Keep unreferenced blobs newer than prune_grace_period (closes #387)

A backup uploads its blobs before the manifest that lists them, and the
process lock covers one user on one machine, so a prune from another host
or user sharing the destination deleted a running backup's blobs.

Prune now deletes an unreferenced blob only once its listed time is older
than the new config key prune_grace_period (default 24h; a negative value
is refused). A newer blob, or one listed with no time, is kept and counted
(blobs_too_new). The rclone backend takes each object's time from the
listing (use_server_modtime).

Decision: a blob a backup reuses from an earlier run that stopped before
its manifest keeps its old time and is not protected; the docs and config
comments say so.

Model: opus-5-5

ARCHITECTURE.md

@@ -279,8 +279,8 @@ Key methods:
 - `Scan(ctx, path, snapshotID)` → Main scan operation
 
 Scan phases:
-1. **Phase 0**: Detect deleted files from previous snapshots
-2. **Phase 1**: Walk directory, identify files needing processing
+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)
 
 #### SnapshotManager
@@ -311,6 +311,8 @@ Repository interfaces:
 
 ```
 CreateSnapshot(opts)
+    │
+    ├─► markDestination()   // Refuse a destination without the marker, then write it
     │
     ├─► PruneDatabase()   // Critical: avoid dedup errors
     │
@@ -320,13 +322,13 @@ CreateSnapshot(opts)
     │       │
     │       ├─► scanner.Scan(ctx, path, snapshotID)
     │       │       │
-    │       │       ├─► Phase 0: detectDeletedFiles()
-    │       │       │
     │       │       ├─► 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()
@@ -357,7 +359,7 @@ CreateSnapshot(opts)
 
 ## Deduplication Strategy
 
-1. **File-level**: Files unchanged since last backup are skipped (metadata comparison: size, mtime, mode, uid, gid), unless the file lists a chunk that no uploaded blob holds; such a file is re-chunked
+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
 
 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.
 
@@ -367,6 +369,8 @@ CreateSnapshot(opts)
 
 ```
 bucket/
+├── vaultik-destination.txt          # Marker written by every backup
+│
 ├── blobs/
 │   └── {hash[0:2]}/
 │       └── {hash[2:4]}/
@@ -378,6 +382,16 @@ bucket/
         └── manifest.json.zst        # Blob list (for pruning/verification)
 ```
 
+A destination without `vaultik-destination.txt` 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. The `file://`
+backend never creates its destination directory, so a write to a missing one
+fails.
+
 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

Dockerfile

@@ -5,7 +5,16 @@
 FROM golangci/golangci-lint:v2.14.0@sha256:ad862ba6b3798cbe0fd9fd7408d498fd74fbd2623a92406b2fd3898faf0bf98f AS lint
 WORKDIR /src
 COPY go.mod go.sum ./
-RUN go mod download
+# Without the time limit a stalled connection to the module proxy holds
+# this step, and every build on the host sharing it, until the kernel
+# drops the connection, which can take an hour. The limit is well above
+# the three minutes a working download can take on a busy build host.
+RUN for attempt in 1 2 3; do \
+        timeout 300 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 300s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 # `golangci-lint run` silently ignores an unknown top-level key in
 # .golangci.yml, such as a misspelt `linters:`; `config verify` fails on it.
@@ -18,7 +27,13 @@ RUN golangci-lint run --config .golangci.yml ./...
 FROM golang:1.26.1@sha256:cd78d88e00afadbedd272f977d375a6247455f3a4b1178f8ae8bbcb201743a8a AS test
 WORKDIR /src
 COPY go.mod go.sum ./
-RUN go mod download
+# Limited and retried as in the lint phase above.
+RUN for attempt in 1 2 3; do \
+        timeout 300 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 300s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 RUN go test -timeout 90s -race -cover ./... || \
     { echo "--- Rerunning with -v for details ---"; \
@@ -36,7 +51,13 @@ RUN apk add --no-cache git
 RUN git config --system --add safe.directory /src
 WORKDIR /src
 COPY go.mod go.sum ./
-RUN go mod download
+# Limited and retried as in the lint phase above.
+RUN for attempt in 1 2 3; do \
+        timeout 300 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 300s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 
 # The VERSION build arg when one is given, otherwise

README.md

@@ -24,6 +24,10 @@ grep 'public key' vaultik_backup_private_key.txt
 vaultik config set age_recipients.0 age1YOUR_PUBLIC_KEY_HERE
 vaultik config set storage_url "file:///Volumes/usbstick/mybackup"
 
+# create the destination directory; vaultik never creates it, so a backup
+# to a volume that is not plugged in fails instead of filling the disk
+mkdir /Volumes/usbstick/mybackup
+
 # macOS only: grant your terminal app Full Disk Access first
 # (System Settings → Privacy & Security → Full Disk Access), otherwise
 # the backup will abort with a permission error on protected directories
@@ -92,6 +96,14 @@ vaultik snapshot restore <snapshot-id> ~/vaultik-restore
 # 0 3 * * * vaultik snapshot create --cron --prune --keep-newer-than 4w
 ```
 
+On a destination shared between hosts or users, the prune in that cron line
+relies on `prune_grace_period` (default `24h`) to leave the blobs of a backup
+another one is still running, so it must be longer than the longest backup
+run. It does not protect a blob that backup reuses from an earlier run that
+stopped before its manifest: the prune can delete such a blob, and the new
+snapshot then cannot be restored. To not depend on the grace period, give each
+host its own prefix or directory. See [locking](#locking).
+
 ## restoring on another machine
 
 Restoring on a host that never ran the backup — a replacement machine
@@ -178,7 +190,7 @@ vaultik version
 * `--verbose`, `-v`: Enable verbose output (on stderr — see below)
 * `--debug`: Enable debug output (on stderr — see below)
 * `--quiet`, `-q`: Suppress non-error output (also suppresses startup banner)
-* `--skip-errors`: Skip files that cannot be read when creating a snapshot, or that cannot be restored when restoring, instead of aborting. Packing and storage errors while creating a snapshot (which would leave a chunk recorded but not stored) still abort the run.
+* `--skip-errors`: Skip files that cannot be read when creating a snapshot, or that cannot be restored when restoring, instead of aborting. Packing and storage errors while creating a snapshot (which would leave a chunk recorded but not stored) still abort the run. A run that skipped anything still finishes but exits non-zero: `snapshot create` keeps the snapshot and counts each file or directory it skipped as failed. A file removed during the backup is also counted as failed but does not change the exit status.
 
 ### locking
 
@@ -188,7 +200,22 @@ process-wide lock at `$XDG_DATA_HOME/vaultik/vaultik.pid`
 (`~/.local/share/vaultik/vaultik.pid` on Linux) for the whole run. Only
 one of them runs at a time: a second one exits immediately with an
 "already running" error rather than waiting, so two writers can never
-corrupt the local index or the destination store.
+corrupt the local index.
+
+The lock covers one user on one machine. It does not stop another user,
+or another host, from writing to the same destination store at the same
+time. A backup uploads its blobs before the manifest that lists them, so a
+prune from elsewhere at that moment finds blobs no snapshot lists. It
+deletes such a blob only once it is older than `prune_grace_period`
+(default `24h`), which therefore must be longer than the longest backup run
+of any host or user sharing the destination. The grace period does not
+protect a blob a backup reuses from an earlier run that stopped before its
+manifest: a backup skips the upload of a blob already on the destination, so
+that blob keeps the time of its first upload, and a prune from elsewhere
+during the backup can delete it; the new snapshot then cannot be restored.
+`remote nuke` deletes everything regardless. To not depend on the grace
+period, give each host or user its own prefix (or directory) on the
+destination.
 
 Read-only commands — `info`, `snapshot list`, `snapshot verify`, and
 `remote info` — do not take the lock and are never blocked, so they run
@@ -271,8 +298,17 @@ existing file. Created with mode `0600` since it will contain credentials.
 **`config set`**: Set a scalar config value by dotted YAML path
 (e.g. `vaultik config set compression_level 9`,
 `vaultik config set storage_url "file:///mnt/backups"`). Comments and
-formatting in the file are preserved; intermediate maps are created as
-needed.
+formatting in the file are preserved. Missing keys along the path are
+created: as a list when the next path element is a number (so
+`vaultik config set age_recipients.0 age1...` works on a file without
+`age_recipients`), otherwise as a map. If the config loads, a change
+after which it would not, such as `vaultik config set compression_level 99`,
+is refused and the file is left unchanged. A config that does not load,
+such as the one `config init` writes before `storage_url` is set or one
+with a `${ENV:...}` reference to a variable not set in this shell, is
+checked one setting at a time: only a change after which the changed
+setting cannot be read, such as one giving it the wrong type (a single
+value where a list belongs or a word where a number belongs), is refused.
 
 **`snapshot create`**: Perform incremental backup of configured snapshots.
 * Optional snapshot names argument to create specific snapshots (default: all)
@@ -283,18 +319,28 @@ needed.
 * `--cron`: Silent on total success; warnings and errors are still printed
   (for crontab)
 * `--prune`: After backup, drop older snapshots of each backed-up name and
-  remove orphaned blobs from remote storage. By default keeps only the latest
-  snapshot per name; use `--keep-newer-than` for a rolling window.
+  remove orphaned blobs older than `prune_grace_period` from remote storage,
+  as `prune` does. By default keeps only the latest
+  snapshot per name; use `--keep-newer-than` for a rolling window. When
+  `--skip-errors` skipped a file or directory it could not read or list, the
+  prune does not run and every older snapshot is kept, since one of them may
+  be the only one that holds what was skipped; the command says so and exits
+  non-zero. A file removed during the backup does not stop the prune.
 * `--keep-newer-than <duration>`: With `--prune`, keep snapshots newer than
   this duration instead of only the latest (e.g. `4w`, `30d`, `6mo`, `1y`).
   Without `--prune`, or with a value that is not a valid duration, the
   command fails before the backup starts.
+* Fails before it writes anything when a `file://` destination directory
+  does not exist, or when the local index holds snapshots and the
+  destination has no `vaultik-destination.txt` (see
+  [remote storage layout](#remote-storage-layout))
 
 **`snapshot list`**: Show every snapshot known to this host — the union
 of the local index and the backup destination store — with timestamps
 and three sizes per snapshot (compressed remote size; total
-uncompressed chunk size; size of chunks newly referenced by that
-snapshot).
+uncompressed chunk size; size of the chunks that snapshot stored that
+were not stored before). Both chunk sizes are recorded when the
+snapshot's backup runs.
 
 Listing the destination store does **not** require the age secret key,
 so it works in vaultik's intended configuration, where the backed-up
@@ -307,24 +353,33 @@ only the hash, so hostname and snapshot name exist solely in the local
 index and in the encrypted per-snapshot database. Snapshots found only
 on the destination store are therefore listed as
 `<remote only:<abbreviated remote key>>` and show `<remote only>` in
-the uncompressed and "new chunk" columns, which can only be computed
-from the local index. Their timestamp and compressed size are real,
+the uncompressed and "new chunk" columns, which come only from the
+local index. Their timestamp and compressed size are real,
 read from the manifest.
 
 Snapshots in the local index with no counterpart on the destination
 store are reported below the table as drift, with the `vaultik prune`
-invocation that reconciles them.
+invocation that reconciles them. If the local index is bound to a
+different destination than the configured one (after a change to
+`storage_url`, or `s3.endpoint`, `s3.bucket` or `s3.prefix`), the
+command says so instead and gives the same two ways out that `prune`
+gives: revert the configuration, or run `vaultik database delete`.
 
 If the destination store cannot be listed (unmounted volume,
 permission denied, network down), the command warns, falls back to the
-local index alone, and still exits zero.
+local index alone, and still exits zero. It does the same when the
+destination has no `vaultik-destination.txt` while the local index
+holds snapshots stored there, as when a volume mounted at the
+destination is unplugged and leaves its empty mount point behind.
 * `--json`: Output in JSON format. Each entry carries `locally_tracked`
   (whether the snapshot is in the local index), `remote_key` (the full
   64-character storage key), and `remote_present` (whether it was seen
   on the destination store, or `null` if the destination could not be
-  listed). Warnings about an unlistable destination, unreadable
-  manifests, and a truncated listing all go to stderr through the
-  logger, so stdout stays a single parseable document.
+  listed). Warnings about an unlistable destination, a local index
+  bound to a different destination (with both destinations and the
+  same two ways out), unreadable manifests, and a truncated listing all
+  go to stderr through the logger, so stdout stays a single parseable
+  document.
 
 **`snapshot verify`**: Verify snapshot integrity.
 * Default (shallow): checks that every blob the manifest lists is present in
@@ -350,16 +405,21 @@ snapshot from the local index and strips the snapshot's metadata from
 the backup destination store. Blobs are NOT touched — deleting blobs
 requires reading every remaining remote manifest (the destination store
 may hold snapshots this host doesn't know about), which is what
-`vaultik prune` does. On success the command prints the exact `vaultik
-prune` invocation to run as a follow-up. An ID that names no snapshot in
-the local index or on the destination store is an error; with
-`--local-only`, the snapshot must be in the local index. Local row
+`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;
+with `--local-only`, the snapshot must be in the local index. Local row
 cleanup (files, chunks, blobs the snapshot was the last referrer for)
-runs automatically. If the destination store is unreachable, the local-DB
-removal still completes and a warning is emitted; run `vaultik snapshot
-remove <snapshot-id>` again once the store is reachable to remove the
-snapshot's metadata from it (`vaultik prune` does not). To wipe everything
-on the destination in one go, use `vaultik remote nuke --force`.
+runs automatically. If removing the snapshot's metadata from the
+destination store fails (for example because the store is unreachable,
+or the local index holds snapshots and the destination has no
+`vaultik-destination.txt`), the local-DB removal still completes and a
+warning is emitted; run `vaultik snapshot remove <snapshot-id>` again
+once the store is reachable to remove the snapshot's metadata from it
+(`vaultik prune` does not). For a snapshot that is only on the
+destination store, that failure is an error. To wipe everything on the
+destination in one go, use `vaultik remote nuke --force`.
 * `--local-only`: Skip remote cleanup; only touch the local index
 * `--dry-run`: Show what would be deleted without deleting
 * `--force`: Skip confirmation prompt
@@ -372,10 +432,14 @@ on the destination in one go, use `vaultik remote nuke --force`.
   store — its remote key (or an unambiguous leading part of it) as shown
   by `snapshot list`. See
   [restoring on another machine](#restoring-on-another-machine).
-* Optional path arguments to restore specific files/directories (default: all)
+* Optional path arguments to restore specific files/directories (default: all).
+  A path that matches nothing in the snapshot is named in a warning, and the
+  command exits non-zero after restoring the paths that match.
 * Preserves file permissions, timestamps, ownership (ownership requires root),
   symlinks, and empty directories
-* `--verify`: After restoring, verify every file's chunk hashes match
+* `--verify`: After restoring, verify every restored file's chunk hashes
+  match. A file skipped under `--skip-errors` is reported as failed to
+  restore and is not verified.
 
 **`prune`**: Tidy up everything that isn't needed. Runs three passes:
 (1) reconcile the local index against the destination store — any
@@ -383,13 +447,28 @@ local snapshot whose remote metadata is missing is dropped from the
 local index; (2) delete orphaned local rows (files, chunks, blobs no
 longer referenced by any completed snapshot); (3) list every remote
 manifest on the destination store to compute the still-referenced blob
-set and delete any blob not in that set. Step (3) reads all remote
-manifests — network cost scales with the number of snapshots. `snapshot
-create --prune` runs the same cleanup automatically; this is the
-manual entry point for the same work.
+set and delete any blob not in that set that is older than
+`prune_grace_period` (default `24h`). A newer one is kept and counted: it
+may belong to a backup still running, on this host or another sharing the
+destination store, that has not yet uploaded its manifest. The grace period
+does not protect a blob such a backup reuses from an earlier run that stopped
+before its manifest: that blob keeps the time of its first upload, so it can
+be deleted. Step (3) reads all remote manifests — network cost scales with
+the number of snapshots.
+`snapshot create --prune` runs the same cleanup automatically; this is the
+manual entry point for the same work. If the destination store cannot
+be listed, or the local index holds snapshots and the destination has
+no `vaultik-destination.txt` (an unplugged volume's empty mount point),
+prune fails and deletes nothing. `snapshot purge` does the same. If a
+blob cannot be deleted, the command (and `snapshot create --prune`)
+exits non-zero.
 * `--force`: Skip confirmation prompt
 * `--json`: Output stats as JSON. Also skips the confirmation prompt, as
-  `--force` does.
+  `--force` does. A blob that cannot be deleted is counted in
+  `blobs_failed`, and the document is printed before the non-zero exit.
+  A blob kept for being newer than `prune_grace_period`, or because the
+  listing of the destination gives no time for it, is counted in
+  `blobs_too_new` and not in `blobs_found`.
 
 **`info`**: Display system configuration, storage settings, encryption
 recipients, and local database statistics.
@@ -408,10 +487,18 @@ under `blobs/` whose name is not a blob hash is not counted as a blob,
 since `prune` never deletes it.
 * `--json`: Output as JSON
 
-**`remote nuke`**: Delete every snapshot's metadata and every blob from the
-backup destination store, leaving the bucket prefix empty. Destructive and
-irreversible. This is the single supported way to wipe the entire
-destination store.
+**`remote nuke`**: Delete every object under `metadata/` and `blobs/` on
+the backup destination store, whatever its name, and then
+`vaultik-destination.txt`, leaving the bucket prefix empty. Destructive
+and irreversible. This is the single supported way to wipe the entire
+destination store. A snapshot is removed from the local index only once
+its metadata is deleted from the destination store. If the destination
+store cannot be listed, or the local index holds snapshots and the
+destination has no `vaultik-destination.txt` (an unplugged volume's
+empty mount point), the command fails before it removes anything. If a
+listing afterwards still finds anything under `metadata/` or `blobs/`,
+the command names each object left and exits non-zero; running it again
+deletes what is left.
 * `--force`: Required to confirm destruction.
 
 **`database delete`**: Delete the local SQLite state database file
@@ -433,7 +520,15 @@ object store. Credentials are read from `s3.access_key_id` and
 `s3.secret_access_key` in the config file.
 
 **Local filesystem** (`file:///path/to/backup`): Stores blobs and metadata on
-a local or mounted filesystem. Useful for testing or backing up to a NAS.
+a local or mounted filesystem. Useful for testing or backing up to a NAS. The
+path must be absolute; a relative one is refused when the config loads.
+Percent-escapes in the path are decoded, so `my%20stick` and `my stick` name
+the same directory. The directory must exist before the first backup: vaultik
+never creates it. A directory on a removable volume is gone while the volume is
+not mounted, so a backup then fails instead of writing to the disk under the
+mount point. The mount point itself stays behind when the volume is unplugged:
+there only `vaultik-destination.txt` stops such a backup, and only while the
+local index holds snapshots.
 
 **Rclone** (`rclone://remote/path`): Uses rclone's 70+ supported cloud
 providers. Requires rclone to be configured separately (`rclone config`).
@@ -448,6 +543,14 @@ being written, a killed upload can leave a truncated object under its name,
 which a later backup takes for complete. A leftover `.partial` file is ignored
 and can be deleted.
 
+A destination shared between hosts or users, each with its own local index,
+relies on `prune_grace_period` to keep one's prune from deleting the blobs of
+another's backup still running, so it must be longer than the longest backup
+run. It does not protect a blob that backup reuses from an earlier run that
+stopped before its manifest, which the prune can delete. To not depend on the
+grace period, give each host its own prefix (or directory). See
+[locking](#locking).
+
 Legacy S3 configuration via `s3.*` fields (endpoint, bucket, prefix, etc.) is
 still supported for backward compatibility. `storage_url` takes precedence if
 both are set.
@@ -464,6 +567,7 @@ For an implementation-level view of the internals — the data model, the
 
 ```
 <bucket>/<prefix>/
+├── vaultik-destination.txt     # Marker written by every backup
 ├── blobs/
 │   └── <aa>/<bb>/<full_blob_hash>
 └── metadata/
@@ -472,6 +576,16 @@ For an implementation-level view of the internals — the data model, the
         └── manifest.json.zst   # Unencrypted blob list (for pruning)
 ```
 
+* `vaultik-destination.txt` holds one fixed line of text, with no host
+  name, path or time. A destination without it is treated as not
+  present while the local index holds snapshots stored there:
+  `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 an
+  unplugged volume's empty mount point from being taken for a
+  destination whose snapshots are all gone. Mount
+  the volume if it is removable; if the destination was emptied on
+  purpose, run `vaultik database delete` and back up again.
 * Blobs are two-level directory sharded using the first 4 hex chars of the blob hash
 * `db.zst.age` is a binary SQLite database (zstd compressed, age encrypted)
   containing all file metadata, chunk mappings, and relationships for the snapshot
@@ -523,7 +637,8 @@ format does and does not protect.
 1. List all snapshot manifests
 2. Build set of all referenced blob hashes
 3. List all blobs in storage
-4. Delete any blob not in the referenced set
+4. Delete any blob not in the referenced set that is older than
+   `prune_grace_period`
 
 ### chunking and deduplication
 
@@ -568,6 +683,7 @@ complete annotated example also lives in
 | `chunk_size` | `10MB` | Average chunk size for content-defined chunking |
 | `blob_size_limit` | `10GB` | Maximum blob size before splitting. Must be at least four times `chunk_size` (the largest chunk the chunker can emit), otherwise a single-chunk blob could exceed the limit. 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` |
 | `compression_level` | `3` | zstd compression level (1-19) |
+| `prune_grace_period` | `24h` | How old a blob that no snapshot lists must be before `prune` deletes it. A backup uploads its blobs before the manifest that lists them, so on a destination shared between hosts or users this must be longer than the longest backup run (see [locking](#locking)). It does not protect a blob a backup reuses from an earlier run that stopped before its manifest: that blob keeps the time of its first upload, so a prune can delete it. A duration such as `36h` or `90m`; `0s` deletes such blobs at once |
 | `hostname` | system hostname | Hostname used in snapshot IDs |
 | `index_path` | platform data dir | Local SQLite index path |
 
@@ -691,8 +807,10 @@ the `NO_COLOR` environment variable is unset (https://no-color.org/).
   reports and tables (`version`, `info`, `remote info`, `snapshot
   verify`, `snapshot list`) but never the machine-consumed `config get`
   value or the `--json` documents, which a script depends on. The
-  `database delete` confirmation prompt is also written this way and
-  always shown: it is an interactive exchange the operator must see.
+  confirmation prompts of `database delete`, `prune`, `snapshot purge`
+  and `snapshot remove`, each with what it asks to delete, are also
+  written this way and always shown: they are an interactive exchange
+  the operator must see.
 
 `internal/ui` writes to stdout; it is the output the user asked for. The
 exceptions are the startup banner and the error a failed command ends

TODO.md

@@ -15,13 +15,229 @@ pre-1.0
 # Next Step
 
 The 1.0 work is complete on `next`: the scope settled on
-[issue #125](https://git.eeqj.de/sneak/vaultik/issues/125) was the work
-already planned for 1.0, and all of it has landed. The mechanism to cut
-the tag exists and is exercised; what is left is merging `next` to
-`main` and tagging, both the owner's.
+[issue #125](https://git.eeqj.de/sneak/vaultik/issues/125) has landed,
+and so has every fix from the defect hunt before the release candidate
+([issue #288](https://git.eeqj.de/sneak/vaultik/issues/288)). What is
+left is merging `next` to `main` and tagging the release, both the
+owner's; a `v*` tag runs the release build in
+`.gitea/workflows/release.yml`.
 
 # Completed Steps
 
+- 2026-10-14: Stopped `snapshot create --prune` pruning after a backup
+  that skipped files it could not read
+  ([issue #382](https://git.eeqj.de/sneak/vaultik/issues/382)). Such a
+  prune could delete the only snapshot still holding those files.
+
+- 2026-10-14: Made `snapshot create --skip-errors` exit non-zero,
+  instead of 0, when it skips anything it cannot read
+  ([issue #310](https://git.eeqj.de/sneak/vaultik/issues/310)).
+
+- 2026-10-14: Kept the local index when `remote nuke` cannot list or
+  find the destination store
+  ([issue #292](https://git.eeqj.de/sneak/vaultik/issues/292)). Nuke
+  had cleared the local snapshot records anyway; it now fails before
+  removing anything.
+
+- 2026-10-14: Stopped counting directories and unchanged symlinks as
+  files to back up
+  ([issue #305](https://git.eeqj.de/sneak/vaultik/issues/305)).
+
+- 2026-10-14: Refused a backup destination that an unplugged volume
+  left empty, as the owner ruled
+  ([issue #239](https://git.eeqj.de/sneak/vaultik/issues/239)). Every
+  backup now writes `vaultik-destination.txt` at the destination root,
+  and while the local index holds snapshots stored there, a destination
+  without it is treated as not present.
+
+- 2026-10-14: Warned about restored ownership only when an entry's
+  owner or group differs from the recorded one
+  ([issue #316](https://git.eeqj.de/sneak/vaultik/issues/316)); a
+  restore by a non-root user had warned even when none did.
+
+- 2026-10-14: Made `snapshot list` report a changed destination instead
+  of advising `prune`, which refuses in that case
+  ([issue #317](https://git.eeqj.de/sneak/vaultik/issues/317)).
+
+- 2026-10-14: Counted a file as deleted by a backup only if the
+  previous snapshot of the same name listed it
+  ([issue #304](https://git.eeqj.de/sneak/vaultik/issues/304)); files
+  that snapshot did not list had been counted too.
+
+- 2026-10-14: Named each prefix once in restore, verify and purge
+  errors, some of which repeated it
+  ([issue #327](https://git.eeqj.de/sneak/vaultik/issues/327)). A deep
+  verify log line that named S3 whatever the destination now names
+  storage.
+
+- 2026-10-14: Refused credentials in a storage URL's endpoint or rclone
+  remote name, which were accepted and could then be printed
+  ([issue #322](https://git.eeqj.de/sneak/vaultik/issues/322)).
+
+- 2026-10-14: Ran `$EDITOR` through the shell in `config edit`, so a
+  value with arguments, such as `code --wait`, no longer fails
+  ([issue #315](https://git.eeqj.de/sneak/vaultik/issues/315)).
+
+- 2026-10-14: Made `snapshot remove` fail, instead of exiting 0, when
+  it cannot delete the metadata of a snapshot that is only on the
+  destination store
+  ([issue #340](https://git.eeqj.de/sneak/vaultik/issues/340)). The
+  README and the help now say when it prints the `vaultik prune`
+  follow-up.
+
+- 2026-10-14: Labelled the totals in the local index cleanup before
+  each backup as totals
+  ([issue #306](https://git.eeqj.de/sneak/vaultik/issues/306)). Each
+  was shown as `(N remain)`, which read as rows left to clean up.
+
+- 2026-10-13: Freed a cached blob once no file still to restore needs
+  it ([issue #324](https://git.eeqj.de/sneak/vaultik/issues/324)); one
+  also used by a file outside the selected paths had stayed in the
+  temporary cache until the restore ended.
+
+- 2026-10-13: Kept the whole startup error when it contains `): `
+  ([issue #320](https://git.eeqj.de/sneak/vaultik/issues/320)). Such
+  an error was cut at its last `): `, so the `blob_size_limit` one lost
+  the setting's name.
+
+- 2026-10-13: Silenced the status lines of `prune` and snapshot removal
+  under `--cron` and `--quiet`
+  ([issue #294](https://git.eeqj.de/sneak/vaultik/issues/294)).
+
+- 2026-10-13: Read the rclone config file that `RCLONE_CONFIG` names
+  ([issue #321](https://git.eeqj.de/sneak/vaultik/issues/321)); a
+  remote defined only there had been reported as not found.
+
+- 2026-10-13: Made `/**/` in an exclude pattern match zero directories,
+  as in gitignore
+  ([issue #338](https://git.eeqj.de/sneak/vaultik/issues/338));
+  `a/**/*.log` had missed `a/x.log`.
+
+- 2026-10-13: Made the `chunk_size` and `blob_size_limit` defaults and
+  the `chunk_size` minimum 1000-based, like the size parser and the
+  docs ([issue #313](https://git.eeqj.de/sneak/vaultik/issues/313)).
+
+- 2026-10-13: Kept a snapshot's remote metadata when `snapshot purge`
+  cannot remove it from the local index
+  ([issue #337](https://git.eeqj.de/sneak/vaultik/issues/337)); the
+  purge had deleted the metadata anyway and exited 0.
+
+- 2026-10-13: Made `snapshot restore` fail for a requested path that
+  matches nothing, which it had ignored
+  ([issue #309](https://git.eeqj.de/sneak/vaultik/issues/309)).
+
+- 2026-10-13: Made `remote nuke` delete every object under `metadata/`
+  and `blobs/`
+  ([issue #291](https://git.eeqj.de/sneak/vaultik/issues/291)); it had
+  left some, such as a stray `.DS_Store`, while reporting the store
+  empty.
+
+- 2026-10-13: Made `config set` create a list for an index under a
+  missing key
+  ([issue #308](https://git.eeqj.de/sneak/vaultik/issues/308)); it had
+  written a map, after which no command could load the config.
+
+- 2026-10-13: Listed a `file://` destination through a symlinked
+  `metadata/` or `blobs/`, which had listed as empty
+  ([issue #296](https://git.eeqj.de/sneak/vaultik/issues/296)). A link
+  to a missing target now fails the listing
+  ([issue #332](https://git.eeqj.de/sneak/vaultik/issues/332)).
+
+- 2026-10-13: Made `snapshot verify --deep --json` report its blob
+  counts ([issue #314](https://git.eeqj.de/sneak/vaultik/issues/314));
+  a failed run had reported 0 missing and 0 verified.
+
+- 2026-10-13: Limited and retried the module download in the Docker
+  build ([issue #351](https://git.eeqj.de/sneak/vaultik/issues/351)),
+  which had hung for as long as a stalled connection to the module
+  proxy lasted.
+
+- 2026-10-13: Recorded each snapshot's uncompressed size when its
+  backup runs
+  ([issue #360](https://git.eeqj.de/sneak/vaultik/issues/360)). A later
+  backup could change an older snapshot's UNCOMPRESSED SIZE in
+  `snapshot list`.
+
+- 2026-10-13: Stopped `snapshot restore --skip-errors --verify`
+  reporting a file that failed to restore again as a checksum failure
+  ([issue #323](https://git.eeqj.de/sneak/vaultik/issues/323)).
+
+- 2026-10-13: Recorded the stored size of a blob a backup finds already
+  stored, not the size of its local copy, which can differ, so
+  `snapshot verify` could fail on a good snapshot
+  ([issue #300](https://git.eeqj.de/sneak/vaultik/issues/300)).
+
+- 2026-10-13: Stopped the local index and the blob upload step ending
+  the process on a failed close of query results or a failed temp file
+  cleanup ([issue #345](https://git.eeqj.de/sneak/vaultik/issues/345));
+  both are now returned as errors.
+
+- 2026-10-13: Closed the snapshot database's storage reader too when
+  restore or deep verify is interrupted
+  ([issue #356](https://git.eeqj.de/sneak/vaultik/issues/356)).
+
+- 2026-10-13: Made `prune` and `remote nuke` fail when a delete on the
+  backup destination store fails
+  ([issue #290](https://git.eeqj.de/sneak/vaultik/issues/290)).
+
+- 2026-10-13: Made an S3 listing that the server cuts short an error
+  ([issue #297](https://git.eeqj.de/sneak/vaultik/issues/297)); `prune`
+  had taken it as complete and could delete blobs still in use.
+
+- 2026-10-13: Made shallow `snapshot verify` report a blob it could not
+  check apart from a missing one
+  ([issue #318](https://git.eeqj.de/sneak/vaultik/issues/318)); any
+  error checking a blob had counted as missing, its cause dropped.
+
+- 2026-10-13: Restored the other files when a file's chunk is in no
+  blob under `--skip-errors`
+  ([issue #326](https://git.eeqj.de/sneak/vaultik/issues/326)); that
+  file had stopped the restore before anything was restored.
+
+- 2026-10-13: Recorded each snapshot's new chunk size when its backup
+  runs ([issue #311](https://git.eeqj.de/sneak/vaultik/issues/311)).
+  `snapshot list` could show a NEW CHUNK SIZE of 0 B for a snapshot
+  whose change was a modified file.
+
+- 2026-10-13: Made a backup interrupted during a local index write exit
+  as interrupted
+  ([issue #301](https://git.eeqj.de/sneak/vaultik/issues/301)); it could
+  end with `FATAL: failed to rollback transaction` and exit 1 instead of
+  130.
+
+- 2026-10-13: Printed `snapshot list` and `snapshot purge` sizes like
+  every other command
+  ([issue #312](https://git.eeqj.de/sneak/vaultik/issues/312)). They,
+  and the `bytes` value in terminal log lines, were 1024-based but
+  labelled KB or MB, and now use 1000-based units.
+
+- 2026-10-13: Closed each blob's storage reader when restore or deep
+  verify is interrupted, which ends a stalled read unless it is stuck in
+  the kernel ([issue #325](https://git.eeqj.de/sneak/vaultik/issues/325)).
+
+- 2026-10-13: Made a mistyped subcommand under `snapshot`, `config`,
+  `remote`, `database` or `completion` an error
+  ([issue #319](https://git.eeqj.de/sneak/vaultik/issues/319)); each had
+  printed its help and exited 0.
+
+- 2026-10-13: Made an exclude pattern starting with `**/` match
+  directly under the snapshot path too, as in gitignore
+  ([issue #303](https://git.eeqj.de/sneak/vaultik/issues/303)).
+
+- 2026-10-13: Refused a relative `file://` storage URL and decoded its
+  percent-escapes
+  ([issue #299](https://git.eeqj.de/sneak/vaultik/issues/299)); a
+  relative path had moved with the working directory.
+
+- 2026-10-13: Removed the blob temp file when a backup stops part-way
+  ([issue #302](https://git.eeqj.de/sneak/vaultik/issues/302)); the
+  half-written blob had stayed in `$TMPDIR`.
+
+- 2026-10-13: Checked `--keep-newer-than` before a backup starts
+  ([issue #307](https://git.eeqj.de/sneak/vaultik/issues/307)); an
+  invalid value had been read only by the prune after the backup, so
+  retention never ran.
+
 - 2026-10-13: Stopped `remote info` counting a file under `blobs/` that
   is not a blob as an orphaned blob
   ([issue #295](https://git.eeqj.de/sneak/vaultik/issues/295)). A stray

cmd/vaultik/dockermoddownload_test.go

@@ -0,0 +1,45 @@
+package main_test
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDockerfileBoundsEveryModuleDownload fails unless every
+// `go mod download` in the Dockerfile runs under a time limit, is tried
+// a fixed number of times, and then fails the build with a message that
+// names the module download (issue #351). Like the version tests, this
+// parses the Dockerfile; that a stalled download really is cut off and
+// retried was checked by hand.
+func TestDockerfileBoundsEveryModuleDownload(t *testing.T) {
+	t.Parallel()
+
+	var downloads []string
+
+	for _, instruction := range instructions(t, productDockerfile) {
+		if strings.Contains(instruction, "go mod download") {
+			downloads = append(downloads, instruction)
+		}
+	}
+
+	require.NotEmpty(t, downloads, "%s must download modules",
+		productDockerfile)
+
+	for _, download := range downloads {
+		assert.Contains(t, download, "timeout 300 go mod download",
+			"%s: a module download must have a time limit: %s",
+			productDockerfile, download)
+		assert.Contains(t, download, "for attempt in 1 2 3",
+			"%s: a module download must be retried: %s",
+			productDockerfile, download)
+		assert.Contains(t, download, "module download failed",
+			"%s: a module download that never succeeds must say so: %s",
+			productDockerfile, download)
+		assert.True(t, strings.HasSuffix(download, "exit 1"),
+			"%s: a module download that never succeeds must fail the"+
+				" build: %s", productDockerfile, download)
+	}
+}

config.example.yml

@@ -302,7 +302,8 @@ storage_url: "rclone://myremote/path/to/backups"
 
 # Average chunk size for content-defined chunking
 # Smaller chunks = better deduplication but more metadata
-# Supports: 10MB, 5M, 1GB, 500KB, 64MiB, etc.
+# Minimum 1MB
+# Supports: 10MB, 5M, 1GB, 64MiB, etc.
 # Default: 10MB
 #chunk_size: 10MB
 
@@ -327,6 +328,18 @@ storage_url: "rclone://myremote/path/to/backups"
 # Higher = better compression but slower
 # Default: 3
 compression_level: 5
+
+# How old a blob that no snapshot lists must be before prune deletes it
+# A backup uploads its blobs before the manifest that lists them, so on a
+# destination shared between hosts or users this must be longer than the
+# longest backup run
+# It does not protect a blob a backup reuses from an earlier run that stopped
+# before its manifest: that blob keeps the time of its first upload, so a
+# prune can delete it
+# Supports: 36h, 90m, etc.; 0s deletes them at once
+# Default: 24h
+#prune_grace_period: 24h
+
 # Hostname to use in backup metadata
 # Default: system hostname
 #hostname: myserver

docs/DATAMODEL.md

@@ -90,7 +90,7 @@ Stores information about packed, compressed, and encrypted blob files.
 - `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
+- `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)
 
 **Purpose:** Tracks blob lifecycle from creation through upload. The UUID primary key allows immediate association of chunks with blobs.
@@ -119,8 +119,10 @@ Tracks backup snapshots.
 - `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
@@ -183,7 +185,7 @@ Tracks blob upload metrics.
    - `UPDATE blobs SET blob_hash = ?, finished_ts = ?` - Finalize blob after packing
 
 4. **Upload**
-   - `UPDATE blobs SET uploaded_ts = ?` - Mark blob as uploaded
+   - `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
 
@@ -226,7 +228,8 @@ The restore process doesn't use the local database. Instead:
 
 1. **Identify Unreferenced Blobs**
    - Query blobs not referenced by any remaining snapshot
-   - Delete from S3 and local database
+   - Delete from S3, once older than `prune_grace_period`, and from the local
+     database
 
 ### 6. Incomplete Snapshot Cleanup
 

docs/REPOSTRUCTURE.md

@@ -4,7 +4,7 @@ This document describes the structure and organization of data stored in the S3
 
 ## Overview
 
-Vaultik stores all backup data in an S3-compatible object store. The repository consists of two main components:
+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)
 
@@ -12,6 +12,7 @@ Vaultik stores all backup data in an S3-compatible object store. The repository
 
 ```
 <bucket>/<prefix>/
+├── vaultik-destination.txt
 ├── blobs/
 │   └── <hash[0:2]>/
 │       └── <hash[2:4]>/
@@ -24,6 +25,10 @@ Vaultik stores all backup data in an S3-compatible object store. The repository
 
 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).
 
+## Marker File (`vaultik-destination.txt`)
+
+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. A destination without it is treated as not present while the local index holds snapshots stored there: `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.
+
 ## Blobs Directory (`blobs/`)
 
 ### Structure
@@ -161,7 +166,7 @@ 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
 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. Prune requires exclusive access to the destination: running it during a concurrent backup can race a snapshot whose manifest is not yet written, so do not prune while a backup is in progress
+4. It deletes an unreferenced blob only once the blob is older than `prune_grace_period` (default `24h`). A backup uploads its blobs before the manifest that lists them, 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
 
 ## Restoration Requirements
 

go.mod

@@ -22,6 +22,7 @@ require (
 	github.com/spf13/cobra v1.10.1
 	github.com/spf13/pflag v1.0.10
 	github.com/stretchr/testify v1.11.1
+	go.uber.org/dig v1.19.0
 	go.uber.org/fx v1.24.0
 	golang.org/x/sync v0.18.0
 	golang.org/x/sys v0.38.0
@@ -256,7 +257,6 @@ require (
 	go.opentelemetry.io/otel/metric v1.38.0 // indirect
 	go.opentelemetry.io/otel/trace v1.38.0 // indirect
 	go.shabbyrobe.org/gocovmerge v0.0.0-20230507111327-fa4f82cfbf4d // indirect
-	go.uber.org/dig v1.19.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
 	go.uber.org/zap v1.27.0 // indirect
 	go.yaml.in/yaml/v2 v2.4.3 // indirect

internal/blob/packer.go

@@ -245,6 +245,17 @@ func (p *Packer) Flush(ctx context.Context) error {
 	return nil
 }
 
+// DiscardCurrentBlob closes and removes the temp file of the blob being
+// packed, if any, and drops that blob. A caller that stops before Flush
+// must call it, or the temp file is left behind. Thread-safe.
+func (p *Packer) DiscardCurrentBlob() {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+
+	p.cleanupTempFile()
+	p.currentBlob = nil
+}
+
 // FinalizeBlob finalizes the current blob being assembled.
 // This compresses the accumulated chunks, encrypts the result, and computes
 // the content-addressed hash. The finalized blob is either passed to the

internal/cli/app.go

@@ -8,12 +8,12 @@ import (
 	"errors"
 	"fmt"
 	"path/filepath"
-	"strings"
 	"sync"
 	"time"
 
 	"github.com/adrg/xdg"
 	"github.com/spf13/cobra"
+	"go.uber.org/dig"
 	"go.uber.org/fx"
 	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/database"
@@ -146,16 +146,12 @@ func (e *startupError) Error() string {
 //	could not build arguments for function "Y" (file:line): failed to build U:
 //	received non-nil error from function "Z" (file:line): <real error>
 //
-// Users care about the real error, not the DI plumbing. We strip everything
-// up through the last "): " (which is always the close-paren of an fx
-// function-location annotation followed by the wrapped error).
+// Users care about the real error, not the DI plumbing. dig.RootCause,
+// from the library fx is built on, unwraps those layers and returns the
+// real error whole. Cutting the text at "): " instead would also cut a
+// real error that contains it, such as the blob_size_limit one.
 func cleanStartupError(err error) error {
-	msg := err.Error()
-	if idx := strings.LastIndex(msg, "): "); idx >= 0 {
-		msg = msg[idx+3:]
-	}
-
-	return &startupError{msg: msg}
+	return &startupError{msg: dig.RootCause(err).Error()}
 }
 
 // RunApp starts the fx application, blocks until it is asked to stop, and

internal/cli/app_test.go

@@ -2,58 +2,65 @@ package cli //nolint:testpackage // needs access to unexported cleanStartupError
 
 import (
 	"errors"
+	"os"
 	"path/filepath"
 	"testing"
 
+	"go.uber.org/fx"
+	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/pidlock"
 )
 
+// TestCleanStartupError starts a real fx app whose config has a
+// blob_size_limit below the minimum. fx puts annotations naming the
+// functions it was building in front of the config error; only those
+// may be removed, although the config error itself contains "): ".
 func TestCleanStartupError(t *testing.T) {
 	t.Parallel()
 
-	tests := []struct {
-		name string
-		in   string
-		want string
-	}{
-		{
-			name: "real fx error chain",
-			in: `could not build arguments for function ` +
-				`"sneak.berlin/go/vaultik/internal/cli".newSnapshotCreateCommand.func1.1 ` +
-				`(/Users/user/dev/vaultik/internal/cli/snapshot.go:71): ` +
-				`failed to build *vaultik.Vaultik: ` +
-				`could not build arguments for function ` +
-				`"sneak.berlin/go/vaultik/internal/vaultik".New ` +
-				`(/Users/user/dev/vaultik/internal/vaultik/vaultik.go:59): ` +
-				`failed to build storage.Storer: ` +
-				`received non-nil error from function ` +
-				`"sneak.berlin/go/vaultik/internal/storage".NewStorer ` +
-				`(/Users/user/dev/vaultik/internal/storage/module.go:23): ` +
-				`creating base path: mkdir /Volumes/BACKUPS: permission denied`,
-			want: `creating base path: mkdir /Volumes/BACKUPS: permission denied`,
-		},
-		{
-			name: "no fx wrapping",
-			in:   "plain error",
-			want: "plain error",
-		},
-		{
-			name: "single fx wrapping",
-			in:   `received non-nil error from function "foo" (file.go:1): underlying problem`,
-			want: "underlying problem",
-		},
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte("snapshots:\n"+
+		"  test:\n"+
+		"    paths: [/tmp/vaultik-test-source]\n"+
+		"storage_url: file:///tmp/vaultik-test-storage\n"+
+		"chunk_size: 10MB\n"+
+		"blob_size_limit: 30MB\n"), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
 	}
 
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			t.Parallel()
+	app := fx.New(
+		fx.NopLogger,
+		fx.Supply(config.Path(path)),
+		config.Module,
+		fx.Invoke(func(*config.Config) {}),
+	)
 
-			//nolint:err113 // test constructs errors from table input
-			got := cleanStartupError(errors.New(tt.in)).Error()
-			if got != tt.want {
-				t.Errorf("got %q, want %q", got, tt.want)
-			}
-		})
+	err = app.Start(t.Context())
+	if err == nil {
+		t.Fatal("app started; want a blob_size_limit error")
+	}
+
+	want := "failed to load config: invalid config: blob_size_limit must " +
+		"be at least the largest chunk the chunker can emit (chunk_size " +
+		"times the FastCDC size spread): need at least 40000000 bytes"
+
+	got := cleanStartupError(err).Error()
+	if got != want {
+		t.Errorf("got %q, want %q", got, want)
+	}
+}
+
+// TestCleanStartupErrorNotFromFx checks that an error fx did not wrap is
+// shown unchanged.
+func TestCleanStartupErrorNotFromFx(t *testing.T) {
+	t.Parallel()
+
+	//nolint:err113 // a one-off error standing in for a startup failure
+	got := cleanStartupError(errors.New("plain error")).Error()
+	if got != "plain error" {
+		t.Errorf("got %q, want %q", got, "plain error")
 	}
 }
 

internal/cli/config.go

@@ -8,6 +8,7 @@ import (
 	"os/exec"
 	"path/filepath"
 	"reflect"
+	"slices"
 	"strconv"
 	"strings"
 	"unicode/utf8"
@@ -44,6 +45,9 @@ var (
 	errNeedNumericIndex = errors.New("key is a list; use a numeric index")
 	errIndexOutOfRange  = errors.New("index out of range")
 	errNotMapOrList     = errors.New("key is not a map or list")
+	errWouldNotLoad     = errors.New(
+		"vaultik would not load the config after this change, " +
+			"so the file was not changed")
 )
 
 const defaultConfigTemplate = `# vaultik configuration
@@ -220,7 +224,7 @@ storage_url: ""
 
 # Average chunk size for content-defined chunking (FastCDC).
 # Smaller = better deduplication but more metadata overhead.
-# Accepts: 1MB, 10M, 64KB, etc.
+# Minimum 1MB. Accepts: 1MB, 10M, 64MiB, etc.
 # Default: 10MB
 # chunk_size: 10MB
 
@@ -235,6 +239,16 @@ storage_url: ""
 # Default: 3
 # compression_level: 3
 
+# How old a blob that no snapshot lists must be before prune deletes it.
+# A backup uploads its blobs before the manifest that lists them, so on a
+# destination shared between hosts or users this must be longer than the
+# longest backup run. It does not protect a blob a backup reuses from an
+# earlier run that stopped before its manifest: that blob keeps the time of
+# its first upload, so a prune can delete it.
+# Accepts: 36h, 90m, etc. 0s deletes them at once.
+# Default: 24h
+# prune_grace_period: 24h
+
 # Hostname used in snapshot IDs. Default: system hostname.
 # hostname: myserver
 
@@ -251,6 +265,12 @@ func NewConfigCommand() *cobra.Command {
 		Use:   "config",
 		Short: "Manage the configuration file",
 		Long:  "Commands for creating, editing, and querying the vaultik config file.",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	cmd.AddCommand(newConfigInitCommand())
@@ -323,8 +343,15 @@ func newConfigEditCommand() *cobra.Command {
 				editor = "vi"
 			}
 
+			// Run $EDITOR through the shell, as git does, so a value
+			// with arguments such as "code --wait" works. The path
+			// goes in as an argument ("$@"), not into the command
+			// text, so a path with spaces or quotes arrives whole.
+			// sh -c takes the word after the command as $0, which
+			// is why editor appears twice.
 			//nolint:gosec // G204: launching the operator's own $EDITOR is the point
-			ed := exec.CommandContext(cmd.Context(), editor, path)
+			ed := exec.CommandContext(cmd.Context(),
+				"sh", "-c", editor+` "$@"`, editor, path)
 			ed.Stdin = os.Stdin
 			ed.Stdout = os.Stdout
 			ed.Stderr = os.Stderr
@@ -385,8 +412,17 @@ func newConfigSetCommand() *cobra.Command {
 		Use:   "set <key> <value>",
 		Short: "Set a config value by dotted path (e.g. compression_level 5)",
 		Long: `Sets a scalar config value addressed by dotted YAML path and writes
-the file back, preserving comments and formatting. Intermediate maps
-are created as needed.
+the file back, preserving comments and formatting. Missing keys along
+the path are created: as a list when the next path element is a
+number, otherwise as a map.
+
+If the config loads, a change after which it would not, such as
+compression_level 99, is refused and the file is left unchanged. A
+config that does not load, such as the one config init writes before
+storage_url is set or one with a ${ENV:...} reference to a variable
+not set here, is checked one setting at a time: only a change after
+which the changed setting cannot be read, such as one giving it the
+wrong type, is refused.
 
 Examples:
   vaultik config set storage_url "file:///mnt/backups"
@@ -416,7 +452,14 @@ func writeConfigSet(out *ui.Writer, path, key, value string) error {
 		return err
 	}
 
-	err = yamlPathSet(root, strings.Split(key, "."), value)
+	oldData, err := marshalConfigYAML(root)
+	if err != nil {
+		return fmt.Errorf("marshaling config: %w", err)
+	}
+
+	keys := strings.Split(key, ".")
+
+	err = yamlPathSet(root, keys, value)
 	if err != nil {
 		return err
 	}
@@ -426,6 +469,21 @@ func writeConfigSet(out *ui.Writer, path, key, value string) error {
 		return fmt.Errorf("marshaling config: %w", err)
 	}
 
+	setting, err := yamlPathOnly(root, keys)
+	if err != nil {
+		return err
+	}
+
+	settingData, err := marshalConfigYAML(setting)
+	if err != nil {
+		return fmt.Errorf("marshaling config: %w", err)
+	}
+
+	err = checkConfigChange(oldData, data, settingData)
+	if err != nil {
+		return fmt.Errorf("%w: %w", errWouldNotLoad, err)
+	}
+
 	err = os.WriteFile(path, data, configFileMode)
 	if err != nil {
 		return fmt.Errorf("writing config file: %w", err)
@@ -447,6 +505,58 @@ func writeConfigSet(out *ui.Writer, path, key, value string) error {
 	return nil
 }
 
+// checkConfigChange returns the reason vaultik would not load newData, the
+// config after a change, if the change brings that failure in. If oldData,
+// the config before it, loads, the new one must load too. A config that
+// does not load, such as one still being filled in or one with a ${...}
+// reference that cannot be filled in here, is checked one setting at a
+// time: settingData, the changed setting on its own, must decode into the
+// config type.
+func checkConfigChange(oldData, newData, settingData []byte) error {
+	oldCfg, err := config.Parse(oldData)
+	if err == nil && oldCfg.Validate() == nil {
+		newCfg, err := config.Parse(newData)
+		if err != nil {
+			return err
+		}
+
+		return newCfg.Validate()
+	}
+
+	_, err = config.Parse(settingData)
+
+	return err
+}
+
+// yamlPathOnly returns a document holding only the value at the dotted key
+// path in root, inside the maps and lists that lead to it there.
+func yamlPathOnly(root *yaml.Node, keys []string) (*yaml.Node, error) {
+	node, err := yamlPathGet(root, keys)
+	if err != nil {
+		return nil, err
+	}
+
+	for i, key := range slices.Backward(keys) {
+		parent, err := yamlPathGet(root, keys[:i])
+		if err != nil {
+			return nil, err
+		}
+
+		if parent.Kind == yaml.SequenceNode {
+			node = &yaml.Node{
+				Kind: yaml.SequenceNode, Content: []*yaml.Node{node},
+			}
+		} else {
+			keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: key}
+			node = &yaml.Node{
+				Kind: yaml.MappingNode, Content: []*yaml.Node{keyNode, node},
+			}
+		}
+	}
+
+	return &yaml.Node{Kind: yaml.DocumentNode, Content: []*yaml.Node{node}}, nil
+}
+
 // marshalConfigYAML renders a config document tree with 2-space indentation,
 // matching defaultConfigTemplate. yaml.Marshal defaults to 4 spaces, which
 // would reindent the whole file on the first `config set` despite the promise
@@ -553,8 +663,8 @@ func yamlPathGet(root *yaml.Node, keys []string) (*yaml.Node, error) {
 	return node, nil
 }
 
-// yamlPathSet navigates a dotted key path, creating intermediate maps as
-// needed, and sets the final key to the given scalar value. Numeric path
+// yamlPathSet navigates a dotted key path, creating missing keys as needed,
+// and sets the final key to the given scalar value. Numeric path
 // components index into sequences; an index equal to the sequence length
 // appends a new element (e.g. "age_recipients.1" on a 1-element list).
 func yamlPathSet(root *yaml.Node, keys []string, value string) error {
@@ -567,12 +677,12 @@ func yamlPathSet(root *yaml.Node, keys []string, value string) error {
 		node = node.Content[0]
 	}
 
-	for i, key := range keys {
+	for i := range keys {
 		last := i == len(keys)-1
 
 		switch node.Kind {
 		case yaml.MappingNode:
-			node = yamlSetInMapping(node, key, value, last)
+			node = yamlSetInMapping(node, keys, i, value, last)
 		case yaml.SequenceNode:
 			next, err := yamlSetInSequence(node, keys, i, value, last)
 			if err != nil {
@@ -641,14 +751,16 @@ func yamlField(typ reflect.Type, key string) (reflect.StructField, bool) {
 	return reflect.StructField{}, false
 }
 
-// yamlSetInMapping resolves (creating if needed) the value node for key
-// within a mapping node, setting it to value when it is the final path
-// element, and returns the node to descend into.
-func yamlSetInMapping(node *yaml.Node, key, value string, last bool) *yaml.Node {
+// yamlSetInMapping resolves (creating if needed) the value node for the
+// path element keys[i] within a mapping node, setting it to value when it
+// is the final path element, and returns the node to descend into.
+func yamlSetInMapping(
+	node *yaml.Node, keys []string, i int, value string, last bool,
+) *yaml.Node {
 	var valueNode *yaml.Node
 
 	for j := 0; j+1 < len(node.Content); j += 2 {
-		if node.Content[j].Value == key {
+		if node.Content[j].Value == keys[i] {
 			valueNode = node.Content[j+1]
 
 			break
@@ -656,13 +768,8 @@ func yamlSetInMapping(node *yaml.Node, key, value string, last bool) *yaml.Node
 	}
 
 	if valueNode == nil {
-		keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: key}
-
-		valueNode = &yaml.Node{Kind: yaml.MappingNode}
-		if last {
-			valueNode = &yaml.Node{Kind: yaml.ScalarNode, Value: value}
-		}
-
+		keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: keys[i]}
+		valueNode = newPathNode(keys, i, value)
 		node.Content = append(node.Content, keyNode, valueNode)
 	} else if last {
 		setScalar(valueNode, value)
@@ -690,12 +797,7 @@ func yamlSetInSequence(
 	}
 
 	if idx == len(node.Content) {
-		newNode := &yaml.Node{Kind: yaml.MappingNode}
-		if last {
-			newNode = &yaml.Node{Kind: yaml.ScalarNode, Value: value}
-		}
-
-		node.Content = append(node.Content, newNode)
+		node.Content = append(node.Content, newPathNode(keys, i, value))
 	} else if last {
 		setScalar(node.Content[idx], value)
 	}
@@ -703,6 +805,22 @@ func yamlSetInSequence(
 	return node.Content[idx], nil
 }
 
+// newPathNode returns the node to create for the missing path element
+// keys[i]: the value when it is the final path element, otherwise an empty
+// list when the next path element is a number and an empty map when not.
+func newPathNode(keys []string, i int, value string) *yaml.Node {
+	if i == len(keys)-1 {
+		return &yaml.Node{Kind: yaml.ScalarNode, Value: value}
+	}
+
+	_, err := strconv.Atoi(keys[i+1])
+	if err == nil {
+		return &yaml.Node{Kind: yaml.SequenceNode}
+	}
+
+	return &yaml.Node{Kind: yaml.MappingNode}
+}
+
 // setScalar overwrites a node in place with a plain scalar value.
 func setScalar(n *yaml.Node, value string) {
 	n.Kind = yaml.ScalarNode

internal/cli/config_test.go

@@ -58,14 +58,17 @@ 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
 // `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.
 func TestConfigSetRecipientOnFreshConfig(t *testing.T) {
 	t.Parallel()
 
-	const recipient = "age1278m9q7dp3chsh2dcy82qk27v047zywyvtxwnj4cvt0z65jw6a7q5dqhfj"
-
 	path := filepath.Join(t.TempDir(), "config.yml")
 
 	err := os.WriteFile(path, []byte(defaultConfigTemplate), configFileMode)
@@ -75,7 +78,7 @@ func TestConfigSetRecipientOnFreshConfig(t *testing.T) {
 
 	out := ui.NewWithColor(&bytes.Buffer{}, false)
 
-	err = writeConfigSet(out, path, "age_recipients.0", recipient)
+	err = writeConfigSet(out, path, "age_recipients.0", testRecipient)
 	if err != nil {
 		t.Fatalf("config set age_recipients.0: %v", err)
 	}
@@ -90,8 +93,211 @@ func TestConfigSetRecipientOnFreshConfig(t *testing.T) {
 		t.Fatalf("config.Load: %v", err)
 	}
 
-	if len(cfg.AgeRecipients) != 1 || cfg.AgeRecipients[0] != recipient {
-		t.Errorf("age_recipients = %v, want [%s]", cfg.AgeRecipients, recipient)
+	if len(cfg.AgeRecipients) != 1 || cfg.AgeRecipients[0] != testRecipient {
+		t.Errorf("age_recipients = %v, want [%s]", cfg.AgeRecipients, testRecipient)
+	}
+
+	err = writeConfigSet(out, path, "age_recipients.0", "notakey")
+	if err == nil || !strings.Contains(err.Error(), "not a valid recipient") {
+		t.Errorf("config set age_recipients.0 notakey gives error %v, "+
+			"want one saying it is not a valid recipient", err)
+	}
+
+	_, err = config.Load(path)
+	if err != nil {
+		t.Errorf("config.Load after the refused change: %v", err)
+	}
+}
+
+// configWithoutRecipients loads, but has no age_recipients key and no
+// snapshot named work.
+const configWithoutRecipients = `storage_url: file:///mnt/backups
+snapshots:
+  home:
+    paths:
+      - "~"
+`
+
+// configWithUnsetReference loads only where VAULTIK_TEST_UNSET_SECRET is
+// set, as a backup job's environment might set it. No test sets it.
+const configWithUnsetReference = configWithoutRecipients + `s3:
+  secret_access_key: ${ENV:VAULTIK_TEST_UNSET_SECRET}
+`
+
+// configWithLevelReference returns configWithoutRecipients with a ${FILE:...}
+// reference on compression_level to a file holding 9, so that the config
+// loads once the reference is filled in.
+func configWithLevelReference(t *testing.T) string {
+	t.Helper()
+
+	levelPath := filepath.Join(t.TempDir(), "level")
+
+	err := os.WriteFile(levelPath, []byte("9"), configFileMode)
+	if err != nil {
+		t.Fatalf("write level: %v", err)
+	}
+
+	return configWithoutRecipients +
+		"compression_level: ${FILE:" + levelPath + "}\n"
+}
+
+// TestConfigSetIndexUnderMissingKey checks that a list index under a key
+// the file does not have creates a list, and that the result loads.
+func TestConfigSetIndexUnderMissingKey(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		key   string
+		value string
+		field func(cfg *config.Config) []string
+	}{
+		{"age_recipients.0", testRecipient,
+			func(cfg *config.Config) []string { return cfg.AgeRecipients }},
+		{"snapshots.work.paths.0", "/srv/work",
+			func(cfg *config.Config) []string { return cfg.Snapshots["work"].Paths }},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.key, func(t *testing.T) {
+			t.Parallel()
+
+			path := filepath.Join(t.TempDir(), "config.yml")
+
+			err := os.WriteFile(path, []byte(configWithoutRecipients), configFileMode)
+			if err != nil {
+				t.Fatalf("write config: %v", err)
+			}
+
+			err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+				path, tt.key, tt.value)
+			if err != nil {
+				t.Fatalf("config set %s: %v", tt.key, err)
+			}
+
+			cfg, err := config.Load(path)
+			if err != nil {
+				t.Fatalf("config.Load: %v", err)
+			}
+
+			got := tt.field(cfg)
+			if len(got) != 1 || got[0] != tt.value {
+				t.Errorf("%s gives %v, want [%s]", tt.key, got, tt.value)
+			}
+		})
+	}
+}
+
+// TestConfigSetKeepsFileThatWouldNotLoad checks that config set refuses a
+// change after which the config would not load, with an error giving the
+// reason, and leaves the file as it was. On the file `config init` writes,
+// which does not load until storage_url is set, and on one that cannot be
+// read here, only a value of the wrong type is refused.
+func TestConfigSetKeepsFileThatWouldNotLoad(t *testing.T) {
+	t.Parallel()
+
+	levelReference := configWithLevelReference(t)
+
+	tests := []struct {
+		config string
+		key    string
+		value  string
+		reason string
+	}{
+		// Index 1 of the empty list the missing key gives.
+		{configWithoutRecipients, "age_recipients.1", testRecipient,
+			"index out of range"},
+		// A list set to a single value.
+		{configWithoutRecipients, "snapshots.home.paths", "/srv",
+			"into []string"},
+		{configWithoutRecipients, "chunk_size", "fast",
+			"invalid size format"},
+		// Values of the right type that fail validation.
+		{configWithoutRecipients, "compression_level", "99",
+			"compression_level must be between 1 and 19"},
+		{configWithoutRecipients, "chunk_size", "1KB",
+			"chunk_size must be at least 1MB"},
+		// With its reference filled in, the config loads.
+		{levelReference, "blob_size_limit", "1MB",
+			"blob_size_limit must be at least"},
+		// The type check on a config that does not load yet.
+		{defaultConfigTemplate, "snapshots.apps.paths", "/srv",
+			"into []string"},
+		// The type check on a config that cannot be read here.
+		{configWithUnsetReference, "snapshots.home.paths", "/x",
+			"cannot unmarshal !!str `/x`"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.key+" "+tt.value, func(t *testing.T) {
+			t.Parallel()
+
+			path := filepath.Join(t.TempDir(), "config.yml")
+
+			err := os.WriteFile(path, []byte(tt.config), configFileMode)
+			if err != nil {
+				t.Fatalf("write config: %v", err)
+			}
+
+			err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+				path, tt.key, tt.value)
+			if err == nil || !strings.Contains(err.Error(), tt.reason) {
+				t.Errorf("config set %s %s gives error %v, want one saying %q",
+					tt.key, tt.value, err, tt.reason)
+			}
+
+			data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+			if err != nil {
+				t.Fatalf("read config: %v", err)
+			}
+
+			if string(data) != tt.config {
+				t.Errorf("config set %s changed the file to:\n%s",
+					tt.key, data)
+			}
+		})
+	}
+}
+
+// TestConfigSetRefusesOnlyItsOwnFailure checks that setting another key
+// works on a config with a ${...} reference on compression_level, which
+// loads because loading fills the reference in, on one whose
+// compression_level already fails to load, and on one with a reference
+// that cannot be filled in here.
+func TestConfigSetRefusesOnlyItsOwnFailure(t *testing.T) {
+	t.Parallel()
+
+	configs := map[string]string{
+		"reference":       configWithLevelReference(t),
+		"failing":         configWithoutRecipients + "compression_level: fast\n",
+		"unset reference": configWithUnsetReference,
+	}
+
+	for name, configYAML := range configs {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			path := filepath.Join(t.TempDir(), "config.yml")
+
+			err := os.WriteFile(path, []byte(configYAML), configFileMode)
+			if err != nil {
+				t.Fatalf("write config: %v", err)
+			}
+
+			err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+				path, "hostname", "myhost")
+			if err != nil {
+				t.Fatalf("config set hostname: %v", err)
+			}
+
+			data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+			if err != nil {
+				t.Fatalf("read config: %v", err)
+			}
+
+			if !strings.Contains(string(data), "hostname: myhost") {
+				t.Errorf("config set hostname wrote:\n%s", data)
+			}
+		})
 	}
 }
 
@@ -441,6 +647,30 @@ func TestWriteConfigSetTightensMode(t *testing.T) {
 	}
 }
 
+// TestConfigEditEditorWithArgument checks that an $EDITOR carrying an
+// argument works and that a config path with a space reaches it whole.
+// The "editor" is cp with the replacement file as its argument, so the
+// config ends up with the replacement's content only if both arrived.
+func TestConfigEditEditorWithArgument(t *testing.T) {
+	dir := t.TempDir()
+	path := seedFile(t, dir, "my config.yml", "compression_level: 3\n")
+	edited := seedFile(t, dir, "edited.yml", "compression_level: 9\n")
+
+	setRootFlags(t, RootFlags{ConfigPath: path})
+	t.Setenv("EDITOR", "cp '"+edited+"'")
+
+	mustExecute(t, newConfigEditCommand())
+
+	data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+	if err != nil {
+		t.Fatalf("reading config back: %v", err)
+	}
+
+	if string(data) != "compression_level: 9\n" {
+		t.Errorf("config after edit = %q, want the edited content", data)
+	}
+}
+
 func splitPath(s string) []string {
 	return strings.Split(s, ".")
 }

internal/cli/database.go

@@ -15,6 +15,12 @@ func NewDatabaseCommand() *cobra.Command {
 		Use:   "database",
 		Short: "Manage the local state database",
 		Long:  `Commands for managing the local SQLite state database.`,
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	cmd.AddCommand(

internal/cli/entry_prune_json_test.go

@@ -101,7 +101,8 @@ func TestEntryPruneJSONStdoutIsExactlyOneDocument(t *testing.T) {
 // writeHermeticPruneConfig builds a config over a temp directory with an
 // empty destination directory and, if seedStale is set, creates the
 // index database up front with one snapshot record that has no
-// counterpart on the destination store. Returns the config path.
+// counterpart on the destination store, and gives the destination the
+// marker a backup writes. Returns the config path.
 func writeHermeticPruneConfig(t *testing.T, seedStale bool) string {
 	t.Helper()
 
@@ -131,6 +132,11 @@ func writeHermeticPruneConfig(t *testing.T, seedStale bool) string {
 
 	if seedStale {
 		seedStaleSnapshotRecord(t, indexPath)
+
+		// While the local index holds a snapshot, prune fails on a
+		// destination without the marker a backup writes.
+		require.NoError(t, os.WriteFile(
+			filepath.Join(storeDir, "vaultik-destination.txt"), nil, 0o600))
 	}
 
 	return configPath

internal/cli/entry_prune_quiet_test.go

@@ -0,0 +1,127 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// flagCron is used to build the argument vectors below.
+const flagCron = "--cron"
+
+// plantUnreferencedBlob writes a blob that no snapshot references to the
+// destination store at storeDir, dated long before any prune grace period,
+// so that a prune deletes it. Returns its path.
+func plantUnreferencedBlob(t *testing.T, storeDir string) string {
+	t.Helper()
+
+	hash := strings.Repeat("b", 64)
+	path := filepath.Join(storeDir, "blobs", hash[:2], hash[2:4], hash)
+
+	require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o750))
+	require.NoError(t, os.WriteFile(path, []byte("unreferenced"), 0o600))
+
+	longAgo := time.Unix(0, 0)
+	require.NoError(t, os.Chtimes(path, longAgo, longAgo))
+
+	return path
+}
+
+// TestEntrySnapshotCreatePruneIsSilent checks that a successful
+// `snapshot create --prune` under --cron or --quiet prints nothing on
+// stdout or stderr, whether or not the prune finds a blob to delete.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreatePruneIsSilent(t *testing.T) {
+	for _, testCase := range []struct {
+		name      string
+		args      []string
+		plantBlob bool
+	}{
+		{
+			name: "cron, no blob to delete",
+			args: []string{cmdSnapshot, cmdCreate, flagCron, flagPrune},
+		},
+		{
+			name:      "cron, blob to delete",
+			args:      []string{cmdSnapshot, cmdCreate, flagCron, flagPrune},
+			plantBlob: true,
+		},
+		{
+			name: "cron with keep-newer-than",
+			args: []string{
+				cmdSnapshot, cmdCreate, flagCron, flagPrune,
+				flagKeepNewerThan, "4w",
+			},
+			plantBlob: true,
+		},
+		{
+			name:      "quiet",
+			args:      []string{flagQuiet, cmdSnapshot, cmdCreate, flagPrune},
+			plantBlob: true,
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath, storeDir := writeBackupConfig(t)
+
+			var blob string
+			if testCase.plantBlob {
+				blob = plantUnreferencedBlob(t, storeDir)
+			}
+
+			code, stdout, stderr := runEntry(t,
+				append([]string{flagConfig, configPath}, testCase.args...)...)
+
+			assert.Equal(t, 0, code)
+			assert.Empty(t, stdout)
+			assert.Empty(t, stderr)
+
+			if testCase.plantBlob {
+				assert.NoFileExists(t, blob, "the prune must delete the blob")
+			}
+		})
+	}
+}
+
+// TestEntryPruneQuietIsSilent checks that a successful `prune --force`
+// under --quiet prints nothing on stdout or stderr, both when there is
+// nothing to remove and when it removes a stale local snapshot record
+// and an unreferenced blob.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntryPruneQuietIsSilent(t *testing.T) {
+	for _, testCase := range []struct {
+		name               string
+		staleRecordAndBlob bool
+	}{
+		{name: "nothing to remove"},
+		{name: "stale record and blob to remove", staleRecordAndBlob: true},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath := writeHermeticPruneConfig(t, testCase.staleRecordAndBlob)
+
+			var blob string
+			if testCase.staleRecordAndBlob {
+				// writeHermeticPruneConfig puts the store beside the config.
+				blob = plantUnreferencedBlob(t,
+					filepath.Join(filepath.Dir(configPath), "store"))
+			}
+
+			code, stdout, stderr := runEntry(t,
+				flagConfig, configPath, flagQuiet, cmdPrune, flagForce)
+
+			assert.Equal(t, 0, code)
+			assert.Empty(t, stdout)
+			assert.Empty(t, stderr)
+
+			if testCase.staleRecordAndBlob {
+				assert.NoFileExists(t, blob, "the prune must delete the blob")
+			}
+		})
+	}
+}

internal/cli/entry_snapshot_create_skip_errors_test.go

@@ -0,0 +1,30 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// TestEntrySnapshotCreateSkipErrorsExitsNonZeroWhenItSkips backs up,
+// under --skip-errors, a source directory that does not exist, which the
+// walk skips. The command exits non-zero and reports the backup as
+// completed, not as a failed snapshot creation. See
+// https://git.eeqj.de/sneak/vaultik/issues/310.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreateSkipErrorsExitsNonZeroWhenItSkips(t *testing.T) {
+	// writeBackupConfig would create the source directory.
+	configPath := writeHermeticPruneConfig(t, false)
+
+	code, stdout, stderr := runEntry(t,
+		flagConfig, configPath, "--skip-errors", cmdSnapshot, cmdCreate)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stdout,
+		"Files: 1 examined, 0 backed up, 0 unchanged, 1 failed.")
+	assert.NotContains(t, stdout, "Finished successfully.")
+	assert.Contains(t, stderr,
+		"Backup completed, but 1 file(s) could not be read")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+}

internal/cli/entry_snapshot_purge_test.go

@@ -0,0 +1,75 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// Words and flags of `snapshot purge` used to build argument vectors
+// below.
+const (
+	cmdPurge      = "purge"
+	flagOlderThan = "--older-than"
+	flagForce     = "--force"
+)
+
+// TestEntrySnapshotPurgeKeepsMetadataWhenLocalIndexDeleteFails makes
+// the local index refuse to delete an old snapshot and checks that
+// `snapshot purge` exits non-zero and leaves that snapshot's metadata on
+// the destination store. It used to log the error, delete the metadata
+// and exit 0.
+//
+// `snapshot create --prune` runs the same purge, but cannot be driven
+// this way: the refusal also breaks the backup's metadata export, which
+// deletes the other snapshots from its copy of the local index.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotPurgeKeepsMetadataWhenLocalIndexDeleteFails(t *testing.T) {
+	configPath := writeHermeticPruneConfig(t, false)
+	dir := filepath.Dir(configPath)
+	indexPath := filepath.Join(dir, "index.sqlite")
+
+	// seedStaleSnapshotRecord writes an old snapshot to the local index
+	// only; the manifest planted here puts it on the destination store
+	// too, so the purge selects it rather than dropping it as stale.
+	seedStaleSnapshotRecord(t, indexPath)
+
+	// While the local index holds a snapshot, purge fails on a
+	// destination without the marker a backup writes.
+	require.NoError(t, os.WriteFile(
+		filepath.Join(dir, "store", "vaultik-destination.txt"), nil, 0o600))
+
+	manifestDir := filepath.Join(dir, "store", "metadata",
+		snapshot.RemoteSnapshotKey(stalePruneSnapshotID))
+	require.NoError(t, os.MkdirAll(manifestDir, 0o750))
+
+	manifestPath := filepath.Join(manifestDir, "manifest.json.zst")
+	require.NoError(t, os.WriteFile(manifestPath, []byte("stub"), 0o600))
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, indexPath)
+	require.NoError(t, err)
+
+	_, err = db.Conn().ExecContext(ctx, `
+		CREATE TRIGGER refuse_snapshot_delete BEFORE DELETE ON snapshots
+		BEGIN SELECT RAISE(ABORT, 'simulated index error'); END`)
+	require.NoError(t, err)
+	require.NoError(t, db.Close())
+
+	code, _, stderr := runEntry(t, flagConfig, configPath,
+		cmdSnapshot, cmdPurge, flagOlderThan, "30d", flagForce)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr, "simulated index error",
+		"the purge must fail at the local-index delete")
+	assert.FileExists(t, manifestPath,
+		"the snapshot is still in the local index, so its metadata must stay")
+}

internal/cli/entry_unknown_subcommand_test.go

@@ -0,0 +1,43 @@
+package cli //nolint:testpackage // shares runEntry and the command-word constants
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// TestEntryGroupRejectsUnknownSubcommand runs each group command with a
+// mistyped subcommand, as a crontab line might carry one. It must fail
+// with exit 1 and name the word, as `vaultik bogus` does at the top
+// level, instead of printing the group's help and exiting 0 having done
+// nothing. The group name alone still prints its help.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryGroupRejectsUnknownSubcommand(t *testing.T) {
+	for _, testCase := range []struct {
+		group    string
+		mistyped string
+	}{
+		{group: cmdSnapshot, mistyped: "craete"},
+		{group: "config", mistyped: "inti"},
+		{group: cmdRemote, mistyped: "ifno"},
+		{group: "database", mistyped: "delte"},
+		{group: "completion", mistyped: "nosuchshell"},
+	} {
+		t.Run(testCase.group, func(t *testing.T) {
+			code, stdout, stderr := runEntry(t,
+				testCase.group, testCase.mistyped)
+
+			assert.Equal(t, 1, code)
+			assert.Empty(t, stdout)
+			assert.Contains(t, stderr,
+				`unknown command "`+testCase.mistyped+`"`)
+
+			code, stdout, _ = runEntry(t, testCase.group)
+
+			assert.Equal(t, 0, code)
+			assert.Contains(t, stdout,
+				programName+" "+testCase.group+" [command]")
+		})
+	}
+}

internal/cli/prune.go

@@ -20,7 +20,12 @@ Local cleanup drops 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
-set.
+set that is older than prune_grace_period (default 24h). A newer one
+is kept: it may belong to a backup still running, on this host or
+another sharing the destination store, that has not yet uploaded its
+manifest. The grace period does not protect a blob such a backup
+reuses from an earlier run that stopped before its manifest: that
+blob keeps the time of its first upload, so it can be deleted.
 
 Snapshot create --prune runs the same cleanup automatically; this
 command is the manual entry point for the same work (e.g. after a

internal/cli/remote.go

@@ -18,6 +18,12 @@ func NewRemoteCommand() *cobra.Command {
 		Use:   "remote",
 		Short: "Remote storage management commands",
 		Long:  "Commands for inspecting and managing remote storage",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add subcommands

internal/cli/root.go

@@ -74,6 +74,21 @@ on the source system.`,
 		NewDatabaseCommand(),
 	)
 
+	// cobra adds the completion command only when Execute runs, and with
+	// no Run, so it skipped its own cobra.NoArgs and a mistyped shell name
+	// printed help and exited 0. Adding it here lets it have a Run. Its
+	// scripts go to os.Stdout as it is now: a later SetOut does not
+	// redirect them.
+	cmd.InitDefaultCompletionCmd()
+
+	for _, sub := range cmd.Commands() {
+		if sub.Name() == "completion" {
+			sub.Run = func(completion *cobra.Command, _ []string) {
+				_ = completion.Help()
+			}
+		}
+	}
+
 	return cmd
 }
 

internal/cli/snapshot.go

@@ -41,6 +41,12 @@ func NewSnapshotCommand() *cobra.Command {
 		Use:   "snapshot",
 		Short: "Snapshot management commands",
 		Long:  "Commands for creating, listing, and managing snapshots",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add subcommands
@@ -109,6 +115,13 @@ on Linux), then /etc/vaultik/config.yml.`,
 					return
 				}
 
+				if errors.Is(err, vaultik.ErrFilesUnreadable) {
+					log.Error("Backup completed with unreadable files", "error", err)
+					ReportErrorf("Backup completed, but %v", err)
+
+					return
+				}
+
 				log.Error("Snapshot creation failed", "error", err)
 				ReportErrorf("Snapshot creation failed: %v", err)
 			})
@@ -119,7 +132,8 @@ on Linux), then /etc/vaultik/config.yml.`,
 		"Run in cron mode (silent unless warning or error)")
 	cmd.Flags().BoolVar(&opts.Prune, "prune", false,
 		"After backup, drop older snapshots of the same name and remove "+
-			"orphaned blobs")
+			"orphaned blobs older than prune_grace_period (not after a "+
+			"backup that skipped unreadable files)")
 	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) "+
@@ -264,8 +278,9 @@ By default, this removes the snapshot from the local index database and
 strips the snapshot's metadata from the backup destination store. Blobs
 are NOT touched: deleting them requires reading every remaining remote
 manifest (the destination store may hold snapshots this host doesn't
-know about), which is what 'vaultik prune' does. On success the command
-prints the exact 'vaultik prune' invocation to run as a follow-up.
+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.
@@ -274,10 +289,12 @@ 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
 must then be in the local index.
 
-If the remote is unreachable, the local-database removal still completes
-and a warning is emitted; run 'vaultik snapshot remove <snapshot-id>' again
-once the destination store is reachable to remove the snapshot's metadata
-from it ('vaultik prune' does not).
+If removing the snapshot's metadata from the destination store fails (for
+example because the remote is unreachable), the local-database removal
+still completes and a warning is emitted; run 'vaultik snapshot remove
+<snapshot-id>' again once the destination store is reachable to remove
+the snapshot's metadata from it ('vaultik prune' does not). For a
+snapshot that is only on the destination store, that failure is an error.
 
 To wipe the entire destination store and start over, use 'vaultik remote
 nuke --force' — it is the single supported entry point for that.`,

internal/cli/snapshot_restore.go

@@ -30,6 +30,8 @@ This command will restore files from the specified snapshot to the
 target directory.
 If no paths are specified, all files are restored.
 If paths are specified, only matching files/directories are restored.
+A path that matches nothing in the snapshot is named in a warning, and
+the command exits non-zero after restoring the paths that match.
 
 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

internal/config/config.go

@@ -4,12 +4,14 @@
 package config
 
 import (
+	"bytes"
 	"errors"
 	"fmt"
 	"os"
 	"path/filepath"
 	"sort"
 	"strings"
+	"time"
 
 	"filippo.io/age"
 	"git.eeqj.de/sneak/smartconfig"
@@ -29,13 +31,16 @@ const appName = "vaultik"
 //nolint:gosec // G101: marker for detecting a pasted secret key, not a credential
 const secretKeyPrefix = "AGE-SECRET-KEY-"
 
-// Defaults and validation bounds for tunable settings.
+// Defaults and validation bounds for tunable settings. MB and GB are
+// 1000-based, as the size parser reads them; the S3 part sizes are S3's own
+// limits, which are 1024-based (MiB, GiB).
 const (
-	defaultBlobSizeLimit    = Size(10 * 1024 * 1024 * 1024) // 10GB
-	defaultChunkSize        = Size(10 * 1024 * 1024)        // 10MB
+	defaultBlobSizeLimit    = Size(10 * 1000 * 1000 * 1000) // 10GB
+	defaultChunkSize        = Size(10 * 1000 * 1000)        // 10MB
 	defaultS3PartSize       = Size(5 * 1024 * 1024)         // 5MiB
 	defaultCompressionLevel = 3
-	minChunkSize            = 1024 * 1024 // 1MB
+	defaultPruneGracePeriod = 24 * time.Hour
+	minChunkSize            = 1000 * 1000 // 1MB
 	minCompressionLevel     = 1
 	maxCompressionLevel     = 19
 	// S3 accepts a multipart upload part from 5MiB to 5GiB.
@@ -61,6 +66,11 @@ var (
 	errBadS3PartSize    = errors.New("s3.part_size must be between 5MiB and 5GiB")
 	errBadStorageScheme = errors.New(
 		"storage_url must start with s3://, file://, or rclone://")
+	errFileURLNotAbsolute = errors.New(
+		"a file:// storage_url needs an absolute path, " +
+			"as in file:///path/to/backup")
+	errNegativePruneGracePeriod = errors.New(
+		"prune_grace_period must not be negative")
 	errStorageNotConfigured = errors.New(
 		"storage not configured; set storage_url or provide s3.endpoint + " +
 			"s3.bucket + credentials")
@@ -181,6 +191,13 @@ type Config struct {
 	Snapshots        map[string]SnapshotConfig `yaml:"snapshots"`
 	CompressionLevel int                       `yaml:"compression_level"`
 
+	// PruneGracePeriod is how old a blob no manifest lists must be before
+	// a prune deletes it. A backup lists its blobs in a manifest only
+	// once it has uploaded them all, so a younger one may belong to a
+	// backup still running, on this host or another sharing the
+	// destination store.
+	PruneGracePeriod time.Duration `yaml:"prune_grace_period"`
+
 	// StorageURL specifies the storage backend using a URL format.
 	// Takes precedence over S3Config if set.
 	// Supported formats:
@@ -230,14 +247,48 @@ func New(path Path) (*Config, error) {
 	return cfg, nil
 }
 
-// Load reads and parses the configuration file from the specified path.
-// It applies default values for optional fields, performs environment variable
-// substitution using smartconfig, and validates the configuration.
-// The configuration file should be in YAML format. Returns an error if the file
-// cannot be read, parsed, or if validation fails.
+// Load reads the configuration file at path, parses it with Parse, warns
+// when others can read the file, and validates the configuration.
+// Returns an error if the file cannot be read, parsed, or if validation
+// fails.
 func Load(path string) (*Config, error) {
+	data, err := os.ReadFile(path) //nolint:gosec // G304: config path is operator-supplied
+	if err != nil {
+		return nil, fmt.Errorf("failed to load config file: %w", err)
+	}
+
+	cfg, err := Parse(data)
+	if err != nil {
+		return nil, err
+	}
+
+	// Check config file permissions (warn if world or group readable)
+	//nolint:gosec // G703: config path is operator-supplied by design
+	info, statErr := os.Stat(path)
+	if statErr == nil {
+		mode := info.Mode().Perm()
+		if mode&0044 != 0 { // group or world readable
+			log.Warn(cfg.readableByOthersWarning(),
+				"path", path,
+				"mode", fmt.Sprintf("%04o", mode),
+				"recommendation", "chmod 600 "+path)
+		}
+	}
+
+	err = cfg.Validate()
+	if err != nil {
+		return nil, fmt.Errorf("invalid config: %w", err)
+	}
+
+	return cfg, nil
+}
+
+// Parse reads YAML config data as Load does, without validating it. It
+// applies default values for optional fields, fills in ${...} references
+// using smartconfig, and expands ~ in paths.
+func Parse(data []byte) (*Config, error) {
 	// Load config using smartconfig for interpolation
-	sc, err := smartconfig.NewFromConfigPath(path)
+	sc, err := smartconfig.NewFromReader(bytes.NewReader(data))
 	if err != nil {
 		return nil, fmt.Errorf("failed to load config file: %w", err)
 	}
@@ -248,6 +299,7 @@ func Load(path string) (*Config, error) {
 		ChunkSize:        defaultChunkSize,
 		IndexPath:        filepath.Join(xdg.DataHome, appName, "index.sqlite"),
 		CompressionLevel: defaultCompressionLevel,
+		PruneGracePeriod: defaultPruneGracePeriod,
 		S3:               S3Config{PartSize: defaultS3PartSize},
 	}
 
@@ -299,24 +351,6 @@ func Load(path string) (*Config, error) {
 		cfg.S3.Region = "us-east-1"
 	}
 
-	// Check config file permissions (warn if world or group readable)
-	//nolint:gosec // G703: config path is operator-supplied by design
-	info, statErr := os.Stat(path)
-	if statErr == nil {
-		mode := info.Mode().Perm()
-		if mode&0044 != 0 { // group or world readable
-			log.Warn(cfg.readableByOthersWarning(),
-				"path", path,
-				"mode", fmt.Sprintf("%04o", mode),
-				"recommendation", "chmod 600 "+path)
-		}
-	}
-
-	err = cfg.Validate()
-	if err != nil {
-		return nil, fmt.Errorf("invalid config: %w", err)
-	}
-
 	return cfg, nil
 }
 
@@ -334,6 +368,7 @@ func Load(path string) (*Config, error) {
 //     exceeds the configured limit
 //   - Compression level must be between 1 and 19
 //   - S3 part size must be between 5MiB and 5GiB, the part sizes S3 accepts
+//   - Prune grace period must not be negative
 //
 // Returns an error describing the first validation failure encountered.
 func (c *Config) Validate() error {
@@ -383,6 +418,10 @@ func (c *Config) Validate() error {
 		return errBadS3PartSize
 	}
 
+	if c.PruneGracePeriod < 0 {
+		return errNegativePruneGracePeriod
+	}
+
 	return nil
 }
 
@@ -461,11 +500,18 @@ func (c *Config) validateStorage() error {
 
 // validateStorageURL validates URL-based storage configuration. File and
 // rclone URLs need no credentials; S3 URLs require the legacy s3.*
-// credential fields.
+// credential fields. A file URL must name an absolute path: a relative one
+// would resolve against the working directory, so the destination would
+// move with it.
 func (c *Config) validateStorageURL() error {
 	switch {
 	case strings.HasPrefix(c.StorageURL, "file://"):
-		// File storage doesn't need S3 credentials
+		// storage.ParseStorageURL applies the same rule, but config
+		// cannot import storage, which imports config.
+		if !strings.HasPrefix(c.StorageURL, "file:///") {
+			return errFileURLNotAbsolute
+		}
+
 		return nil
 	case strings.HasPrefix(c.StorageURL, "rclone://"):
 		// Rclone storage uses rclone's own config

internal/config/config_test.go

@@ -6,6 +6,7 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	"sneak.berlin/go/vaultik/internal/chunker"
 	"sneak.berlin/go/vaultik/internal/log"
@@ -337,6 +338,177 @@ func TestLoadS3PartSize(t *testing.T) {
 	})
 }
 
+// TestDocumentedSizeDefaults checks that the defaults are the documented
+// chunk_size 10MB and blob_size_limit 10GB as the parser reads those values
+// (1000-based), so writing a default out in the config file changes nothing.
+func TestDocumentedSizeDefaults(t *testing.T) {
+	t.Parallel()
+
+	cfg, err := Parse([]byte("chunk_size: 10MB\nblob_size_limit: 10GB\n"))
+	if err != nil {
+		t.Fatalf("Parse() unexpected error: %v", err)
+	}
+
+	if cfg.ChunkSize != defaultChunkSize {
+		t.Errorf("chunk_size: 10MB is %d bytes, the default is %d",
+			cfg.ChunkSize, defaultChunkSize)
+	}
+
+	if cfg.BlobSizeLimit != defaultBlobSizeLimit {
+		t.Errorf("blob_size_limit: 10GB is %d bytes, the default is %d",
+			cfg.BlobSizeLimit, defaultBlobSizeLimit)
+	}
+}
+
+// TestDocumentedSizeExamples checks the 1MB minimum chunk_size and that
+// every size the README, config.example.yml and the config init template
+// list as an example is accepted.
+func TestDocumentedSizeExamples(t *testing.T) {
+	t.Parallel()
+
+	const withoutSizes = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	tests := []struct {
+		setting string
+		wantErr error
+	}{
+		{"chunk_size: 1MB", nil},
+		{"chunk_size: 999999", errChunkSizeTooSmall},
+		{"chunk_size: 5M", nil},
+		{"chunk_size: 10M", nil},
+		{"chunk_size: 10MB", nil},
+		{"chunk_size: 64MiB", nil},
+		{"chunk_size: 1GB", nil},
+		// Four times the default chunk_size, the documented minimum.
+		{"blob_size_limit: 40MB", nil},
+		{"blob_size_limit: 500MB", nil},
+		{"blob_size_limit: 1GB", nil},
+		{"blob_size_limit: 1GiB", nil},
+		{"blob_size_limit: 10G", nil},
+		{"blob_size_limit: 10GB", nil},
+		{"s3: {part_size: 5MiB}", nil},
+		{"s3: {part_size: 10MB}", nil},
+		{"s3: {part_size: 16MiB}", nil},
+		{"s3: {part_size: 100MiB}", nil},
+		{"s3: {part_size: 5GiB}", nil},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.setting, func(t *testing.T) {
+			t.Parallel()
+
+			cfg, err := Parse([]byte(withoutSizes + tt.setting + "\n"))
+			if err != nil {
+				t.Fatalf("Parse() unexpected error: %v", err)
+			}
+
+			err = cfg.Validate()
+			if !errors.Is(err, tt.wantErr) {
+				t.Fatalf("Validate() error = %v, want %v", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+// TestValidatePruneGracePeriod checks that a negative prune_grace_period
+// is refused rather than acting like 0s, which deletes unreferenced blobs
+// at once, and that 0s and the documented examples are accepted.
+func TestValidatePruneGracePeriod(t *testing.T) {
+	t.Parallel()
+
+	const withoutGracePeriod = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	tests := []struct {
+		setting string
+		wantErr error
+	}{
+		{"prune_grace_period: 0s", nil},
+		{"prune_grace_period: 90m", nil},
+		{"prune_grace_period: 36h", nil},
+		{"prune_grace_period: -1h", errNegativePruneGracePeriod},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.setting, func(t *testing.T) {
+			t.Parallel()
+
+			cfg, err := Parse([]byte(withoutGracePeriod + tt.setting + "\n"))
+			if err != nil {
+				t.Fatalf("Parse() unexpected error: %v", err)
+			}
+
+			err = cfg.Validate()
+			if !errors.Is(err, tt.wantErr) {
+				t.Fatalf("Validate() error = %v, want %v", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+// TestLoadPruneGracePeriod checks that a config file without
+// prune_grace_period loads with the documented 24h. The value is written out
+// here rather than taken from defaultPruneGracePeriod, so that a change of the
+// default fails this test.
+func TestLoadPruneGracePeriod(t *testing.T) {
+	t.Parallel()
+
+	const withoutGracePeriod = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(withoutGracePeriod), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	cfg, err := Load(path)
+	if err != nil {
+		t.Fatalf("Load() unexpected error: %v", err)
+	}
+
+	if cfg.PruneGracePeriod != 24*time.Hour {
+		t.Errorf("prune_grace_period = %s, want 24h", cfg.PruneGracePeriod)
+	}
+}
+
+// TestLoadRejectsRelativeFileURL checks that a file:// storage_url with a
+// relative path fails at load. Accepted, it would resolve against the
+// working directory, so a backup run from another directory would write
+// to a different destination and still succeed.
+func TestLoadRejectsRelativeFileURL(t *testing.T) {
+	t.Parallel()
+
+	const relativeConfig = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file://dest\n"
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(relativeConfig), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	_, err = Load(path)
+	if !errors.Is(err, errFileURLNotAbsolute) {
+		t.Fatalf("Load() error = %v, want errFileURLNotAbsolute", err)
+	}
+
+	if !strings.Contains(err.Error(), "absolute path") {
+		t.Errorf("error %q does not say to use an absolute path", err.Error())
+	}
+}
+
 // TestValidateAgeRecipients checks that recipients are parsed at config load
 // (a bad entry fails immediately, not mid-backup) and that no invalid entry —
 // least of all a pasted secret key — is echoed in the error. An empty list

internal/database/blob_chunks.go

@@ -58,12 +58,7 @@ func (r *BlobChunkRepository) GetByBlobID(
 		return nil, fmt.Errorf("querying blob chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var blobChunks []*BlobChunk
 

internal/database/blobs.go

@@ -83,12 +83,7 @@ func (r *BlobRepository) GetAll(ctx context.Context) (map[string]*Blob, error) {
 		return nil, fmt.Errorf("querying blobs: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	out := make(map[string]*Blob)
 
@@ -159,13 +154,16 @@ func (r *BlobRepository) UpdateFinished(
 	return nil
 }
 
-// UpdateUploaded marks a blob as uploaded
+// UpdateUploaded marks a blob as uploaded and records the size of the
+// stored object as its compressed size. That size can differ from the
+// local copy's when an identical blob was already stored with another
+// compression level or other recipients.
 func (r *BlobRepository) UpdateUploaded(
-	ctx context.Context, tx *sql.Tx, id string,
+	ctx context.Context, tx *sql.Tx, id string, storedSize int64,
 ) error {
 	query := `
-		UPDATE blobs 
-		SET uploaded_ts = ?
+		UPDATE blobs
+		SET uploaded_ts = ?, compressed_size = ?
 		WHERE id = ?
 	`
 
@@ -173,9 +171,9 @@ func (r *BlobRepository) UpdateUploaded(
 
 	var err error
 	if tx != nil {
-		_, err = tx.ExecContext(ctx, query, now, id)
+		_, err = tx.ExecContext(ctx, query, now, storedSize, id)
 	} else {
-		_, err = r.db.ExecWithLog(ctx, query, now, id)
+		_, err = r.db.ExecWithLog(ctx, query, now, storedSize, id)
 	}
 
 	if err != nil {

internal/database/blobs_test.go

@@ -122,8 +122,8 @@ func TestBlobRepositoryUpdates(t *testing.T) {
 		t.Errorf("expected compressed size 500, got %d", updated.CompressedSize)
 	}
 
-	// Test UpdateUploaded
-	err = repo.UpdateUploaded(ctx, nil, blob.ID.String())
+	// Test UpdateUploaded, with a stored object larger than the local copy
+	err = repo.UpdateUploaded(ctx, nil, blob.ID.String(), 600)
 	if err != nil {
 		t.Fatalf("failed to update blob as uploaded: %v", err)
 	}
@@ -141,6 +141,10 @@ func TestBlobRepositoryUpdates(t *testing.T) {
 	if uploaded.UploadedTS.Before(now.Add(-1 * time.Second)) {
 		t.Error("uploaded timestamp should be around test time")
 	}
+
+	if uploaded.CompressedSize != 600 {
+		t.Errorf("expected the stored size 600, got %d", uploaded.CompressedSize)
+	}
 }
 
 func TestBlobRepositoryDuplicate(t *testing.T) {

internal/database/chunk_files.go

@@ -61,12 +61,7 @@ func (r *ChunkFileRepository) GetByChunkHash(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -87,12 +82,7 @@ func (r *ChunkFileRepository) GetByFilePath(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -112,12 +102,7 @@ func (r *ChunkFileRepository) GetByFileID(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }

internal/database/chunks.go

@@ -107,12 +107,7 @@ func (r *ChunkRepository) GetByHashes(
 		return nil, fmt.Errorf("querying chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 
@@ -152,12 +147,7 @@ func (r *ChunkRepository) ListUnpacked(
 		return nil, fmt.Errorf("querying unpacked chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 

internal/database/chunks_ext.go

@@ -38,12 +38,7 @@ func (r *ChunkRepository) list(ctx context.Context, query string) ([]*Chunk, err
 		return nil, fmt.Errorf("querying chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 

internal/database/errors.go

@@ -1,12 +0,0 @@
-package database
-
-import (
-	"fmt"
-	"os"
-)
-
-// Fatalf prints an error message to stderr and exits with status 1
-func Fatalf(format string, args ...any) {
-	fmt.Fprintf(os.Stderr, "FATAL: "+format+"\n", args...)
-	os.Exit(1)
-}

internal/database/file_chunks.go

@@ -62,12 +62,7 @@ func (r *FileChunkRepository) GetByPath(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanFileChunks(rows)
 }
@@ -88,12 +83,7 @@ func (r *FileChunkRepository) GetByFileID(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanFileChunks(rows)
 }
@@ -117,12 +107,7 @@ func (r *FileChunkRepository) GetByPathTx(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	fileChunks, err := r.scanFileChunks(rows)
 	LogSQL("GetByPathTx", "Complete", path, "count", len(fileChunks))

internal/database/files.go

@@ -175,12 +175,7 @@ func (r *FileRepository) ListModifiedSince(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -256,12 +251,46 @@ func (r *FileRepository) ListUnderPath(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
+	defer func() { _ = rows.Close() }()
+
+	var files []*File
+
+	for rows.Next() {
+		file, err := r.scanFileRows(rows)
 		if err != nil {
-			Fatalf("failed to close rows: %v", err)
+			return nil, fmt.Errorf("scanning file: %w", err)
 		}
-	}()
+
+		files = append(files, file)
+	}
+
+	return files, rows.Err()
+}
+
+// ListUnderPathInSnapshot returns the files that ListUnderPath returns
+// for path and that the snapshot lists, in no particular order.
+func (r *FileRepository) ListUnderPathInSnapshot(
+	ctx context.Context, path, snapshotID string,
+) ([]*File, error) {
+	path = strings.TrimRight(path, "/")
+	dirPrefix := path + "/"
+
+	query := `
+		SELECT f.id, f.path, f.source_path, f.mtime, f.mtime_nsec, f.size,
+			f.mode, f.uid, f.gid, f.link_target
+		FROM files f
+		JOIN snapshot_files sf ON sf.file_id = f.id
+		WHERE sf.snapshot_id = ?
+		  AND (f.path = ? OR substr(f.path, 1, length(?)) = ?)
+	`
+
+	rows, err := r.db.conn.QueryContext(ctx, query,
+		snapshotID, path, dirPrefix, dirPrefix)
+	if err != nil {
+		return nil, fmt.Errorf("querying files: %w", err)
+	}
+
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -304,12 +333,7 @@ func (r *FileRepository) ListIDsWithChunksNotInUploadedBlobs(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var ids []types.FileID
 
@@ -340,12 +364,7 @@ func (r *FileRepository) ListAll(ctx context.Context) ([]*File, error) {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 

internal/database/files_test.go

@@ -212,6 +212,65 @@ func TestFileRepositoryListUnderPath(t *testing.T) {
 	}
 }
 
+// ListUnderPathInSnapshot leaves out files outside path and files the
+// snapshot does not list, though another snapshot lists them.
+func TestFileRepositoryListUnderPathInSnapshot(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	files := database.NewFileRepository(db)
+	snapshots := database.NewSnapshotRepository(db)
+
+	listed := map[string][]string{
+		"one": {"/src/a.txt", "/src2/c.txt"},
+		"two": {"/src/b.txt"},
+	}
+
+	for snapshotID, paths := range listed {
+		err := snapshots.Create(ctx, nil, &database.Snapshot{
+			ID:        types.SnapshotID(snapshotID),
+			Hostname:  "host",
+			StartedAt: time.Now(),
+		})
+		if err != nil {
+			t.Fatalf("failed to create snapshot %s: %v", snapshotID, err)
+		}
+
+		for _, path := range paths {
+			err := files.Create(ctx, nil, &database.File{
+				Path:  types.FilePath(path),
+				MTime: time.Now().Truncate(time.Second),
+				Mode:  0644,
+			})
+			if err != nil {
+				t.Fatalf("failed to create %s: %v", path, err)
+			}
+
+			err = snapshots.AddFile(ctx, nil, snapshotID, path)
+			if err != nil {
+				t.Fatalf("failed to add %s to %s: %v", path, snapshotID, err)
+			}
+		}
+	}
+
+	found, err := files.ListUnderPathInSnapshot(ctx, "/src", "one")
+	if err != nil {
+		t.Fatalf("failed to list files: %v", err)
+	}
+
+	got := make([]string, 0, len(found))
+	for _, f := range found {
+		got = append(got, f.Path.String())
+	}
+
+	if !slices.Equal(got, []string{"/src/a.txt"}) {
+		t.Errorf("got %q, want only /src/a.txt", got)
+	}
+}
+
 func TestFileRepositorySymlink(t *testing.T) {
 	t.Parallel()
 

internal/database/repositories.go

@@ -3,6 +3,7 @@ package database
 import (
 	"context"
 	"database/sql"
+	"errors"
 	"fmt"
 )
 
@@ -60,22 +61,24 @@ func (r *Repositories) WithTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
+			// The panic is the failure to report, so a rollback error
+			// is dropped.
+			_ = tx.Rollback()
 
 			panic(p)
-		} else if err != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
 		}
 	}()
 
 	err = fn(ctx, tx)
 	if err != nil {
+		// Once ctx is cancelled, database/sql rolls the transaction
+		// back itself and Rollback returns sql.ErrTxDone.
+		rollbackErr := tx.Rollback()
+		if rollbackErr != nil && !errors.Is(rollbackErr, sql.ErrTxDone) {
+			return fmt.Errorf("%w; rolling back transaction: %w",
+				err, rollbackErr)
+		}
+
 		return err
 	}
 
@@ -103,22 +106,24 @@ func (r *Repositories) WithReadTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
+			// The panic is the failure to report, so a rollback error
+			// is dropped.
+			_ = tx.Rollback()
 
 			panic(p)
-		} else if err != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
 		}
 	}()
 
 	err = fn(ctx, tx)
 	if err != nil {
+		// Once ctx is cancelled, database/sql rolls the transaction
+		// back itself and Rollback returns sql.ErrTxDone.
+		rollbackErr := tx.Rollback()
+		if rollbackErr != nil && !errors.Is(rollbackErr, sql.ErrTxDone) {
+			return fmt.Errorf("%w; rolling back transaction: %w",
+				err, rollbackErr)
+		}
+
 		return err
 	}
 

internal/database/repositories_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"database/sql"
 	"errors"
+	"strings"
 	"testing"
 	"time"
 
@@ -235,6 +236,125 @@ func TestRepositoriesTransactionRollback(t *testing.T) {
 	}
 }
 
+// When the context is cancelled during the body, database/sql rolls the
+// transaction back itself, so WithTx's own rollback finds it already
+// done. WithTx must return the body's error alone, without that rollback.
+func TestRepositoriesTransactionContextCancelled(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	repos := database.NewRepositories(db)
+
+	err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		file := &database.File{
+			Path:  "/test/cancelled_file.txt",
+			MTime: time.Now().Truncate(time.Second),
+			Size:  1024,
+			Mode:  0644,
+			UID:   1000,
+			GID:   1000,
+		}
+
+		err := repos.Files.Create(ctx, tx, file)
+		if err != nil {
+			return err
+		}
+
+		cancel()
+
+		// database/sql rolls back from a goroutine of its own; wait
+		// until it has, so the test does not depend on that race.
+		for {
+			_, err = tx.ExecContext(context.WithoutCancel(ctx), "SELECT 1")
+			if errors.Is(err, sql.ErrTxDone) {
+				return ctx.Err()
+			}
+
+			time.Sleep(time.Millisecond)
+		}
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got: %v", err)
+	}
+
+	if errors.Is(err, sql.ErrTxDone) {
+		t.Fatalf("the already-done rollback was reported: %v", err)
+	}
+
+	file, err := repos.Files.GetByPath(
+		context.Background(), "/test/cancelled_file.txt",
+	)
+	if err != nil {
+		t.Fatalf("error checking for file: %v", err)
+	}
+
+	if file != nil {
+		t.Error("file should not exist after the cancelled transaction")
+	}
+}
+
+// A body that ends the transaction itself with ROLLBACK makes WithTx's
+// own rollback fail. WithTx must return that failure together with the
+// body's error.
+func TestRepositoriesTransactionRollbackFails(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		_, err := tx.ExecContext(ctx, "ROLLBACK")
+		if err != nil {
+			return err
+		}
+
+		return errIntentionalRollback
+	})
+	if !errors.Is(err, errIntentionalRollback) {
+		t.Fatalf("expected the body's error, got: %v", err)
+	}
+
+	if !strings.Contains(err.Error(), "rolling back transaction") {
+		t.Errorf("expected the failed rollback in the error, got: %v", err)
+	}
+}
+
+// When the body ends the transaction itself with ROLLBACK and then
+// panics, WithTx's own rollback fails; the panic must still reach the
+// caller.
+func TestRepositoriesTransactionPanicRollbackFails(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	defer func() {
+		if recover() == nil {
+			t.Error("expected the body's panic to reach the caller")
+		}
+	}()
+
+	_ = repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		_, err := tx.ExecContext(ctx, "ROLLBACK")
+		if err != nil {
+			return err
+		}
+
+		panic(errIntentionalRollback)
+	})
+}
+
 func TestRepositoriesReadTransaction(t *testing.T) {
 	t.Parallel()
 
@@ -291,3 +411,41 @@ func TestRepositoriesReadTransaction(t *testing.T) {
 		t.Error("expected to retrieve file in read transaction")
 	}
 }
+
+// When the context is cancelled during a read transaction's body,
+// database/sql rolls the transaction back itself, so WithReadTx's own
+// rollback finds it already done. WithReadTx must return the body's error
+// alone, without that rollback.
+func TestRepositoriesReadTransactionContextCancelled(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	repos := database.NewRepositories(db)
+
+	err := repos.WithReadTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		cancel()
+
+		// database/sql rolls back from a goroutine of its own; wait
+		// until it has, so the test does not depend on that race.
+		for {
+			_, err := tx.ExecContext(context.WithoutCancel(ctx), "SELECT 1")
+			if errors.Is(err, sql.ErrTxDone) {
+				return ctx.Err()
+			}
+
+			time.Sleep(time.Millisecond)
+		}
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got: %v", err)
+	}
+
+	if errors.Is(err, sql.ErrTxDone) {
+		t.Fatalf("the already-done rollback was reported: %v", err)
+	}
+}

internal/database/schema/001.sql

@@ -86,8 +86,10 @@ CREATE TABLE IF NOT EXISTS snapshots (
     completed_at INTEGER,
     file_count INTEGER NOT NULL DEFAULT 0,
     chunk_count INTEGER NOT NULL DEFAULT 0,
+    new_chunk_size INTEGER NOT NULL DEFAULT 0,
     blob_count INTEGER NOT NULL DEFAULT 0,
     total_size INTEGER NOT NULL DEFAULT 0,
+    uncompressed_chunk_size INTEGER NOT NULL DEFAULT 0,
     blob_size INTEGER NOT NULL DEFAULT 0,
     blob_uncompressed_size INTEGER NOT NULL DEFAULT 0,
     compression_ratio REAL NOT NULL DEFAULT 1.0,

internal/database/snapshots.go

@@ -120,7 +120,11 @@ func (r *SnapshotRepository) UpdateCounts(
 	return nil
 }
 
-// UpdateExtendedStats updates extended statistics for a snapshot
+// UpdateExtendedStats updates extended statistics for a snapshot. It also
+// records the total plaintext size of the distinct chunks the snapshot's
+// files reference. Only the snapshot's own backup can take that total: a
+// file's row is shared by every snapshot that lists it, and a later
+// backup of a changed file rewrites the row with the new chunks.
 func (r *SnapshotRepository) UpdateExtendedStats(
 	ctx context.Context,
 	tx *sql.Tx,
@@ -129,6 +133,7 @@ func (r *SnapshotRepository) UpdateExtendedStats(
 	compressionLevel int,
 	uploadBytes int64,
 	uploadDurationMs int64,
+	newChunkSize int64,
 ) error {
 	compressionRatio, err := r.extendedCompressionRatio(
 		ctx, tx, snapshotID, blobUncompressedSize,
@@ -143,18 +148,29 @@ func (r *SnapshotRepository) UpdateExtendedStats(
 		    compression_ratio = ?,
 		    compression_level = ?,
 		    upload_bytes = ?,
-		    upload_duration_ms = ?
+		    upload_duration_ms = ?,
+		    new_chunk_size = ?,
+		    uncompressed_chunk_size = (
+		        SELECT COALESCE(SUM(c.size), 0)
+		        FROM (
+		            SELECT DISTINCT fc.chunk_hash
+		            FROM snapshot_files sf
+		            JOIN file_chunks fc ON sf.file_id = fc.file_id
+		            WHERE sf.snapshot_id = snapshots.id
+		        ) sc
+		        JOIN chunks c ON sc.chunk_hash = c.chunk_hash
+		    )
 		WHERE id = ?
 	`
 
 	if tx != nil {
 		_, err = tx.ExecContext(ctx, query,
 			blobUncompressedSize, compressionRatio, compressionLevel,
-			uploadBytes, uploadDurationMs, snapshotID)
+			uploadBytes, uploadDurationMs, newChunkSize, snapshotID)
 	} else {
 		_, err = r.db.ExecWithLog(ctx, query,
 			blobUncompressedSize, compressionRatio, compressionLevel,
-			uploadBytes, uploadDurationMs, snapshotID)
+			uploadBytes, uploadDurationMs, newChunkSize, snapshotID)
 	}
 
 	if err != nil {
@@ -239,12 +255,7 @@ func (r *SnapshotRepository) GetOnlySnapshot(ctx context.Context) (*Snapshot, er
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	snapshots, err := r.scanSnapshotRows(rows)
 	if err != nil {
@@ -279,12 +290,7 @@ func (r *SnapshotRepository) ListRecent(
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -499,12 +505,7 @@ func (r *SnapshotRepository) GetBlobHashes(
 		return nil, fmt.Errorf("querying blob hashes: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var blobs []string
 
@@ -568,71 +569,42 @@ func (r *SnapshotRepository) GetSnapshotBlobSizes(
 	return compressed, uncompressed, nil
 }
 
-// GetSnapshotUncompressedChunkSize returns the sum of plaintext sizes of all unique
-// chunks referenced by a snapshot (via snapshot_files → file_chunks → chunks).
+// GetSnapshotUncompressedChunkSize returns the total plaintext size of the
+// distinct chunks the snapshot's files reference, as the snapshot's backup
+// recorded it (see UpdateExtendedStats).
 func (r *SnapshotRepository) GetSnapshotUncompressedChunkSize(
 	ctx context.Context, snapshotID string,
 ) (int64, error) {
-	query := `
-		SELECT COALESCE(SUM(c.size), 0)
-		FROM (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshot_files sf
-			JOIN file_chunks fc ON sf.file_id = fc.file_id
-			WHERE sf.snapshot_id = ?
-		) sc
-		JOIN chunks c ON sc.chunk_hash = c.chunk_hash
-	`
+	query := `SELECT uncompressed_chunk_size FROM snapshots WHERE id = ?`
 
-	var totalSize int64
+	var uncompressedChunkSize int64
 
-	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&totalSize)
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&uncompressedChunkSize)
 	if err != nil {
 		return 0, fmt.Errorf("querying uncompressed chunk size: %w", err)
 	}
 
-	return totalSize, nil
+	return uncompressedChunkSize, nil
 }
 
-// GetSnapshotNewChunkSize returns the sum of plaintext sizes of chunks that are
-// referenced by this snapshot but not by any earlier completed snapshot known to
-// the local database. The result is the marginal uncompressed data this snapshot
-// added to the dedup pool — i.e., the delta from prior snapshots.
+// GetSnapshotNewChunkSize returns the total plaintext size of the chunks the
+// snapshot's backup stored that were not stored before, as the backup
+// recorded it. It is not derived from the snapshot's files: a file's row is
+// shared by every snapshot that lists it and is rewritten when the file
+// changes, so it shows only the file's latest chunks.
 func (r *SnapshotRepository) GetSnapshotNewChunkSize(
 	ctx context.Context, snapshotID string,
 ) (int64, error) {
-	query := `
-		WITH this_snap_chunks AS (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshot_files sf
-			JOIN file_chunks fc ON sf.file_id = fc.file_id
-			WHERE sf.snapshot_id = ?
-		),
-		prior_chunks AS (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshots s
-			JOIN snapshot_files sf ON sf.snapshot_id = s.id
-			JOIN file_chunks fc ON fc.file_id = sf.file_id
-			WHERE s.completed_at IS NOT NULL
-			  AND s.id != ?
-			  AND s.started_at < (SELECT started_at FROM snapshots WHERE id = ?)
-		)
-		SELECT COALESCE(SUM(c.size), 0)
-		FROM chunks c
-		JOIN this_snap_chunks t ON c.chunk_hash = t.chunk_hash
-		WHERE c.chunk_hash NOT IN (SELECT chunk_hash FROM prior_chunks)
-	`
+	query := `SELECT new_chunk_size FROM snapshots WHERE id = ?`
 
-	var totalSize int64
+	var newChunkSize int64
 
-	err := r.db.conn.QueryRowContext(
-		ctx, query, snapshotID, snapshotID, snapshotID,
-	).Scan(&totalSize)
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&newChunkSize)
 	if err != nil {
 		return 0, fmt.Errorf("querying new chunk size: %w", err)
 	}
 
-	return totalSize, nil
+	return newChunkSize, nil
 }
 
 // GetIncompleteSnapshots returns all snapshots that haven't been completed
@@ -653,12 +625,7 @@ func (r *SnapshotRepository) GetIncompleteSnapshots(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -681,12 +648,7 @@ func (r *SnapshotRepository) GetIncompleteByHostname(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	// Same column set as every other multi-row snapshot query, so the
 	// shared scanner applies — including its timestamp normalization.

internal/log/tty_handler.go

@@ -10,6 +10,8 @@ import (
 	"sync"
 	"time"
 	"unicode"
+
+	"github.com/dustin/go-humanize"
 )
 
 // groupSeparator joins an open group path to an attribute key. This
@@ -304,18 +306,13 @@ func formatDuration(d time.Duration) string {
 	}
 }
 
-// formatBytes formats bytes in a human-readable way
+// formatBytes formats a byte count with humanize.Bytes, as every command
+// prints one. humanize.Bytes takes no negative count, so one is printed
+// as it is.
 func formatBytes(b int64) string {
-	const unit = 1024
-	if b < unit {
+	if b < 0 {
 		return fmt.Sprintf("%d B", b)
 	}
 
-	div, exp := int64(unit), 0
-	for n := b / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-
-	return fmt.Sprintf("%.1f %cB", float64(b)/float64(div), "KMGTPE"[exp])
+	return humanize.Bytes(uint64(b))
 }

internal/log/tty_handler_test.go

@@ -200,7 +200,8 @@ func TestTTYHandlerWithGroupQualifiesKeys(t *testing.T) {
 func TestTTYHandlerByteFormattingSurvivesGrouping(t *testing.T) {
 	t.Parallel()
 
-	const oneAndAHalfKiB = 1536
+	// Printed as "1.0 MB", in powers of 1000 like every other byte count.
+	const oneMegabyte = 1_000_000
 
 	for name, testCase := range map[string]struct {
 		derive func(*slog.Logger) *slog.Logger
@@ -223,13 +224,13 @@ func TestTTYHandlerByteFormattingSurvivesGrouping(t *testing.T) {
 			var buf bytes.Buffer
 
 			logger := slog.New(log.NewTTYHandler(&buf, debugHandlerOptions()))
-			testCase.derive(logger).Info("uploaded", "bytes", oneAndAHalfKiB)
+			testCase.derive(logger).Info("uploaded", "bytes", oneMegabyte)
 
 			line := ansiEscape.ReplaceAllString(buf.String(), "")
 
-			assert.Contains(t, line, testCase.key+"=1.5 KB",
+			assert.Contains(t, line, testCase.key+"=1.0 MB",
 				"a byte count must be human-readable however it is qualified")
-			assert.NotContains(t, line, strconv.Itoa(oneAndAHalfKiB),
+			assert.NotContains(t, line, strconv.Itoa(oneMegabyte),
 				"the raw number must not survive the formatting")
 		})
 	}

internal/s3/client.go

@@ -8,6 +8,7 @@ import (
 	"io"
 	"strings"
 	"sync/atomic"
+	"time"
 
 	"github.com/aws/aws-sdk-go-v2/aws"
 	"github.com/aws/aws-sdk-go-v2/config"
@@ -18,6 +19,13 @@ import (
 	"github.com/aws/smithy-go/logging"
 )
 
+// ErrNoContinuationToken is returned by ListObjects and ListObjectsStream
+// when the server marks a page of a listing truncated but sends no
+// continuation token to fetch the rest with, as a server answering V2
+// listings V1-style does.
+var ErrNoContinuationToken = errors.New(
+	"server marked the listing truncated but sent no continuation token")
+
 // Client wraps the AWS S3 client for vaultik operations.
 // It provides a simplified interface for S3 operations with automatic
 // prefix handling and connection management. All operations are performed
@@ -214,7 +222,7 @@ func (c *Client) ListObjects(ctx context.Context, prefix string) ([]string, erro
 	})
 
 	for paginator.HasMorePages() {
-		page, err := paginator.NextPage(ctx)
+		page, err := nextListPage(ctx, paginator)
 		if err != nil {
 			return nil, err
 		}
@@ -235,6 +243,26 @@ func (c *Client) ListObjects(ctx context.Context, prefix string) ([]string, erro
 	return keys, nil
 }
 
+// nextListPage fetches the next page of a listing. The SDK paginator takes
+// a page marked truncated without a continuation token for the last page
+// and stops without an error, so such a page is ErrNoContinuationToken
+// here instead.
+func nextListPage(
+	ctx context.Context, paginator *s3.ListObjectsV2Paginator,
+) (*s3.ListObjectsV2Output, error) {
+	page, err := paginator.NextPage(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	truncated := aws.ToBool(page.IsTruncated)
+	if truncated && aws.ToString(page.NextContinuationToken) == "" {
+		return nil, ErrNoContinuationToken
+	}
+
+	return page, nil
+}
+
 // HeadObject checks if an object exists in S3.
 // Returns true if the object exists, false otherwise.
 // The key is automatically prefixed with the configured prefix.
@@ -275,7 +303,10 @@ func IsNotFound(err error) bool {
 type ObjectInfo struct {
 	Key  string
 	Size int64
-	Err  error
+	// ModTime is the object's LastModified time. Only ListObjectsStream
+	// sets it.
+	ModTime time.Time
+	Err     error
 }
 
 // ListObjectsStream lists objects with the given prefix and returns a channel.
@@ -300,7 +331,7 @@ func (c *Client) ListObjectsStream(
 		})
 
 		for paginator.HasMorePages() {
-			page, err := paginator.NextPage(ctx)
+			page, err := nextListPage(ctx, paginator)
 			if err != nil {
 				ch <- ObjectInfo{Err: err}
 
@@ -316,8 +347,9 @@ func (c *Client) ListObjectsStream(
 					}
 
 					ch <- ObjectInfo{
-						Key:  key,
-						Size: *obj.Size,
+						Key:     key,
+						Size:    *obj.Size,
+						ModTime: aws.ToTime(obj.LastModified),
 					}
 				}
 			}

internal/snapshot/blob_temp_file_test.go

@@ -0,0 +1,62 @@
+//nolint:testpackage // exercises the unexported cleanupBlobTempFile
+package snapshot
+
+import (
+	"errors"
+	"os"
+	"testing"
+
+	"github.com/spf13/afero"
+	"sneak.berlin/go/vaultik/internal/blob"
+)
+
+// A blob temp file that cannot be closed or removed must make the cleanup
+// return an error, not end the process.
+func TestCleanupBlobTempFileReturnsErrors(t *testing.T) {
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	s := &Scanner{fs: fs}
+
+	tests := []struct {
+		name string
+		// prepare runs before the cleanup and makes its close or remove
+		// fail.
+		prepare func(tempFile afero.File) error
+		want    error
+	}{
+		{
+			name:    "close fails",
+			prepare: func(tempFile afero.File) error { return tempFile.Close() },
+			want:    os.ErrClosed,
+		},
+		{
+			name: "remove fails",
+			prepare: func(tempFile afero.File) error {
+				return fs.Remove(tempFile.Name())
+			},
+			want: os.ErrNotExist,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			tempFile, err := afero.TempFile(fs, t.TempDir(), "blob-*.tmp")
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = tt.prepare(tempFile)
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = s.cleanupBlobTempFile(&blob.WithReader{TempFile: tempFile})
+			if !errors.Is(err, tt.want) {
+				t.Fatalf("expected an error wrapping %v, got: %v", tt.want, err)
+			}
+		})
+	}
+}

internal/snapshot/exclude_test.go

@@ -531,3 +531,101 @@ func TestExcludePatterns_UnanchoredPatternFile(t *testing.T) {
 	require.Equal(t, 1, result.FilesScanned,
 		"Unanchored pattern for file should match anywhere")
 }
+
+func TestExcludePatterns_DoubleStarDirectoryAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a leading **/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"**/projectname"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// **/projectname should exclude BOTH:
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 2
+	require.Equal(t, 2, result.FilesScanned,
+		"**/ directory pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_DoubleStarFileAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a leading **/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"**/*.txt"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// **/*.txt should exclude ALL .txt files:
+	// - /backup/file.txt
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 3
+	require.Equal(t, 1, result.FilesScanned,
+		"**/ file pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_AnchoredDoubleStarAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, /**/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"/**/projectname"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// /**/projectname should exclude BOTH:
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 2
+	require.Equal(t, 2, result.FilesScanned,
+		"/**/ pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_DoubleStarInMiddleMatchesZeroDirectories(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a/**/b matches a/b as well as a/c/b. The last part
+	// is a glob: with a plain name the glob library already matched a/b.
+	fs := &afero.MemMapFs{}
+	files := []string{"/backup/a/x.log", "/backup/a/c/y.log", "/backup/a/c/kept.txt"}
+
+	for _, path := range files {
+		require.NoError(t, fs.MkdirAll(filepath.Dir(path), 0755))
+		require.NoError(t, afero.WriteFile(fs, path, []byte(path), 0644))
+	}
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"a/**/*.log"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// a/**/*.log should exclude /backup/a/x.log and /backup/a/c/y.log
+	// Total files: 3, excluded: 2
+	require.Equal(t, 1, result.FilesScanned,
+		"a/**/*.log should match a/x.log as well as a/c/y.log")
+}

internal/snapshot/module.go

@@ -16,6 +16,9 @@ type ScannerParams struct {
 	Fs             afero.Fs
 	Exclude        []string // Exclude patterns (combined global + snapshot-specific)
 	SkipErrors     bool     // Skip file read errors (log loudly but continue)
+	// PreviousSnapshotID is the most recent completed snapshot of the
+	// same name, "" for none.
+	PreviousSnapshotID string
 }
 
 // Module exports backup functionality as an fx module.
@@ -44,17 +47,18 @@ func provideScannerFactory(
 		}
 
 		return NewScanner(ScannerConfig{
-			FS:               params.Fs,
-			ChunkSize:        cfg.ChunkSize.Int64(),
-			Repositories:     repos,
-			Storage:          storer,
-			MaxBlobSize:      cfg.BlobSizeLimit.Int64(),
-			CompressionLevel: cfg.CompressionLevel,
-			AgeRecipients:    cfg.AgeRecipients,
-			EnableProgress:   params.EnableProgress,
-			UI:               params.UI,
-			Exclude:          excludes,
-			SkipErrors:       params.SkipErrors,
+			FS:                 params.Fs,
+			ChunkSize:          cfg.ChunkSize.Int64(),
+			Repositories:       repos,
+			Storage:            storer,
+			MaxBlobSize:        cfg.BlobSizeLimit.Int64(),
+			CompressionLevel:   cfg.CompressionLevel,
+			AgeRecipients:      cfg.AgeRecipients,
+			EnableProgress:     params.EnableProgress,
+			UI:                 params.UI,
+			Exclude:            excludes,
+			SkipErrors:         params.SkipErrors,
+			PreviousSnapshotID: params.PreviousSnapshotID,
 		})
 	}
 }

internal/snapshot/progress_test.go

@@ -78,8 +78,7 @@ func TestProgressPercentWithSecondPathSmaller(t *testing.T) {
 
 	stats := progress.GetStats()
 
-	// Directories count toward the total size but produce no chunks, so
-	// the percentage ends just under 100%.
+	// With both paths counted, the percentage must not exceed 100%.
 	percent := float64(stats.BytesProcessed.Load()) /
 		float64(stats.TotalSize.Load()) * 100
 	if percent > 100 {

internal/snapshot/scanner.go

@@ -58,6 +58,10 @@ type Scanner struct {
 	compressionLevel int
 	ageRecipient     string
 	snapshotID       string // Current snapshot being processed
+	// previousSnapshotID is the most recent completed snapshot of the
+	// same name, "" for none. Only a file it lists can be counted as
+	// deleted.
+	previousSnapshotID string
 	// currentSourcePath is the source directory being scanned, stored with
 	// each file record.
 	currentSourcePath string
@@ -129,13 +133,20 @@ type ScannerConfig struct {
 	// SkipErrors skips files that cannot be opened or read (log loudly but
 	// continue); packer, database, encryption, and upload errors still abort.
 	SkipErrors bool
+	// PreviousSnapshotID is the most recent completed snapshot of the
+	// same name, "" for none.
+	PreviousSnapshotID string
 }
 
 // ScanResult contains the results of a scan operation. Files and bytes
 // are counted per file: BytesScanned is the size of the new and changed
 // files, BytesSkipped that of the unchanged ones. FilesFailed counts the
-// new and changed files that phase 2 could not store; FilesScanned
-// includes them and BytesScanned does not.
+// new and changed files that phase 2 could not store, each entry removed
+// after the walk listed its directory and, under --skip-errors, each
+// entry the walk could not read, such as a directory it could not list;
+// FilesScanned includes them and BytesScanned does not. ChunksCreated
+// counts the chunks that were not already stored, and NewChunkSize is
+// their total size.
 type ScanResult struct {
 	FilesScanned   int
 	FilesSkipped   int
@@ -145,12 +156,18 @@ type ScanResult struct {
 	BytesSkipped   int64
 	BytesDeleted   int64
 	ChunksCreated  int
+	NewChunkSize   int64
 	BlobsCreated   int
 	BlobsUploaded  int
 	BytesUploaded  int64
 	UploadDuration time.Duration
 	StartTime      time.Time
 	EndTime        time.Time
+
+	// FilesUnreadable is the part of FilesFailed that --skip-errors
+	// skipped because it could not be read or listed; the rest are files
+	// removed after the walk listed them.
+	FilesUnreadable int
 }
 
 // NewScanner creates a new scanner instance
@@ -204,6 +221,7 @@ func NewScanner(cfg ScannerConfig) *Scanner {
 		compiledExclude:    compiledExclude,
 		progress:           progress,
 		skipErrors:         cfg.SkipErrors,
+		previousSnapshotID: cfg.PreviousSnapshotID,
 		ui:                 uiw,
 		pendingChunkHashes: make(map[string]struct{}),
 	}
@@ -255,8 +273,9 @@ func (s *Scanner) Scan(
 
 	filesToProcess := scanResult.FilesToProcess
 
-	// Phase 1b: Detect deleted files by comparing DB against scanned files
-	err = s.detectDeletedFilesFromMap(ctx, knownFiles, existingFiles, result)
+	// Phase 1b: Detect deleted files by comparing the previous snapshot
+	// of this name against scanned files
+	err = s.detectDeletedFiles(ctx, path, existingFiles, result)
 	if err != nil {
 		return nil, fmt.Errorf("detecting deleted files: %w", err)
 	}
@@ -272,6 +291,11 @@ func (s *Scanner) Scan(
 		}
 	}
 
+	err = s.recordDirectories(ctx, scanResult.Directories)
+	if err != nil {
+		return nil, err
+	}
+
 	// Summarize scan phase results and update progress
 	s.summarizeScanPhase(result, filesToProcess)
 
@@ -436,9 +460,8 @@ func (s *Scanner) summarizeScanPhase(
 }
 
 // loadKnownFiles loads the known files at and beneath path from the
-// database into a map for fast lookup. Every loaded file the scan does
-// not find is counted as deleted. This avoids per-file database queries
-// during the scan phase.
+// database into a map for fast lookup. This avoids per-file database
+// queries during the scan phase.
 func (s *Scanner) loadKnownFiles(
 	ctx context.Context, path string,
 ) (map[string]*database.File, error) {
@@ -844,14 +867,15 @@ func (s *Scanner) batchInsertFileData(
 // ScanPhaseResult contains the results of the scan phase
 type ScanPhaseResult struct {
 	FilesToProcess   []*FileToProcess
-	UnchangedFileIDs []types.FileID // IDs of unchanged files to associate with snapshot
+	UnchangedFileIDs []types.FileID   // IDs of unchanged files to associate with snapshot
+	Directories      []*FileToProcess // Every directory found, to record
 }
 
 // scanPhase performs the initial directory scan to identify files to process
 // It uses the pre-loaded knownFiles map for fast change detection without DB
 // queries. It also populates existingFiles map for deletion detection.
-// Returns files needing processing and IDs of unchanged files for snapshot
-// association.
+// Returns files needing processing, IDs of unchanged files for snapshot
+// association, and the directories found.
 func (s *Scanner) scanPhase(
 	ctx context.Context,
 	path string,
@@ -859,8 +883,10 @@ func (s *Scanner) scanPhase(
 	existingFiles map[string]struct{},
 	knownFiles map[string]*database.File,
 ) (*ScanPhaseResult, error) {
-	// Use known file count as estimate for progress (accurate for subsequent backups)
-	estimatedTotal := int64(len(knownFiles))
+	// Use known file count as estimate for progress (accurate for
+	// subsequent backups). The walk does not count directories, so the
+	// estimate leaves them out.
+	estimatedTotal := countNonDirectories(knownFiles)
 
 	collector := &scanCollector{}
 
@@ -871,8 +897,28 @@ func (s *Scanner) scanPhase(
 	log.Debug("Starting directory walk", "path", path)
 
 	walkFn := func(filePath string, info os.FileInfo, err error) error {
+		// Check exclude patterns first, so an excluded entry that is gone
+		// or cannot be looked up is neither counted nor reported. For
+		// directories, skip the entire subtree. info is nil when the
+		// walk could not look the entry up.
+		if s.shouldExclude(filePath, path) {
+			if info != nil && info.IsDir() {
+				return filepath.SkipDir
+			}
+
+			return nil
+		}
+
+		// The source path is in no directory listing, so its absence is
+		// an error rather than a removal.
+		if errors.Is(err, os.ErrNotExist) && filePath != path {
+			countRemovedDuringWalk(filePath, existingFiles, result)
+
+			return nil
+		}
+
 		if err != nil {
-			return s.handleWalkError(filePath, err)
+			return s.handleWalkError(filePath, err, result)
 		}
 
 		// Check context cancellation
@@ -882,18 +928,9 @@ func (s *Scanner) scanPhase(
 		default:
 		}
 
-		// Check exclude patterns - for directories, skip the entire subtree
-		if s.shouldExclude(filePath, path) {
-			if info.IsDir() {
-				return filepath.SkipDir
-			}
-
-			return nil
-		}
-
 		// Handle symlinks and directories
 		handled, err := s.recordSpecialEntry(
-			filePath, info, existingFiles, collector, result)
+			filePath, info, existingFiles, knownFiles, collector, result)
 		if handled {
 			return err
 		}
@@ -907,9 +944,10 @@ func (s *Scanner) scanPhase(
 		existingFiles[filePath] = struct{}{}
 
 		// Check file against in-memory map (no DB query!)
-		file, needsProcessing := s.checkFileInMemory(filePath, info, knownFiles)
+		file := s.buildRegularFileEntry(filePath, info)
+		needsProcessing := s.checkFileInMemory(file, knownFiles)
 
-		filesScanned, changedCount := collector.recordRegularFile(
+		filesScanned, changedCount := collector.recordFile(
 			filePath, info, file, needsProcessing)
 
 		// Update result stats
@@ -933,27 +971,45 @@ func (s *Scanner) scanPhase(
 	return &ScanPhaseResult{
 		FilesToProcess:   collector.filesToProcess,
 		UnchangedFileIDs: collector.unchangedFileIDs,
+		Directories:      collector.directories,
 	}, nil
 }
 
+// countNonDirectories returns how many of files are not directories.
+func countNonDirectories(files map[string]*database.File) int64 {
+	var count int64
+
+	for _, f := range files {
+		if f.Mode&uint32(os.ModeDir) == 0 {
+			count++
+		}
+	}
+
+	return count
+}
+
 // recordSpecialEntry records symlinks and directories (which have no
-// data to chunk) and reports whether it handled the entry. For a symlink
-// whose target cannot be read it returns handleWalkError's result.
+// data to chunk) and reports whether it handled the entry. A symlink is
+// counted and checked for changes as a regular file is; a directory is
+// set aside to be recorded and is not counted. For a symlink whose
+// target cannot be read it returns handleWalkError's result.
 func (s *Scanner) recordSpecialEntry(
 	filePath string, info os.FileInfo,
 	existingFiles map[string]struct{},
+	knownFiles map[string]*database.File,
 	collector *scanCollector, result *ScanResult,
 ) (bool, error) {
 	// Handle symlinks
 	if info.Mode()&os.ModeSymlink != 0 {
 		file, err := s.buildSymlinkEntry(filePath, info)
 		if err != nil {
-			return true, s.handleWalkError(filePath, err)
+			return true, s.handleWalkError(filePath, err, result)
 		}
 
 		existingFiles[filePath] = struct{}{}
-		collector.addToProcess(filePath, info, file)
-		s.updateScanEntryStats(result, true, info)
+		needsProcessing := s.checkFileInMemory(file, knownFiles)
+		collector.recordFile(filePath, info, file, needsProcessing)
+		s.updateScanEntryStats(result, needsProcessing, info)
 
 		return true, nil
 	}
@@ -963,7 +1019,7 @@ func (s *Scanner) recordSpecialEntry(
 	if info.IsDir() {
 		file := s.buildDirectoryEntry(filePath, info)
 		existingFiles[filePath] = struct{}{}
-		collector.addToProcess(filePath, info, file)
+		collector.addDirectory(filePath, info, file)
 
 		return true, nil
 	}
@@ -971,15 +1027,38 @@ func (s *Scanner) recordSpecialEntry(
 	return false, nil
 }
 
+// countRemovedDuringWalk counts an entry that was in its directory's
+// listing but was gone when the walk looked it up, like a file removed
+// before phase 2 reads it: as examined and failed, and not as an error,
+// with or without --skip-errors.
+func countRemovedDuringWalk(
+	filePath string, existingFiles map[string]struct{}, result *ScanResult,
+) {
+	log.Warn("File was deleted during the directory walk, skipping",
+		"path", filePath)
+
+	// The walk found it, so it is not also counted as deleted.
+	existingFiles[filePath] = struct{}{}
+	result.FilesScanned++
+	result.FilesFailed++
+}
+
 // handleWalkError deals with a filesystem error surfaced by the walk:
-// skipped (with a UI notice) under --skip-errors, fatal otherwise.
-func (s *Scanner) handleWalkError(filePath string, err error) error {
+// under --skip-errors the entry is skipped with a UI notice and counted
+// as examined, failed and unreadable, otherwise the error is fatal.
+func (s *Scanner) handleWalkError(
+	filePath string, err error, result *ScanResult,
+) error {
 	if s.skipErrors {
 		log.Error("Failed to access file (skipping due to --skip-errors)",
 			"path", filePath, "error", err)
 		s.ui.Errorf("Failed to access %s: %v. Skipping (--skip-errors).",
 			s.ui.Path(filePath), err)
 
+		result.FilesScanned++
+		result.FilesFailed++
+		result.FilesUnreadable++
+
 		return nil // Continue scanning
 	}
 
@@ -993,28 +1072,29 @@ type scanCollector struct {
 	mu               sync.Mutex
 	filesToProcess   []*FileToProcess
 	unchangedFileIDs []types.FileID // Just IDs - no new records needed
+	directories      []*FileToProcess
 	filesScanned     int64
 }
 
-// addToProcess queues a file entry for processing.
-func (c *scanCollector) addToProcess(
+// addDirectory sets a directory aside to be recorded. It is not counted
+// as a file.
+func (c *scanCollector) addDirectory(
 	path string, info os.FileInfo, file *database.File,
 ) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
-	c.filesToProcess = append(c.filesToProcess, &FileToProcess{
+	c.directories = append(c.directories, &FileToProcess{
 		Path:     path,
 		FileInfo: info,
 		File:     file,
 	})
-	c.filesScanned++
 }
 
-// recordRegularFile records a regular file as changed (to process) or
-// unchanged (snapshot association only) and returns the running scanned
-// and changed counts.
-func (c *scanCollector) recordRegularFile(
+// recordFile records a regular file or symlink as changed (to process)
+// or unchanged (snapshot association only) and returns the running
+// scanned and changed counts.
+func (c *scanCollector) recordFile(
 	path string, info os.FileInfo, file *database.File, needsProcessing bool,
 ) (int64, int) {
 	c.mu.Lock()
@@ -1117,7 +1197,28 @@ func (s *Scanner) printScanProgressLine(
 	}
 }
 
-// buildSymlinkEntry creates a File record for a symlink.
+// buildRegularFileEntry creates a File record for a regular file, with
+// no ID: checkFileInMemory sets it.
+func (s *Scanner) buildRegularFileEntry(path string, info os.FileInfo) *database.File {
+	var uid, gid uint32
+	if stat, ok := info.Sys().(*syscall.Stat_t); ok {
+		uid = stat.Uid
+		gid = stat.Gid
+	}
+
+	return &database.File{
+		Path:       types.FilePath(path),
+		SourcePath: types.SourcePath(s.currentSourcePath),
+		MTime:      info.ModTime(),
+		Size:       info.Size(),
+		Mode:       uint32(info.Mode()),
+		UID:        uid,
+		GID:        gid,
+	}
+}
+
+// buildSymlinkEntry creates a File record for a symlink, with no ID:
+// checkFileInMemory sets it.
 func (s *Scanner) buildSymlinkEntry(
 	path string, info os.FileInfo,
 ) (*database.File, error) {
@@ -1133,7 +1234,6 @@ func (s *Scanner) buildSymlinkEntry(
 	}
 
 	return &database.File{
-		ID:         types.NewFileID(),
 		Path:       types.FilePath(path),
 		SourcePath: types.SourcePath(s.currentSourcePath),
 		MTime:      info.ModTime(),
@@ -1165,6 +1265,22 @@ 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.
+func (s *Scanner) recordDirectories(
+	ctx context.Context, directories []*FileToProcess,
+) error {
+	for _, dir := range directories {
+		err := s.recordNonRegularFile(ctx, dir)
+		if err != nil {
+			return fmt.Errorf("recording directory %s: %w", dir.Path, err)
+		}
+	}
+
+	return nil
+}
+
 // recordNonRegularFile writes a symlink or directory entry to the database
 // and associates it with the current snapshot. No chunking is performed.
 func (s *Scanner) recordNonRegularFile(ctx context.Context, ftp *FileToProcess) error {
@@ -1178,62 +1294,36 @@ func (s *Scanner) recordNonRegularFile(ctx context.Context, ftp *FileToProcess)
 	})
 }
 
-// checkFileInMemory checks if a file needs processing using the in-memory map
+// checkFileInMemory reports whether a regular file or symlink needs
+// processing, and gives file the ID of its known record, or a new ID for
+// a new path.
 // No database access is performed - this is purely CPU/memory work
 func (s *Scanner) checkFileInMemory(
-	path string, info os.FileInfo, knownFiles map[string]*database.File,
-) (*database.File, bool) {
-	var uid, gid uint32
-	if stat, ok := info.Sys().(*syscall.Stat_t); ok {
-		uid = stat.Uid
-		gid = stat.Gid
-	}
-
-	// Check against in-memory map first to get existing ID if available
-	existingFile, exists := knownFiles[path]
-
-	// Create file record with ID set upfront
-	// For new files, generate UUID immediately so it's available for chunk associations
-	// For existing files, reuse the existing ID
-	var fileID types.FileID
-	if exists {
-		fileID = existingFile.ID
-	} else {
-		fileID = types.NewFileID()
-	}
-
-	file := &database.File{
-		ID:         fileID,
-		Path:       types.FilePath(path),
-		SourcePath: types.SourcePath(s.currentSourcePath),
-		MTime:      info.ModTime(),
-		Size:       info.Size(),
-		Mode:       uint32(info.Mode()),
-		UID:        uid,
-		GID:        gid,
-	}
+	file *database.File, knownFiles map[string]*database.File,
+) bool {
+	existingFile, exists := knownFiles[file.Path.String()]
 
 	// New file - needs processing
 	if !exists {
-		return file, true
+		file.ID = types.NewFileID()
+
+		return true
 	}
 
+	file.ID = existingFile.ID
+
 	// No uploaded blob holds one of its chunks (see loadFilesToRechunk)
-	if _, rechunk := s.filesToRechunk[fileID]; rechunk {
-		return file, true
+	if _, rechunk := s.filesToRechunk[file.ID]; rechunk {
+		return true
 	}
 
 	// Check if file has changed
-	if existingFile.Size != file.Size ||
+	return existingFile.Size != file.Size ||
 		!existingFile.MTime.Equal(file.MTime) ||
 		existingFile.Mode != file.Mode ||
 		existingFile.UID != file.UID ||
-		existingFile.GID != file.GID {
-		return file, true
-	}
-
-	// File unchanged
-	return file, false
+		existingFile.GID != file.GID ||
+		existingFile.LinkTarget != file.LinkTarget
 }
 
 // batchAddFilesToSnapshot adds existing file IDs to the snapshot association table
@@ -1295,6 +1385,11 @@ func (s *Scanner) batchAddFilesToSnapshot(
 func (s *Scanner) processPhase(
 	ctx context.Context, filesToProcess []*FileToProcess, result *ScanResult,
 ) error {
+	// On an error or cancellation the blob being packed is never
+	// finalized, so its temp file must be removed here. After a
+	// successful flush there is no such blob and this does nothing.
+	defer s.packer.DiscardCurrentBlob()
+
 	// Calculate total bytes to process
 	var totalBytes int64
 	for _, f := range filesToProcess {
@@ -1355,17 +1450,16 @@ func (s *Scanner) processPhase(
 	return s.finalizeProcessPhase(ctx, result)
 }
 
-// processFileWithErrorHandling records a directory or symlink, or wraps
+// processFileWithErrorHandling records a symlink, or wraps
 // processFileStreaming for a regular file with error recovery for deleted
 // files and skip-errors mode. Returns (skipped, error).
 func (s *Scanner) processFileWithErrorHandling(
 	ctx context.Context, fileToProcess *FileToProcess, result *ScanResult,
 ) (bool, error) {
-	// A directory or symlink has no data to open or read; it is only
-	// recorded in the local index. An error recording it stops the run
-	// even under --skip-errors, or the snapshot would complete without it.
-	mode := os.FileMode(fileToProcess.File.Mode)
-	if mode&os.ModeSymlink != 0 || mode.IsDir() {
+	// A symlink has no data to open or read; it is only recorded in the
+	// local index. An error recording it stops the run even under
+	// --skip-errors, or the snapshot would complete without it.
+	if os.FileMode(fileToProcess.File.Mode)&os.ModeSymlink != 0 {
 		err := s.recordNonRegularFile(ctx, fileToProcess)
 		if err != nil {
 			return false, fmt.Errorf("processing file %s: %w", fileToProcess.Path, err)
@@ -1406,6 +1500,7 @@ func (s *Scanner) processFileWithErrorHandling(
 				s.ui.Path(fileToProcess.Path), err)
 
 			countFailedFile(fileToProcess, result)
+			result.FilesUnreadable++
 
 			return true, nil
 		}
@@ -1509,7 +1604,8 @@ func (s *Scanner) finalizeProcessPhase(ctx context.Context, result *ScanResult)
 			// snapshot references has uploaded_ts set, so deduplication and
 			// interrupted-run repair treat these blobs as trustworthy.
 			err = s.repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
-				err := s.repos.Blobs.UpdateUploaded(ctx, tx, b.ID)
+				err := s.repos.Blobs.UpdateUploaded(
+					ctx, tx, b.ID, b.Compressed)
 				if err != nil {
 					return fmt.Errorf("marking blob uploaded: %w", err)
 				}
@@ -1545,23 +1641,28 @@ func (s *Scanner) handleBlobReady(
 	blobPath := fmt.Sprintf("blobs/%s/%s/%s",
 		finishedBlob.Hash[:2], finishedBlob.Hash[2:4], finishedBlob.Hash)
 
-	blobExists, err := s.uploadBlobIfNeeded(
+	storedSize, blobExists, err := s.uploadBlobIfNeeded(
 		ctx, blobPath, blobWithReader, startTime, result)
 	if err != nil {
-		s.cleanupBlobTempFile(blobWithReader)
+		cleanupErr := s.cleanupBlobTempFile(blobWithReader)
 
-		return fmt.Errorf("uploading blob %s: %w", finishedBlob.Hash, err)
+		return errors.Join(
+			fmt.Errorf("uploading blob %s: %w", finishedBlob.Hash, err), cleanupErr)
 	}
 
-	err = s.recordBlobMetadata(ctx, finishedBlob, blobExists, startTime)
+	err = s.recordBlobMetadata(
+		ctx, finishedBlob, storedSize, blobExists, startTime)
 	if err != nil {
-		s.cleanupBlobTempFile(blobWithReader)
+		cleanupErr := s.cleanupBlobTempFile(blobWithReader)
 
+		return errors.Join(err, cleanupErr)
+	}
+
+	err = s.cleanupBlobTempFile(blobWithReader)
+	if err != nil {
 		return err
 	}
 
-	s.cleanupBlobTempFile(blobWithReader)
-
 	// Chunks from this blob are now committed to DB - remove from pending set
 	s.removePendingChunkHashes(blobWithReader.InsertedChunkHashes)
 
@@ -1575,20 +1676,22 @@ func (s *Scanner) handleBlobReady(
 }
 
 // uploadBlobIfNeeded uploads the blob to storage if it doesn't already
-// exist, returns whether it existed
+// exist. It returns the size of the object in storage, which for an
+// existing object can differ from the local copy's, and whether it
+// already existed.
 func (s *Scanner) uploadBlobIfNeeded(
 	ctx context.Context,
 	blobPath string,
 	blobWithReader *blob.WithReader,
 	startTime time.Time,
 	result *ScanResult,
-) (bool, error) {
+) (int64, bool, error) {
 	finishedBlob := blobWithReader.FinishedBlob
 
 	// Check if blob already exists (deduplication after restart)
 	destination := s.storage.Info().Location
 
-	_, err := s.storage.Stat(ctx, blobPath)
+	stored, err := s.storage.Stat(ctx, blobPath)
 	if err == nil {
 		log.Info("Blob already exists in storage, skipping upload",
 			"hash", finishedBlob.Hash,
@@ -1597,7 +1700,7 @@ func (s *Scanner) uploadBlobIfNeeded(
 			s.ui.Hex(finishedBlob.Hash), s.ui.Size(finishedBlob.Compressed),
 			s.ui.Path(destination))
 
-		return true, nil
+		return stored.Size, true, nil
 	}
 
 	s.ui.Beginf("Uploading blob %s (%s) to %s.",
@@ -1611,7 +1714,7 @@ func (s *Scanner) uploadBlobIfNeeded(
 	if err != nil {
 		log.Error("Failed to upload blob", "hash", finishedBlob.Hash, "error", err)
 
-		return false, fmt.Errorf("uploading blob to storage: %w", err)
+		return 0, false, fmt.Errorf("uploading blob to storage: %w", err)
 	}
 
 	uploadDuration := time.Since(startTime)
@@ -1642,7 +1745,7 @@ func (s *Scanner) uploadBlobIfNeeded(
 		stats.BytesUploaded.Add(finishedBlob.Compressed)
 	}
 
-	return false, nil
+	return finishedBlob.Compressed, false, nil
 }
 
 // makeUploadProgressCallback creates a progress callback for blob uploads.
@@ -1712,6 +1815,7 @@ func (s *Scanner) makeUploadProgressCallback(
 func (s *Scanner) recordBlobMetadata(
 	ctx context.Context,
 	finishedBlob *blob.FinishedBlob,
+	storedSize int64,
 	blobExists bool,
 	startTime time.Time,
 ) error {
@@ -1723,7 +1827,8 @@ func (s *Scanner) recordBlobMetadata(
 	uploadDuration := time.Since(startTime)
 
 	return s.repos.WithTx(ctx, func(txCtx context.Context, tx *sql.Tx) error {
-		err := s.repos.Blobs.UpdateUploaded(txCtx, tx, finishedBlob.ID)
+		err := s.repos.Blobs.UpdateUploaded(
+			txCtx, tx, finishedBlob.ID, storedSize)
 		if err != nil {
 			return fmt.Errorf("updating blob upload timestamp: %w", err)
 		}
@@ -1754,20 +1859,22 @@ func (s *Scanner) recordBlobMetadata(
 }
 
 // cleanupBlobTempFile closes and removes the blob's temporary file
-func (s *Scanner) cleanupBlobTempFile(blobWithReader *blob.WithReader) {
+func (s *Scanner) cleanupBlobTempFile(blobWithReader *blob.WithReader) error {
 	if blobWithReader.TempFile != nil {
 		tempName := blobWithReader.TempFile.Name()
 
 		err := blobWithReader.TempFile.Close()
 		if err != nil {
-			log.Fatal("Failed to close temp file", "file", tempName, "error", err)
+			return fmt.Errorf("closing blob temp file: %w", err)
 		}
 
 		err = s.fs.Remove(tempName)
 		if err != nil {
-			log.Fatal("Failed to remove temp file", "file", tempName, "error", err)
+			return fmt.Errorf("removing blob temp file: %w", err)
 		}
 	}
+
+	return nil
 }
 
 // streamingChunkInfo tracks chunk metadata collected during streaming
@@ -1861,11 +1968,12 @@ func (s *Scanner) processFileStreaming(
 	return nil
 }
 
-// updateChunkStats counts a chunk that was not already stored. The scan
-// result's file counts, BytesScanned and BytesSkipped are not touched
-// here: the scan phase counts each file once.
+// updateChunkStats counts a chunk that was not already stored, and its
+// size. The scan result's file counts, BytesScanned and BytesSkipped are
+// not touched here: the scan phase counts each file once.
 func (s *Scanner) updateChunkStats(chunkSize int64, result *ScanResult) {
 	result.ChunksCreated++
+	result.NewChunkSize += chunkSize
 
 	if s.progress != nil {
 		s.progress.GetStats().ChunksCreated.Add(1)
@@ -1933,21 +2041,29 @@ func (s *Scanner) queueFileForBatchInsert(
 	})
 }
 
-// detectDeletedFilesFromMap finds files that existed in previous snapshots
-// but no longer exist. Uses pre-loaded maps to avoid any filesystem or
-// database access.
-func (s *Scanner) detectDeletedFilesFromMap(
+// detectDeletedFiles counts as deleted each file at or beneath path that
+// the previous snapshot of this name lists, the walk did not find, and
+// is gone from disk. A file that is excluded, or inside an excluded
+// directory, is not counted: the walk skips it whether or not it is
+// still on disk.
+func (s *Scanner) detectDeletedFiles(
 	ctx context.Context,
-	knownFiles map[string]*database.File,
+	path string,
 	existingFiles map[string]struct{},
 	result *ScanResult,
 ) error {
-	if len(knownFiles) == 0 {
+	if s.previousSnapshotID == "" {
 		return nil
 	}
 
-	// Check each known file against the enumerated set (no filesystem access needed)
-	for path, file := range knownFiles {
+	previousFiles, err := s.repos.Files.ListUnderPathInSnapshot(
+		ctx, path, s.previousSnapshotID)
+	if err != nil {
+		return fmt.Errorf("listing files of snapshot %s: %w",
+			s.previousSnapshotID, err)
+	}
+
+	for _, file := range previousFiles {
 		// Check context cancellation periodically
 		select {
 		case <-ctx.Done():
@@ -1955,13 +2071,22 @@ func (s *Scanner) detectDeletedFilesFromMap(
 		default:
 		}
 
-		// Check if the file exists in our enumerated set
-		if _, exists := existingFiles[path]; !exists {
-			// File has been deleted
-			result.FilesDeleted++
-			result.BytesDeleted += file.Size
-			log.Debug("Detected deleted file", "path", path, "size", file.Size)
+		filePath := file.Path.String()
+
+		_, exists := existingFiles[filePath]
+		if exists || s.excludedOrInExcludedDir(filePath, path) {
+			continue
 		}
+
+		// Under --skip-errors the walk also misses what it cannot reach,
+		// such as the files in a directory it cannot open.
+		if !s.goneFromDisk(filePath) {
+			continue
+		}
+
+		result.FilesDeleted++
+		result.BytesDeleted += file.Size
+		log.Debug("Detected deleted file", "path", filePath, "size", file.Size)
 	}
 
 	if result.FilesDeleted > 0 {
@@ -1972,6 +2097,36 @@ func (s *Scanner) detectDeletedFilesFromMap(
 	return nil
 }
 
+// excludedOrInExcludedDir reports whether filePath, which is at or
+// beneath rootPath, or any directory between rootPath and it, matches an
+// exclude pattern. A pattern such as */.local/share/Trash matches the
+// directory but not the files inside it, which the walk skips too.
+func (s *Scanner) excludedOrInExcludedDir(filePath, rootPath string) bool {
+	for p := filePath; len(p) > len(rootPath); p = filepath.Dir(p) {
+		if s.shouldExclude(p, rootPath) {
+			return true
+		}
+	}
+
+	return false
+}
+
+// goneFromDisk reports whether looking filePath up on disk finds that
+// it does not exist. Like the walk, it does not follow a symlink, so a
+// symlink whose target is gone is still on disk. A file whose directory
+// was replaced by a regular file is gone too, though the lookup reports
+// "not a directory".
+func (s *Scanner) goneFromDisk(filePath string) bool {
+	var err error
+	if lstater, ok := s.fs.(afero.Lstater); ok {
+		_, _, err = lstater.LstatIfPossible(filePath)
+	} else {
+		_, err = s.fs.Stat(filePath)
+	}
+
+	return errors.Is(err, os.ErrNotExist) || errors.Is(err, syscall.ENOTDIR)
+}
+
 // wrapPermissionError augments permission errors with platform-specific
 // remediation instructions. On macOS, TCC-protected directories (Calendars,
 // Reminders, Photos, etc.) return EPERM unless the running application has
@@ -2013,26 +2168,61 @@ func compileExcludePatterns(patterns []string) []compiledPattern {
 		// Remove trailing slash if present (directory indicator)
 		pattern = strings.TrimSuffix(pattern, "/")
 
-		// Compile the glob pattern
-		// For patterns without path separators, we need to match them as components
-		// e.g., ".git" should match ".git" anywhere in the path
-		g, err := glob.Compile(pattern, '/')
-		if err != nil {
-			log.Warn("Invalid exclude pattern, skipping", "pattern", p, "error", err)
-
-			continue
+		// A leading "**/" demands a "/" before the name and so misses an
+		// entry directly under the source path. It means "at every depth",
+		// which is what an unanchored pattern already does, so drop the
+		// prefix and match unanchored: as in gitignore, "**/x" and
+		// "/**/x" then also match a top-level "x".
+		if rest, found := strings.CutPrefix(pattern, "**/"); found {
+			pattern = rest
+			anchored = false
 		}
 
-		compiled = append(compiled, compiledPattern{
-			pattern:  g,
-			anchored: anchored,
-			original: p,
-		})
+		// Likewise "/**/" inside a pattern demands two slashes, so
+		// "a/**/*.log" misses "a/x.log". As in gitignore it stands for
+		// zero or more directories, so also compile the pattern with it
+		// made a single "/". The glob library's {a,b} is no help: with a
+		// "**" inside it in mid-pattern, it misses paths it should match.
+		for _, variant := range doubleStarVariants(pattern) {
+			// For patterns without path separators, we need to match them as components
+			// e.g., ".git" should match ".git" anywhere in the path
+			g, err := glob.Compile(variant, '/')
+			if err != nil {
+				log.Warn("Invalid exclude pattern, skipping", "pattern", p, "error", err)
+
+				break
+			}
+
+			compiled = append(compiled, compiledPattern{
+				pattern:  g,
+				anchored: anchored,
+				original: p,
+			})
+		}
 	}
 
 	return compiled
 }
 
+// doubleStarVariants returns pattern once for every way of keeping each
+// "/**/" in it or making it a single "/".
+func doubleStarVariants(pattern string) []string {
+	const doubleStar = "/**/"
+
+	before, after, found := strings.Cut(pattern, doubleStar)
+	if !found {
+		return []string{pattern}
+	}
+
+	var variants []string
+
+	for _, rest := range doubleStarVariants(after) {
+		variants = append(variants, before+doubleStar+rest, before+"/"+rest)
+	}
+
+	return variants
+}
+
 // shouldExclude checks if a path should be excluded based on exclude patterns
 // filePath is the full path to the file
 // rootPath is the root of the backup source directory

internal/snapshot/scanner_estimate_test.go

@@ -0,0 +1,28 @@
+//nolint:testpackage // exercises the unexported countNonDirectories helper
+package snapshot
+
+import (
+	"os"
+	"testing"
+
+	"sneak.berlin/go/vaultik/internal/database"
+)
+
+// The walk counts regular files and symlinks, not directories, so the
+// estimate of its count made from the local index's records leaves the
+// directories out. See https://git.eeqj.de/sneak/vaultik/issues/305.
+func TestCountNonDirectories(t *testing.T) {
+	t.Parallel()
+
+	known := map[string]*database.File{
+		"/src":       {Mode: uint32(os.ModeDir | 0o755)},
+		"/src/sub":   {Mode: uint32(os.ModeDir | 0o755)},
+		"/src/a.txt": {Mode: 0o644},
+		"/src/link":  {Mode: uint32(os.ModeSymlink | 0o777), LinkTarget: "a.txt"},
+	}
+
+	got := countNonDirectories(known)
+	if got != 2 {
+		t.Errorf("countNonDirectories = %d, want 2", got)
+	}
+}

internal/snapshot/skip_errors_test.go

@@ -381,3 +381,60 @@ func TestScannerCancelStopsSkipErrorsRun(t *testing.T) {
 		})
 	}
 }
+
+// TestScannerRemovesBlobTempFileWhenRunStops checks that a run stopped
+// part-way, by a file it cannot read or by a cancellation, removes the
+// temp file of the blob it was packing.
+func TestScannerRemovesBlobTempFileWhenRunStops(t *testing.T) {
+	t.Parallel()
+
+	// a.txt is backed up first and starts a blob; the run stops at b.txt.
+	const target = "/source/b.txt"
+
+	tests := []struct {
+		name  string
+		newFs func(cancel context.CancelFunc) afero.Fs
+	}{
+		{
+			name: "file cannot be read",
+			newFs: func(context.CancelFunc) afero.Fs {
+				return &readFailFs{Fs: afero.NewMemMapFs(), target: target}
+			},
+		},
+		{
+			name: "cancelled",
+			newFs: func(cancel context.CancelFunc) afero.Fs {
+				return &cancelOnOpenFs{
+					Fs: afero.NewMemMapFs(), target: target, cancel: cancel,
+				}
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			ctx, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			fs := tt.newFs(cancel)
+			writeSkipErrorTestFile(t, fs, "/source/a.txt", "first file content")
+			writeSkipErrorTestFile(t, fs, target, "second file content")
+
+			_, err := runSkipErrorScan(ctx, t, fs, "/source", false, nil)
+			if err == nil {
+				t.Fatal("expected the scan to stop with an error, got nil")
+			}
+
+			left, err := afero.Glob(fs, filepath.Join(os.TempDir(), "vaultik-blob-*"))
+			if err != nil {
+				t.Fatalf("listing blob temp files: %v", err)
+			}
+
+			if len(left) != 0 {
+				t.Fatalf("expected no blob temp file left, got %v", left)
+			}
+		})
+	}
+}

internal/snapshot/snapshot.go

@@ -199,6 +199,7 @@ func (sm *SnapshotManager) UpdateSnapshotStatsExtended(
 			stats.CompressionLevel,
 			stats.BytesUploaded,
 			stats.UploadDurationMs,
+			stats.NewChunkSize,
 		)
 	})
 }
@@ -906,6 +907,7 @@ type ExtendedBackupStats struct {
 	BlobUncompressedSize int64 // Total uncompressed size of all referenced blobs
 	CompressionLevel     int   // Compression level used for this snapshot
 	UploadDurationMs     int64 // Total milliseconds spent uploading to S3
+	NewChunkSize         int64 // Total size of the chunks counted in ChunksCreated
 }
 
 // deleteOtherSnapshots deletes all snapshots except the current one

internal/storage/conformance_test.go

@@ -8,6 +8,7 @@ import (
 	"reflect"
 	"sort"
 	"testing"
+	"time"
 
 	"sneak.berlin/go/vaultik/internal/storage"
 )
@@ -19,15 +20,17 @@ import (
 // constructor, so the contract is defined once.
 //
 // It exercises the public Storer interface: round-trip, stat, list with
-// prefix filtering, overwrite, delete, delete-of-missing, and not-found on
-// Get and Stat. Each section takes its own fresh backend instance, so the
-// order of sections never matters and no section sees another's objects.
+// prefix filtering, the modification time ListStream reports, overwrite,
+// delete, delete-of-missing, and not-found on Get and Stat. Each section
+// takes its own fresh backend instance, so the order of sections never
+// matters and no section sees another's objects.
 func runStorerConformance(t *testing.T, newStorer func(*testing.T) storage.Storer) {
 	t.Helper()
 
 	conformanceRoundTrip(t, newStorer(t))
 	conformanceOverwrite(t, newStorer(t))
 	conformanceList(t, newStorer(t))
+	conformanceListStreamModTime(t, newStorer(t))
 	conformanceDelete(t, newStorer(t))
 	conformanceNotFound(t, newStorer(t))
 }
@@ -118,6 +121,35 @@ func conformanceList(t *testing.T, s storage.Storer) {
 	}
 }
 
+// conformanceListStreamModTime checks that ListStream gives an object
+// just written a modification time of about now, which prune compares
+// with its grace period.
+func conformanceListStreamModTime(t *testing.T, s storage.Storer) {
+	t.Helper()
+
+	ctx := context.Background()
+
+	err := s.Put(ctx, "blobs/dd/written-now", bytes.NewReader([]byte("data")))
+	if err != nil {
+		t.Fatalf("Put: %v", err)
+	}
+
+	var listed []storage.ObjectInfo
+
+	for info := range s.ListStream(ctx, "blobs/") {
+		listed = append(listed, info)
+	}
+
+	if len(listed) != 1 || listed[0].Err != nil {
+		t.Fatalf("ListStream(\"blobs/\") = %v, want the one object", listed)
+	}
+
+	if time.Since(listed[0].ModTime).Abs() > time.Minute {
+		t.Errorf("ListStream ModTime = %v, want within a minute of now",
+			listed[0].ModTime)
+	}
+}
+
 // conformanceDelete checks that Delete removes an object and that deleting
 // a missing key is not an error.
 func conformanceDelete(t *testing.T, s storage.Storer) {

internal/storage/faultstore/faultstore.go

@@ -1,15 +1,16 @@
 // Package faultstore provides a storage.Storer wrapper that injects
 // faults on demand, so tests can reproduce the failure modes a real
 // backend exhibits: an upload that fails partway, a backend that reports
-// success while storing nothing, and reads that return corrupt or
-// truncated bytes. It is the seam called for by the fault-injection
-// tests (sneak/vaultik issue 72) and is meant to be reused by future
-// tests rather than re-implemented per case.
+// success while storing nothing, reads that return corrupt or truncated
+// bytes, and a delete the backend refuses. It is the seam called for by
+// the fault-injection tests (sneak/vaultik issue 72) and is meant to be
+// reused by future tests rather than re-implemented per case.
 //
-// The wrapper delegates every method to the inner Storer. Two hooks
-// change that: OnPut decides the fate of each write, and OnGet decides
-// how each read's bytes are returned. Both are keyed by the object key,
-// so a test can fault only blobs, only metadata, or a single object.
+// The wrapper delegates every method to the inner Storer. Three hooks
+// change that: OnPut decides the fate of each write, OnGet decides how
+// each read's bytes are returned, and OnDelete decides whether each
+// delete fails. All are keyed by the object key, so a test can fault
+// only blobs, only metadata, or a single object.
 package faultstore
 
 import (
@@ -25,6 +26,10 @@ import (
 // ErrInjectedUpload is returned by a Put the OnPut hook chose to fail.
 var ErrInjectedUpload = errors.New("faultstore: injected upload failure")
 
+// ErrInjectedDelete is returned by a Delete the OnDelete hook chose to
+// fail.
+var ErrInjectedDelete = errors.New("faultstore: injected delete failure")
+
 // PutAction is the disposition OnPut assigns to a write.
 type PutAction int
 
@@ -66,6 +71,11 @@ type Storer struct {
 	// OnGet, when set, is consulted for every Get with the object key
 	// and damages the returned bytes accordingly.
 	OnGet func(key string) GetFault
+
+	// OnDelete, when set, is consulted before every Delete with the
+	// object key. Returning true fails the delete and leaves the object
+	// in place.
+	OnDelete func(key string) bool
 }
 
 // New wraps inner. inner must be non-nil.
@@ -154,8 +164,12 @@ func (f *Storer) Stat(ctx context.Context, key string) (*storage.ObjectInfo, err
 	return f.inner.Stat(ctx, key)
 }
 
-// Delete delegates unchanged.
+// Delete removes the object unless OnDelete fails the delete.
 func (f *Storer) Delete(ctx context.Context, key string) error {
+	if f.OnDelete != nil && f.OnDelete(key) {
+		return fmt.Errorf("%w for %q", ErrInjectedDelete, key)
+	}
+
 	return f.inner.Delete(ctx, key)
 }
 

internal/storage/file.go

@@ -21,9 +21,13 @@ type FileStorer struct {
 // NewFileStorer creates a new filesystem storage backend.
 //
 // Construction is intentionally cheap and does not touch the filesystem.
-// The basePath is recorded; the directory is created lazily on first
-// write. Write operations (Put/PutWithProgress) call MkdirAll for the
-// per-blob parent directory, which also covers basePath on first use.
+// The backend never creates basePath: the user does, so that when
+// basePath is a directory on an unplugged volume, a write fails instead
+// of landing on the disk under the mount point. This does not hold when
+// basePath is the mount point itself, which stays behind when its volume
+// is unplugged.
+// Put and PutWithProgress fail on a missing basePath and create the
+// directories below it as needed.
 // Get and Stat report a key under a missing basePath as ErrNotFound.
 // List and ListStream fail on a missing basePath, because listing it as
 // an empty store would make `prune` drop every local snapshot record.
@@ -120,9 +124,10 @@ func (f *FileStorer) Delete(_ context.Context, key string) error {
 	return nil
 }
 
-// List returns all keys with the given prefix. It fails when the
-// destination directory is missing; a missing prefix under it is an
-// empty listing.
+// List returns all keys with the given prefix, reading through symlinked
+// directories. It fails when the destination directory is missing, or
+// when the prefix or a directory above it is a symlink whose target is
+// missing; a prefix that is truly missing is an empty listing.
 func (f *FileStorer) List(ctx context.Context, prefix string) ([]string, error) {
 	var keys []string
 
@@ -133,8 +138,7 @@ func (f *FileStorer) List(ctx context.Context, prefix string) ([]string, error)
 
 	basePath := f.fullPath(prefix)
 
-	// Check if the prefix exists
-	exists, err := afero.Exists(f.fs, basePath)
+	exists, err := f.prefixExists(prefix)
 	if err != nil {
 		return nil, fmt.Errorf("checking path: %w", err)
 	}
@@ -143,7 +147,7 @@ func (f *FileStorer) List(ctx context.Context, prefix string) ([]string, error)
 		return keys, nil // Empty list for non-existent prefix
 	}
 
-	err = afero.Walk(f.fs, basePath, func(path string, info os.FileInfo, err error) error {
+	err = f.walk(basePath, func(path string, info os.FileInfo, err error) error {
 		if err != nil {
 			return err
 		}
@@ -176,7 +180,9 @@ func (f *FileStorer) List(ctx context.Context, prefix string) ([]string, error)
 }
 
 // ListStream returns a channel of ObjectInfo for large result sets. Like
-// List, it sends an error when the destination directory is missing.
+// List, it reads through symlinked directories and sends an error when
+// the destination directory is missing, or when the prefix or a directory
+// above it is a symlink whose target is missing.
 func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan ObjectInfo {
 	ch := make(chan ObjectInfo)
 
@@ -192,8 +198,7 @@ func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan Objec
 
 		basePath := f.fullPath(prefix)
 
-		// Check if the prefix exists
-		exists, err := afero.Exists(f.fs, basePath)
+		exists, err := f.prefixExists(prefix)
 		if err != nil {
 			ch <- ObjectInfo{Err: fmt.Errorf("checking path: %w", err)}
 
@@ -204,7 +209,7 @@ func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan Objec
 			return // Empty channel for non-existent prefix
 		}
 
-		_ = afero.Walk(f.fs, basePath, func(path string, info os.FileInfo, err error) error {
+		_ = f.walk(basePath, func(path string, info os.FileInfo, err error) error {
 			// Check context cancellation
 			select {
 			case <-ctx.Done():
@@ -230,8 +235,9 @@ func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan Objec
 				// Normalize path separators
 				relPath = strings.ReplaceAll(relPath, string(filepath.Separator), "/")
 				ch <- ObjectInfo{
-					Key:  relPath,
-					Size: info.Size(),
+					Key:     relPath,
+					Size:    info.Size(),
+					ModTime: info.ModTime(),
 				}
 			}
 
@@ -243,11 +249,12 @@ func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan Objec
 }
 
 // DeletePartialUploads removes every file under prefix whose name ends in
-// tempSuffix. A missing prefix has none to remove.
+// tempSuffix. Like List, it reads through symlinked directories and fails
+// on one whose target is missing. A missing prefix has none to remove.
 func (f *FileStorer) DeletePartialUploads(ctx context.Context, prefix string) error {
 	basePath := f.fullPath(prefix)
 
-	exists, err := afero.Exists(f.fs, basePath)
+	exists, err := f.prefixExists(prefix)
 	if err != nil {
 		return fmt.Errorf("checking path: %w", err)
 	}
@@ -256,7 +263,7 @@ func (f *FileStorer) DeletePartialUploads(ctx context.Context, prefix string) er
 		return nil
 	}
 
-	err = afero.Walk(f.fs, basePath, func(path string, info os.FileInfo, err error) error {
+	err = f.walk(basePath, func(path string, info os.FileInfo, err error) error {
 		if err != nil {
 			return err
 		}
@@ -297,7 +304,19 @@ func (f *FileStorer) writeAtomic(
 	path := f.fullPath(key)
 	dir := filepath.Dir(path)
 
-	err := f.fs.MkdirAll(dir, storageDirPerm)
+	// Checked on every write, so a destination that disappears part-way
+	// through a backup fails it rather than being recreated by MkdirAll.
+	_, err := f.fs.Stat(f.basePath)
+	if os.IsNotExist(err) {
+		return fmt.Errorf("destination directory does not exist; "+
+			"create it, or mount the volume it is on: %w", err)
+	}
+
+	if err != nil {
+		return fmt.Errorf("checking destination directory: %w", err)
+	}
+
+	err = f.fs.MkdirAll(dir, storageDirPerm)
 	if err != nil {
 		return fmt.Errorf("creating directories: %w", err)
 	}
@@ -352,6 +371,54 @@ func (f *FileStorer) writeAtomic(
 	return nil
 }
 
+// walk is afero.Walk, except that it reads through a symlinked directory
+// the way Get and Stat do, where afero.Walk reports the link as one file.
+// afero.Walk uses Lstat, which does not follow a link, whenever the
+// filesystem has one; the struct below carries only the afero.Fs methods,
+// so the walk uses Stat instead. A link back up the tree is followed until
+// the operating system refuses the path for too many levels of links, and
+// the walk then fails.
+func (f *FileStorer) walk(root string, fn filepath.WalkFunc) error {
+	return afero.Walk(struct{ afero.Fs }{f.fs}, root, fn)
+}
+
+// prefixExists reports whether prefix exists under the destination
+// directory. It checks each directory on the way down to prefix in turn,
+// and the first one Stat cannot find decides: if it is not there at all,
+// prefix is missing; if it is a symlink whose target is missing, such as
+// a link to a volume that is not mounted, prefixExists fails, since
+// listing it as empty would let prune delete every blob.
+func (f *FileStorer) prefixExists(prefix string) (bool, error) {
+	lstater, canLstat := f.fs.(afero.Lstater)
+	path := f.basePath
+
+	for name := range strings.SplitSeq(prefix, "/") {
+		path = filepath.Join(path, name)
+
+		_, err := f.fs.Stat(path)
+		if err == nil {
+			continue
+		}
+
+		if !os.IsNotExist(err) {
+			return false, err
+		}
+
+		// Lstat does not follow a symlink at path, so it finds the link
+		// that Stat could not follow.
+		if canLstat {
+			_, _, lstatErr := lstater.LstatIfPossible(path)
+			if lstatErr == nil {
+				return false, err
+			}
+		}
+
+		return false, nil
+	}
+
+	return true, nil
+}
+
 // fullPath returns the full filesystem path for a key.
 func (f *FileStorer) fullPath(key string) string {
 	return filepath.Join(f.basePath, key)

internal/storage/file_backend_test.go

@@ -4,7 +4,10 @@ import (
 	"context"
 	"errors"
 	"io/fs"
+	"os"
 	"path/filepath"
+	"reflect"
+	"strings"
 	"testing"
 
 	"sneak.berlin/go/vaultik/internal/storage"
@@ -60,3 +63,272 @@ func TestFileStorerListMissingDestination(t *testing.T) {
 		t.Errorf("ListStream error = %v, want a not-exist error", streamErr)
 	}
 }
+
+// The two directories at the top of a destination, either of which the
+// tests below replace with a symlink.
+const (
+	metadataDir = "metadata"
+	blobsDir    = "blobs"
+)
+
+// moveBehindSymlink moves base/dir to a directory elsewhere and puts a
+// symlink to it in its place.
+func moveBehindSymlink(t *testing.T, base, dir string) {
+	t.Helper()
+
+	linkPath := filepath.Join(base, dir)
+	movedPath := filepath.Join(t.TempDir(), dir)
+
+	err := os.Rename(linkPath, movedPath)
+	if err != nil {
+		t.Fatalf("moving %s away: %v", dir, err)
+	}
+
+	err = os.Symlink(movedPath, linkPath)
+	if err != nil {
+		t.Fatalf("linking %s: %v", dir, err)
+	}
+}
+
+// TestFileStorerListsThroughSymlinkedDirectory checks that List and
+// ListStream read through a symlinked metadata/ or blobs/ directory, as Get
+// and Stat do, instead of listing the link as one key. Listing the whole
+// destination meets the link part-way through the walk; listing its prefix
+// starts the walk at the link. A prefix missing behind the link lists as
+// empty.
+func TestFileStorerListsThroughSymlinkedDirectory(t *testing.T) {
+	t.Parallel()
+
+	for _, dir := range []string{metadataDir, blobsDir} {
+		t.Run(dir, func(t *testing.T) {
+			t.Parallel()
+
+			base := t.TempDir()
+			key := dir + "/aa/bb/object"
+
+			s, err := storage.NewFileStorer(base)
+			if err != nil {
+				t.Fatalf("NewFileStorer: %v", err)
+			}
+
+			err = s.Put(context.Background(), key, strings.NewReader("data"))
+			if err != nil {
+				t.Fatalf("Put: %v", err)
+			}
+
+			moveBehindSymlink(t, base, dir)
+
+			want := []string{key}
+
+			for _, prefix := range []string{"", dir + "/"} {
+				got := listSorted(t, s, prefix)
+				if !reflect.DeepEqual(got, want) {
+					t.Errorf("List(%q) = %v, want %v", prefix, got, want)
+				}
+
+				got = listStreamKeys(t, s, prefix)
+				if !reflect.DeepEqual(got, want) {
+					t.Errorf("ListStream(%q) = %v, want %v", prefix, got, want)
+				}
+			}
+
+			absent := dir + "/absent/"
+
+			if got := listSorted(t, s, absent); len(got) != 0 {
+				t.Errorf("List(%q) = %v, want empty", absent, got)
+			}
+
+			if got := listStreamKeys(t, s, absent); len(got) != 0 {
+				t.Errorf("ListStream(%q) = %v, want empty", absent, got)
+			}
+		})
+	}
+}
+
+// TestFileStorerListFailsOnSymlinkWithMissingTarget checks that List and
+// ListStream fail on a symlinked metadata/ or blobs/ directory whose target
+// is missing, such as a link to a volume that is not mounted, as they do on
+// a missing destination directory, instead of listing it as empty. That
+// holds for a prefix below the link too.
+func TestFileStorerListFailsOnSymlinkWithMissingTarget(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+
+	for _, dir := range []string{metadataDir, blobsDir} {
+		t.Run(dir, func(t *testing.T) {
+			t.Parallel()
+
+			base := t.TempDir()
+
+			s, err := storage.NewFileStorer(base)
+			if err != nil {
+				t.Fatalf("NewFileStorer: %v", err)
+			}
+
+			unmounted := filepath.Join(t.TempDir(), "unmounted")
+
+			err = os.Symlink(unmounted, filepath.Join(base, dir))
+			if err != nil {
+				t.Fatalf("Symlink: %v", err)
+			}
+
+			for _, prefix := range []string{"", dir + "/", dir + "/aa/"} {
+				keys, err := s.List(ctx, prefix)
+				if !errors.Is(err, fs.ErrNotExist) {
+					t.Errorf("List(%q) = %v, %v; want a not-exist error",
+						prefix, keys, err)
+				}
+
+				var streamErr error
+
+				for object := range s.ListStream(ctx, prefix) {
+					if object.Err != nil {
+						streamErr = object.Err
+					}
+				}
+
+				if !errors.Is(streamErr, fs.ErrNotExist) {
+					t.Errorf("ListStream(%q) error = %v, want a not-exist error",
+						prefix, streamErr)
+				}
+			}
+		})
+	}
+}
+
+// TestFileStorerListFailsOnSymlinkLoop checks that a symlink pointing back
+// at the destination makes List and ListStream fail instead of walking it
+// forever.
+func TestFileStorerListFailsOnSymlinkLoop(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+	base := t.TempDir()
+
+	s, err := storage.NewFileStorer(base)
+	if err != nil {
+		t.Fatalf("NewFileStorer: %v", err)
+	}
+
+	err = s.Put(ctx, testBlobKey, strings.NewReader("data"))
+	if err != nil {
+		t.Fatalf("Put: %v", err)
+	}
+
+	err = os.Symlink(base, filepath.Join(base, "blobs", "loop"))
+	if err != nil {
+		t.Fatalf("Symlink: %v", err)
+	}
+
+	keys, err := s.List(ctx, "blobs/")
+	if err == nil {
+		t.Errorf("List = %d keys and no error, want an error", len(keys))
+	}
+
+	var streamErr error
+
+	for object := range s.ListStream(ctx, "blobs/") {
+		if object.Err != nil {
+			streamErr = object.Err
+		}
+	}
+
+	if streamErr == nil {
+		t.Error("ListStream sent no error, want one")
+	}
+}
+
+// TestFileStorerDeletePartialUploadsThroughSymlinkedDirectory checks that
+// DeletePartialUploads, which remote nuke uses, removes a leftover temp
+// file behind a symlinked metadata/ or blobs/ directory and keeps the
+// object at the real key.
+func TestFileStorerDeletePartialUploadsThroughSymlinkedDirectory(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+
+	for _, dir := range []string{metadataDir, blobsDir} {
+		t.Run(dir, func(t *testing.T) {
+			t.Parallel()
+
+			base := t.TempDir()
+			key := dir + "/aa/bb/object"
+
+			s, err := storage.NewFileStorer(base)
+			if err != nil {
+				t.Fatalf("NewFileStorer: %v", err)
+			}
+
+			err = s.Put(ctx, key, strings.NewReader("data"))
+			if err != nil {
+				t.Fatalf("Put: %v", err)
+			}
+
+			leftover := filepath.Join(base, key+"-123456.partial")
+
+			err = os.WriteFile(leftover, []byte("half"), 0o600)
+			if err != nil {
+				t.Fatalf("writing leftover temp file: %v", err)
+			}
+
+			moveBehindSymlink(t, base, dir)
+
+			err = s.DeletePartialUploads(ctx, dir+"/")
+			if err != nil {
+				t.Fatalf("DeletePartialUploads: %v", err)
+			}
+
+			// leftover runs through the link.
+			_, err = os.Stat(leftover)
+			if !os.IsNotExist(err) {
+				t.Errorf("leftover temp file was not removed: %v", err)
+			}
+
+			_, err = s.Stat(ctx, key)
+			if err != nil {
+				t.Errorf("Stat of the real key: %v", err)
+			}
+		})
+	}
+}
+
+// TestFileStorerPutMissingDestination checks that a write fails when the
+// destination directory has gone, even after earlier writes to it
+// succeeded, and that it does not create the directory again.
+func TestFileStorerPutMissingDestination(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+	base := filepath.Join(t.TempDir(), "usbstick")
+
+	err := os.Mkdir(base, 0o750)
+	if err != nil {
+		t.Fatalf("Mkdir: %v", err)
+	}
+
+	s, err := storage.NewFileStorer(base)
+	if err != nil {
+		t.Fatalf("NewFileStorer: %v", err)
+	}
+
+	err = s.Put(ctx, "blobs/aa/bb/first", strings.NewReader("first"))
+	if err != nil {
+		t.Fatalf("first Put: %v", err)
+	}
+
+	err = os.Rename(base, base+"-unplugged")
+	if err != nil {
+		t.Fatalf("Rename: %v", err)
+	}
+
+	err = s.Put(ctx, "blobs/aa/bb/second", strings.NewReader("second"))
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("Put = %v, want a not-exist error", err)
+	}
+
+	_, err = os.Stat(base)
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("Put created the destination directory again (Stat = %v)", err)
+	}
+}

internal/storage/rclone.go

@@ -7,10 +7,12 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"os"
 	"strings"
 	"time"
 
 	"github.com/rclone/rclone/fs"
+	"github.com/rclone/rclone/fs/config"
 	"github.com/rclone/rclone/fs/config/configfile"
 	"github.com/rclone/rclone/fs/operations"
 
@@ -33,6 +35,15 @@ type RcloneStorer struct {
 // The remote parameter is the rclone remote name (as configured via `rclone config`).
 // The path parameter is the path within the remote.
 func NewRcloneStorer(ctx context.Context, remote, path string) (*RcloneStorer, error) {
+	// rclone reads RCLONE_CONFIG only as the value of its --config flag,
+	// which vaultik does not have, so it is applied here as that flag would.
+	if configPath, ok := os.LookupEnv("RCLONE_CONFIG"); ok {
+		err := config.SetConfigPath(configPath)
+		if err != nil {
+			return nil, fmt.Errorf("using RCLONE_CONFIG %q: %w", configPath, err)
+		}
+	}
+
 	// Install the default config file handler
 	configfile.Install()
 
@@ -42,6 +53,14 @@ func NewRcloneStorer(ctx context.Context, remote, path string) (*RcloneStorer, e
 		rclonePath += path
 	}
 
+	// On S3, Swift and some other remotes rclone keeps a time of its own in
+	// each object's metadata, which costs one request per object to read.
+	// rclone's use_server_modtime setting makes ModTime the time the
+	// listing gives instead: when the object was stored, which is what
+	// prune compares with prune_grace_period.
+	ctx, ci := fs.AddConfig(ctx)
+	ci.UseServerModTime = true
+
 	// Create the rclone filesystem
 	fsys, err := fs.NewFs(ctx, rclonePath)
 	if err != nil {
@@ -200,8 +219,9 @@ func (r *RcloneStorer) ListStream(
 
 			if prefix == "" || strings.HasPrefix(key, prefix) {
 				ch <- ObjectInfo{
-					Key:  key,
-					Size: obj.Size(),
+					Key:     key,
+					Size:    obj.Size(),
+					ModTime: obj.ModTime(ctx),
 				}
 			}
 		})

internal/storage/rclone_test.go

@@ -5,12 +5,19 @@ import (
 	"context"
 	"errors"
 	"io"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"path/filepath"
 	"strings"
+	"sync/atomic"
 	"testing"
+	"time"
 
+	"github.com/johannesboyne/gofakes3"
+	"github.com/johannesboyne/gofakes3/backend/s3mem"
 	"github.com/rclone/rclone/fs"
+	"github.com/rclone/rclone/fs/config"
 	"github.com/rclone/rclone/fs/config/configmap"
 	"sneak.berlin/go/vaultik/internal/storage"
 )
@@ -390,3 +397,110 @@ func TestNewRcloneStorerUnknownRemote(t *testing.T) {
 		t.Errorf("NewRcloneStorer error = %v, want ErrRemoteNotFound", err)
 	}
 }
+
+// TestNewRcloneStorerReadsRcloneConfigEnv checks that a remote defined only
+// in the file RCLONE_CONFIG names is found and written to, as rclone itself
+// would.
+func TestNewRcloneStorerReadsRcloneConfigEnv(t *testing.T) {
+	// rclone keeps the config file path in a process-global variable, which
+	// NewRcloneStorer leaves pointing at this test's file.
+	previous := config.GetConfigPath()
+
+	t.Cleanup(func() { _ = config.SetConfigPath(previous) })
+
+	confFile := filepath.Join(t.TempDir(), "rclone.conf")
+
+	err := os.WriteFile(confFile, []byte("[loc]\ntype = local\n"), 0o600)
+	if err != nil {
+		t.Fatalf("writing rclone config: %v", err)
+	}
+
+	t.Setenv("RCLONE_CONFIG", confFile)
+
+	dir := t.TempDir()
+	ctx := context.Background()
+
+	s, err := storage.NewRcloneStorer(ctx, "loc", dir)
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	err = s.Put(ctx, testBlobKey, strings.NewReader("blob-bytes"))
+	if err != nil {
+		t.Fatalf("Put: %v", err)
+	}
+
+	_, err = os.Stat(filepath.Join(dir, testBlobKey))
+	if err != nil {
+		t.Errorf("object is not in the remote's directory: %v", err)
+	}
+}
+
+// TestRcloneListStreamOnS3MakesNoRequestPerObject checks that listing an
+// rclone S3 remote takes each object's time from the listing itself. By
+// default rclone reads the time it keeps in each object's metadata, one
+// request per object, and every command that lists the destination store
+// would slow down as it grows.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneListStreamOnS3MakesNoRequestPerObject(t *testing.T) {
+	backend := s3mem.New()
+
+	err := backend.CreateBucket(s3TestBucket)
+	if err != nil {
+		t.Fatalf("create bucket: %v", err)
+	}
+
+	const body = "blob-bytes"
+
+	_, err = backend.PutObject(s3TestBucket, testBlobKey, nil,
+		strings.NewReader(body), int64(len(body)))
+	if err != nil {
+		t.Fatalf("seed object: %v", err)
+	}
+
+	var headRequests atomic.Int64
+
+	fakeS3 := gofakes3.New(backend).Server()
+	srv := httptest.NewServer(http.HandlerFunc(
+		func(w http.ResponseWriter, r *http.Request) {
+			if r.Method == http.MethodHead {
+				headRequests.Add(1)
+			}
+
+			fakeS3.ServeHTTP(w, r)
+		}))
+	t.Cleanup(srv.Close)
+
+	remote := ":s3,provider=Other" +
+		",access_key_id=" + s3TestAccessKeyID +
+		",secret_access_key=" + s3TestSecretAccessKey +
+		",endpoint='" + srv.URL + "'"
+	ctx := context.Background()
+
+	s, err := storage.NewRcloneStorer(ctx, remote, s3TestBucket)
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	headRequests.Store(0)
+
+	var listed []storage.ObjectInfo
+
+	for info := range s.ListStream(ctx, "blobs/") {
+		listed = append(listed, info)
+	}
+
+	if len(listed) != 1 || listed[0].Err != nil {
+		t.Fatalf("ListStream(\"blobs/\") = %v, want the one object", listed)
+	}
+
+	if time.Since(listed[0].ModTime).Abs() > time.Minute {
+		t.Errorf("ListStream ModTime = %v, want within a minute of now",
+			listed[0].ModTime)
+	}
+
+	if n := headRequests.Load(); n != 0 {
+		t.Errorf("ListStream made %d HEAD request(s), want none", n)
+	}
+}

internal/storage/s3.go

@@ -89,9 +89,10 @@ func (s *S3Storer) ListStream(ctx context.Context, prefix string) <-chan ObjectI
 
 		for info := range s.client.ListObjectsStream(ctx, prefix, false) {
 			ch <- ObjectInfo{
-				Key:  info.Key,
-				Size: info.Size,
-				Err:  info.Err,
+				Key:     info.Key,
+				Size:    info.Size,
+				ModTime: info.ModTime,
+				Err:     info.Err,
 			}
 		}
 	}()

internal/storage/s3_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"errors"
+	"io"
 	"net/http"
 	"net/http/httptest"
 	"slices"
@@ -22,13 +23,16 @@ import (
 // s3TestBucket is the bucket created for each in-process S3 server.
 const s3TestBucket = "test-bucket"
 
-// Credentials for the tests that build a storer from a config.Config. The
-// in-process S3 server accepts any.
+// Credentials for the tests' S3 clients. The test servers accept any.
 const (
 	s3TestAccessKeyID     = "key"
 	s3TestSecretAccessKey = "secret"
 )
 
+// s3TestRegion is the region of the S3 clients built here with
+// s3.NewClient. The test servers ignore it.
+const s3TestRegion = "us-east-1"
+
 // newS3Storer builds an s3:// backend backed by a fresh in-process
 // S3 server. It reuses the same in-memory S3 harness (gofakes3 + s3mem
 // over httptest) that internal/s3 and the not-found regression test use,
@@ -55,7 +59,7 @@ func newS3Storer(t *testing.T) storage.Storer {
 		Bucket:          s3TestBucket,
 		AccessKeyID:     "test",
 		SecretAccessKey: "test",
-		Region:          "us-east-1",
+		Region:          s3TestRegion,
 	})
 	if err != nil {
 		t.Fatalf("new client: %v", err)
@@ -93,6 +97,68 @@ func TestS3StorerMissingKeyMapsToErrNotFound(t *testing.T) {
 	}
 }
 
+// TestS3ListingCutShortIsAnError checks that List and ListStream fail on a
+// listing page marked truncated that carries no continuation token, as a
+// server answering V2 listings V1-style sends it. Taken as the last page,
+// it would hide every later key, and prune would treat the blobs of the
+// snapshots listed there as unreferenced.
+func TestS3ListingCutShortIsAnError(t *testing.T) {
+	t.Parallel()
+
+	const (
+		prefix        = "metadata/"
+		truncatedPage = `<?xml version="1.0" encoding="UTF-8"?>
+<ListBucketResult xmlns="http://s3.amazonaws.com/doc/2006-03-01/">
+  <Name>test-bucket</Name>
+  <Prefix>metadata/</Prefix>
+  <KeyCount>1</KeyCount>
+  <MaxKeys>1</MaxKeys>
+  <IsTruncated>true</IsTruncated>
+  <Contents>
+    <Key>metadata/snap/manifest.json.zst</Key>
+    <Size>8</Size>
+  </Contents>
+</ListBucketResult>`
+	)
+
+	srv := httptest.NewServer(http.HandlerFunc(
+		func(w http.ResponseWriter, _ *http.Request) {
+			_, _ = io.WriteString(w, truncatedPage)
+		}))
+	t.Cleanup(srv.Close)
+
+	client, err := s3.NewClient(context.Background(), s3.Config{
+		Endpoint:        srv.URL,
+		Bucket:          s3TestBucket,
+		AccessKeyID:     s3TestAccessKeyID,
+		SecretAccessKey: s3TestSecretAccessKey,
+		Region:          s3TestRegion,
+	})
+	if err != nil {
+		t.Fatalf("new client: %v", err)
+	}
+
+	storer := storage.NewS3Storer(client)
+	ctx := context.Background()
+
+	keys, err := storer.List(ctx, prefix)
+	if !errors.Is(err, s3.ErrNoContinuationToken) {
+		t.Errorf("List = %q, %v; want ErrNoContinuationToken", keys, err)
+	}
+
+	var streamErr error
+
+	for obj := range storer.ListStream(ctx, prefix) {
+		if obj.Err != nil {
+			streamErr = obj.Err
+		}
+	}
+
+	if !errors.Is(streamErr, s3.ErrNoContinuationToken) {
+		t.Errorf("ListStream error = %v, want ErrNoContinuationToken", streamErr)
+	}
+}
+
 // TestS3URLPrefixKeyLayout pins the bucket keys an s3:// URL reads and
 // writes: the README's remote storage layout, with the prefix joined to
 // each key by one "/". s3://b/p and s3://b/p/ must be the same

internal/storage/storer.go

@@ -14,6 +14,7 @@ import (
 	"context"
 	"errors"
 	"io"
+	"time"
 )
 
 // ErrNotFound is returned when an object does not exist.
@@ -27,7 +28,10 @@ type ProgressCallback func(bytesTransferred int64) error
 type ObjectInfo struct {
 	Key  string // Object key/path
 	Size int64  // Size in bytes
-	Err  error  // Error for streaming results (nil on success)
+	// ModTime is when the object was written, as the store reports it.
+	// Only ListStream sets it; Stat leaves it zero.
+	ModTime time.Time
+	Err     error // Error for streaming results (nil on success)
 }
 
 // Info provides human-readable storage configuration.
@@ -41,7 +45,8 @@ type Info struct {
 // directory for filesystem).
 type Storer interface {
 	// Put stores data at the specified key.
-	// Parent directories are created automatically for filesystem backends.
+	// The filesystem backend fails when its destination directory does
+	// not exist, and creates the directories below it as needed.
 	Put(ctx context.Context, key string, data io.Reader) error
 
 	// PutWithProgress stores data with progress reporting.
@@ -66,7 +71,8 @@ type Storer interface {
 	// For large result sets, prefer ListStream.
 	List(ctx context.Context, prefix string) ([]string, error)
 
-	// ListStream returns a channel of ObjectInfo for large result sets.
+	// ListStream returns a channel of ObjectInfo, with each object's key,
+	// size and modification time, for large result sets.
 	// The channel is closed when listing completes.
 	// If an error occurs during listing, the final item will have Err set.
 	ListStream(ctx context.Context, prefix string) <-chan ObjectInfo

internal/storage/url.go

@@ -17,8 +17,10 @@ const (
 
 // Sentinel errors for storage URL parsing.
 var (
-	ErrEmptyStorageURL   = errors.New("storage URL is empty")
-	ErrEmptyFilePath     = errors.New("file URL path is empty")
+	ErrEmptyStorageURL  = errors.New("storage URL is empty")
+	ErrEmptyFilePath    = errors.New("file URL path is empty")
+	ErrRelativeFilePath = errors.New(
+		"file URL needs an absolute path, as in file:///path/to/backup")
 	ErrMissingBucket     = errors.New("s3 URL missing bucket name")
 	ErrMissingRemote     = errors.New("rclone URL missing remote name")
 	ErrUnsupportedScheme = errors.New(
@@ -28,6 +30,12 @@ var (
 		"storage URL must not carry credentials; " +
 			"set s3.access_key_id and s3.secret_access_key in the config instead")
 	ErrURLUnknownParam = errors.New("unknown query parameter in storage URL")
+	ErrURLBadQuery     = errors.New(
+		"storage URL query cannot be decoded; " +
+			"check it for a ; or a % not followed by two hex digits")
+	ErrRcloneRemoteParams = errors.New(
+		"rclone URL must not carry parameters in the remote name; " +
+			"set them on the remote with rclone config instead")
 )
 
 // URL represents a parsed storage URL.
@@ -51,17 +59,8 @@ func ParseStorageURL(rawURL string) (*URL, error) {
 		return nil, ErrEmptyStorageURL
 	}
 
-	// Handle file:// URLs
 	if after, ok := strings.CutPrefix(rawURL, "file://"); ok {
-		path := after
-		if path == "" {
-			return nil, ErrEmptyFilePath
-		}
-
-		return &URL{
-			Scheme: schemeFile,
-			Prefix: path,
-		}, nil
+		return parseFileURL(after)
 	}
 
 	if strings.HasPrefix(rawURL, "s3://") {
@@ -75,9 +74,31 @@ func ParseStorageURL(rawURL string) (*URL, error) {
 	return nil, ErrUnsupportedScheme
 }
 
+// parseFileURL parses the path that follows file://. The path must be
+// absolute: a relative one would resolve against the working directory,
+// so the destination would move with it. Percent-escapes are decoded, so
+// my%20stick names the directory "my stick"; a literal space is kept.
+func parseFileURL(rawPath string) (*URL, error) {
+	if rawPath == "" {
+		return nil, ErrEmptyFilePath
+	}
+
+	if !strings.HasPrefix(rawPath, "/") {
+		return nil, ErrRelativeFilePath
+	}
+
+	path, err := url.PathUnescape(rawPath)
+	if err != nil {
+		return nil, fmt.Errorf("invalid file URL path: %w", err)
+	}
+
+	return &URL{Scheme: schemeFile, Prefix: path}, nil
+}
+
 // parseS3URL parses an s3://bucket/prefix URL. It rejects credentials in
-// the userinfo and any query parameter other than endpoint, region and
-// ssl, so a credential-bearing URL is never stored or echoed.
+// the userinfo or in the endpoint parameter's userinfo, a query that
+// cannot be decoded, and any query parameter other than endpoint, region
+// and ssl, so a credential-bearing URL is never stored or echoed.
 func parseS3URL(rawURL string) (*URL, error) {
 	u, err := url.Parse(rawURL)
 	if err != nil {
@@ -93,13 +114,26 @@ func parseS3URL(rawURL string) (*URL, error) {
 		return nil, ErrMissingBucket
 	}
 
-	query := u.Query()
+	// u.Query() would silently drop a part it cannot decode, endpoint
+	// included. The decode error is not wrapped: it can quote the value.
+	query, err := url.ParseQuery(u.RawQuery)
+	if err != nil {
+		return nil, ErrURLBadQuery
+	}
 
 	err = rejectUnknownParams(query, "endpoint", "region", "ssl")
 	if err != nil {
 		return nil, err
 	}
 
+	// A host name never contains "@", so one in an endpoint, with or
+	// without a scheme, can only introduce userinfo.
+	for _, endpoint := range query["endpoint"] {
+		if strings.Contains(endpoint, "@") {
+			return nil, ErrURLCredentials
+		}
+	}
+
 	return &URL{
 		Scheme:   schemeS3,
 		Bucket:   bucket,
@@ -111,9 +145,23 @@ func parseS3URL(rawURL string) (*URL, error) {
 }
 
 // parseRcloneURL parses an rclone://remote/path URL. rclone:// takes no
-// query parameters, so credentials in the userinfo and any parameter at
+// query parameters, so credentials in the userinfo, parameters in the
+// remote name, a query that cannot be decoded and any query parameter at
 // all are rejected rather than silently ignored.
 func parseRcloneURL(rawURL string) (*URL, error) {
+	// The remote name runs up to the first "/", "?" or "#". It never
+	// contains a comma; one starts rclone's remote,param=value form,
+	// whose values can be credentials. This is checked before url.Parse,
+	// whose error for such a name can print part of it.
+	remote := strings.TrimPrefix(rawURL, "rclone://")
+	if end := strings.IndexAny(remote, "/?#"); end >= 0 {
+		remote = remote[:end]
+	}
+
+	if strings.Contains(remote, ",") {
+		return nil, ErrRcloneRemoteParams
+	}
+
 	u, err := url.Parse(rawURL)
 	if err != nil {
 		return nil, wrapParseError(err)
@@ -123,12 +171,19 @@ func parseRcloneURL(rawURL string) (*URL, error) {
 		return nil, ErrURLCredentials
 	}
 
-	remote := u.Host
+	remote = u.Host
 	if remote == "" {
 		return nil, ErrMissingRemote
 	}
 
-	err = rejectUnknownParams(u.Query())
+	// As in parseS3URL: u.Query() would drop a part it cannot decode, and
+	// the decode error can quote the value.
+	query, err := url.ParseQuery(u.RawQuery)
+	if err != nil {
+		return nil, ErrURLBadQuery
+	}
+
+	err = rejectUnknownParams(query)
 	if err != nil {
 		return nil, err
 	}

internal/storage/url_parse_test.go

@@ -14,7 +14,10 @@ import (
 func TestParseStorageURLValid(t *testing.T) {
 	t.Parallel()
 
-	const bucket = "mybucket"
+	const (
+		bucket     = "mybucket"
+		fileScheme = "file"
+	)
 
 	cases := []struct {
 		name string
@@ -24,7 +27,17 @@ func TestParseStorageURLValid(t *testing.T) {
 		{
 			name: "file absolute path",
 			raw:  "file:///var/backups/vaultik",
-			want: &storage.URL{Scheme: "file", Prefix: "/var/backups/vaultik"},
+			want: &storage.URL{Scheme: fileScheme, Prefix: "/var/backups/vaultik"},
+		},
+		{
+			name: "file percent-escaped path is decoded",
+			raw:  "file:///Volumes/my%20stick/backup",
+			want: &storage.URL{Scheme: fileScheme, Prefix: "/Volumes/my stick/backup"},
+		},
+		{
+			name: "file path with a literal space",
+			raw:  "file:///Volumes/my stick/backup",
+			want: &storage.URL{Scheme: fileScheme, Prefix: "/Volumes/my stick/backup"},
 		},
 		{
 			name: "s3 bucket and prefix, ssl defaults on",
@@ -90,6 +103,7 @@ func TestParseStorageURLErrors(t *testing.T) {
 	}{
 		{"empty url", "", storage.ErrEmptyStorageURL},
 		{"file empty path", "file://", storage.ErrEmptyFilePath},
+		{"file relative path", "file://dest", storage.ErrRelativeFilePath},
 		{"s3 missing bucket", "s3://", storage.ErrMissingBucket},
 		{"s3 missing bucket with path", "s3:///justprefix", storage.ErrMissingBucket},
 		{"rclone missing remote", "rclone://", storage.ErrMissingRemote},
@@ -136,6 +150,19 @@ func TestParseStorageURLRejectsCredentials(t *testing.T) {
 			wantErr: storage.ErrURLCredentials,
 			secrets: []string{key, secret, "mybucket"},
 		},
+		{
+			name: "s3 userinfo in endpoint",
+			raw: "s3://mybucket?endpoint=https://" +
+				key + ":" + secret + "@minio.example.com",
+			wantErr: storage.ErrURLCredentials,
+			secrets: []string{key, secret},
+		},
+		{
+			name:    "s3 userinfo in endpoint without scheme",
+			raw:     "s3://mybucket?endpoint=" + key + ":" + secret + "@minio.example.com",
+			wantErr: storage.ErrURLCredentials,
+			secrets: []string{key, secret},
+		},
 		{
 			name:    "s3 unknown query param",
 			raw:     "s3://mybucket?access_key=" + key + "&secret=" + secret,
@@ -189,6 +216,112 @@ func TestParseStorageURLRejectsCredentials(t *testing.T) {
 	}
 }
 
+// TestParseStorageURLRejectsRcloneRemoteParams checks that an rclone
+// remote name carrying parameters (rclone's "remote,param=value" form,
+// which can hold a credential) is rejected without echoing the value,
+// including the URLs that url.Parse itself refuses.
+func TestParseStorageURLRejectsRcloneRemoteParams(t *testing.T) {
+	t.Parallel()
+
+	cases := []struct {
+		name    string
+		raw     string
+		secrets []string // must not appear in the error message
+	}{
+		{
+			name:    "configured remote",
+			raw:     "rclone://gdrive,client_secret=topsecret/backups",
+			secrets: []string{"topsecret"},
+		},
+		{
+			name:    "remote defined inline",
+			raw:     "rclone://:s3,access_key_id=KEY,secret_access_key=SECRET/bucket",
+			secrets: []string{"KEY", "SECRET"},
+		},
+		{
+			name:    "colon in a parameter value",
+			raw:     "rclone://gdrive,client_secret=ab:cd/backups",
+			secrets: []string{"cd"},
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			_, err := storage.ParseStorageURL(tc.raw)
+			if !errors.Is(err, storage.ErrRcloneRemoteParams) {
+				t.Fatalf("ParseStorageURL(%q) error = %v, want %v",
+					tc.raw, err, storage.ErrRcloneRemoteParams)
+			}
+
+			for _, secret := range tc.secrets {
+				if strings.Contains(err.Error(), secret) {
+					t.Errorf("error message leaked %q: %v", secret, err.Error())
+				}
+			}
+		})
+	}
+}
+
+// TestParseStorageURLRejectsUndecodableQuery checks that an s3 or rclone
+// URL whose query cannot be decoded is rejected, not read with that part
+// of the query left out, and that the error does not echo it.
+func TestParseStorageURLRejectsUndecodableQuery(t *testing.T) {
+	t.Parallel()
+
+	const (
+		key    = "AKIAKEY"
+		secret = "topsecret"
+	)
+
+	cases := []struct {
+		name string
+		raw  string
+	}{
+		{
+			name: "semicolon",
+			raw: "s3://mybucket/prefix?endpoint=https://" + key + ":" + secret +
+				"@minio.example.com:9000;region=us-east-1",
+		},
+		{
+			name: "bad percent escape",
+			raw: "s3://mybucket/prefix?endpoint=https://" + key + ":" + secret +
+				"%zz@minio.example.com:9000",
+		},
+		{
+			name: "rclone semicolon",
+			raw:  "rclone://gdrive/backups?token=" + secret + ";x=1",
+		},
+		{
+			name: "rclone bad percent escape",
+			raw:  "rclone://gdrive/backups?token=" + secret + "%zz",
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			_, err := storage.ParseStorageURL(tc.raw)
+			if err == nil {
+				t.Fatalf("ParseStorageURL(%q) returned no error", tc.raw)
+			}
+
+			if !errors.Is(err, storage.ErrURLBadQuery) {
+				t.Errorf("ParseStorageURL(%q) error = %v, want %v",
+					tc.raw, err, storage.ErrURLBadQuery)
+			}
+
+			for _, s := range []string{key, secret, "minio.example.com"} {
+				if strings.Contains(err.Error(), s) {
+					t.Errorf("error message leaked %q: %v", s, err.Error())
+				}
+			}
+		})
+	}
+}
+
 // TestParseStorageURLParseFailureHidesURL checks that when url.Parse
 // itself fails, the wrapped error carries only the inner cause, not the
 // *url.Error whose text embeds the raw (possibly credential-bearing) URL.

internal/vaultik/blob_fetch.go

@@ -103,6 +103,9 @@ func (v *Vaultik) FetchAndDecryptBlob(
 }
 
 // FetchBlob downloads a blob and returns a reader for the encrypted data.
+// The reader is closed as soon as ctx is cancelled, which ends a read
+// blocked inside it. A storage error is returned as is: restore and deep
+// verify name the blob in their own errors.
 func (v *Vaultik) FetchBlob(
 	ctx context.Context, blobHash string,
 ) (io.ReadCloser, error) {
@@ -117,8 +120,50 @@ func (v *Vaultik) FetchBlob(
 
 	rc, err := v.Storage.Get(ctx, blobPath)
 	if err != nil {
-		return nil, fmt.Errorf("downloading blob %s: %w", shortHash(blobHash), err)
+		return nil, err
 	}
 
-	return rc, nil
+	return newCloseOnCancelReader(ctx, rc), nil
+}
+
+// closeOnCancelReader closes the reader it wraps when ctx is cancelled.
+// Not every backend's reader stops on cancel by itself (the file
+// backend's does not), and a read stalled on a slow or hung source would
+// otherwise outlast the shutdown timeout, so the process would exit
+// before restore or deep verify removed its decrypted temporary files.
+type closeOnCancelReader struct {
+	io.ReadCloser
+
+	ctx  context.Context //nolint:containedctx // Read reports its error
+	stop func() bool
+}
+
+func newCloseOnCancelReader(
+	ctx context.Context, rc io.ReadCloser,
+) *closeOnCancelReader {
+	return &closeOnCancelReader{
+		ReadCloser: rc,
+		ctx:        ctx,
+		stop:       context.AfterFunc(ctx, func() { _ = rc.Close() }),
+	}
+}
+
+// Read returns the context's error, not the closed reader's, once ctx
+// has been cancelled.
+func (r *closeOnCancelReader) Read(p []byte) (int, error) {
+	n, err := r.ReadCloser.Read(p)
+	if err != nil && r.ctx.Err() != nil {
+		return n, r.ctx.Err()
+	}
+
+	return n, err
+}
+
+// Close closes the wrapped reader, unless the cancel already closed it.
+func (r *closeOnCancelReader) Close() error {
+	if !r.stop() {
+		return nil
+	}
+
+	return r.ReadCloser.Close()
 }

internal/vaultik/blobcache.go

@@ -39,9 +39,9 @@ type blobDiskCacheEntry struct {
 // internally and will fall back to LRU eviction if curBytes exceeds
 // maxBytes. Restore passes math.MaxInt64 as maxBytes and drives
 // eviction itself via Delete() through restoreSweeper, which deletes
-// each blob the moment every file that references its chunks has been
-// written. LRU never fires under that configuration; it is kept as a
-// safety net for callers that don't manage eviction themselves.
+// each blob once no file still to be restored in this run needs it.
+// LRU never fires under that configuration; it is kept as a safety net
+// for callers that don't manage eviction themselves.
 //
 // Get/ReadAt/peak-Len counters are debugging instrumentation used by
 // tests to assert that the restore code path uses ReadAt rather than

internal/vaultik/changed_file_restore_test.go

@@ -137,7 +137,7 @@ func TestBackupAfterRemovingNewestSnapshotRestoresChangedFile(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 	changedPath := filepath.Join(dataDir, "appended.bin")
@@ -184,7 +184,7 @@ func TestBackupAfterInterruptedRunRestoresChangedFile(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 	changedPath := filepath.Join(dataDir, "appended.bin")

internal/vaultik/deep_verify_test.go

@@ -1,15 +1,20 @@
 package vaultik_test
 
 import (
+	"bytes"
 	"context"
+	"crypto/rand"
 	"io"
 	"os"
+	"path"
 	"path/filepath"
 	"testing"
 
 	"github.com/spf13/afero"
 	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
 	"sneak.berlin/go/vaultik/internal/ui"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
@@ -30,7 +35,7 @@ func TestDeepVerifyAcceptsHealthyAndRejectsCorruptBlob(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	chunkSize := int64(64 * 1024)
@@ -77,6 +82,180 @@ func TestDeepVerifyAcceptsHealthyAndRejectsCorruptBlob(t *testing.T) {
 		"deep verify should fail on a corrupted blob")
 }
 
+// TestDeepVerifyJSONCountsBlobs checks the verified and missing counts
+// that deep verify reports under --json: on a healthy snapshot, with a
+// damaged blob, and with missing blobs. Deep verify stops at the first
+// damaged blob, so the blobs read before it are the ones verified; when
+// blobs are missing it reads no blob contents, but counts every missing
+// blob.
+func TestDeepVerifyJSONCountsBlobs(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	ctx := context.Background()
+	storeDir := t.TempDir()
+
+	cfg, storer, snapshotID, keys := backUpSeveralBlobs(ctx, t, storeDir)
+
+	healthy, err := deepVerifyJSON(ctx, t, cfg, storer, snapshotID)
+	require.NoError(t, err)
+	require.Equal(t, "ok", healthy.Status)
+	require.Len(t, keys, healthy.BlobCount)
+	require.Equal(t, healthy.BlobCount, healthy.Verified)
+	require.Zero(t, healthy.Missing)
+
+	// Damage the last blob read, keeping its length, so every blob before
+	// it is read and passes first.
+	lastPath := filepath.Join(storeDir, keys[len(keys)-1])
+	blob, err := afero.ReadFile(fs, lastPath)
+	require.NoError(t, err)
+
+	blob[len(blob)/2] ^= 0xff
+	require.NoError(t, afero.WriteFile(fs, lastPath, blob, 0o600))
+
+	got, err := deepVerifyJSON(ctx, t, cfg, storer, snapshotID)
+	require.Error(t, err)
+	require.Equal(t, "failed", got.Status)
+	require.Equal(t, healthy.BlobCount-1, got.Verified)
+	require.Zero(t, got.Missing)
+
+	// Delete the first blob and the damaged last one: both must be
+	// counted as missing, not only the first one found.
+	var deletedSize int64
+
+	for _, key := range []string{keys[0], keys[len(keys)-1]} {
+		info, statErr := storer.Stat(ctx, key)
+		require.NoError(t, statErr)
+
+		deletedSize += info.Size
+
+		require.NoError(t, storer.Delete(ctx, key))
+	}
+
+	got, err = deepVerifyJSON(ctx, t, cfg, storer, snapshotID)
+	require.Error(t, err)
+	require.Equal(t, "failed", got.Status)
+	require.Equal(t, 2, got.Missing)
+	require.Equal(t, deletedSize, got.MissingSize)
+	require.Zero(t, got.Verified)
+}
+
+// TestDeepVerifyJSONCountsWrongSizeBlob checks that under --deep --json a
+// blob of the wrong size is counted in mismatched and fails the run, and
+// that the check goes on past it to count a missing blob after it. The
+// store is on disk, so the message must not call it S3.
+func TestDeepVerifyJSONCountsWrongSizeBlob(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	ctx := context.Background()
+	storeDir := t.TempDir()
+
+	cfg, storer, snapshotID, keys := backUpSeveralBlobs(ctx, t, storeDir)
+
+	// Make the first blob checked longer, and delete the last one.
+	firstPath := filepath.Join(storeDir, keys[0])
+	blob, err := afero.ReadFile(fs, firstPath)
+	require.NoError(t, err)
+
+	blob = append(blob, []byte("extra")...)
+	require.NoError(t, afero.WriteFile(fs, firstPath, blob, 0o600))
+	require.NoError(t, storer.Delete(ctx, keys[len(keys)-1]))
+
+	got, err := deepVerifyJSON(ctx, t, cfg, storer, snapshotID)
+	require.Error(t, err)
+	require.Equal(t, "failed", got.Status)
+	require.Equal(t, 1, got.Mismatched)
+	require.Equal(t, 1, got.Missing)
+	require.Zero(t, got.Verified)
+	require.Contains(t, got.ErrorMessage, "with the wrong size")
+	require.NotContains(t, got.ErrorMessage, "S3")
+}
+
+// TestDeepVerifyStopsOnBlobThatCannotBeChecked checks that a blob that is
+// present in storage but cannot be checked is not counted as missing:
+// deep verify stops and fails, saying it could not check the blob, with
+// the storage error.
+func TestDeepVerifyStopsOnBlobThatCannotBeChecked(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := t.TempDir()
+
+	cfg, storer, snapshotID, keys := backUpSeveralBlobs(ctx, t, storeDir)
+	failing := statFailingStorer{Storer: storer, failKeys: []string{keys[0]}}
+
+	got, err := deepVerifyJSON(ctx, t, cfg, failing, snapshotID)
+	require.Error(t, err)
+	require.Equal(t, "failed", got.Status)
+	require.Zero(t, got.Missing)
+	require.Contains(t, got.ErrorMessage,
+		"could not check blob "+path.Base(keys[0]))
+	require.Contains(t, got.ErrorMessage, errStatDenied.Error())
+}
+
+// backUpSeveralBlobs backs up random data, which does not compress or
+// deduplicate, so it fills at least three blobs in a store on disk at
+// storeDir. It returns the blobs' storage keys sorted, which is hash
+// order, the order deep verify reads them in.
+func backUpSeveralBlobs(
+	ctx context.Context, t *testing.T, storeDir string,
+) (*config.Config, *storage.FileStorer, string, []string) {
+	t.Helper()
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "source")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	chunkSize := int64(32 * 1024)
+	maxBlobSize := int64(128 * 1024)
+
+	data := make([]byte, maxBlobSize*4)
+	_, err := rand.Read(data)
+	require.NoError(t, err)
+	require.NoError(t, fs.MkdirAll(dataDir, 0o755))
+	require.NoError(t, afero.WriteFile(fs,
+		filepath.Join(dataDir, "data.bin"), data, 0o644))
+
+	cfg, storer, snapshotID := runFileStorageBackup(
+		ctx, t, fs, dataDir, storeDir, dbPath, chunkSize, maxBlobSize)
+
+	keys := storedBlobKeys(t, storeDir)
+	require.GreaterOrEqual(t, len(keys), 3)
+
+	return cfg, storer, snapshotID, keys
+}
+
+// deepVerifyJSON runs deep verify with --json on snapshotID and returns
+// the decoded result and the error RunDeepVerify returned.
+func deepVerifyJSON(
+	ctx context.Context, t *testing.T,
+	cfg *config.Config, storer storage.Storer, snapshotID string,
+) (vaultik.VerifyResult, error) {
+	t.Helper()
+
+	var out bytes.Buffer
+
+	v := &vaultik.Vaultik{
+		Config:  cfg,
+		Storage: storer,
+		Fs:      afero.NewOsFs(),
+		Stdout:  &out,
+		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
+	}
+	v.SetContext(ctx)
+
+	err := v.RunDeepVerify(snapshotID,
+		&vaultik.VerifyOptions{Deep: true, JSON: true})
+
+	return decodeVerifyResult(t, out.Bytes()), err
+}
+
 // corruptOneBlob flips a middle byte of the first blob file found under
 // blobsDir, leaving the file length unchanged.
 func corruptOneBlob(t *testing.T, fs afero.Fs, blobsDir string) {

internal/vaultik/destination_marker.go

@@ -0,0 +1,90 @@
+package vaultik
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+
+	"sneak.berlin/go/vaultik/internal/storage"
+)
+
+// destinationMarkerKey is the object every backup writes at the root of
+// the backup destination store. While the local index holds a snapshot
+// stored there, a destination without it is treated as not present: an
+// unplugged volume leaves an empty directory at its mount point, which
+// would otherwise read as a store whose snapshots are all gone.
+const destinationMarkerKey = "vaultik-destination.txt"
+
+// destinationMarkerText is the whole content of the marker. It names no
+// host, path or time.
+const destinationMarkerText = "This is a vaultik backup destination store.\n"
+
+// errDestinationNotPresent is returned when the local index holds a
+// snapshot stored on the backup destination store and the store has no
+// marker.
+var errDestinationNotPresent = errors.New("backup destination store is not present")
+
+// checkDestinationMarker returns errDestinationNotPresent when the local
+// index holds a snapshot stored on the backup destination store and the
+// store has no marker. With no snapshot in the local index, as on a
+// first backup or on another machine, there is nothing to check; nor
+// when the local index is bound to a different destination than the
+// configured one, since its snapshots are stored there.
+func (v *Vaultik) checkDestinationMarker() error {
+	ids, err := v.localSnapshotIDs()
+	if err != nil {
+		return fmt.Errorf("listing local snapshots: %w", err)
+	}
+
+	if len(ids) == 0 {
+		return nil
+	}
+
+	// Config is nil only in tests, which then record no binding (see
+	// EnsureStorageBinding).
+	if v.Config != nil {
+		stored, configured, err := v.storageBinding()
+		if err != nil {
+			return err
+		}
+
+		if stored != "" && stored != configured {
+			return nil
+		}
+	}
+
+	_, err = v.Storage.Stat(v.ctx, destinationMarkerKey)
+	if errors.Is(err, storage.ErrNotFound) {
+		return fmt.Errorf("%w: %s has no %s, but the local index holds "+
+			"snapshots stored there; mount the volume if it is removable, "+
+			"or if the destination was emptied on purpose, run "+
+			"'vaultik database delete' and back up again",
+			errDestinationNotPresent, v.Storage.Info().Location,
+			destinationMarkerKey)
+	}
+
+	if err != nil {
+		return fmt.Errorf("checking for %s: %w", destinationMarkerKey, err)
+	}
+
+	return nil
+}
+
+// markDestination checks the backup destination store with
+// checkDestinationMarker and then writes the marker, so a first backup
+// marks its destination and a later one writes the same text again.
+func (v *Vaultik) markDestination() error {
+	err := v.checkDestinationMarker()
+	if err != nil {
+		return err
+	}
+
+	err = v.Storage.Put(v.ctx, destinationMarkerKey,
+		strings.NewReader(destinationMarkerText))
+	if err != nil {
+		return fmt.Errorf("writing %s to the backup destination store: %w",
+			destinationMarkerKey, err)
+	}
+
+	return nil
+}

internal/vaultik/destination_marker_test.go

@@ -0,0 +1,297 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// These tests cover https://git.eeqj.de/sneak/vaultik/issues/239: a
+// file:// destination at a volume's mount point leaves an empty directory
+// behind when the volume is unplugged. While the local index holds
+// snapshots, a destination without vaultik-destination.txt is not present,
+// so no command may write to it or conclude from it that the local
+// snapshots are gone.
+
+// markerFile is the name of the file a backup writes at the root of the
+// destination.
+const markerFile = "vaultik-destination.txt"
+
+// backUpThenUnplugMountPoint backs up to a file:// destination, then
+// moves the destination directory away and leaves an empty directory in
+// its place, as unplugging a volume mounted there would. It returns the
+// path of that empty directory.
+func backUpThenUnplugMountPoint(
+	ctx context.Context, t *testing.T,
+) (*vaultik.Vaultik, *database.Repositories, *bytes.Buffer, string) {
+	t.Helper()
+
+	storeDir := filepath.Join(t.TempDir(), "mnt")
+	v, repos, out := backUpToFileDestination(ctx, t, storeDir)
+
+	require.NoError(t, os.Rename(storeDir, storeDir+"-unplugged"))
+	require.NoError(t, os.Mkdir(storeDir, 0o750))
+
+	return v, repos, out, storeDir
+}
+
+// assertUntouched checks that the local index still holds its one
+// snapshot and that nothing was written to the empty mount point.
+func assertUntouched(
+	ctx context.Context, t *testing.T,
+	repos *database.Repositories, storeDir string,
+) {
+	t.Helper()
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 1, "no local snapshot record may be deleted")
+
+	entries, err := os.ReadDir(storeDir)
+	require.NoError(t, err)
+	assert.Empty(t, entries, "nothing may be written to the empty mount point")
+}
+
+// TestFirstBackupWritesMarker checks that a backup writes the marker
+// file at the root of the destination.
+func TestFirstBackupWritesMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	storeDir := t.TempDir()
+	backUpToFileDestination(context.Background(), t, storeDir)
+
+	assert.FileExists(t, filepath.Join(storeDir, markerFile))
+}
+
+// TestBackupRefusesDestinationWithoutMarker checks that snapshot create
+// --prune writes nothing and drops no local snapshot record when the
+// destination has no marker.
+func TestBackupRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, _, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		Cron:  true,
+		Prune: true,
+	})
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestBackupChecksMarkerBeforeDroppingIncompleteSnapshots checks that
+// snapshot create writes nothing when the local index holds only an
+// incomplete snapshot and the destination has no marker. The backup drops
+// incomplete snapshots before it scans; checked after that, the local
+// index would hold none and nothing would be checked.
+func TestBackupChecksMarkerBeforeDroppingIncompleteSnapshots(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := t.TempDir()
+	v, repos := newFileDestinationVaultik(ctx, t, storeDir)
+
+	require.NoError(t, repos.Snapshots.Create(ctx, nil, &database.Snapshot{
+		ID:        "host_first_2026-01-02T03:04:05Z",
+		Hostname:  "host",
+		StartedAt: time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC),
+	}))
+
+	err := backUp(v, "first")
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestPruneRefusesDestinationWithoutMarker checks that prune fails and
+// deletes no local snapshot record when the destination has no marker.
+func TestPruneRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, _, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.Prune(&vaultik.PruneOptions{Force: true})
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestPurgeRefusesDestinationWithoutMarker checks that snapshot purge
+// fails and deletes no local snapshot record when the destination has no
+// marker.
+func TestPurgeRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, _, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		KeepLatest: true,
+		Force:      true,
+	})
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestNukeRemoteRefusesDestinationWithoutMarker checks that remote nuke
+// fails, deletes no local snapshot record, and neither reports the
+// destination empty nor advises running nuke again when the destination
+// has no marker.
+func TestNukeRemoteRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.NukeRemote(true)
+	require.ErrorContains(t, err, markerFile)
+	assert.NotContains(t, out.String(), "now empty")
+	assert.NotContains(t, out.String(), "vaultik remote nuke --force")
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestRemoveSnapshotWarnsWhenDestinationHasNoMarker checks that snapshot
+// remove warns that the metadata could not be removed from a destination
+// without a marker, instead of reporting that it was.
+func TestRemoveSnapshotWarnsWhenDestinationHasNoMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out, _ := backUpThenUnplugMountPoint(ctx, t)
+
+	result, err := v.RemoveSnapshot(localSnapshotID(ctx, t, repos, "first"),
+		&vaultik.RemoveOptions{Force: true})
+	require.NoError(t, err)
+
+	assert.False(t, result.RemoteRemoved)
+	assert.Contains(t, out.String(),
+		"Could not remove snapshot metadata from remote")
+	assert.Contains(t, out.String(), markerFile)
+	assert.NotContains(t, out.String(),
+		"Removed snapshot metadata from remote storage")
+}
+
+// TestRemoveSnapshotOnlyOnDestinationFailsWithoutMarker checks that
+// snapshot remove, for a snapshot that is only on the destination, fails
+// saying the destination is not present when it has no marker, instead
+// of saying the snapshot does not exist.
+func TestRemoveSnapshotOnlyOnDestinationFailsWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "mnt")
+	v, repos, _ := backUpToFileDestination(ctx, t, storeDir)
+	require.NoError(t, backUp(v, "second"))
+
+	id := localSnapshotID(ctx, t, repos, "first")
+	_, err := v.RemoveSnapshot(id,
+		&vaultik.RemoveOptions{Force: true, LocalOnly: true})
+	require.NoError(t, err)
+
+	// Unplug the volume: an empty directory stays at its mount point.
+	require.NoError(t, os.Rename(storeDir, storeDir+"-unplugged"))
+	require.NoError(t, os.Mkdir(storeDir, 0o750))
+
+	_, err = v.RemoveSnapshot(id, &vaultik.RemoveOptions{Force: true})
+	require.ErrorContains(t, err, markerFile)
+	require.ErrorContains(t, err, "mount the volume")
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestListSnapshotsWarnsWhenDestinationHasNoMarker checks that snapshot
+// list warns and shows the local index alone when the destination has no
+// marker, without reporting the local snapshot as missing from it.
+func TestListSnapshotsWarnsWhenDestinationHasNoMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out, _ := backUpThenUnplugMountPoint(ctx, t)
+	id := localSnapshotID(ctx, t, repos, "first")
+
+	require.NoError(t, v.ListSnapshots(false))
+
+	assert.Contains(t, out.String(), "Could not list backup destination store")
+	assert.Contains(t, out.String(), markerFile)
+	assert.Contains(t, out.String(), id)
+	assert.NotContains(t, out.String(), "not found in backup destination store")
+}
+
+// TestListSnapshotsAfterStorageURLChangeToEmptyDirectory checks that
+// snapshot list, after storage_url is changed to an empty directory,
+// warns only that the local index is bound to a different destination:
+// the local snapshots are stored at the old one, so the new one's lack
+// of a marker says nothing. With --json the local row is not on the new
+// destination, so remote_present is false, not unknown.
+func TestListSnapshotsAfterStorageURLChangeToEmptyDirectory(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out := backUpToFileDestination(ctx, t, t.TempDir())
+	id := localSnapshotID(ctx, t, repos, "first")
+
+	newDir := t.TempDir()
+	store, err := storage.NewFileStorer(newDir)
+	require.NoError(t, err)
+
+	v.Storage = store
+	v.Config.StorageURL = "file://" + newDir
+
+	require.NoError(t, v.ListSnapshots(false))
+
+	assert.Contains(t, out.String(), id)
+	assert.Contains(t, out.String(),
+		"local index is bound to a different backup destination")
+	assert.NotContains(t, out.String(), "Could not list backup destination store")
+	assert.NotContains(t, out.String(), markerFile)
+
+	out.Reset()
+	require.NoError(t, v.ListSnapshots(true))
+
+	rows := decodeListJSON(t, out.String())
+	require.Len(t, rows, 1)
+	require.NotNil(t, rows[0].RemotePresent)
+	assert.False(t, *rows[0].RemotePresent)
+}
+
+// TestBackupAfterNukeRemoteSucceeds checks that remote nuke leaves the
+// destination ready for a new first backup, which writes the marker
+// again.
+func TestBackupAfterNukeRemoteSucceeds(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	storeDir := t.TempDir()
+	v, _, _ := backUpToFileDestination(context.Background(), t, storeDir)
+
+	require.NoError(t, v.NukeRemote(true))
+	assert.NoFileExists(t, filepath.Join(storeDir, markerFile))
+
+	require.NoError(t, backUp(v, "second"))
+	assert.FileExists(t, filepath.Join(storeDir, markerFile))
+}

internal/vaultik/destination_validation_test.go

@@ -161,13 +161,14 @@ func TestVerifyManifestBlobsRejectsShortHash(t *testing.T) {
 		Blobs: []snapshot.BlobInfo{{Hash: "abc", CompressedSize: 1}},
 	}
 
-	verified, missing, mismatched, missingSize, err :=
+	verified, missing, mismatched, missingSize, unchecked, err :=
 		v.verifyManifestBlobs(manifest, &VerifyOptions{JSON: true})
 	require.ErrorIs(t, err, errInvalidBlobHash)
 	require.Zero(t, verified)
 	require.Zero(t, missing)
 	require.Zero(t, mismatched)
 	require.Zero(t, missingSize)
+	require.Empty(t, unchecked)
 }
 
 // TestVerifyBlobChunksRejectsNegativeLength proves a blob_chunks row with a

internal/vaultik/fault_injection_test.go

@@ -192,7 +192,7 @@ func assertRestoreRejectsDamagedBlob(
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -238,7 +238,7 @@ func TestDeepVerifyCatchesLyingBackend(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	ctx := context.Background()
@@ -290,7 +290,7 @@ func TestInterruptedBlobUploadRecordsNoUploadedBlob(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	ctx := context.Background()
@@ -298,10 +298,6 @@ func TestInterruptedBlobUploadRecordsNoUploadedBlob(t *testing.T) {
 
 	writeFaultSourceTree(t, fs, dataDir)
 
-	// No upload succeeds, so nothing creates the destination directory.
-	// It must exist for the listing below to show that no blob survived.
-	require.NoError(t, os.Mkdir(storeDir, 0o750))
-
 	inner, err := storage.NewFileStorer(storeDir)
 	require.NoError(t, err)
 
@@ -365,7 +361,7 @@ func TestBackupRetryAfterInterruptedUploadIsRestorable(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -429,7 +425,7 @@ func TestBackupSurvivesMetadataExportInterruption(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	ctx := context.Background()
@@ -445,6 +441,10 @@ func TestBackupSurvivesMetadataExportInterruption(t *testing.T) {
 
 	repos := database.NewRepositories(db)
 
+	// snapshot create writes the marker; this backup runs its steps one
+	// by one, so the marker is written here.
+	require.NoError(t, inner.Put(ctx, markerFile, strings.NewReader("")))
+
 	// Back up and complete with a working backend.
 	sm := newFaultSnapshotManager(fs, inner, cfg, repos)
 	scanner := newFaultScanner(fs, inner, cfg, repos)
@@ -493,8 +493,14 @@ func TestBackupSurvivesMetadataExportInterruption(t *testing.T) {
 		"shallow verify must report the missing manifest")
 
 	// snapshot list must not crash on the partial snapshot.
+	var listOut strings.Builder
+
+	reader.UI = ui.NewWithColor(&listOut, false)
 	require.NoError(t, reader.ListSnapshots(false),
 		"snapshot list must tolerate a partially-exported snapshot")
+	assert.NotContains(t, listOut.String(),
+		"Could not list backup destination store",
+		"snapshot list must read the partial snapshot from the destination")
 }
 
 // Scenario 2, repair: the process dies during the metadata export of a
@@ -511,7 +517,7 @@ func TestBackupCompletesOnlyAfterMetadataExport(t *testing.T) {
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -661,17 +667,18 @@ func faultScannerFactory(
 ) snapshot.ScannerFactory {
 	return func(params snapshot.ScannerParams) *snapshot.Scanner {
 		return snapshot.NewScanner(snapshot.ScannerConfig{
-			FS:               params.Fs,
-			Storage:          storer,
-			ChunkSize:        faultChunkSize,
-			MaxBlobSize:      faultMaxBlobSize,
-			CompressionLevel: cfg.CompressionLevel,
-			AgeRecipients:    cfg.AgeRecipients,
-			Repositories:     repos,
-			EnableProgress:   params.EnableProgress,
-			UI:               params.UI,
-			Exclude:          params.Exclude,
-			SkipErrors:       params.SkipErrors,
+			FS:                 params.Fs,
+			Storage:            storer,
+			ChunkSize:          faultChunkSize,
+			MaxBlobSize:        faultMaxBlobSize,
+			CompressionLevel:   cfg.CompressionLevel,
+			AgeRecipients:      cfg.AgeRecipients,
+			Repositories:       repos,
+			EnableProgress:     params.EnableProgress,
+			UI:                 params.UI,
+			Exclude:            params.Exclude,
+			SkipErrors:         params.SkipErrors,
+			PreviousSnapshotID: params.PreviousSnapshotID,
 		})
 	}
 }
@@ -686,7 +693,7 @@ func TestRestoreReportsDiskFull(t *testing.T) {
 	osFS := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 

internal/vaultik/helpers.go

@@ -89,22 +89,6 @@ type SnapshotInfo struct {
 	RemotePresent    *bool            `json:"remote_present"`
 }
 
-// formatBytes formats bytes in a human-readable format
-func formatBytes(bytes int64) string {
-	const unit = 1024
-	if bytes < unit {
-		return fmt.Sprintf("%d B", bytes)
-	}
-
-	div, exp := int64(unit), 0
-	for n := bytes / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-
-	return fmt.Sprintf("%.1f %cB", float64(bytes)/float64(div), "KMGTPE"[exp])
-}
-
 // parseSnapshotTimestamp extracts the timestamp from a snapshot ID
 // Format: hostname_snapshotname_2026-01-12T14:41:15Z
 func parseSnapshotTimestamp(snapshotID string) (time.Time, error) {

internal/vaultik/info.go

@@ -457,9 +457,9 @@ func (v *Vaultik) scanRemoteBlobStorage(
 		return err
 	}
 
-	for _, size := range allBlobs {
+	for _, blob := range allBlobs {
 		result.TotalBlobCount++
-		result.TotalBlobSize += size
+		result.TotalBlobSize += blob.Size
 	}
 
 	// A blob named only by a manifest that could not be read, or by one
@@ -474,10 +474,10 @@ func (v *Vaultik) scanRemoteBlobStorage(
 		orphanedSize  int64
 	)
 
-	for hash, size := range allBlobs {
+	for hash, blob := range allBlobs {
 		if _, referenced := referencedBlobs[hash]; !referenced {
 			orphanedCount++
-			orphanedSize += size
+			orphanedSize += blob.Size
 		}
 	}
 

internal/vaultik/integration_test.go

@@ -771,7 +771,7 @@ func TestEndToEndFileStorage(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -964,7 +964,7 @@ func TestDedupOnlySnapshotRestores(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 

internal/vaultik/missing_destination_test.go

@@ -24,13 +24,11 @@ import (
 // that is not plugged in, cannot be listed. It is not an empty store, so
 // no command may conclude from it that the local snapshots are gone.
 
-// backUpToFileDestination backs up the snapshot named "first" to a
-// file:// destination at storeDir, which need not exist yet. Everything
-// the returned Vaultik prints after the backup goes to the returned
-// buffer.
-func backUpToFileDestination(
+// newFileDestinationVaultik returns a Vaultik that backs up the
+// snapshot named "first" to a file:// destination at storeDir.
+func newFileDestinationVaultik(
 	ctx context.Context, t *testing.T, storeDir string,
-) (*vaultik.Vaultik, *database.Repositories, *bytes.Buffer) {
+) (*vaultik.Vaultik, *database.Repositories) {
 	t.Helper()
 
 	osFs := afero.NewOsFs()
@@ -49,8 +47,22 @@ func backUpToFileDestination(
 
 	repos := database.NewRepositories(db)
 	cfg := changedFileConfig(dataDir, dbPath)
-	v := newBackupVaultik(ctx, cfg, store, repos, db, osFs)
 
+	return newBackupVaultik(ctx, cfg, store, repos, db, osFs), repos
+}
+
+// backUpToFileDestination creates the directory storeDir, as a user must
+// before the first backup, and backs up the snapshot named "first" to a
+// file:// destination there. Everything the returned Vaultik prints
+// after the backup goes to the returned buffer.
+func backUpToFileDestination(
+	ctx context.Context, t *testing.T, storeDir string,
+) (*vaultik.Vaultik, *database.Repositories, *bytes.Buffer) {
+	t.Helper()
+
+	require.NoError(t, os.MkdirAll(storeDir, 0o750))
+
+	v, repos := newFileDestinationVaultik(ctx, t, storeDir)
 	require.NoError(t, backUp(v, "first"))
 
 	out := &bytes.Buffer{}
@@ -75,21 +87,28 @@ func backUpThenUnplug(
 	return v, repos, out
 }
 
-// TestFirstBackupCreatesDestinationDirectory checks that a first backup
-// to a destination directory that does not exist yet creates it, and
-// that the destination can be listed afterwards.
-func TestFirstBackupCreatesDestinationDirectory(t *testing.T) {
+// TestBackupFailsWhenDestinationDirectoryMissing checks that a backup to
+// a file:// destination whose directory does not exist fails before it
+// records a snapshot, and does not create the directory.
+func TestBackupFailsWhenDestinationDirectoryMissing(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
 
 	ctx := context.Background()
-	storeDir := filepath.Join(t.TempDir(), "volume", "backup")
-	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+	volume := filepath.Join(t.TempDir(), "volume")
+	v, repos := newFileDestinationVaultik(ctx, t, filepath.Join(volume, "backup"))
 
-	require.NoError(t, v.ListSnapshots(false))
+	err := backUp(v, "first")
+	require.ErrorIs(t, err, fs.ErrNotExist)
+	require.ErrorContains(t, err, "create it, or mount the volume it is on")
 
-	assert.NotContains(t, out.String(), "Could not list backup destination store")
-	assert.NotContains(t, out.String(), "not found in backup destination store")
+	_, err = os.Stat(volume)
+	require.ErrorIs(t, err, fs.ErrNotExist,
+		"the backup must not create the destination directory")
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	assert.Empty(t, snapshots)
 }
 
 // TestListSnapshotsWarnsWhenDestinationMissing checks that snapshot list
@@ -169,3 +188,41 @@ func TestPurgeSaysListingFailedOnceWhenDestinationMissing(t *testing.T) {
 	assert.Equal(t, 1, strings.Count(err.Error(), "listing remote snapshots"),
 		err.Error())
 }
+
+// TestNukeRemoteKeepsLocalIndexWhenDestinationMissing checks that remote
+// nuke fails on a destination it cannot list, leaves the local index
+// unchanged, and neither reports the destination empty nor advises
+// running nuke again.
+func TestNukeRemoteKeepsLocalIndexWhenDestinationMissing(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out := backUpThenUnplug(ctx, t)
+
+	filesBefore, err := repos.Files.ListAll(ctx)
+	require.NoError(t, err)
+	require.NotEmpty(t, filesBefore)
+
+	blobsBefore, err := repos.Blobs.GetAll(ctx)
+	require.NoError(t, err)
+	require.NotEmpty(t, blobsBefore)
+
+	err = v.NukeRemote(true)
+	require.ErrorIs(t, err, fs.ErrNotExist)
+	require.ErrorContains(t, err, "listing remote snapshots")
+	assert.NotContains(t, out.String(), "now empty")
+	assert.NotContains(t, out.String(), "vaultik remote nuke --force")
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 1, "nuke must delete no local snapshot record")
+
+	files, err := repos.Files.ListAll(ctx)
+	require.NoError(t, err)
+	assert.Len(t, files, len(filesBefore), "nuke must delete no local file record")
+
+	blobs, err := repos.Blobs.GetAll(ctx)
+	require.NoError(t, err)
+	assert.Len(t, blobs, len(blobsBefore), "nuke must delete no local blob record")
+}

internal/vaultik/new_chunk_size_test.go

@@ -0,0 +1,133 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"maps"
+	"path/filepath"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/chunker"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestNewChunkSizeOfChangedFileStaysFixed covers
+// https://git.eeqj.de/sneak/vaultik/issues/311. A backup whose only change
+// is a modified file lists the size of the chunks that change added as its
+// NEW CHUNK SIZE, and a later backup that changes the file again leaves
+// every earlier snapshot's figure as it was. The file's row in the local
+// index is shared by all three snapshots and rewritten by each backup.
+func TestNewChunkSizeOfChangedFileStaysFixed(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "src")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+	changedPath := filepath.Join(dataDir, "changed.bin")
+
+	ctx := context.Background()
+	files := writeFaultSourceTree(t, fs, dataDir)
+	cfg := changedFileConfig(dataDir, dbPath)
+
+	appendRandomBytes(t, fs, files, changedPath, appendedFileSize)
+
+	store, err := storage.NewFileStorer(t.TempDir())
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	repos := database.NewRepositories(db)
+	v := newBackupVaultik(ctx, cfg, store, repos, db, fs)
+
+	require.NoError(t, backUp(v, "first"))
+
+	firstFiles := maps.Clone(files)
+	firstID := localSnapshotID(ctx, t, repos, "first")
+
+	appendRandomBytes(t, fs, files, changedPath, faultChunkSize)
+	require.NoError(t, backUp(v, "second"))
+
+	secondID := localSnapshotID(ctx, t, repos, "second")
+	secondAdded := sizeOfNewChunks(t, files, firstFiles)
+	require.Positive(t, secondAdded)
+	assert.Equal(t, secondAdded, listedNewChunkSizes(t, v)[secondID],
+		"the second snapshot must count the chunks the change added")
+
+	appendRandomBytes(t, fs, files, changedPath, faultChunkSize)
+	require.NoError(t, backUp(v, "third"))
+
+	sizes := listedNewChunkSizes(t, v)
+	assert.Equal(t, sizeOfNewChunks(t, firstFiles, nil), sizes[firstID],
+		"a later backup must not change the first snapshot's figure")
+	assert.Equal(t, secondAdded, sizes[secondID],
+		"a later backup must not change the second snapshot's figure")
+}
+
+// sizeOfNewChunks returns the total size of the distinct chunks of the
+// files in after that no file in before has, cut the way the backup cuts
+// them.
+func sizeOfNewChunks(t *testing.T, after, before map[string][]byte) int64 {
+	t.Helper()
+
+	known := make(map[string]bool)
+	for _, chunk := range chunksOf(t, before) {
+		known[chunk.Hash] = true
+	}
+
+	var total int64
+
+	for _, chunk := range chunksOf(t, after) {
+		if !known[chunk.Hash] {
+			known[chunk.Hash] = true
+			total += chunk.Size
+		}
+	}
+
+	return total
+}
+
+// chunksOf returns the chunks of every file in files.
+func chunksOf(t *testing.T, files map[string][]byte) []chunker.Chunk {
+	t.Helper()
+
+	c := chunker.NewChunker(faultChunkSize)
+
+	var all []chunker.Chunk
+
+	for _, content := range files {
+		chunks, err := c.ChunkReader(bytes.NewReader(content))
+		require.NoError(t, err)
+
+		all = append(all, chunks...)
+	}
+
+	return all
+}
+
+// listedNewChunkSizes runs `snapshot list --json` and returns each listed
+// snapshot's new_chunk_size by snapshot ID.
+func listedNewChunkSizes(t *testing.T, v *vaultik.Vaultik) map[string]int64 {
+	t.Helper()
+
+	var out bytes.Buffer
+
+	v.Stdout = &out
+	require.NoError(t, v.ListSnapshots(true))
+
+	sizes := make(map[string]int64)
+	for _, row := range decodeListJSON(t, out.String()) {
+		sizes[row.ID] = row.NewChunkSize
+	}
+
+	return sizes
+}

internal/vaultik/nuke_remote_test.go

@@ -5,25 +5,28 @@ import (
 	"io/fs"
 	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"sneak.berlin/go/vaultik/internal/log"
 	"sneak.berlin/go/vaultik/internal/snapshot"
+	"sneak.berlin/go/vaultik/internal/storage/faultstore"
 )
 
 // TestNukeRemoteLeavesNoFiles checks that remote nuke leaves no file
 // under a file:// destination, including the `.partial` files uploads
 // killed part-way leave next to a snapshot's metadata and next to where a
-// blob would have been.
+// blob would have been, a snapshot's database whose manifest was never
+// uploaded, and a file under blobs/ whose name is not a blob hash.
 func TestNukeRemoteLeavesNoFiles(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
 
 	ctx := context.Background()
 	storeDir := filepath.Join(t.TempDir(), "store")
-	v, repos, _ := backUpToFileDestination(ctx, t, storeDir)
+	v, repos, out := backUpToFileDestination(ctx, t, storeDir)
 
 	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
 	require.NoError(t, err)
@@ -36,6 +39,9 @@ func TestNukeRemoteLeavesNoFiles(t *testing.T) {
 			"db.zst.age-123456.partial"),
 		filepath.Join(storeDir, "blobs", hash[:2], hash[2:4],
 			hash+"-123456.partial"),
+		filepath.Join(storeDir, "metadata",
+			snapshot.RemoteSnapshotKey("no-manifest"), "db.zst.age"),
+		filepath.Join(storeDir, "blobs", "0b", "01", ".DS_Store"),
 	}
 
 	for _, leftover := range leftovers {
@@ -61,4 +67,135 @@ func TestNukeRemoteLeavesNoFiles(t *testing.T) {
 		})
 	require.NoError(t, err)
 	assert.Empty(t, files)
+
+	// Nuke deletes the blobs itself, so advice to run prune is wrong.
+	assert.Contains(t, out.String(), "Backup destination store is now empty.")
+	assert.NotContains(t, out.String(), "vaultik prune")
+}
+
+// TestNukeRemoteFailsWhenADeleteFails checks that remote nuke returns an
+// error when a snapshot's metadata or a blob cannot be deleted, does not
+// report the store empty, and advises running nuke again: prune deletes
+// no metadata.
+func TestNukeRemoteFailsWhenADeleteFails(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	for _, refusedPrefix := range []string{"metadata/", "blobs/"} {
+		t.Run(refusedPrefix, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+			storeDir := filepath.Join(t.TempDir(), "store")
+			v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+			store := faultstore.New(v.Storage)
+			store.OnDelete = func(key string) bool {
+				return strings.HasPrefix(key, refusedPrefix)
+			}
+			v.Storage = store
+
+			require.Error(t, v.NukeRemote(true))
+
+			assert.NotContains(t, out.String(), "now empty")
+			assert.Contains(t, out.String(), "vaultik remote nuke --force")
+			assert.NotContains(t, out.String(), "vaultik prune")
+		})
+	}
+}
+
+// TestNukeRemoteKeepsLocalSnapshotWhoseMetadataIsNotDeleted checks that
+// remote nuke keeps a snapshot in the local index while its metadata
+// cannot be deleted from the destination store: dropped from the local
+// index, it would be left there as a remote-only snapshot without its
+// name.
+func TestNukeRemoteKeepsLocalSnapshotWhoseMetadataIsNotDeleted(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, repos, _ := backUpToFileDestination(ctx, t, storeDir)
+
+	store := faultstore.New(v.Storage)
+	store.OnDelete = func(key string) bool {
+		return strings.HasPrefix(key, "metadata/")
+	}
+	v.Storage = store
+
+	require.Error(t, v.NukeRemote(true))
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 1)
+}
+
+// TestNukeRemoteNamesWhatIsLeft checks that remote nuke, when an object
+// it does not recognise cannot be deleted, does not report the store
+// empty but names that object and fails.
+func TestNukeRemoteNamesWhatIsLeft(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+	const leftKey = "blobs/0b/01/.DS_Store"
+
+	leftPath := filepath.Join(storeDir, filepath.FromSlash(leftKey))
+	require.NoError(t, os.MkdirAll(filepath.Dir(leftPath), 0o750))
+	require.NoError(t, os.WriteFile(leftPath, []byte("finder"), 0o600))
+
+	store := faultstore.New(v.Storage)
+	store.OnDelete = func(key string) bool { return key == leftKey }
+	v.Storage = store
+
+	require.Error(t, v.NukeRemote(true))
+
+	assert.Contains(t, out.String(), leftKey)
+	assert.NotContains(t, out.String(), "now empty")
+	assert.Contains(t, out.String(), "vaultik remote nuke --force")
+}
+
+// TestNukeRemoteRefusesAnotherDestinationWithoutRetryAdvice checks that
+// remote nuke, when the local index belongs to a different destination
+// than the configured one, fails without advising to run nuke again:
+// that run would fail the same way.
+func TestNukeRemoteRefusesAnotherDestinationWithoutRetryAdvice(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+	v.Config.StorageURL = "file://" + filepath.Join(t.TempDir(), "elsewhere")
+
+	err := v.NukeRemote(true)
+
+	require.ErrorContains(t, err, "bound to a different backup destination")
+	assert.NotContains(t, out.String(), "vaultik remote nuke --force")
+}
+
+// TestNukeRemoteFailsWhenALocalSnapshotCannotBeRemoved checks that a
+// snapshot remote nuke cannot remove from the local index fails the
+// run, and that nuke then does not report the destination store empty.
+func TestNukeRemoteFailsWhenALocalSnapshotCannotBeRemoved(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+	_, err := v.DB.Conn().ExecContext(ctx, `
+		CREATE TRIGGER refuse_snapshot_delete BEFORE DELETE ON snapshots
+		BEGIN SELECT RAISE(ABORT, 'simulated index error'); END`)
+	require.NoError(t, err)
+
+	require.Error(t, v.NukeRemote(true))
+
+	assert.NotContains(t, out.String(), "now empty")
+	assert.Contains(t, out.String(), "vaultik remote nuke --force")
 }

internal/vaultik/prune.go

@@ -5,8 +5,10 @@ import (
 	"errors"
 	"fmt"
 	"strings"
+	"time"
 
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
 )
 
 // PruneOptions contains options for the prune command
@@ -19,6 +21,19 @@ type PruneOptions struct {
 var errNukeRequiresForce = errors.New(
 	"nuke requires --force (this deletes ALL remote snapshots and blobs)")
 
+// errBlobDeleteFailed reports unreferenced blobs prune could not delete.
+var errBlobDeleteFailed = errors.New("blob(s) could not be deleted")
+
+// errObjectsLeft reports objects remote nuke still finds on the backup
+// destination store after deleting.
+var errObjectsLeft = errors.New(
+	"object(s) still on the backup destination store")
+
+// nukeCommandHint is the command suggested when a remote nuke could not
+// delete everything: running it again deletes what is left, which
+// `vaultik prune` does not do for snapshot metadata.
+const nukeCommandHint = "vaultik remote nuke --force"
+
 // metadataDirName is the top-level remote directory holding snapshot
 // metadata.
 const metadataDirName = "metadata"
@@ -27,10 +42,12 @@ const metadataDirName = "metadata"
 // (blobs/<aa>/<bb>/<hash>).
 const blobKeyParts = 4
 
-// NukeRemote deletes every snapshot's metadata and every blob from remote
-// storage, along with any object an upload cut off part-way left under a
-// temporary `.partial` name. After this returns successfully the bucket
-// prefix is empty and the next backup starts from scratch.
+// NukeRemote deletes every object under metadata/ and blobs/ in remote
+// storage, whatever its name, along with any object an upload cut off
+// part-way left under a temporary `.partial` name, and last the
+// destination marker. It returns successfully only when a listing
+// afterwards finds nothing left under metadata/ and blobs/, and the next
+// backup then starts from scratch.
 //
 // Refuses to run unless force is true. The caller is responsible for
 // confirming with the user.
@@ -39,6 +56,37 @@ func (v *Vaultik) NukeRemote(force bool) error {
 		return errNukeRequiresForce
 	}
 
+	// Checked before anything is deleted: running nuke again would fail
+	// the same way, so the advice below must not follow these errors.
+	// The listing also fails on a destination without the marker while
+	// the local index holds snapshots.
+	err := v.EnsureStorageBinding()
+	if err != nil {
+		return err
+	}
+
+	_, err = v.listAllRemoteSnapshotKeys()
+	if err != nil {
+		return err
+	}
+
+	err = v.emptyDestinationStore()
+	if err != nil {
+		v.UI.Warningf("Run '%s' again to delete what is left on the "+
+			"backup destination store.", nukeCommandHint)
+
+		return err
+	}
+
+	v.UI.Completef("Backup destination store is now empty.")
+
+	return nil
+}
+
+// emptyDestinationStore does the deleting for NukeRemote and stops at
+// the first step that fails. It also fails when objects are still
+// listed at the end, naming each one.
+func (v *Vaultik) emptyDestinationStore() error {
 	v.UI.Beginf("Removing all snapshot metadata from backup destination store.")
 
 	_, err := v.RemoveAllSnapshots(&RemoveOptions{Force: true})
@@ -46,25 +94,67 @@ func (v *Vaultik) NukeRemote(force bool) error {
 		return fmt.Errorf("removing all snapshots: %w", err)
 	}
 
-	v.UI.Beginf("Removing any blobs still present in backup destination store.")
+	v.UI.Beginf("Removing everything still present in backup destination store.")
 
-	err = v.PruneBlobs(&PruneOptions{Force: true})
-	if err != nil {
-		return fmt.Errorf("pruning blobs: %w", err)
+	var left []string
+
+	for _, prefix := range []string{"metadata/", "blobs/"} {
+		keys, err := v.deleteEverythingUnder(prefix)
+		if err != nil {
+			return err
+		}
+
+		left = append(left, keys...)
 	}
 
-	// The file and rclone listings skip `.partial` objects, so the two
-	// steps above never delete them.
-	for _, prefix := range []string{"metadata/", "blobs/"} {
-		err = v.Storage.DeletePartialUploads(v.ctx, prefix)
+	for _, key := range left {
+		v.UI.Warningf("Still on the backup destination store: %s", key)
+	}
+
+	if len(left) > 0 {
+		return fmt.Errorf("%d %w", len(left), errObjectsLeft)
+	}
+
+	// Last, once the local snapshot records are gone: the next backup
+	// then has none and writes the marker again.
+	err = v.Storage.Delete(v.ctx, destinationMarkerKey)
+	if err != nil {
+		return fmt.Errorf("deleting %s: %w", destinationMarkerKey, err)
+	}
+
+	return nil
+}
+
+// deleteEverythingUnder deletes every object under prefix, whatever its
+// name, and returns the keys a listing afterwards still finds. A delete
+// that fails is logged and the others still run; that object then
+// shows up in the listing afterwards.
+func (v *Vaultik) deleteEverythingUnder(prefix string) ([]string, error) {
+	keys, err := v.Storage.List(v.ctx, prefix)
+	if err != nil {
+		return nil, fmt.Errorf("listing %s: %w", prefix, err)
+	}
+
+	for _, key := range keys {
+		err = v.Storage.Delete(v.ctx, key)
 		if err != nil {
-			return fmt.Errorf("deleting partial uploads: %w", err)
+			log.Error("Failed to delete object", "key", key, "error", err)
 		}
 	}
 
-	v.UI.Completef("Backup destination store is now empty.")
+	// The file and rclone listings skip `.partial` objects, so the
+	// deletes above never reach them.
+	err = v.Storage.DeletePartialUploads(v.ctx, prefix)
+	if err != nil {
+		return nil, fmt.Errorf("deleting partial uploads: %w", err)
+	}
 
-	return nil
+	left, err := v.Storage.List(v.ctx, prefix)
+	if err != nil {
+		return nil, fmt.Errorf("listing %s: %w", prefix, err)
+	}
+
+	return left, nil
 }
 
 // PruneBlobsResult contains the result of a blob prune operation
@@ -75,6 +165,10 @@ type PruneBlobsResult struct {
 	BlobsDeleted int   `json:"blobs_deleted"`
 	BlobsFailed  int   `json:"blobs_failed,omitempty"`
 	BytesFreed   int64 `json:"bytes_freed"`
+	// BlobsTooNew counts the unreferenced blobs kept because they are
+	// newer than the prune grace period, or of unknown age. BlobsFound
+	// does not include them.
+	BlobsTooNew int `json:"blobs_too_new"`
 }
 
 // Prune removes orphaned data from the local index database AND
@@ -120,9 +214,15 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 		return err
 	}
 
-	unreferencedBlobs, totalSize := v.findUnreferencedBlobs(allBlobs, allBlobsReferenced)
+	unreferencedBlobs, totalSize, tooNew := v.findUnreferencedBlobs(
+		allBlobs, allBlobsReferenced)
 
-	result := &PruneBlobsResult{BlobsFound: len(unreferencedBlobs)}
+	result := &PruneBlobsResult{
+		BlobsFound:  len(unreferencedBlobs),
+		BlobsTooNew: tooNew,
+	}
+
+	v.reportBlobsTooNew(tooNew, opts)
 
 	if len(unreferencedBlobs) == 0 {
 		log.Info("No unreferenced blobs found")
@@ -131,7 +231,7 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 			return v.outputPruneBlobsJSON(result)
 		}
 
-		v.printlnStdout("No unreferenced blobs to remove.")
+		v.UI.Infof("No unreferenced blobs to remove.")
 
 		return nil
 	}
@@ -139,12 +239,17 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 	log.Info("Found unreferenced blobs",
 		"count", len(unreferencedBlobs), "total_size", ubytes(totalSize))
 
-	if !opts.JSON {
-		v.stdoutf("Found %d unreferenced blob(s) totaling %s\n",
+	if opts.Force && !opts.JSON {
+		v.UI.Infof("Found %d unreferenced blob(s) totaling %s.",
 			len(unreferencedBlobs), ubytes(totalSize))
 	}
 
 	if !opts.Force && !opts.JSON {
+		// The count, the prompt and its answer bypass the UI: they are an
+		// interactive exchange the operator must see, so --quiet does not
+		// silence them.
+		v.stdoutf("Found %d unreferenced blob(s) totaling %s.\n",
+			len(unreferencedBlobs), ubytes(totalSize))
 		v.stdoutf("\nDelete %d unreferenced blob(s)? [y/N] ", len(unreferencedBlobs))
 
 		var confirm string
@@ -166,14 +271,17 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 	v.deleteUnreferencedBlobs(unreferencedBlobs, allBlobs, result)
 
 	if opts.JSON {
-		return v.outputPruneBlobsJSON(result)
+		err = v.outputPruneBlobsJSON(result)
+		if err != nil {
+			return err
+		}
+	} else {
+		v.UI.Infof("Deleted %d blob(s) totaling %s.",
+			result.BlobsDeleted, ubytes(result.BytesFreed))
 	}
 
-	v.stdoutf("\nDeleted %d blob(s) totaling %s\n",
-		result.BlobsDeleted, ubytes(result.BytesFreed))
-
 	if result.BlobsFailed > 0 {
-		v.stdoutf("Failed to delete %d blob(s)\n", result.BlobsFailed)
+		return fmt.Errorf("%d %w", result.BlobsFailed, errBlobDeleteFailed)
 	}
 
 	return nil
@@ -248,11 +356,12 @@ func (v *Vaultik) listUniqueSnapshotIDs() ([]string, error) {
 	return snapshotIDs, nil
 }
 
-// listAllRemoteBlobs returns a map of all blob hashes to their sizes in remote storage
-func (v *Vaultik) listAllRemoteBlobs() (map[string]int64, error) {
+// listAllRemoteBlobs returns a map of all blob hashes in remote storage
+// to their listing entries.
+func (v *Vaultik) listAllRemoteBlobs() (map[string]storage.ObjectInfo, error) {
 	log.Info("Listing all blobs in storage")
 
-	allBlobs := make(map[string]int64)
+	allBlobs := make(map[string]storage.ObjectInfo)
 	blobObjectCh := v.Storage.ListStream(v.ctx, "blobs/")
 
 	for object := range blobObjectCh {
@@ -276,7 +385,7 @@ func (v *Vaultik) listAllRemoteBlobs() (map[string]int64, error) {
 			continue
 		}
 
-		allBlobs[parts[3]] = object.Size
+		allBlobs[parts[3]] = object
 	}
 
 	log.Info("Found blobs in storage", "count", len(allBlobs))
@@ -284,30 +393,59 @@ func (v *Vaultik) listAllRemoteBlobs() (map[string]int64, error) {
 	return allBlobs, nil
 }
 
-// findUnreferencedBlobs returns blob hashes not referenced by any
-// manifest and their total size.
+// findUnreferencedBlobs returns the hashes of the blobs no manifest
+// references that are older than the prune grace period, and their total
+// size. It also returns how many unreferenced blobs it left out for
+// being newer than that, or of unknown age because the listing gave no
+// time for them.
 func (v *Vaultik) findUnreferencedBlobs(
-	allBlobs map[string]int64, referenced map[string]bool,
-) ([]string, int64) {
+	allBlobs map[string]storage.ObjectInfo, referenced map[string]bool,
+) ([]string, int64, int) {
 	var (
 		unreferenced []string
 		totalSize    int64
+		tooNew       int
 	)
 
-	for hash, size := range allBlobs {
-		if !referenced[hash] {
-			unreferenced = append(unreferenced, hash)
-			totalSize += size
+	cutoff := time.Now().UTC().Add(-v.Config.PruneGracePeriod)
+
+	for hash, blob := range allBlobs {
+		if referenced[hash] {
+			continue
 		}
+
+		if blob.ModTime.IsZero() || blob.ModTime.After(cutoff) {
+			tooNew++
+
+			continue
+		}
+
+		unreferenced = append(unreferenced, hash)
+		totalSize += blob.Size
 	}
 
-	return unreferenced, totalSize
+	return unreferenced, totalSize, tooNew
+}
+
+// reportBlobsTooNew says how many unreferenced blobs prune kept for being
+// newer than the grace period or of unknown age. Under --json the
+// document carries the count instead.
+func (v *Vaultik) reportBlobsTooNew(tooNew int, opts *PruneOptions) {
+	if tooNew == 0 || opts.JSON {
+		return
+	}
+
+	v.UI.Infof("Kept %d unreferenced blob(s) newer than "+
+		"prune_grace_period (%s), or of unknown age: a backup still "+
+		"running may need them.",
+		tooNew, v.Config.PruneGracePeriod)
 }
 
 // deleteUnreferencedBlobs deletes the given blobs from storage and
 // populates the result.
 func (v *Vaultik) deleteUnreferencedBlobs(
-	unreferencedBlobs []string, allBlobs map[string]int64, result *PruneBlobsResult,
+	unreferencedBlobs []string, allBlobs map[string]storage.ObjectInfo,
+	result *PruneBlobsResult,
 ) {
 	log.Info("Deleting unreferenced blobs")
 
@@ -322,7 +460,7 @@ func (v *Vaultik) deleteUnreferencedBlobs(
 		}
 
 		result.BlobsDeleted++
-		result.BytesFreed += allBlobs[hash]
+		result.BytesFreed += allBlobs[hash].Size
 
 		if (i+1)%progressLogEvery == 0 || i == len(unreferencedBlobs)-1 {
 			log.Info("Deletion progress",

internal/vaultik/prune_count_test.go

@@ -1,13 +1,17 @@
 package vaultik //nolint:testpackage // exercises unexported count helpers
 
 import (
+	"bytes"
 	"context"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+	"sneak.berlin/go/vaultik/internal/ui"
 )
 
 // TestTableCountForReportSurfacesReadFailure is the regression guard for
@@ -76,3 +80,59 @@ func TestCountDiffUnknownWhenEitherSideUnknown(t *testing.T) {
 	assert.Nil(t, countDiff(&before, nil), "unknown after yields unknown delta")
 	assert.Nil(t, countDiff(nil, nil))
 }
+
+// TestPruneDatabaseSummaryLabelsTotals removes the only incomplete
+// snapshot from an index that also holds a complete one. The figure in
+// brackets is every snapshot left in the local index, so the line must
+// say so rather than read as incomplete snapshots that remain.
+func TestPruneDatabaseSummaryLabelsTotals(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, ":memory:")
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	repos := database.NewRepositories(db)
+
+	startedAt := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	completedAt := startedAt.Add(time.Minute)
+
+	require.NoError(t, repos.Snapshots.Create(ctx, nil, &database.Snapshot{
+		ID:          "host_complete_2026-01-02T03:04:05Z",
+		Hostname:    "host",
+		StartedAt:   startedAt,
+		CompletedAt: &completedAt,
+	}))
+	require.NoError(t, repos.Snapshots.Create(ctx, nil, &database.Snapshot{
+		ID:        "host_incomplete_2026-01-02T03:04:05Z",
+		Hostname:  "host",
+		StartedAt: startedAt,
+	}))
+
+	var out bytes.Buffer
+
+	v := &Vaultik{
+		DB:           db,
+		Repositories: repos,
+		SnapshotManager: snapshot.NewSnapshotManager(
+			snapshot.SnapshotManagerParams{Repos: repos}),
+		UI: ui.NewWithColor(&out, false),
+	}
+	v.SetContext(ctx)
+
+	result, err := v.PruneDatabase()
+	require.NoError(t, err)
+	require.Equal(t, int64(1), result.SnapshotsDeleted)
+
+	assert.Contains(t, out.String(), "Incomplete snapshots: 1 removed "+
+		"(total snapshots left in the local index: 1).")
+	assert.Contains(t, out.String(), "Orphaned files: 0 removed "+
+		"(total files left in the local index: 0).")
+	assert.Contains(t, out.String(), "Orphaned chunks: 0 removed "+
+		"(total chunks left in the local index: 0).")
+	assert.Contains(t, out.String(), "Orphaned blobs: 0 removed "+
+		"(total blobs left in the local index: 0).")
+}

internal/vaultik/prune_delete_failure_test.go

@@ -0,0 +1,98 @@
+package vaultik_test
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage/faultstore"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// refuseBlobDeletes is an OnDelete hook that fails every blob delete.
+func refuseBlobDeletes(key string) bool {
+	return strings.HasPrefix(key, "blobs/")
+}
+
+// TestPruneBlobs_FailedDeleteReturnsError checks that prune returns an
+// error, so the command exits non-zero, when an unreferenced blob cannot
+// be deleted.
+func TestPruneBlobs_FailedDeleteReturnsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+
+	// The store holds no manifest, so nothing references this blob.
+	addBlob(t, env.store.testStorer, testBlobHashA)
+
+	store := faultstore.New(env.store)
+	store.OnDelete = refuseBlobDeletes
+	env.v.Storage = store
+
+	err := env.v.PruneBlobs(&vaultik.PruneOptions{Force: true})
+
+	require.ErrorContains(t, err, "1 blob(s) could not be deleted")
+}
+
+// TestPruneBlobs_JSONFailedDeleteReturnsError checks that prune --json
+// prints its document and then returns an error when an unreferenced
+// blob cannot be deleted.
+func TestPruneBlobs_JSONFailedDeleteReturnsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	addBlob(t, env.store.testStorer, testBlobHashA)
+
+	store := faultstore.New(env.store)
+	store.OnDelete = refuseBlobDeletes
+	env.v.Storage = store
+
+	err := env.v.PruneBlobs(&vaultik.PruneOptions{JSON: true})
+
+	require.Error(t, err, "prune must fail when a blob cannot be deleted")
+
+	var result vaultik.PruneBlobsResult
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &result),
+		"stdout must hold the JSON document, got:\n%s", env.stdout.String())
+	assert.Equal(t, 1, result.BlobsFailed)
+	assert.Equal(t, 0, result.BlobsDeleted)
+}
+
+// TestCreateSnapshotPrune_FailedBlobDeleteReturnsError checks that the
+// prune step of snapshot create --prune fails the run when an
+// unreferenced blob cannot be deleted.
+func TestCreateSnapshotPrune_FailedBlobDeleteReturnsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, _ := backUpToFileDestination(ctx, t, storeDir)
+
+	// No snapshot references this blob, so the prune step deletes it.
+	orphan := filepath.Join(storeDir, "blobs", testBlobHashA[:2],
+		testBlobHashA[2:4], testBlobHashA)
+	require.NoError(t, os.MkdirAll(filepath.Dir(orphan), 0o750))
+	require.NoError(t, os.WriteFile(orphan, []byte("orphan"), 0o600))
+
+	store := faultstore.New(v.Storage)
+	store.OnDelete = refuseBlobDeletes
+	v.Storage = store
+
+	err := v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		Cron:  true,
+		Prune: true,
+	})
+
+	require.ErrorContains(t, err, "1 blob(s) could not be deleted")
+	assert.FileExists(t, orphan)
+}

internal/vaultik/prune_grace_period_test.go

@@ -0,0 +1,166 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"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/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/storage/faultstore"
+	"sneak.berlin/go/vaultik/internal/ui"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestPruneDuringBackupKeepsItsBlobs runs a prune from another host
+// sharing the destination store at the moment a backup has uploaded its
+// blobs but not yet its manifest, so no manifest lists those blobs. The
+// prune must leave them, and the backup must then restore.
+func TestPruneDuringBackupKeepsItsBlobs(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "src")
+	storeDir := t.TempDir()
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	ctx := context.Background()
+	cfg := faultTestConfig()
+	testFiles := writeFaultSourceTree(t, fs, dataDir)
+
+	inner, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	otherHostCfg := faultTestConfig()
+	otherHostCfg.PruneGracePeriod = time.Hour
+	otherHost := newReaderVaultik(ctx, otherHostCfg, inner, nil, fs)
+
+	var (
+		pruned   bool
+		pruneErr error
+	)
+
+	store := faultstore.New(inner)
+	store.OnPut = func(key string) faultstore.PutAction {
+		if strings.HasSuffix(key, "/manifest.json.zst") {
+			pruned = true
+			pruneErr = otherHost.PruneBlobs(&vaultik.PruneOptions{Force: true})
+		}
+
+		return faultstore.PutNormal
+	}
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+
+	repos := database.NewRepositories(db)
+
+	id := fullFaultBackup(ctx, t, fs, store, cfg, repos, dataDir, dbPath, "shared")
+	require.NoError(t, db.Close())
+
+	require.True(t, pruned, "the prune must run before the manifest upload")
+	require.NoError(t, pruneErr)
+
+	v := newReaderVaultik(ctx, cfg, inner, nil, fs)
+	require.NoError(t, v.Restore(&vaultik.RestoreOptions{
+		SnapshotID: id,
+		TargetDir:  restoreDir,
+		Verify:     true,
+	}), "a prune during the backup must not delete the backup's blobs")
+
+	assertRestoredTree(t, fs, restoreDir, testFiles)
+}
+
+// TestPruneKeepsUnreferencedBlobWithNoTime checks that a blob no
+// manifest lists, whose listing gives no modification time (as an S3
+// listing entry without LastModified does), is kept and counted with
+// the blobs too new to delete: its age is unknown.
+func TestPruneKeepsUnreferencedBlobWithNoTime(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+
+	// Written into the mock store's map directly rather than through
+	// Put, so its listing gives no time.
+	blobKey := "blobs/" + testBlobHashA[:2] + "/" + testBlobHashA[2:4] +
+		"/" + testBlobHashA
+	env.store.data[blobKey] = []byte("blob")
+
+	require.NoError(t, env.v.PruneBlobs(&vaultik.PruneOptions{JSON: true}))
+
+	assert.True(t, env.store.hasKey(blobKey),
+		"a blob of unknown age must be kept")
+
+	var result vaultik.PruneBlobsResult
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &result))
+	assert.Equal(t, 0, result.BlobsDeleted)
+	assert.Equal(t, 1, result.BlobsTooNew)
+}
+
+// TestPruneDeletesUnreferencedBlobOlderThanGracePeriod checks that the
+// grace period only delays deletion: a blob no manifest lists is deleted
+// once it is older than the grace period, and kept and counted until
+// then.
+func TestPruneDeletesUnreferencedBlobOlderThanGracePeriod(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := t.TempDir()
+
+	store, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	oldKey := "blobs/" + testBlobHashA[:2] + "/" + testBlobHashA[2:4] +
+		"/" + testBlobHashA
+	newKey := "blobs/" + testBlobHashB[:2] + "/" + testBlobHashB[2:4] +
+		"/" + testBlobHashB
+
+	for _, key := range []string{oldKey, newKey} {
+		require.NoError(t, store.Put(ctx, key, strings.NewReader("blob")))
+	}
+
+	gracePeriod := time.Hour
+	beforeGracePeriod := time.Now().Add(-gracePeriod - time.Minute)
+	require.NoError(t, os.Chtimes(filepath.Join(storeDir, oldKey),
+		beforeGracePeriod, beforeGracePeriod))
+
+	stdout := &bytes.Buffer{}
+	v := &vaultik.Vaultik{
+		Config:  &config.Config{PruneGracePeriod: gracePeriod},
+		Storage: store,
+		Stdout:  stdout,
+		UI:      ui.NewWithColor(stdout, false),
+	}
+	v.SetContext(ctx)
+
+	require.NoError(t, v.PruneBlobs(&vaultik.PruneOptions{JSON: true}))
+
+	_, err = store.Stat(ctx, oldKey)
+	require.ErrorIs(t, err, storage.ErrNotFound,
+		"a blob older than the grace period must be deleted")
+
+	_, err = store.Stat(ctx, newKey)
+	require.NoError(t, err, "a blob newer than the grace period must be kept")
+
+	var result vaultik.PruneBlobsResult
+
+	require.NoError(t, json.Unmarshal(stdout.Bytes(), &result))
+	assert.Equal(t, 1, result.BlobsDeleted)
+	assert.Equal(t, 1, result.BlobsTooNew)
+}

internal/vaultik/purge_local_remote_consistency_test.go

@@ -14,6 +14,7 @@ import (
 	"sneak.berlin/go/vaultik/internal/log"
 	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/types"
+	"sneak.berlin/go/vaultik/internal/ui"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
@@ -34,6 +35,9 @@ func setupConsistencyTest(
 	repos := database.NewRepositories(db)
 	mockStorage := NewMockStorer()
 
+	// A backup wrote to this destination, so it carries the marker.
+	require.NoError(t, mockStorage.Put(ctx, markerFile, strings.NewReader("")))
+
 	for _, id := range snapshotIDs {
 		parts := strings.Split(id, "_")
 		startedAt, err := time.Parse(time.RFC3339, parts[len(parts)-1])
@@ -58,13 +62,16 @@ func setupConsistencyTest(
 		require.NoError(t, err)
 	}
 
+	stdout := &bytes.Buffer{}
+
 	v := &vaultik.Vaultik{
 		Storage:      mockStorage,
 		Repositories: repos,
 		DB:           db,
-		Stdout:       &bytes.Buffer{},
+		Stdout:       stdout,
 		Stderr:       &bytes.Buffer{},
 		Stdin:        &bytes.Buffer{},
+		UI:           ui.NewWithColor(stdout, false),
 	}
 	v.SetContext(ctx)
 

internal/vaultik/purge_per_name_test.go

@@ -14,6 +14,7 @@ import (
 	"sneak.berlin/go/vaultik/internal/log"
 	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/types"
+	"sneak.berlin/go/vaultik/internal/ui"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
@@ -44,6 +45,9 @@ func setupPurgeTest(
 	repos := database.NewRepositories(db)
 	mockStorage := NewMockStorer()
 
+	// A backup wrote to this destination, so it carries the marker.
+	require.NoError(t, mockStorage.Put(ctx, markerFile, strings.NewReader("")))
+
 	// Insert each snapshot into the DB and create remote metadata stubs.
 	// Use timestamps parsed from snapshot IDs for realistic ordering.
 	for _, id := range snapshotIDs {
@@ -86,6 +90,7 @@ func setupPurgeTest(
 		Stdout:       stdout,
 		Stderr:       stderr,
 		Stdin:        stdin,
+		UI:           ui.NewWithColor(stdout, false),
 	}
 	v.SetContext(ctx)
 
@@ -279,7 +284,8 @@ func TestPurgeOlderThan_TooLargeDeletesNothing(t *testing.T) {
 		OlderThan: "1000y",
 		Force:     true,
 	})
-	require.Error(t, err)
+	require.EqualError(t, err,
+		`invalid duration: "1000y" is longer than about 292 years`)
 
 	err = v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		OlderThan: "292y",

internal/vaultik/remove_quiet_test.go

@@ -0,0 +1,152 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/ui"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestPurgeSnapshots_QuietSuppressesReport checks that `snapshot purge
+// --keep-latest --force` under --quiet deletes the older snapshot and
+// prints nothing.
+func TestPurgeSnapshots_QuietSuppressesReport(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	v := setupPurgeTest(t, snapHostname, []string{snapHomeT0, snapHomeT1})
+
+	var out bytes.Buffer
+
+	v.Stdout = &out
+	v.UI = ui.NewWithColor(&out, false)
+	v.UI.SetQuiet(true)
+
+	require.NoError(t, v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		KeepLatest: true,
+		Force:      true,
+	}))
+
+	assert.Equal(t, []string{snapHomeT1}, listRemainingSnapshots(t, v))
+	assert.Empty(t, out.String())
+}
+
+// TestPurgeSnapshots_QuietPromptListsSnapshots checks that `snapshot
+// purge --keep-latest` without --force still lists the snapshots its
+// prompt asks to delete under --quiet.
+func TestPurgeSnapshots_QuietPromptListsSnapshots(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	v := setupPurgeTest(t, snapHostname, []string{snapHomeT0, snapHomeT1})
+
+	var out bytes.Buffer
+
+	v.Stdout = &out
+	v.UI = ui.NewWithColor(&out, false)
+	v.UI.SetQuiet(true)
+
+	// Stdin is empty, so the prompt reads no answer and cancels.
+	require.NoError(t, v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		KeepLatest: true,
+	}))
+
+	assert.Contains(t, out.String(), snapHomeT0)
+	assert.Contains(t, out.String(), "Delete 1 snapshot(s)? [y/N]")
+	assert.Len(t, listRemainingSnapshots(t, v), 2)
+}
+
+// TestPruneBlobs_QuietPromptShowsBlobTotal checks that `prune` without
+// --force still states the count and total size of the blobs its prompt
+// asks to delete under --quiet.
+func TestPruneBlobs_QuietPromptShowsBlobTotal(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	env.v.UI.SetQuiet(true)
+
+	// The store holds no manifest, so nothing references this blob.
+	addBlob(t, env.store.testStorer, testBlobHashA)
+
+	// Stdin is empty, so the prompt reads no answer and cancels.
+	require.NoError(t, env.v.PruneBlobs(&vaultik.PruneOptions{}))
+
+	assert.Contains(t, env.stdout.String(),
+		"Found 1 unreferenced blob(s) totaling")
+	assert.Contains(t, env.stdout.String(),
+		"Delete 1 unreferenced blob(s)? [y/N]")
+}
+
+// TestRemoveSnapshot_QuietSuppressesReport checks that `snapshot remove`
+// under --quiet prints nothing with --dry-run or with --force, and that
+// the --json document still prints.
+func TestRemoveSnapshot_QuietSuppressesReport(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	newQuietRemover := func() (*vaultik.TestVaultik, *testStorer) {
+		store := newTestStorer()
+		addManifest(t, store, "snapshot-001", nil)
+
+		// A backup wrote to this destination, so it carries the marker.
+		require.NoError(t, store.Put(context.Background(), markerFile,
+			bytes.NewReader(nil)))
+
+		tv := vaultik.NewForTesting(store)
+		tv.Repositories = newLocalIndex(t, "snapshot-001")
+		tv.UI.SetQuiet(true)
+
+		return tv, store
+	}
+
+	tv, store := newQuietRemover()
+
+	_, err := tv.RemoveSnapshot("snapshot-001",
+		&vaultik.RemoveOptions{DryRun: true})
+	require.NoError(t, err)
+	assert.Empty(t, tv.Stdout.String(),
+		"--dry-run must print nothing under --quiet")
+
+	_, err = tv.RemoveSnapshot("snapshot-001",
+		&vaultik.RemoveOptions{Force: true})
+	require.NoError(t, err)
+	assert.False(t, store.hasKey(remoteKeyPath("snapshot-001")))
+	assert.Empty(t, tv.Stdout.String(),
+		"--force must print nothing under --quiet")
+
+	tv, _ = newQuietRemover()
+
+	_, err = tv.RemoveSnapshot("snapshot-001",
+		&vaultik.RemoveOptions{JSON: true})
+	require.NoError(t, err)
+
+	var result vaultik.RemoveResult
+
+	require.NoError(t, json.Unmarshal(tv.Stdout.Bytes(), &result),
+		"the --json document must still print under --quiet")
+	assert.True(t, result.RemoteRemoved)
+}
+
+// TestNukeRemote_QuietSuppressesReport checks that `remote nuke --force`
+// under --quiet prints nothing while it removes a snapshot's metadata
+// and its blobs.
+func TestNukeRemote_QuietSuppressesReport(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+	v.UI.SetQuiet(true)
+
+	require.NoError(t, v.NukeRemote(true))
+	assert.Empty(t, out.String())
+}

internal/vaultik/remove_snapshot_test.go

@@ -17,6 +17,7 @@ import (
 	"sneak.berlin/go/vaultik/internal/log"
 	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/storage/faultstore"
 	"sneak.berlin/go/vaultik/internal/types"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
@@ -25,11 +26,14 @@ import (
 type testStorer struct {
 	mu   sync.Mutex
 	data map[string][]byte
+	// putTimes holds when Put stored each object; ListStream reports it.
+	putTimes map[string]time.Time
 }
 
 func newTestStorer() *testStorer {
 	return &testStorer{
-		data: make(map[string][]byte),
+		data:     make(map[string][]byte),
+		putTimes: make(map[string]time.Time),
 	}
 }
 
@@ -43,6 +47,7 @@ func (s *testStorer) Put(_ context.Context, key string, reader io.Reader) error
 	}
 
 	s.data[key] = data
+	s.putTimes[key] = time.Now().UTC()
 
 	return nil
 }
@@ -86,6 +91,7 @@ func (s *testStorer) Delete(_ context.Context, key string) error {
 	defer s.mu.Unlock()
 
 	delete(s.data, key)
+	delete(s.putTimes, key)
 
 	return nil
 }
@@ -119,8 +125,9 @@ func (s *testStorer) ListStream(
 		for key, data := range s.data {
 			if prefix == "" || strings.HasPrefix(key, prefix) {
 				ch <- storage.ObjectInfo{
-					Key:  key,
-					Size: int64(len(data)),
+					Key:     key,
+					Size:    int64(len(data)),
+					ModTime: s.putTimes[key],
 				}
 			}
 		}
@@ -303,6 +310,10 @@ func TestRemoveSnapshot_DefaultRemovesMetadataNotBlobs(t *testing.T) {
 	addBlob(t, store, blobUnique)
 	addBlob(t, store, blobShared)
 
+	// A backup wrote to this destination, so it carries the marker.
+	require.NoError(t, store.Put(context.Background(), markerFile,
+		bytes.NewReader(nil)))
+
 	tv := vaultik.NewForTesting(store)
 	tv.Repositories = newLocalIndex(t, "snapshot-001", "snapshot-002")
 
@@ -350,7 +361,7 @@ func TestRemoveSnapshot_DryRun(t *testing.T) {
 	assert.True(t, store.hasKey("blobs/aa/aa/"+blobA))
 	assert.True(t, store.hasKey(remoteKeyPath("snapshot-001")))
 
-	assert.Contains(t, tv.Stdout.String(), "[Dry run - no changes made]")
+	assert.Contains(t, tv.Stdout.String(), "Dry run: no changes made.")
 }
 
 // TestRemoveSnapshot_UnknownIDIsError checks that an ID naming no
@@ -372,6 +383,10 @@ func TestRemoveSnapshot_UnknownIDIsError(t *testing.T) {
 			store := newTestStorer()
 			addManifest(t, store, "snapshot-001", nil)
 
+			// A backup wrote to this destination, so it carries the marker.
+			require.NoError(t, store.Put(context.Background(), markerFile,
+				bytes.NewReader(nil)))
+
 			tv := vaultik.NewForTesting(store)
 			tv.Repositories = newLocalIndex(t, "snapshot-001")
 
@@ -392,7 +407,12 @@ func TestRemoveSnapshot_NoMetadataOnDestination(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
 
-	tv := vaultik.NewForTesting(newTestStorer())
+	// A backup wrote to this destination, so it carries the marker.
+	store := newTestStorer()
+	require.NoError(t, store.Put(context.Background(), markerFile,
+		bytes.NewReader(nil)))
+
+	tv := vaultik.NewForTesting(store)
 	tv.Repositories = newLocalIndex(t, "snapshot-001")
 
 	result, err := tv.RemoveSnapshot("snapshot-001",
@@ -431,6 +451,40 @@ func TestRemoveSnapshot_NotInLocalIndex(t *testing.T) {
 	assert.Contains(t, out, "Removed snapshot metadata from remote storage")
 }
 
+// TestRemoveSnapshot_NotInLocalIndexDeleteFails checks that removing a
+// snapshot only the destination store holds fails when its metadata
+// cannot be deleted there, under --force and under --json: nothing was
+// removed, so the command must not succeed.
+func TestRemoveSnapshot_NotInLocalIndexDeleteFails(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	for name, opts := range map[string]vaultik.RemoveOptions{
+		"force": {Force: true},
+		"json":  {JSON: true},
+	} {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			store := newTestStorer()
+			addManifest(t, store, "snapshot-001", nil)
+
+			faulty := faultstore.New(store)
+			faulty.OnDelete = func(string) bool { return true }
+
+			tv := vaultik.NewForTesting(faulty)
+			tv.Repositories = newLocalIndex(t)
+
+			result, err := tv.RemoveSnapshot("snapshot-001", &opts)
+
+			require.ErrorIs(t, err, faultstore.ErrInjectedDelete)
+			assert.False(t, result.RemoteRemoved)
+			assert.True(t, store.hasKey(remoteKeyPath("snapshot-001")))
+			assert.Empty(t, tv.Stdout.String())
+		})
+	}
+}
+
 func TestRemoveAllSnapshots_RequiresForce(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -476,7 +530,8 @@ func TestRemoveAllSnapshots_WithForce(t *testing.T) {
 	out := tv.Stdout.String()
 	assert.Contains(t, out, "Removed 2 snapshot(s)")
 	assert.Contains(t, out, "Removed snapshot metadata from remote storage")
-	assert.Contains(t, out, "vaultik prune")
+	// Remote nuke, the only caller, deletes the blobs next itself.
+	assert.NotContains(t, out, "vaultik prune")
 }
 
 func TestRemoveAllSnapshots_DryRun(t *testing.T) {
@@ -503,7 +558,7 @@ func TestRemoveAllSnapshots_DryRun(t *testing.T) {
 
 	assert.Equal(t, initialCount, store.keyCount())
 
-	assert.Contains(t, tv.Stdout.String(), "[Dry run - no changes made]")
+	assert.Contains(t, tv.Stdout.String(), "Dry run: no changes made.")
 }
 
 func TestRemoveAllSnapshots_NoSnapshots(t *testing.T) {

internal/vaultik/restore.go

@@ -12,6 +12,7 @@ import (
 	"path/filepath"
 	"slices"
 	"strings"
+	"syscall"
 	"time"
 
 	"filippo.io/age"
@@ -27,6 +28,7 @@ import (
 // Sentinel errors for restore failures.
 var (
 	errFilesFailedRestore    = errors.New("file(s) failed to restore")
+	errPathsNotInSnapshot    = errors.New("requested path(s) not in the snapshot")
 	errFilesFailedVerify     = errors.New("files failed verification")
 	errDecryptionKeyRequired = errors.New(
 		"decryption key required for restore\n\n" +
@@ -103,11 +105,12 @@ type RestoreResult struct {
 	BlobsDownloaded int
 	BytesDownloaded int64
 	Duration        time.Duration
+	FilesFailed     int
+	FailedFiles     []string // Paths of files that failed to restore
 	// Verification results (only populated if Verify option is set)
-	FilesVerified int
-	BytesVerified int64
-	FilesFailed   int
-	FailedFiles   []string // Paths of files that failed verification
+	FilesVerified     int
+	BytesVerified     int64
+	VerifyFailedFiles []string // Paths of restored files that failed verification
 }
 
 // Restore restores files from a snapshot to the target directory
@@ -149,16 +152,20 @@ func (v *Vaultik) Restore(opts *RestoreOptions) error {
 	repos := database.NewRepositories(tempDB)
 
 	// Step 2: Get list of files to restore
-	files, err := v.getFilesToRestore(v.ctx, repos, opts.Paths)
+	files, unmatched, err := v.getFilesToRestore(v.ctx, repos, opts.Paths)
 	if err != nil {
 		return fmt.Errorf("getting files to restore: %w", err)
 	}
 
+	// Named before anything is restored, so the names are shown even
+	// when the restore stops on an error.
+	unmatchedErr := v.reportUnmatchedPaths(unmatched)
+
 	if len(files) == 0 {
 		log.Warn("No files found to restore")
 		v.UI.Warningf("No files found to restore.")
 
-		return nil
+		return unmatchedErr
 	}
 
 	log.Info("Found files to restore", "count", len(files))
@@ -184,6 +191,20 @@ func (v *Vaultik) Restore(opts *RestoreOptions) error {
 
 	result.Duration = time.Since(startTime)
 
+	return v.finishRestore(repos, files, unmatchedErr, opts, result)
+}
+
+// finishRestore prints the restore summary and the post-restore
+// warnings, runs optional verification, and converts any failed-file
+// count into an error. Otherwise it returns unmatchedErr, the error for
+// the requested paths that matched nothing (nil when there are none).
+func (v *Vaultik) finishRestore(
+	repos *database.Repositories,
+	files []*database.File,
+	unmatchedErr error,
+	opts *RestoreOptions,
+	result *RestoreResult,
+) error {
 	log.Info("Restore complete",
 		"files_restored", result.FilesRestored,
 		"bytes_restored", ubytes(result.BytesRestored),
@@ -198,21 +219,8 @@ func (v *Vaultik) Restore(opts *RestoreOptions) error {
 		v.UI.Duration(result.Duration),
 	)
 
-	return v.finishRestore(repos, files, opts, result)
-}
-
-// finishRestore emits the post-restore warnings, runs optional
-// verification, and converts any failed-file count into an error.
-func (v *Vaultik) finishRestore(
-	repos *database.Repositories,
-	files []*database.File,
-	opts *RestoreOptions,
-	result *RestoreResult,
-) error {
 	if os.Geteuid() != 0 {
-		v.UI.Warningf("Restore did not preserve file ownership: chown(2) " +
-			"requires root. Re-run as root (e.g. with sudo) if you need " +
-			"original UID/GID preserved.")
+		v.warnIfOwnershipNotPreserved(files, opts.TargetDir)
 	}
 
 	if result.FilesFailed > 0 {
@@ -235,7 +243,60 @@ func (v *Vaultik) finishRestore(
 		return fmt.Errorf("%d %w", result.FilesFailed, errFilesFailedRestore)
 	}
 
-	return nil
+	return unmatchedErr
+}
+
+// warnIfOwnershipNotPreserved warns once if some entry restored under
+// targetDir has another owner or group on disk than the snapshot
+// records. It is meant for a restore by a normal user, which changes no
+// owner or group: a new entry is owned by that user, and its group is
+// that user's, or the group of the directory it is created in on macOS
+// and in a setgid directory on Linux. An entry that is not on disk, or
+// whose filesystem reports no owner, is skipped.
+func (v *Vaultik) warnIfOwnershipNotPreserved(
+	files []*database.File, targetDir string,
+) {
+	for _, file := range files {
+		targetPath, err := containedRestorePath(
+			v.Fs, targetDir, file.Path.String())
+		if err != nil {
+			continue
+		}
+
+		info, err := lstatIfPossible(v.Fs, targetPath)
+		if err != nil {
+			continue
+		}
+
+		owner, ok := info.Sys().(*syscall.Stat_t)
+		if !ok {
+			continue
+		}
+
+		if owner.Uid != file.UID || owner.Gid != file.GID {
+			v.UI.Warningf("Restore did not preserve file ownership: chown(2) " +
+				"requires root. Re-run as root (e.g. with sudo) if you need " +
+				"original UID/GID preserved.")
+
+			return
+		}
+	}
+}
+
+// reportUnmatchedPaths warns, by name, about each requested path that
+// matched nothing in the snapshot, and returns an error counting them,
+// or nil when there are none.
+func (v *Vaultik) reportUnmatchedPaths(unmatched []string) error {
+	if len(unmatched) == 0 {
+		return nil
+	}
+
+	for _, path := range unmatched {
+		log.Warn("Requested path matches nothing in the snapshot", "path", path)
+		v.UI.Warningf("%s matches nothing in the snapshot.", v.UI.Path(path))
+	}
+
+	return fmt.Errorf("%d %w", len(unmatched), errPathsNotInSnapshot)
 }
 
 // restoreIdentities parses the configured age secret key once into every
@@ -298,22 +359,33 @@ func (v *Vaultik) restoreAllFiles(
 		_ = blobCache.Close()
 	}()
 
-	// Per-restore sweep state: every blob_size_limit/100 bytes written,
-	// scan the cache and delete any blob whose remaining file references
-	// are all already restored.
-	sweeper := newRestoreSweeper(v.ctx, repos, blobCache,
-		v.Config.BlobSizeLimit.Int64()/sweepIntervalDivisor)
-
 	blobByHash, blobIDToHash, err := v.buildBlobIndexes(repos)
 	if err != nil {
 		return nil, err
 	}
 
-	plan, err := newRestorePlan(v.ctx, repos, files, chunkToBlobMap, blobIDToHash)
+	plan, unplanned, err := newRestorePlan(
+		v.ctx, repos, files, chunkToBlobMap, blobIDToHash)
 	if err != nil {
 		return nil, fmt.Errorf("building restore plan: %w", err)
 	}
 
+	// Per-restore sweep state: every blob_size_limit/100 bytes written,
+	// scan the cache and delete any blob that no pending file in the
+	// plan references.
+	sweeper := newRestoreSweeper(plan, blobCache,
+		v.Config.BlobSizeLimit.Int64()/sweepIntervalDivisor)
+
+	// Ranging over files rather than the map reports them in a stable order.
+	for _, file := range files {
+		if planErr, ok := unplanned[file.ID]; ok {
+			err = v.handleRestoreFileError(plan, opts, result, file, file.ID, planErr)
+			if err != nil {
+				return nil, err
+			}
+		}
+	}
+
 	filesByID, totalBytesExpected := indexRestoreFiles(files)
 
 	v.UI.Beginf("Restoring %s files (%s) to %s.",
@@ -391,10 +463,8 @@ func (v *Vaultik) runRestoreLoop(
 			continue
 		}
 
-		// Record the file as restored so the sweeper can free blobs
-		// once all referencing files are done, and drop it from the
-		// plan's indexes so future picks ignore it.
-		session.sweeper.fileRestored(fileID.String())
+		// Drop the file from the plan's indexes so future picks ignore
+		// it and the sweeper can free blobs no pending file needs.
 		plan.finishFile(fileID)
 
 		processed++
@@ -647,20 +717,34 @@ func (v *Vaultik) handleRestoreVerification(
 	opts *RestoreOptions,
 	result *RestoreResult,
 ) error {
-	err := v.verifyRestoredFiles(v.ctx, repos, files, opts.TargetDir, result)
+	// A file that failed to restore was not written and is already
+	// reported as such, so it is not verified.
+	failedToRestore := make(map[string]bool, len(result.FailedFiles))
+	for _, path := range result.FailedFiles {
+		failedToRestore[path] = true
+	}
+
+	restored := make([]*database.File, 0, len(files))
+	for _, file := range files {
+		if !failedToRestore[file.Path.String()] {
+			restored = append(restored, file)
+		}
+	}
+
+	err := v.verifyRestoredFiles(v.ctx, repos, restored, opts.TargetDir, result)
 	if err != nil {
 		return fmt.Errorf("verification failed: %w", err)
 	}
 
-	if result.FilesFailed > 0 {
+	if len(result.VerifyFailedFiles) > 0 {
 		v.UI.Errorf("Verification failed: %s files did not match expected checksums.",
-			v.UI.Count(result.FilesFailed))
+			v.UI.Count(len(result.VerifyFailedFiles)))
 
-		for _, path := range result.FailedFiles {
+		for _, path := range result.VerifyFailedFiles {
 			v.UI.Detailf("%s", v.UI.Path(path))
 		}
 
-		return fmt.Errorf("%d %w", result.FilesFailed, errFilesFailedVerify)
+		return fmt.Errorf("%d %w", len(result.VerifyFailedFiles), errFilesFailedVerify)
 	}
 
 	v.UI.Completef("Verified %s files (%s).",
@@ -691,6 +775,8 @@ func (v *Vaultik) downloadSnapshotDB(
 		return nil, "", fmt.Errorf("downloading %s: %w", dbKey, err)
 	}
 
+	reader = newCloseOnCancelReader(v.ctx, reader)
+
 	defer func() { _ = reader.Close() }()
 
 	// Decrypt and decompress straight from the storage stream, then stream
@@ -812,27 +898,34 @@ func (v *Vaultik) materializeSnapshotDB(
 	return db, tempDir, nil
 }
 
-// getFilesToRestore returns the list of files to restore based on path filters
+// getFilesToRestore returns the list of files to restore based on path
+// filters, and the filters, as given, that matched no file.
 func (v *Vaultik) getFilesToRestore(
 	ctx context.Context, repos *database.Repositories, pathFilters []string,
-) ([]*database.File, error) {
+) ([]*database.File, []string, error) {
 	// If no filters, get all files
 	if len(pathFilters) == 0 {
-		return repos.Files.ListAll(ctx)
+		files, err := repos.Files.ListAll(ctx)
+
+		return files, nil, err
 	}
 
 	// Get files matching the path filters
-	var result []*database.File
+	var (
+		result    []*database.File
+		unmatched []string
+	)
 
 	seen := make(map[string]bool)
 
 	for _, filter := range pathFilters {
-		// Normalize the filter path
-		filter = filepath.Clean(filter)
-
-		files, err := repos.Files.ListUnderPath(ctx, filter)
+		files, err := repos.Files.ListUnderPath(ctx, filepath.Clean(filter))
 		if err != nil {
-			return nil, fmt.Errorf("listing files under %s: %w", filter, err)
+			return nil, nil, fmt.Errorf("listing files under %s: %w", filter, err)
+		}
+
+		if len(files) == 0 {
+			unmatched = append(unmatched, filter)
 		}
 
 		for _, file := range files {
@@ -843,7 +936,7 @@ func (v *Vaultik) getFilesToRestore(
 		}
 	}
 
-	return result, nil
+	return result, unmatched, nil
 }
 
 // buildChunkToBlobMap creates a mapping from chunk hash to blob information
@@ -908,8 +1001,9 @@ type restoreSession struct {
 	// root can chown a file to an arbitrary UID/GID — non-root chown
 	// always fails with EPERM. Attempting it anyway produces N
 	// guaranteed-failed syscalls + N noisy debug lines, so we skip
-	// the call entirely as non-root and emit one warning at the end
-	// of the restore explaining that ownership was not preserved.
+	// the call entirely as non-root; warnIfOwnershipNotPreserved warns
+	// once at the end of the restore if that left any owner or group
+	// other than recorded.
 	runningAsRoot bool
 	// directories holds every restored directory, for
 	// applyDirectoryMetadata to finish after the restore loop.
@@ -1420,8 +1514,8 @@ func (v *Vaultik) verifyRestoredFiles(
 		if err != nil {
 			log.Error("File verification failed", "path", file.Path, "error", err)
 
-			result.FilesFailed++
-			result.FailedFiles = append(result.FailedFiles, file.Path.String())
+			result.VerifyFailedFiles = append(
+				result.VerifyFailedFiles, file.Path.String())
 		}
 
 		bytesProcessed += file.Size
@@ -1437,7 +1531,7 @@ func (v *Vaultik) verifyRestoredFiles(
 	log.Info("Verification complete",
 		"files_verified", result.FilesVerified,
 		"bytes_verified", ubytes(result.BytesVerified),
-		"files_failed", result.FilesFailed,
+		"files_failed", len(result.VerifyFailedFiles),
 	)
 
 	return nil

internal/vaultik/restore_another_machine_test.go

@@ -42,7 +42,7 @@ func TestRestoreOnAnotherMachine(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 

internal/vaultik/restore_interrupt_test.go

@@ -65,7 +65,7 @@ func TestRestoreCleansTempDirOnInterrupt(t *testing.T) {
 	root := t.TempDir()
 
 	dataDir := filepath.Join(root, "source")
-	storeDir := filepath.Join(root, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(root, "restored")
 	dbPath := filepath.Join(root, "index.sqlite")
 
@@ -74,7 +74,8 @@ func TestRestoreCleansTempDirOnInterrupt(t *testing.T) {
 	buildLocalityFixture(t, fs, dataDir)
 
 	cfg, storer, snapshotID := setupLocalityBackup(
-		context.Background(), t, fs, dataDir, storeDir, dbPath)
+		context.Background(), t, fs, dataDir, storeDir, dbPath,
+		localitySourceBytes)
 
 	// Point the "" temp paths (the blob cache directory and the
 	// snapshot-database directory) at a private directory so the test can
@@ -142,14 +143,14 @@ func TestRestoreCleansTempDirOnInterrupt(t *testing.T) {
 }
 
 // scratchEntries returns the vaultik blob-cache and snapshot-database
-// scratch entries currently present in dir.
+// scratch entries (restore's and deep verify's) currently present in dir.
 func scratchEntries(t *testing.T, dir string) []string {
 	t.Helper()
 
 	var matches []string
 
 	for _, pattern := range []string{
-		"vaultik-blobcache-*", "vaultik-restore-*",
+		"vaultik-blobcache-*", "vaultik-restore-*", "vaultik-verify-*",
 	} {
 		found, err := filepath.Glob(filepath.Join(dir, pattern))
 		require.NoError(t, err)
@@ -206,3 +207,141 @@ func TestRestoreSkipErrorsCancelDuringBlobDownload(t *testing.T) {
 	assert.NotContains(t, out.String(), srcPath,
 		"the cancel was reported as a failed file")
 }
+
+// stallingStorer wraps a Storer so that each download of a key chosen by
+// stalls hands over the first half of the object and then stalls,
+// ignoring the context, as a file:// read on a hung USB stick or network
+// share does. Only closing the returned reader ends the stall.
+type stallingStorer struct {
+	storage.Storer
+
+	stalls  func(key string) bool
+	once    sync.Once
+	stalled chan struct{}
+}
+
+func newStallingStorer(
+	inner storage.Storer, stalls func(key string) bool,
+) *stallingStorer {
+	return &stallingStorer{
+		Storer:  inner,
+		stalls:  stalls,
+		stalled: make(chan struct{}),
+	}
+}
+
+func (s *stallingStorer) Get(
+	ctx context.Context, key string,
+) (io.ReadCloser, error) {
+	rc, err := s.Storer.Get(ctx, key)
+	if err != nil || !s.stalls(key) {
+		return rc, err
+	}
+
+	object, err := io.ReadAll(rc)
+	_ = rc.Close()
+
+	if err != nil {
+		return nil, err
+	}
+
+	// The writer is never closed, so once the first half has been read
+	// the next Read blocks until the reader is closed.
+	pr, pw := io.Pipe()
+
+	go func() {
+		_, _ = pw.Write(object[:len(object)/2])
+
+		s.once.Do(func() { close(s.stalled) })
+	}()
+
+	return pr, nil
+}
+
+// TestInterruptDuringStalledReadCleansTempDir stops a restore and a deep
+// verify while a read of a blob or of the snapshot database is stalled
+// on a source that ignores the context. Stopping must end that read, so
+// the operation unwinds and removes its decrypted scratch files well
+// within the shutdown timeout (issues #325 and #356).
+//
+// Not parallel: it points TMPDIR at a private directory (via t.Setenv)
+// so it can assert on exactly the scratch files the operation created.
+func TestInterruptDuringStalledReadCleansTempDir(t *testing.T) {
+	log.Initialize(log.Config{})
+
+	stalledReads := map[string]func(key string) bool{
+		"blob": func(key string) bool {
+			return strings.HasPrefix(key, "blobs/")
+		},
+		"snapshot database": func(key string) bool {
+			return strings.HasSuffix(key, "/db.zst.age")
+		},
+	}
+
+	operations := map[string]func(v *Vaultik, snapshotID, targetDir string) error{
+		"restore": func(v *Vaultik, snapshotID, targetDir string) error {
+			return v.Restore(&RestoreOptions{
+				SnapshotID: snapshotID,
+				TargetDir:  targetDir,
+			})
+		},
+		"deep verify": func(v *Vaultik, snapshotID, _ string) error {
+			return v.RunDeepVerify(snapshotID, &VerifyOptions{Deep: true})
+		},
+	}
+
+	for readName, stalls := range stalledReads {
+		for opName, op := range operations {
+			t.Run(readName+" read during "+opName, func(t *testing.T) {
+				fs := afero.NewOsFs()
+				root := t.TempDir()
+
+				// Random content, so the blob stays large enough to stall
+				// part-way through its compressed, encrypted body.
+				cfg, storer, snapshotID, _ := backupOneFile(context.Background(),
+					t, fs, root, "a.bin", randomBytes(t, 256*1024), 0o644)
+
+				scratch := filepath.Join(root, "scratch")
+				require.NoError(t, fs.MkdirAll(scratch, 0o755))
+				t.Setenv("TMPDIR", scratch)
+
+				stalling := newStallingStorer(storer, stalls)
+
+				ctx, cancel := context.WithCancel(context.Background())
+				defer cancel()
+
+				v := newRestoreVaultik(ctx, cfg, stalling, fs)
+				v.cancel = cancel
+
+				var opErr error
+
+				stop := v.StartOperation(func() {
+					opErr = op(v, snapshotID, filepath.Join(root, "restored"))
+				})
+
+				select {
+				case <-stalling.stalled:
+				case <-time.After(30 * time.Second):
+					t.Fatal("the operation never reached the stalled read")
+				}
+
+				// The snapshot database's temp directory is created just
+				// after its read starts, so it can appear after the stall.
+				require.Eventually(t, func() bool {
+					return len(scratchEntries(t, scratch)) > 0
+				}, 10*time.Second, 10*time.Millisecond,
+					"expected decrypted scratch files to exist mid-read")
+
+				stopCtx, stopCancel := context.WithTimeout(
+					context.Background(), 10*time.Second)
+				defer stopCancel()
+
+				require.True(t, stop(stopCtx),
+					"stop timed out; the stalled read was not interrupted")
+				require.ErrorIs(t, opErr, context.Canceled)
+				require.Empty(t, scratchEntries(t, scratch),
+					"decrypted scratch files remained after the interrupt")
+			})
+		}
+	}
+}

internal/vaultik/restore_locality_test.go

@@ -63,6 +63,10 @@ type localityCopy struct {
 	data []byte
 }
 
+// localitySourceBytes is the size of each source file of the locality
+// fixture.
+const localitySourceBytes = 1024 * 1024
+
 // buildLocalityFixture writes the adversarial source layout described
 // in TestRestoreLocalityAndReadAt: 15 sources of 1 MiB each (which the
 // backup packs into 3 blobs of 5 chunks) plus 9 interleaved-name copies
@@ -73,7 +77,6 @@ func buildLocalityFixture(
 	t.Helper()
 
 	const (
-		srcBytes   = 1024 * 1024
 		srcCount   = 15
 		blobsCount = 3
 		perBlob    = srcCount / blobsCount
@@ -83,7 +86,7 @@ func buildLocalityFixture(
 	for i := range srcCount {
 		s := &localitySource{
 			path: fmt.Sprintf("src-%03d.bin", i+1),
-			data: randomBytes(t, srcBytes),
+			data: randomBytes(t, localitySourceBytes),
 		}
 		sources[i] = s
 		require.NoError(t, afero.WriteFile(
@@ -115,19 +118,20 @@ func buildLocalityFixture(
 
 // setupLocalityBackup runs the backup half of the locality test: a
 // snapshot of dataDir into a file storer with sizes tuned so 15 one-chunk
-// files pack into 3 blobs. The index database is closed before returning
-// so the restore half runs from remote bytes only.
+// files of srcBytes pack into 3 blobs. The index database is closed
+// before returning so the restore half runs from remote bytes only.
 func setupLocalityBackup(
 	ctx context.Context, t *testing.T, fs afero.Fs,
-	dataDir, storeDir, dbPath string,
+	dataDir, storeDir, dbPath string, srcBytes int64,
 ) (*config.Config, *storage.FileStorer, string) {
 	t.Helper()
 
-	// chunkSize avg = 4 MiB makes minSize = 1 MiB, so a 1 MiB file
-	// becomes one chunk. maxBlobSize = 5 MiB packs exactly 5 chunks
-	// per blob, yielding 3 blobs from 15 source files.
-	chunkSize := int64(4 * 1024 * 1024)
-	maxBlobSize := int64(5 * 1024 * 1024)
+	// An average chunk size of four times srcBytes makes the minimum
+	// chunk srcBytes, so each source file becomes one chunk. A blob limit
+	// of five times srcBytes packs exactly 5 chunks per blob, yielding 3
+	// blobs from 15 source files.
+	chunkSize := 4 * srcBytes
+	maxBlobSize := 5 * srcBytes
 
 	storer, err := storage.NewFileStorer(storeDir)
 	require.NoError(t, err)
@@ -197,7 +201,7 @@ func TestRestoreLocalityAndReadAt(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -208,7 +212,7 @@ func TestRestoreLocalityAndReadAt(t *testing.T) {
 	ctx := context.Background()
 
 	cfg, storer, snapshotID := setupLocalityBackup(
-		ctx, t, fs, dataDir, storeDir, dbPath)
+		ctx, t, fs, dataDir, storeDir, dbPath, localitySourceBytes)
 
 	// Wrap the storer so we can count downloads per blob key.
 	counter := newCountingStorer(storer)

internal/vaultik/restore_malformed_db_test.go

@@ -61,8 +61,8 @@ func craftedFile(
 }
 
 // TestRestoreShortChunkHashInNoBlob proves a file whose short chunk hash
-// has no blob_chunks row fails restore planning and the chunk write with
-// an error.
+// has no blob_chunks row is left out of the restore plan with an error,
+// and fails the chunk write with an error.
 func TestRestoreShortChunkHashInNoBlob(t *testing.T) {
 	t.Parallel()
 
@@ -77,9 +77,11 @@ func TestRestoreShortChunkHashInNoBlob(t *testing.T) {
 	chunkToBlobMap, err := v.buildChunkToBlobMap(ctx, repos)
 	require.NoError(t, err)
 
-	_, err = newRestorePlan(ctx, repos, []*database.File{file},
+	plan, unplanned, err := newRestorePlan(ctx, repos, []*database.File{file},
 		chunkToBlobMap, map[string]string{})
-	require.ErrorIs(t, err, errPlanChunkMissing)
+	require.NoError(t, err)
+	require.ErrorIs(t, unplanned[file.ID], errPlanChunkMissing)
+	require.False(t, plan.hasPending())
 
 	fileChunks, err := repos.FileChunks.GetByFileID(ctx, file.ID)
 	require.NoError(t, err)

internal/vaultik/restore_ownership_warning_test.go

@@ -0,0 +1,131 @@
+package vaultik //nolint:testpackage // calls the unexported ownership check
+
+import (
+	"bytes"
+	"context"
+	"os"
+	"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/log"
+	"sneak.berlin/go/vaultik/internal/ui"
+)
+
+// These tests cover https://git.eeqj.de/sneak/vaultik/issues/316: a
+// restore by a normal user warns that ownership was not preserved only
+// when some restored entry has another owner or group on disk than the
+// snapshot records.
+
+const ownershipWarning = "did not preserve file ownership"
+
+// TestOwnershipWarningSilentWhenEntriesMatch checks that a restore whose
+// entries all end up with their recorded owner and group prints no
+// warning.
+func TestOwnershipWarningSilentWhenEntriesMatch(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	targetDir := t.TempDir()
+	owner := ownerOf(t, targetDir)
+
+	out := restoreAsNormalUser(t, targetDir, []*database.File{
+		{Path: "/a", Mode: plainFileMode, UID: owner.Uid, GID: owner.Gid},
+		{Path: "/b", Mode: plainFileMode, UID: owner.Uid, GID: owner.Gid},
+	})
+
+	assert.Empty(t, out)
+}
+
+// TestOwnershipWarningWhenOwnerDiffers checks that the warning is printed
+// when an entry recorded with another owner is restored owned by the
+// restoring user, even though its group matches the recorded one.
+func TestOwnershipWarningWhenOwnerDiffers(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	targetDir := t.TempDir()
+	owner := ownerOf(t, targetDir)
+
+	out := restoreAsNormalUser(t, targetDir, []*database.File{
+		{Path: "/a", Mode: plainFileMode, UID: otherOwnerID, GID: owner.Gid},
+	})
+
+	assert.Equal(t, 1, strings.Count(out, ownershipWarning), out)
+}
+
+// TestOwnershipWarningWhenDirectoryGivesItsGroup checks that the warning
+// is printed, once, when entries recorded with the restoring user's
+// owner and group take another group from the setgid directory they are
+// restored into, as every new entry takes its directory's group on
+// macOS.
+func TestOwnershipWarningWhenDirectoryGivesItsGroup(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	if os.Geteuid() != 0 {
+		t.Skip("giving a directory to another group needs root")
+	}
+
+	targetDir := t.TempDir()
+	user := ownerOf(t, targetDir)
+
+	require.NoError(t, os.Chown(targetDir, -1, int(otherOwnerID)))
+	require.NoError(t, os.Chmod(targetDir, os.ModeSetgid|writableTestMode))
+
+	out := restoreAsNormalUser(t, targetDir, []*database.File{
+		{Path: "/a", Mode: plainFileMode, UID: user.Uid, GID: user.Gid},
+		{Path: "/b", Mode: plainFileMode, UID: user.Uid, GID: user.Gid},
+	})
+
+	assert.Equal(t, 1, strings.Count(out, ownershipWarning), out)
+}
+
+// ownerOf returns the owner and group path has on disk.
+func ownerOf(t *testing.T, path string) *syscall.Stat_t {
+	t.Helper()
+
+	info, err := os.Lstat(path)
+	require.NoError(t, err)
+
+	owner, ok := info.Sys().(*syscall.Stat_t)
+	require.True(t, ok)
+
+	return owner
+}
+
+// restoreAsNormalUser restores rows, all empty regular files, into
+// targetDir without changing any owner or group, as restore does for a
+// normal user, and returns what warnIfOwnershipNotPreserved prints.
+func restoreAsNormalUser(
+	t *testing.T, targetDir string, rows []*database.File,
+) string {
+	t.Helper()
+
+	ctx := context.Background()
+	rows, repos := makeFiles(ctx, t, rows)
+
+	var out bytes.Buffer
+
+	v := newContainmentVaultik(ctx, afero.NewOsFs())
+	v.UI = ui.NewWithColor(&out, false)
+
+	session := &restoreSession{
+		v:      v,
+		ctx:    ctx,
+		repos:  repos,
+		opts:   &RestoreOptions{TargetDir: targetDir},
+		result: &RestoreResult{},
+	}
+	for _, row := range rows {
+		require.NoError(t, session.restoreFile(row))
+	}
+
+	v.warnIfOwnershipNotPreserved(rows, targetDir)
+
+	return out.String()
+}

internal/vaultik/restore_perms_test.go

@@ -107,7 +107,7 @@ func backupOneFile(
 	require.NoError(t, afero.WriteFile(fs, srcPath, content, mode))
 	require.NoError(t, fs.Chmod(srcPath, mode))
 
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	storer, err := storage.NewFileStorer(storeDir)

internal/vaultik/restore_plan.go

@@ -31,7 +31,8 @@ var (
 //     with no further downloads.
 //   - blobFiles: for each blob, the set of pending files referencing
 //     it. Used to short-circuit "when this blob lands, which files
-//     become ready" without a global scan.
+//     become ready" without a global scan, and by the sweeper to tell
+//     whether a cached blob is still needed.
 type restorePlan struct {
 	fileBlobs map[types.FileID]map[string]struct{}
 	blobFiles map[string]map[types.FileID]struct{}
@@ -41,20 +42,25 @@ type restorePlan struct {
 
 // newRestorePlan builds the file→blob index for the given files. Files
 // whose chunks reference no blobs (symlinks, directories) start in the
-// ready queue immediately.
+// ready queue immediately. A file with a chunk that no blob holds cannot
+// be restored: it is left out of the plan and returned, by ID, with the
+// reason.
 func newRestorePlan(
 	ctx context.Context,
 	repos *database.Repositories,
 	files []*database.File,
 	chunkToBlobMap map[string]*database.BlobChunk,
 	blobIDToHash map[string]string,
-) (*restorePlan, error) {
+) (*restorePlan, map[types.FileID]error, error) {
 	p := &restorePlan{
 		fileBlobs: make(map[types.FileID]map[string]struct{}, len(files)),
 		blobFiles: make(map[string]map[types.FileID]struct{}),
 		ready:     make([]types.FileID, 0, len(files)),
 		cached:    make(map[string]struct{}),
 	}
+	unplanned := make(map[types.FileID]error)
+
+nextFile:
 	for _, f := range files {
 		if f.IsSymlink() || f.Mode&uint32(os.ModeDir) != 0 {
 			// No chunks to fetch — restore can run immediately.
@@ -66,7 +72,7 @@ func newRestorePlan(
 
 		fileChunks, err := repos.FileChunks.GetByFileID(ctx, f.ID)
 		if err != nil {
-			return nil, fmt.Errorf("planning %s: %w", f.Path, err)
+			return nil, nil, fmt.Errorf("planning %s: %w", f.Path, err)
 		}
 
 		blobs := make(map[string]struct{})
@@ -74,14 +80,18 @@ func newRestorePlan(
 		for _, fc := range fileChunks {
 			bc, ok := chunkToBlobMap[fc.ChunkHash.String()]
 			if !ok {
-				return nil, fmt.Errorf("planning %s: %w: %s",
-					f.Path, errPlanChunkMissing, shortHash(fc.ChunkHash.String()))
+				unplanned[f.ID] = fmt.Errorf("%w: %s",
+					errPlanChunkMissing, shortHash(fc.ChunkHash.String()))
+
+				continue nextFile
 			}
 
 			hash, ok := blobIDToHash[bc.BlobID.String()]
 			if !ok {
-				return nil, fmt.Errorf("planning %s: %w: %s",
-					f.Path, errPlanBlobIDMissing, bc.BlobID)
+				unplanned[f.ID] = fmt.Errorf("%w: %s",
+					errPlanBlobIDMissing, bc.BlobID)
+
+				continue nextFile
 			}
 
 			blobs[hash] = struct{}{}
@@ -103,7 +113,7 @@ func newRestorePlan(
 		}
 	}
 
-	return p, nil
+	return p, unplanned, nil
 }
 
 // markBlobCached records that the named blob is now resident in the
@@ -228,6 +238,12 @@ func (p *restorePlan) filesReferencingBlob(blobHash string) []types.FileID {
 	return out
 }
 
+// pendingFileUsesBlob reports whether any pending file references the
+// named blob, whether or not the blob is cached.
+func (p *restorePlan) pendingFileUsesBlob(blobHash string) bool {
+	return len(p.blobFiles[blobHash]) > 0
+}
+
 // hasPending reports whether any unfinished files remain.
 func (p *restorePlan) hasPending() bool {
 	return len(p.fileBlobs) > 0

internal/vaultik/restore_plan_test.go

@@ -63,15 +63,6 @@ func TestRunRestoreLoopFailsOnAbandonedFiles(t *testing.T) {
 
 	t.Cleanup(func() { _ = cache.Close() })
 
-	v := &Vaultik{ctx: ctx}
-	session := &restoreSession{
-		v:       v,
-		ctx:     ctx,
-		repos:   repos,
-		sweeper: newRestoreSweeper(ctx, repos, cache, 1),
-		result:  &RestoreResult{},
-	}
-
 	// A file that is still pending but whose uncached-blob set is empty
 	// and which was never queued as ready: the loop can neither restore
 	// nor download it. This is the abandonment the guard must catch.
@@ -83,6 +74,15 @@ func TestRunRestoreLoopFailsOnAbandonedFiles(t *testing.T) {
 		cached:    map[string]struct{}{},
 	}
 
+	v := &Vaultik{ctx: ctx}
+	session := &restoreSession{
+		v:       v,
+		ctx:     ctx,
+		repos:   repos,
+		sweeper: newRestoreSweeper(plan, cache, 1),
+		result:  &RestoreResult{},
+	}
+
 	err = v.runRestoreLoop(session, plan, map[types.FileID]*database.File{}, 0)
 	require.ErrorIs(t, err, errRestoreIncomplete)
 }

internal/vaultik/restore_skip_errors_test.go

@@ -4,7 +4,9 @@ import (
 	"bytes"
 	"context"
 	"fmt"
+	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 
 	"github.com/spf13/afero"
@@ -28,7 +30,8 @@ const (
 )
 
 // missingBlobBackup is a snapshot of single-chunk files spread over two
-// blobs, with one of those blobs deleted from the store.
+// blobs, with one of those blobs lost: deleted from the store, or missing
+// from the snapshot metadata.
 type missingBlobBackup struct {
 	fs         afero.Fs
 	cfg        *config.Config
@@ -37,20 +40,39 @@ type missingBlobBackup struct {
 	restoreDir string
 	// files holds the original content by source path.
 	files map[string][]byte
-	// lost holds the source paths whose chunk is in the deleted blob.
+	// lost holds the source paths whose chunk is in the lost blob.
 	lost map[string]bool
 }
 
 // TestRestoreSkipErrorsSkipsFilesOfMissingBlob restores with SkipErrors
-// after one blob of a two-blob snapshot was deleted. Every file stored in
-// that blob must be reported as failed and left absent, every other file
-// must be restored intact, and Restore must still return an error.
+// after one blob of a two-blob snapshot was deleted from the store.
 func TestRestoreSkipErrorsSkipsFilesOfMissingBlob(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
 
 	ctx := context.Background()
-	backup := backupThenDeleteOneBlob(ctx, t)
+	assertRestoreSkipsLostFiles(ctx, t, backupThenDeleteOneBlob(ctx, t))
+}
+
+// TestRestoreSkipErrorsSkipsFilesWithChunkInNoBlob restores with
+// SkipErrors a snapshot whose metadata lists chunks that no blob holds, as
+// earlier backup defects left behind.
+func TestRestoreSkipErrorsSkipsFilesWithChunkInNoBlob(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	assertRestoreSkipsLostFiles(ctx, t, backupThenDropOneBlobFromMetadata(ctx, t))
+}
+
+// assertRestoreSkipsLostFiles restores backup with SkipErrors. Every file
+// stored in the lost blob must be reported as failed and left absent,
+// every other file must be restored intact, and Restore must still return
+// an error.
+func assertRestoreSkipsLostFiles(
+	ctx context.Context, t *testing.T, backup *missingBlobBackup,
+) {
+	t.Helper()
 
 	var out bytes.Buffer
 
@@ -72,20 +94,145 @@ func TestRestoreSkipErrorsSkipsFilesOfMissingBlob(t *testing.T) {
 
 		if backup.lost[path] {
 			assert.Containsf(t, out.String(), path,
-				"%s needs the deleted blob and must be reported", path)
+				"%s needs the lost blob and must be reported", path)
 			assert.NoFileExists(t, restored)
 
 			continue
 		}
 
 		got, err := afero.ReadFile(backup.fs, restored)
-		require.NoErrorf(t, err, "%s does not need the deleted blob", path)
+		require.NoErrorf(t, err, "%s does not need the lost blob", path)
 		assert.Equalf(t, content, got, "%s restored with wrong content", path)
 	}
 }
 
+// TestRestoreSkipErrorsVerifyChecksOnlyRestoredFiles restores with
+// SkipErrors and Verify after one blob of a two-blob snapshot was
+// deleted. Verification must check only the files that were written, so
+// a file of the deleted blob is reported as failed to restore and not
+// also as failing verification, and Restore must still return an error.
+func TestRestoreSkipErrorsVerifyChecksOnlyRestoredFiles(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	backup := backupThenDeleteOneBlob(ctx, t)
+
+	var out bytes.Buffer
+
+	v := newReaderVaultik(ctx, backup.cfg, backup.storer, nil, backup.fs)
+	v.UI = ui.NewWithColor(&out, false)
+
+	err := v.Restore(&vaultik.RestoreOptions{
+		SnapshotID: backup.snapshotID,
+		TargetDir:  backup.restoreDir,
+		Verify:     true,
+		SkipErrors: true,
+	})
+
+	require.Error(t, err, "restore must fail when files were skipped")
+	assert.Contains(t, err.Error(),
+		fmt.Sprintf("%d file(s) failed to restore", len(backup.lost)))
+
+	written := len(backup.files) - len(backup.lost)
+
+	assert.Contains(t, out.String(), fmt.Sprintf("Verifying %d files", written))
+	assert.Contains(t, out.String(), fmt.Sprintf("Verified %d files", written))
+	assert.NotContains(t, out.String(), "Verification failed")
+}
+
+// TestRestoreSkipErrorsVerifyReportsFailuresApart restores with
+// SkipErrors and Verify after one blob of a two-blob snapshot was
+// deleted, while one file that does restore is written one byte too long.
+// The restore warning must list only the files of the deleted blob, the
+// verification summary only the long file, each path once, and Restore
+// must return an error.
+func TestRestoreSkipErrorsVerifyReportsFailuresApart(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	backup := backupThenDeleteOneBlob(ctx, t)
+
+	var long string
+
+	for path := range backup.files {
+		if !backup.lost[path] {
+			long = path
+
+			break
+		}
+	}
+
+	var out bytes.Buffer
+
+	fs := trailingByteFs{Fs: backup.fs, suffix: long}
+	v := newReaderVaultik(ctx, backup.cfg, backup.storer, nil, fs)
+	v.UI = ui.NewWithColor(&out, false)
+
+	err := v.Restore(&vaultik.RestoreOptions{
+		SnapshotID: backup.snapshotID,
+		TargetDir:  backup.restoreDir,
+		Verify:     true,
+		SkipErrors: true,
+	})
+	require.Error(t, err, "restore must fail when files failed")
+
+	_, afterWarning, found := strings.Cut(out.String(),
+		fmt.Sprintf("%d file(s) failed to restore:", len(backup.lost)))
+	require.Truef(t, found, "no restore warning in:\n%s", out.String())
+
+	warning, summary, found := strings.Cut(afterWarning,
+		"Verification failed: 1 files did not match")
+	require.Truef(t, found, "no verification summary in:\n%s", out.String())
+
+	for path := range backup.lost {
+		assert.Equalf(t, 1, strings.Count(warning, path),
+			"%s must be listed once as failed to restore", path)
+		assert.NotContainsf(t, summary, path,
+			"%s was not restored and must not fail verification", path)
+	}
+
+	assert.NotContainsf(t, warning, long,
+		"%s was restored and must not be listed as failed to restore", long)
+	assert.Equalf(t, 1, strings.Count(summary, long),
+		"%s must be listed once as failing verification", long)
+}
+
+// trailingByteFs appends one byte, on close, to the file it opens whose
+// path ends in suffix, so that file is restored one byte too long and
+// fails verification.
+type trailingByteFs struct {
+	afero.Fs
+
+	suffix string
+}
+
+//nolint:ireturn // afero.Fs.OpenFile is defined to return the interface
+func (fs trailingByteFs) OpenFile(
+	name string, flag int, perm os.FileMode,
+) (afero.File, error) {
+	f, err := fs.Fs.OpenFile(name, flag, perm)
+	if err != nil || !strings.HasSuffix(name, fs.suffix) {
+		return f, err
+	}
+
+	return trailingByteFile{File: f}, nil
+}
+
+type trailingByteFile struct {
+	afero.File
+}
+
+func (f trailingByteFile) Close() error {
+	_, _ = f.Write([]byte{'x'})
+
+	return f.File.Close()
+}
+
 // TestRestoreMissingBlobAbortsWithoutSkipErrors checks that a deleted blob
-// still ends the restore with an error when SkipErrors is not set.
+// still ends the restore with an error when SkipErrors is not set, and
+// that the error names the blob once.
 func TestRestoreMissingBlobAbortsWithoutSkipErrors(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -101,6 +248,32 @@ func TestRestoreMissingBlobAbortsWithoutSkipErrors(t *testing.T) {
 	})
 
 	require.ErrorIs(t, err, storage.ErrNotFound)
+	assert.Equal(t, 1, strings.Count(err.Error(), "downloading blob"),
+		err.Error())
+}
+
+// TestRestoreChunkInNoBlobAbortsWithoutSkipErrors checks that a chunk held
+// by no blob still ends the restore, before any file is written, when
+// SkipErrors is not set.
+func TestRestoreChunkInNoBlobAbortsWithoutSkipErrors(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	backup := backupThenDropOneBlobFromMetadata(ctx, t)
+
+	v := newReaderVaultik(ctx, backup.cfg, backup.storer, nil, backup.fs)
+
+	err := v.Restore(&vaultik.RestoreOptions{
+		SnapshotID: backup.snapshotID,
+		TargetDir:  backup.restoreDir,
+	})
+
+	require.ErrorContains(t, err, "chunk missing from blob map")
+
+	for path := range backup.files {
+		assert.NoFileExists(t, filepath.Join(backup.restoreDir, path))
+	}
 }
 
 // backupThenDeleteOneBlob backs up missingBlobFileCount single-chunk
@@ -110,6 +283,54 @@ func backupThenDeleteOneBlob(
 ) *missingBlobBackup {
 	t.Helper()
 
+	backup, db, lostBlob := backupSingleChunkFiles(ctx, t)
+	require.NoError(t, db.Close())
+
+	require.NoError(t, backup.storer.Delete(ctx, fmt.Sprintf(
+		"blobs/%s/%s/%s", lostBlob[:2], lostBlob[2:4], lostBlob)))
+
+	return backup
+}
+
+// backupThenDropOneBlobFromMetadata backs up missingBlobFileCount
+// single-chunk files, then removes the blob holding the first of them from
+// the snapshot and exports the snapshot metadata again. The exported
+// metadata then lists chunks that no blob holds.
+func backupThenDropOneBlobFromMetadata(
+	ctx context.Context, t *testing.T,
+) *missingBlobBackup {
+	t.Helper()
+
+	backup, db, lostBlob := backupSingleChunkFiles(ctx, t)
+
+	// The uploads row refers to the blob row, so it would keep the export
+	// from deleting it.
+	for _, query := range []string{
+		"DELETE FROM snapshot_blobs WHERE blob_hash = ?",
+		"DELETE FROM uploads WHERE blob_hash = ?",
+	} {
+		_, err := db.Conn().ExecContext(ctx, query, lostBlob)
+		require.NoError(t, err)
+	}
+
+	sm := newFaultSnapshotManager(
+		backup.fs, backup.storer, backup.cfg, database.NewRepositories(db))
+	require.NoError(t,
+		sm.ExportSnapshotMetadata(ctx, db.Path(), backup.snapshotID))
+	require.NoError(t, db.Close())
+
+	return backup
+}
+
+// backupSingleChunkFiles backs up missingBlobFileCount single-chunk files.
+// It returns the backup, with lost naming the files stored in the same
+// blob as the first of them, the still open local index, and the hash of
+// that blob.
+func backupSingleChunkFiles(
+	ctx context.Context, t *testing.T,
+) (*missingBlobBackup, *database.DB, string) {
+	t.Helper()
+
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
@@ -127,7 +348,7 @@ func backupThenDeleteOneBlob(
 		require.NoError(t, afero.WriteFile(fs, path, files[path], 0o644))
 	}
 
-	storer, err := storage.NewFileStorer(filepath.Join(tempDir, "remote"))
+	storer, err := storage.NewFileStorer(t.TempDir())
 	require.NoError(t, err)
 
 	db, err := database.New(ctx, dbPath)
@@ -143,16 +364,11 @@ func backupThenDeleteOneBlob(
 		blobOfFile[path] = blobHashOfSingleChunkFile(ctx, t, repos, path)
 	}
 
-	require.NoError(t, db.Close())
-
-	deleted := blobOfFile[filepath.Join(dataDir, "file-00.bin")]
-	require.NoError(t, storer.Delete(ctx, fmt.Sprintf(
-		"blobs/%s/%s/%s", deleted[:2], deleted[2:4], deleted)))
-
+	lostBlob := blobOfFile[filepath.Join(dataDir, "file-00.bin")]
 	lost := make(map[string]bool)
 
 	for path, hash := range blobOfFile {
-		if hash == deleted {
+		if hash == lostBlob {
 			lost[path] = true
 		}
 	}
@@ -168,7 +384,7 @@ func backupThenDeleteOneBlob(
 		restoreDir: filepath.Join(tempDir, "restored"),
 		files:      files,
 		lost:       lost,
-	}
+	}, db, lostBlob
 }
 
 // blobHashOfSingleChunkFile returns the hash of the blob holding the one

internal/vaultik/restore_snapshot_identity_test.go

@@ -29,7 +29,7 @@ func TestRestoreAndDeepVerifyRejectSwappedDatabase(t *testing.T) {
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 
 	chunkSize := int64(64 * 1024)
 
@@ -100,7 +100,7 @@ func TestDeepVerifyRejectsSwappedDatabaseWithEmptyManifest(t *testing.T) {
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 
 	chunkSize := int64(64 * 1024)
 

internal/vaultik/restore_sweeper.go

@@ -1,46 +1,30 @@
 package vaultik
 
-import (
-	"context"
-	"fmt"
-
-	"sneak.berlin/go/vaultik/internal/database"
-	"sneak.berlin/go/vaultik/internal/log"
-)
-
-// restoreSweeper frees cached blobs once all files that reference any of
-// their chunks have been restored. It works as follows:
+// restoreSweeper frees cached blobs once no file still to be restored in
+// this run needs them. It works as follows:
 //
-//  1. Callers add a file's ID to an in-memory restored set via
-//     fileRestored once the file is fully written to disk.
-//  2. After each chunk is restored, chunkRestored accumulates a running
+//  1. After each chunk is restored, chunkRestored accumulates a running
 //     byte count.
-//  3. When the accumulator crosses a threshold (one hundredth of the
+//  2. When the accumulator crosses a threshold (one hundredth of the
 //     configured blob size — so a sweep runs about a hundred times per
 //     blob's worth of restored bytes), the sweeper iterates every key in
-//     the cache. For each cached blob it asks the DB which files
-//     reference any chunk in that blob, then compares that list against
-//     the in-memory restored set. If any referencing file is missing
-//     from the set the blob is kept; otherwise the cache entry is
-//     deleted.
+//     the cache and deletes each blob that the restore plan no longer
+//     needs.
 //
-// All DB reads happen against the snapshot's temporary metadata DB,
-// which is local, indexed, and not under contention — the queries are
-// cheap and run at most once per blob per sweep interval.
+// The plan drops a file once it is restored or has failed, and holds
+// only the files this run restores, so files outside the requested paths
+// never keep a blob.
 type restoreSweeper struct {
-	ctx        context.Context //nolint:containedctx // ctx bound at construction by design
-	repos      *database.Repositories
+	plan       *restorePlan
 	cache      *blobDiskCache
 	threshold  int64
 	bytesAccum int64
-	restored   map[string]struct{}
 }
 
 // newRestoreSweeper returns a sweeper that triggers eviction every
 // `threshold` bytes restored. Callers should pass blob_size_limit/100.
 func newRestoreSweeper(
-	ctx context.Context,
-	repos *database.Repositories,
+	plan *restorePlan,
 	cache *blobDiskCache,
 	threshold int64,
 ) *restoreSweeper {
@@ -49,21 +33,12 @@ func newRestoreSweeper(
 	}
 
 	return &restoreSweeper{
-		ctx:       ctx,
-		repos:     repos,
+		plan:      plan,
 		cache:     cache,
 		threshold: threshold,
-		restored:  make(map[string]struct{}),
 	}
 }
 
-// fileRestored records a file as fully restored. After this call, any
-// blob whose only remaining references come from files in the restored
-// set may be evicted on the next sweep.
-func (s *restoreSweeper) fileRestored(fileID string) {
-	s.restored[fileID] = struct{}{}
-}
-
 // chunkRestored accounts n bytes against the sweep threshold and runs a
 // sweep if the threshold has been crossed since the last sweep.
 func (s *restoreSweeper) chunkRestored(n int64) {
@@ -76,60 +51,11 @@ func (s *restoreSweeper) chunkRestored(n int64) {
 	s.sweep()
 }
 
-// sweep deletes any cached blob whose chunks are no longer referenced
-// by an unrestored file. Per-blob DB failures are logged and the blob
-// is kept — we'd rather hold a blob longer than risk a re-download.
+// sweep deletes every cached blob that no pending file references.
 func (s *restoreSweeper) sweep() {
 	for _, blobHash := range s.cache.Keys() {
-		needed, err := s.blobStillNeeded(blobHash)
-		if err != nil {
-			log.Debug("sweeper referencing-files query failed",
-				"blob_hash", blobHash[:16], "error", err)
-
-			continue
-		}
-
-		if !needed {
+		if !s.plan.pendingFileUsesBlob(blobHash) {
 			s.cache.Delete(blobHash)
 		}
 	}
 }
-
-// blobStillNeeded returns true if any file that references a chunk in
-// this blob has not yet been restored. On any error the function
-// returns true — keeping the blob is always the safe answer because we
-// can't prove we're done with it.
-func (s *restoreSweeper) blobStillNeeded(blobHash string) (bool, error) {
-	rows, err := s.repos.DB().Conn().QueryContext(s.ctx, `
-		SELECT DISTINCT fc.file_id
-		FROM file_chunks fc
-		JOIN blob_chunks bc ON bc.chunk_hash = fc.chunk_hash
-		JOIN blobs b ON b.id = bc.blob_id
-		WHERE b.blob_hash = ?
-	`, blobHash)
-	if err != nil {
-		return true, fmt.Errorf("querying referencing files: %w", err)
-	}
-
-	defer func() { _ = rows.Close() }()
-
-	for rows.Next() {
-		var fileID string
-
-		err := rows.Scan(&fileID)
-		if err != nil {
-			return true, fmt.Errorf("scanning file_id: %w", err)
-		}
-
-		if _, ok := s.restored[fileID]; !ok {
-			return true, nil
-		}
-	}
-
-	err = rows.Err()
-	if err != nil {
-		return true, err
-	}
-
-	return false, nil
-}

internal/vaultik/restore_sweeper_integration_test.go

@@ -186,7 +186,7 @@ func TestRestoreSweeperEvictsBlobs(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 

internal/vaultik/restore_sweeper_test.go

@@ -0,0 +1,163 @@
+package vaultik //nolint:testpackage // reads the restore blob cache's peak size
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"path/filepath"
+	"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/storage"
+	"sneak.berlin/go/vaultik/internal/ui"
+)
+
+// TestRestoreFreesBlobsNoPendingFileNeeds asserts that the restore cache
+// frees a blob once no file still to be restored in this run needs it.
+// Every original file has a copy with the same content, so every blob
+// holds chunks of copies: a restore that leaves the copies out of its
+// paths, or fails on them, must not keep the blobs for them.
+func TestRestoreFreesBlobsNoPendingFileNeeds(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	// Small files keep the backup and both restores quick.
+	const fileBytes = 16 * 1024
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+
+	dataDir := filepath.Join(tempDir, "source")
+	originalsDir := filepath.Join(dataDir, "originals")
+	copiesDir := filepath.Join(dataDir, "copies")
+	storeDir := t.TempDir()
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	originals := writeFilesWithCopies(t, fs, originalsDir, copiesDir, fileBytes)
+
+	ctx := context.Background()
+
+	cfg, storer, snapshotID := setupLocalityBackup(
+		ctx, t, fs, dataDir, storeDir, dbPath, fileBytes)
+
+	t.Run("restore by path", func(t *testing.T) {
+		t.Parallel()
+
+		restoreDir := filepath.Join(tempDir, "by-path")
+
+		peak, err := restorePeakCachedBlobs(ctx, t, fs, cfg, storer,
+			&RestoreOptions{
+				SnapshotID: snapshotID,
+				TargetDir:  restoreDir,
+				Paths:      []string{originalsDir},
+			})
+		require.NoError(t, err)
+
+		requireRestoredFiles(t, fs, restoreDir, originals)
+		assert.LessOrEqualf(t, peak, 1,
+			"peak cached blobs was %d; blobs were kept for unselected files",
+			peak)
+	})
+
+	t.Run("restore with failed files", func(t *testing.T) {
+		t.Parallel()
+
+		restoreDir := filepath.Join(tempDir, "with-failures")
+
+		// A regular file where the copies directory belongs makes every
+		// copy fail to restore.
+		blocker := filepath.Join(restoreDir, copiesDir)
+		require.NoError(t, fs.MkdirAll(filepath.Dir(blocker), 0o755))
+		require.NoError(t, afero.WriteFile(fs, blocker, nil, 0o644))
+
+		peak, err := restorePeakCachedBlobs(ctx, t, fs, cfg, storer,
+			&RestoreOptions{
+				SnapshotID: snapshotID,
+				TargetDir:  restoreDir,
+				SkipErrors: true,
+			})
+		require.ErrorIs(t, err, errFilesFailedRestore)
+
+		requireRestoredFiles(t, fs, restoreDir, originals)
+		assert.LessOrEqualf(t, peak, 1,
+			"peak cached blobs was %d; blobs were kept for failed files", peak)
+	})
+}
+
+// writeFilesWithCopies writes 15 random files of fileBytes into
+// originalsDir and a copy of each into copiesDir. With
+// setupLocalityBackup given the same size, each file is one chunk and the
+// 15 chunks fill 3 blobs. Returns the originals' content by path.
+func writeFilesWithCopies(
+	t *testing.T, fs afero.Fs, originalsDir, copiesDir string, fileBytes int,
+) map[string][]byte {
+	t.Helper()
+
+	const fileCount = 15
+
+	require.NoError(t, fs.MkdirAll(originalsDir, 0o755))
+	require.NoError(t, fs.MkdirAll(copiesDir, 0o755))
+
+	originals := make(map[string][]byte, fileCount)
+
+	for i := range fileCount {
+		name := fmt.Sprintf("file-%02d.bin", i)
+		data := randomBytes(t, fileBytes)
+		original := filepath.Join(originalsDir, name)
+		originals[original] = data
+
+		require.NoError(t, afero.WriteFile(fs, original, data, 0o644))
+		require.NoError(t, afero.WriteFile(
+			fs, filepath.Join(copiesDir, name), data, 0o644))
+	}
+
+	return originals
+}
+
+// requireRestoredFiles requires every file in want to have been restored
+// under restoreDir with its content intact.
+func requireRestoredFiles(
+	t *testing.T, fs afero.Fs, restoreDir string, want map[string][]byte,
+) {
+	t.Helper()
+
+	for path, data := range want {
+		got, err := afero.ReadFile(fs, filepath.Join(restoreDir, path))
+		require.NoErrorf(t, err, "file missing after restore: %s", path)
+		require.Equalf(t, data, got, "byte mismatch for %s", path)
+	}
+}
+
+// restorePeakCachedBlobs runs a restore and returns the most blobs its
+// cache held at once, together with the restore's error.
+func restorePeakCachedBlobs(
+	ctx context.Context, t *testing.T, fs afero.Fs,
+	cfg *config.Config, storer storage.Storer, opts *RestoreOptions,
+) (int, error) {
+	t.Helper()
+
+	var cache *blobDiskCache
+
+	v := &Vaultik{
+		Config:  cfg,
+		Storage: storer,
+		Fs:      fs,
+		Stdout:  io.Discard,
+		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
+		restoreCacheObserver: func(c *blobDiskCache) {
+			cache = c
+		},
+	}
+	v.SetContext(ctx)
+
+	err := v.Restore(opts)
+
+	require.NotNil(t, cache, "restoreCacheObserver must fire during restore")
+
+	return cache.PeakLen(), err
+}

internal/vaultik/restore_truncated_db_test.go

@@ -33,7 +33,7 @@ func TestRestoreRejectsTruncatedMetadataDB(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")