 3,
+			AgeRecipients:    []string{testAgeRecipient},
+		},
+		fs: afero.NewOsFs(),
+	}
+
+	_, err = sm.cleanSnapshotDB(ctx, dbPath, "exported")
+	if err != nil {
+		t.Fatalf("failed to clean snapshot database: %v", err)
+	}
+
+	cleaned, err := database.New(ctx, dbPath)
+	if err != nil {
+		t.Fatalf("failed to open cleaned database: %v", err)
+	}
+
+	defer func() { _ = cleaned.Close() }()
+
+	cleanedRepos := database.NewRepositories(cleaned)
+
+	for id, want := range map[string][]string{
+		"exported": {"/data/exported"},
+		"other":    nil,
+	} {
+		got, err := cleanedRepos.Snapshots.GetMountPoints(ctx, id)
+		if err != nil {
+			t.Fatalf("failed to read mount points: %v", err)
+		}
+
+		if !slices.Equal(got, want) {
+			t.Errorf("mount points of %s: got %v, want %v", id, got, want)
+		}
+	}
+}
+
 // Two creates of one snapshot name back to back start within one second,
 // the resolution of the timestamp in a snapshot ID. See
 // https://git.eeqj.de/sneak/vaultik/issues/270.

internal/storage/s3_test.go

@@ -161,7 +161,7 @@ func TestS3ListingCutShortIsAnError(t *testing.T) {
 }
 
 // TestS3URLPrefixKeyLayout pins the bucket keys an s3:// URL reads and
-// writes: the README's remote storage layout, with the prefix joined to
+// writes: the layout in docs/REPOSTRUCTURE.md, with the prefix joined to
 // each key by one "/". s3://b/p and s3://b/p/ must be the same
 // destination, or a host that writes the URL the other way finds no
 // snapshots. The listed object is put straight into the bucket, as

internal/vaultik/fault_injection_test.go

@@ -679,6 +679,7 @@ func faultScannerFactory(
 			Exclude:            params.Exclude,
 			SkipErrors:         params.SkipErrors,
 			PreviousSnapshotID: params.PreviousSnapshotID,
+			DeviceNumber:       params.DeviceNumber,
 		})
 	}
 }

internal/vaultik/snapshot.go

@@ -7,6 +7,7 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"slices"
 	"sort"
 	"strconv"
 	"strings"
@@ -52,6 +53,17 @@ var ErrFilesUnreadable = errors.New("file(s) could not be read")
 // returns ErrFilesUnreadable instead.
 var ErrSourcePathEmpty = errors.New("source path(s) were empty")
 
+// ErrVolumeNotMounted is returned when --prune skipped the post-backup
+// prune because a directory that an earlier snapshot of the same name
+// found as a mount point was not found as one in this run: the volume
+// mounted there is not mounted, the run found no entry in it, or the
+// directory is no longer backed up. The snapshots are complete and
+// kept, so the caller can report it apart from a failed backup. A run
+// that also skipped unreadable files or found a configured path empty
+// returns ErrFilesUnreadable or ErrSourcePathEmpty instead.
+var ErrVolumeNotMounted = errors.New(
+	"mount point(s) an earlier snapshot recorded were not found")
+
 // listRecentLimit caps how many snapshot rows are fetched from the
 // local index for listing and purge operations.
 const listRecentLimit = 10000
@@ -142,7 +154,7 @@ func (v *Vaultik) CreateSnapshot(opts *SnapshotCreateOptions) error {
 		return err
 	}
 
-	filesUnreadable, emptyPaths, err := v.createNamedSnapshots(
+	filesUnreadable, emptyPaths, unmountedPaths, err := v.createNamedSnapshots(
 		opts, hostname, snapshotNames)
 	if err != nil {
 		return err
@@ -155,8 +167,10 @@ func (v *Vaultik) CreateSnapshot(opts *SnapshotCreateOptions) error {
 	}
 
 	switch {
-	case opts.Prune && (filesUnreadable > 0 || len(emptyPaths) > 0):
-		return v.skipPostBackupPrune(filesUnreadable, emptyPaths)
+	case opts.Prune && (filesUnreadable > 0 || len(emptyPaths) > 0 ||
+		len(unmountedPaths) > 0):
+		return v.skipPostBackupPrune(
+			filesUnreadable, emptyPaths, unmountedPaths)
 	case opts.Prune:
 		err = v.runPostBackupPrune(snapshotNames, opts.KeepNewerThan)
 		if err != nil {
@@ -169,20 +183,22 @@ func (v *Vaultik) CreateSnapshot(opts *SnapshotCreateOptions) error {
 
 // createNamedSnapshots creates each named snapshot in turn. It returns
 // the number of entries --skip-errors skipped because it could not read
-// or list them, and each configured path in which the scan found
-// nothing, named with its snapshot.
+// or list them, each configured path in which the scan found nothing,
+// and, under --prune, each directory where an earlier snapshot found a
+// volume mounted and this run did not, the paths named with their
+// snapshot.
 func (v *Vaultik) createNamedSnapshots(
 	opts *SnapshotCreateOptions, hostname string, snapshotNames []string,
-) (int, []string, error) {
+) (int, []string, []string, error) {
 	filesUnreadable := 0
 
-	var emptyPaths []string
+	var emptyPaths, unmountedPaths []string
 
 	for snapIdx, snapName := range snapshotNames {
 		stats, err := v.createNamedSnapshot(
 			opts, hostname, snapName, snapIdx+1, len(snapshotNames))
 		if err != nil {
-			return 0, nil, err
+			return 0, nil, nil, err
 		}
 
 		filesUnreadable += stats.totalFilesUnreadable
@@ -190,21 +206,27 @@ func (v *Vaultik) createNamedSnapshots(
 		for _, dir := range stats.emptyPaths {
 			emptyPaths = append(emptyPaths, dir+" of snapshot "+snapName)
 		}
+
+		for _, dir := range stats.unmountedPaths {
+			unmountedPaths = append(unmountedPaths,
+				dir+" of snapshot "+snapName)
+		}
 	}
 
-	return filesUnreadable, emptyPaths, nil
+	return filesUnreadable, emptyPaths, unmountedPaths, nil
 }
 
 // skipPostBackupPrune says why the post-backup prune does not run, naming
 // every cause, and returns the error the run ends with. An older snapshot
 // may be the only one that still holds a file this run could not read,
-// or what an empty path held. The previous snapshot is no guide: after
-// one run with an empty path it is empty too.
+// or what an empty path or a volume that is not mounted held. The
+// previous snapshot is no guide: after one such run it lacks them too.
 func (v *Vaultik) skipPostBackupPrune(
-	filesUnreadable int, emptyPaths []string,
+	filesUnreadable int, emptyPaths, unmountedPaths []string,
 ) error {
 	log.Warn("Skipping post-backup prune",
-		"files_unreadable", filesUnreadable, "empty_paths", emptyPaths)
+		"files_unreadable", filesUnreadable, "empty_paths", emptyPaths,
+		"unmounted_paths", unmountedPaths)
 
 	if len(emptyPaths) > 0 {
 		v.UI.Warningf("Skipped the post-backup prune because nothing was "+
@@ -212,6 +234,17 @@ func (v *Vaultik) skipPostBackupPrune(
 			strings.Join(emptyPaths, ", "))
 	}
 
+	if len(unmountedPaths) > 0 {
+		v.UI.Warningf("Skipped the post-backup prune because this backup "+
+			"did not find a mount point at %s, where an earlier snapshot "+
+			"found one: the volume is not mounted, this backup found no "+
+			"entry in it, or the directory is no longer backed up. Older "+
+			"snapshots are kept. Mount the volume again, or, if that is on "+
+			"purpose, remove the older snapshots by hand with "+
+			"'vaultik snapshot purge'.",
+			strings.Join(unmountedPaths, ", "))
+	}
+
 	if filesUnreadable > 0 {
 		v.UI.Warningf("Skipped the post-backup prune because %d file(s) "+
 			"could not be read; older snapshots are kept.", filesUnreadable)
@@ -219,7 +252,11 @@ func (v *Vaultik) skipPostBackupPrune(
 		return fmt.Errorf("%d %w", filesUnreadable, ErrFilesUnreadable)
 	}
 
-	return fmt.Errorf("%d %w", len(emptyPaths), ErrSourcePathEmpty)
+	if len(emptyPaths) > 0 {
+		return fmt.Errorf("%d %w", len(emptyPaths), ErrSourcePathEmpty)
+	}
+
+	return fmt.Errorf("%d %w", len(unmountedPaths), ErrVolumeNotMounted)
 }
 
 // finishCreateSnapshot ends a run whose snapshots are all complete and
@@ -329,6 +366,10 @@ type snapshotStats struct {
 	// The configured paths in which the scan found nothing.
 	emptyPaths []string
 
+	// Under --prune, the directories where an earlier snapshot of the
+	// same name found a volume mounted and this one did not.
+	unmountedPaths []string
+
 	// The sizes of all blobs the snapshot references, set by
 	// finalizeSnapshotMetadata once snapshot_blobs is populated.
 	blobSize             int64
@@ -394,9 +435,57 @@ func (v *Vaultik) createNamedSnapshot(
 
 	v.printSnapshotSummary(snapshotID, snapshotStartTime, stats)
 
+	if opts.Prune {
+		stats.unmountedPaths, err = v.findUnmountedPaths(snapName, snapshotID)
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	return stats, nil
 }
 
+// findUnmountedPaths returns each directory that an earlier snapshot
+// named snapName in the local index recorded as a mount point and
+// snapshot snapshotID did not, as when the volume mounted there is not
+// mounted.
+func (v *Vaultik) findUnmountedPaths(
+	snapName, snapshotID string,
+) ([]string, error) {
+	found, err := v.Repositories.Snapshots.GetMountPoints(v.ctx, snapshotID)
+	if err != nil {
+		return nil, err
+	}
+
+	snapshots, err := v.Repositories.Snapshots.ListRecent(v.ctx, listRecentLimit)
+	if err != nil {
+		return nil, fmt.Errorf("listing snapshots: %w", err)
+	}
+
+	var unmounted []string
+
+	for _, s := range snapshots {
+		if s.ID.String() == snapshotID ||
+			parseSnapshotName(s.ID.String(), s.Hostname.String()) != snapName {
+			continue
+		}
+
+		recorded, err := v.Repositories.Snapshots.GetMountPoints(
+			v.ctx, s.ID.String())
+		if err != nil {
+			return nil, err
+		}
+
+		for _, dir := range recorded {
+			if !slices.Contains(found, dir) && !slices.Contains(unmounted, dir) {
+				unmounted = append(unmounted, dir)
+			}
+		}
+	}
+
+	return unmounted, nil
+}
+
 // previousSnapshotID returns the ID of the most recent completed snapshot
 // named snapName in the local index, or "" when there is none.
 func (v *Vaultik) previousSnapshotID(snapName string) (string, error) {

internal/vaultik/snapshot_prune_test.go

@@ -4,12 +4,15 @@ import (
 	"context"
 	"os"
 	"path/filepath"
+	"syscall"
 	"testing"
 
+	"github.com/spf13/afero"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
@@ -259,3 +262,152 @@ func TestBackupWithNoEmptySourcePathPrunes(t *testing.T) {
 		})
 	}
 }
+
+// automountFs is the real filesystem with a volume at volumeDir that is
+// mounted on first access, as by an autofs map that shows its mount
+// points before mounting them: listing volumeDir mounts the volume or,
+// while cannotMount, fails with "no such file or directory".
+type automountFs struct {
+	afero.OsFs
+
+	volumeDir   string
+	cannotMount bool
+	mounted     bool
+}
+
+//nolint:ireturn // afero.Fs.Open is defined to return the interface.
+func (f *automountFs) Open(name string) (afero.File, error) {
+	if name == f.volumeDir && f.cannotMount {
+		return nil, &os.PathError{Op: "open", Path: name, Err: syscall.ENOENT}
+	}
+
+	if name == f.volumeDir {
+		f.mounted = true
+	}
+
+	return f.OsFs.Open(name)
+}
+
+// A backup under --prune in which a volume that an earlier snapshot found
+// mounted below the configured path is missing does not prune: run after
+// run, the snapshot that holds the volume's files is kept with everything
+// it stored, and the run fails. Once the volume is back, the prune runs.
+// The volume is mounted on first access and every backup starts with it
+// not mounted, so its directory is on its parent's device until the walk
+// lists it. While the volume is missing, its directory is left empty, or
+// listing it fails with "no such file or directory", as when an autofs
+// volume cannot be mounted, which the walk counts as a directory removed
+// during the backup. See https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestBackupWithVolumeNotMountedDoesNotPrune(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	tests := []struct {
+		name         string
+		listingFails bool
+	}{
+		{name: "directory left empty", listingFails: false},
+		{name: "listing fails", listingFails: true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			backUpWithVolumeNotMounted(t, tt.listingFails)
+		})
+	}
+}
+
+// backUpWithVolumeNotMounted runs the backups of
+// TestBackupWithVolumeNotMountedDoesNotPrune. While the volume is
+// missing, listing its directory fails if listingFails; otherwise the
+// directory is empty.
+func backUpWithVolumeNotMounted(t *testing.T, listingFails bool) {
+	t.Helper()
+
+	// volume is the directory, below the configured path, where the
+	// volume is mounted.
+	const (
+		snapshotName = "data"
+		volume       = "nas"
+	)
+
+	ctx := context.Background()
+	env := newDeletedCountEnv(t, map[string][]byte{
+		volume + "/a.txt": []byte("file a"),
+		"notes.txt":       []byte("notes"),
+	})
+	env.setSnapshot(snapshotName, "")
+	prune := &vaultik.SnapshotCreateOptions{
+		Prune: true, Snapshots: []string{snapshotName},
+	}
+
+	// The volume's directory is on the volume's device, 2, while the
+	// volume is mounted. Everything else is on device 1.
+	volumeDir := filepath.Join(env.src, volume)
+	fs := &automountFs{volumeDir: volumeDir}
+	newScanner := env.v.ScannerFactory
+	env.v.ScannerFactory = func(params snapshot.ScannerParams) *snapshot.Scanner {
+		fs.mounted = false
+		params.Fs = fs
+		params.DeviceNumber = func(info os.FileInfo) (uint64, bool) {
+			if info.Name() == volume && fs.mounted {
+				return 2, true
+			}
+
+			return 1, true
+		}
+
+		return newScanner(params)
+	}
+
+	env.backUp(t, snapshotName)
+
+	firstID := localSnapshotID(ctx, t, env.v.Repositories, snapshotName)
+	stored, err := env.v.Storage.List(ctx, "")
+	require.NoError(t, err)
+
+	// The volume goes missing.
+	fs.cannotMount = listingFails
+
+	if !listingFails {
+		require.NoError(t, os.RemoveAll(volumeDir))
+		require.NoError(t, os.Mkdir(volumeDir, 0o750))
+	}
+
+	for range 2 {
+		env.out.Reset()
+
+		err := env.v.CreateSnapshot(prune)
+		require.ErrorIs(t, err, vaultik.ErrVolumeNotMounted)
+		assert.Contains(t, env.out.String(),
+			"did not find a mount point at "+volumeDir+" of snapshot "+snapshotName)
+		assert.Contains(t, env.out.String(), "'vaultik snapshot purge'")
+		assert.NotContains(t, env.out.String(), "Running post-backup prune.")
+	}
+
+	snapshots, err := env.v.Repositories.Snapshots.ListRecent(
+		ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	require.Len(t, snapshots, 3)
+	assert.Equal(t, firstID, snapshots[2].ID.String(),
+		"the snapshot that holds the volume's files is kept")
+
+	for _, key := range stored {
+		_, err := env.v.Storage.Stat(ctx, key)
+		require.NoError(t, err, "%s is kept", key)
+	}
+
+	// The volume is back.
+	fs.cannotMount = false
+
+	require.NoError(t, os.WriteFile(
+		filepath.Join(volumeDir, "a.txt"), []byte("file a"), 0o600))
+	require.NoError(t, env.v.CreateSnapshot(prune))
+
+	snapshots, err = env.v.Repositories.Snapshots.ListRecent(
+		ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	require.Len(t, snapshots, 1, "the older snapshots are pruned")
+}