Warn about restored ownership only when an entry's owner or group differs (closes #316)

A restore by a normal user always warned that file ownership was not
preserved, even when every entry came out with its recorded owner and
group. The owner and group each restored entry has on disk are now
compared with the recorded ones, and the warning is printed once if any
differ. The comparison is made on disk because on macOS, and in a setgid
directory on Linux, a new entry takes the group of its directory, not
the user's.

Judgement call: an entry that is not on disk after the restore is not compared.
Unverified: not run on a Mac; the directory-group test runs only as root.

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
+1. **Phase 1**: Walk directory, identify files needing processing
-2. **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
@@ -320,13 +320,13 @@ 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()

README.md

@@ -271,8 +271,17 @@ 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
+formatting in the file are preserved. Missing keys along the path are
-needed.
+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)
@@ -314,7 +323,11 @@ 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
@@ -323,9 +336,11 @@ local index alone, and still exits zero.
   (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
+  listed). Warnings about an unlistable destination, a local index
-  manifests, and a truncated listing all go to stderr through the
+  bound to a different destination (with both destinations and the
-  logger, so stdout stays a single parseable document.
+  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
@@ -351,16 +366,20 @@ 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
+`vaultik prune` does. The command prints the `vaultik prune` invocation
-prune` invocation to run as a follow-up. An ID that names no snapshot in
+to run as a follow-up only when it removed the snapshot's metadata from
-the local index or on the destination store is an error; with
+the destination store, and never under `--json`. An ID that names no
-`--local-only`, the snapshot must be in the local index. Local row
+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
+runs automatically. If removing the snapshot's metadata from the
-removal still completes and a warning is emitted; run `vaultik snapshot
+destination store fails (for example because the store is unreachable),
-remove <snapshot-id>` again once the store is reachable to remove the
+the local-DB removal still completes and a warning is emitted; run
-snapshot's metadata from it (`vaultik prune` does not). To wipe everything
+`vaultik snapshot remove <snapshot-id>` again once the store is reachable
-on the destination in one go, use `vaultik remote nuke --force`.
+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
@@ -373,7 +392,9 @@ 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 restored file's chunk hashes
@@ -413,11 +434,12 @@ 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
+**`remote nuke`**: Delete every object under `metadata/` and `blobs/` on
-backup destination store, leaving the bucket prefix empty. Destructive and
+the backup destination store, whatever its name, leaving the bucket prefix
-irreversible. This is the single supported way to wipe the entire
+empty. Destructive and irreversible. This is the single supported way to
-destination store. If anything cannot be deleted, the command exits
+wipe the entire destination store. If a listing afterwards still finds
-non-zero; running it again deletes what is left.
+anything there, 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
@@ -700,8 +722,10 @@ 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
+  confirmation prompts of `database delete`, `prune`, `snapshot purge`
-  always shown: it is an interactive exchange the operator must see.
+  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

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
 

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/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
+// Users care about the real error, not the DI plumbing. dig.RootCause,
-// up through the last "): " (which is always the close-paren of an fx
+// from the library fx is built on, unwraps those layers and returns the
-// function-location annotation followed by the wrapped error).
+// 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()
+	return &startupError{msg: dig.RootCause(err).Error()}
-	if idx := strings.LastIndex(msg, "): "); idx >= 0 {
-		msg = msg[idx+3:]
-	}
-
-	return &startupError{msg: msg}
 }
 
 // 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 {
+	path := filepath.Join(t.TempDir(), "config.yml")
-		name string
+
-		in   string
+	err := os.WriteFile(path, []byte("snapshots:\n"+
-		want string
+		"  test:\n"+
-	}{
+		"    paths: [/tmp/vaultik-test-source]\n"+
-		{
+		"storage_url: file:///tmp/vaultik-test-storage\n"+
-			name: "real fx error chain",
+		"chunk_size: 10MB\n"+
-			in: `could not build arguments for function ` +
+		"blob_size_limit: 30MB\n"), 0o600)
-				`"sneak.berlin/go/vaultik/internal/cli".newSnapshotCreateCommand.func1.1 ` +
+	if err != nil {
-				`(/Users/user/dev/vaultik/internal/cli/snapshot.go:71): ` +
+		t.Fatalf("write config: %v", err)
-				`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",
-		},
 	}
 
-	for _, tt := range tests {
+	app := fx.New(
-		t.Run(tt.name, func(t *testing.T) {
+		fx.NopLogger,
-			t.Parallel()
+		fx.Supply(config.Path(path)),
+		config.Module,
+		fx.Invoke(func(*config.Config) {}),
+	)
 
-			//nolint:err113 // test constructs errors from table input
+	err = app.Start(t.Context())
-			got := cleanStartupError(errors.New(tt.in)).Error()
+	if err == nil {
-			if got != tt.want {
+		t.Fatal("app started; want a blob_size_limit error")
-				t.Errorf("got %q, want %q", got, tt.want)
+	}
-			}
+
-		})
+	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
 
@@ -329,8 +333,15 @@ 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
@@ -391,8 +402,17 @@ 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
+the file back, preserving comments and formatting. Missing keys along
-are created as needed.
+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"
@@ -422,7 +442,14 @@ 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
 	}
@@ -432,6 +459,21 @@ 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)
@@ -453,6 +495,58 @@ 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
@@ -559,8 +653,8 @@ func yamlPathGet(root *yaml.Node, keys []string) (*yaml.Node, error) {
 	return node, nil
 }
 
-// yamlPathSet navigates a dotted key path, creating intermediate maps as
+// yamlPathSet navigates a dotted key path, creating missing keys as needed,
-// needed, and sets the final key to the given scalar value. Numeric path
+// 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 {
@@ -573,12 +667,12 @@ 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 {
@@ -647,14 +741,16 @@ func yamlField(typ reflect.Type, key string) (reflect.StructField, bool) {
 	return reflect.StructField{}, false
 }
 
-// yamlSetInMapping resolves (creating if needed) the value node for key
+// yamlSetInMapping resolves (creating if needed) the value node for the
-// within a mapping node, setting it to value when it is the final path
+// path element keys[i] within a mapping node, setting it to value when it
-// element, and returns the node to descend into.
+// is the final path element, and returns the node to descend into.
-func yamlSetInMapping(node *yaml.Node, key, value string, last bool) *yaml.Node {
+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
@@ -662,13 +758,8 @@ func yamlSetInMapping(node *yaml.Node, key, value string, last bool) *yaml.Node
 	}
 
 	if valueNode == nil {
-		keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: key}
+		keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: keys[i]}
-
+		valueNode = newPathNode(keys, i, value)
-		valueNode = &yaml.Node{Kind: yaml.MappingNode}
-		if last {
-			valueNode = &yaml.Node{Kind: yaml.ScalarNode, Value: value}
-		}
-
 		node.Content = append(node.Content, keyNode, valueNode)
 	} else if last {
 		setScalar(valueNode, value)
@@ -696,12 +787,7 @@ func yamlSetInSequence(
 	}
 
 	if idx == len(node.Content) {
-		newNode := &yaml.Node{Kind: yaml.MappingNode}
+		node.Content = append(node.Content, newPathNode(keys, i, value))
-		if last {
-			newNode = &yaml.Node{Kind: yaml.ScalarNode, Value: value}
-		}
-
-		node.Content = append(node.Content, newNode)
 	} else if last {
 		setScalar(node.Content[idx], value)
 	}
@@ -709,6 +795,22 @@ 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 {
+	if len(cfg.AgeRecipients) != 1 || cfg.AgeRecipients[0] != testRecipient {
-		t.Errorf("age_recipients = %v, want [%s]", cfg.AgeRecipients, recipient)
+		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/entry_prune_quiet_test.go

@@ -0,0 +1,123 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"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, 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))
+
+	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_purge_test.go

@@ -0,0 +1,68 @@
+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)
+
+	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, _, _ := runEntry(t, flagConfig, configPath,
+		cmdSnapshot, cmdPurge, flagOlderThan, "30d", flagForce)
+
+	assert.Equal(t, 1, code)
+	assert.FileExists(t, manifestPath,
+		"the snapshot is still in the local index, so its metadata must stay")
+}

internal/cli/snapshot.go

@@ -270,8 +270,9 @@ 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
+know about), which is what 'vaultik prune' does. The command prints the
-prints the exact 'vaultik prune' invocation to run as a follow-up.
+'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.
@@ -280,10 +281,12 @@ 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
+If removing the snapshot's metadata from the destination store fails (for
-and a warning is emitted; run 'vaultik snapshot remove <snapshot-id>' again
+example because the remote is unreachable), the local-database removal
-once the destination store is reachable to remove the snapshot's metadata
+still completes and a warning is emitted; run 'vaultik snapshot remove
-from it ('vaultik prune' does not).
+<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

@@ -4,6 +4,7 @@
 package config
 
 import (
+	"bytes"
 	"errors"
 	"fmt"
 	"os"
@@ -29,13 +30,15 @@ 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
+	defaultBlobSizeLimit    = Size(10 * 1000 * 1000 * 1000) // 10GB
-	defaultChunkSize        = Size(10 * 1024 * 1024)        // 10MB
+	defaultChunkSize        = Size(10 * 1000 * 1000)        // 10MB
 	defaultS3PartSize       = Size(5 * 1024 * 1024)         // 5MiB
 	defaultCompressionLevel = 3
-	minChunkSize            = 1024 * 1024 // 1MB
+	minChunkSize            = 1000 * 1000 // 1MB
 	minCompressionLevel     = 1
 	maxCompressionLevel     = 19
 	// S3 accepts a multipart upload part from 5MiB to 5GiB.
@@ -233,14 +236,48 @@ func New(path Path) (*Config, error) {
 	return cfg, nil
 }
 
-// Load reads and parses the configuration file from the specified path.
+// Load reads the configuration file at path, parses it with Parse, warns
-// It applies default values for optional fields, performs environment variable
+// when others can read the file, and validates the configuration.
-// substitution using smartconfig, and validates the configuration.
+// Returns an error if the file cannot be read, parsed, or if validation
-// The configuration file should be in YAML format. Returns an error if the file
+// fails.
-// 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)
 	}
@@ -302,24 +339,6 @@ 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
 }
 

internal/config/config_test.go

@@ -337,6 +337,81 @@ 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)
+			}
+		})
+	}
+}
+
 // 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

internal/database/files.go

@@ -267,6 +267,45 @@ func (r *FileRepository) ListUnderPath(
 	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
+
+	for rows.Next() {
+		file, err := r.scanFileRows(rows)
+		if err != nil {
+			return nil, fmt.Errorf("scanning file: %w", err)
+		}
+
+		files = append(files, file)
+	}
+
+	return files, rows.Err()
+}
+
 // ListIDsWithChunksNotInUploadedBlobs returns the IDs of the files whose
 // path starts with prefix and that list at least one chunk held by no
 // blob whose upload has completed (uploaded_ts set). A new snapshot

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/snapshot/exclude_test.go

@@ -578,3 +578,54 @@ func TestExcludePatterns_DoubleStarFileAtEveryDepth(t *testing.T) {
 	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,
+			FS:                 params.Fs,
-			ChunkSize:        cfg.ChunkSize.Int64(),
+			ChunkSize:          cfg.ChunkSize.Int64(),
-			Repositories:     repos,
+			Repositories:       repos,
-			Storage:          storer,
+			Storage:            storer,
-			MaxBlobSize:      cfg.BlobSizeLimit.Int64(),
+			MaxBlobSize:        cfg.BlobSizeLimit.Int64(),
-			CompressionLevel: cfg.CompressionLevel,
+			CompressionLevel:   cfg.CompressionLevel,
-			AgeRecipients:    cfg.AgeRecipients,
+			AgeRecipients:      cfg.AgeRecipients,
-			EnableProgress:   params.EnableProgress,
+			EnableProgress:     params.EnableProgress,
-			UI:               params.UI,
+			UI:                 params.UI,
-			Exclude:          excludes,
+			Exclude:            excludes,
-			SkipErrors:       params.SkipErrors,
+			SkipErrors:         params.SkipErrors,
+			PreviousSnapshotID: params.PreviousSnapshotID,
 		})
 	}
 }

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
@@ -129,6 +133,9 @@ 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
@@ -207,6 +214,7 @@ func NewScanner(cfg ScannerConfig) *Scanner {
 		compiledExclude:    compiledExclude,
 		progress:           progress,
 		skipErrors:         cfg.SkipErrors,
+		previousSnapshotID: cfg.PreviousSnapshotID,
 		ui:                 uiw,
 		pendingChunkHashes: make(map[string]struct{}),
 	}
@@ -258,8 +266,9 @@ func (s *Scanner) Scan(
 
 	filesToProcess := scanResult.FilesToProcess
 
-	// Phase 1b: Detect deleted files by comparing DB against scanned files
+	// Phase 1b: Detect deleted files by comparing the previous snapshot
-	err = s.detectDeletedFilesFromMap(ctx, knownFiles, existingFiles, result)
+	// 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)
 	}
@@ -439,9 +448,8 @@ 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
+// database into a map for fast lookup. This avoids per-file database
-// not find is counted as deleted. This avoids per-file database queries
+// queries during the scan phase.
-// during the scan phase.
 func (s *Scanner) loadKnownFiles(
 	ctx context.Context, path string,
 ) (map[string]*database.File, error) {
@@ -1954,21 +1962,29 @@ func (s *Scanner) queueFileForBatchInsert(
 	})
 }
 
-// detectDeletedFilesFromMap finds files that existed in previous snapshots
+// detectDeletedFiles counts as deleted each file at or beneath path that
-// but no longer exist. Uses pre-loaded maps to avoid any filesystem or
+// the previous snapshot of this name lists, the walk did not find, and
-// database access.
+// is gone from disk. A file that is excluded, or inside an excluded
-func (s *Scanner) detectDeletedFilesFromMap(
+// 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)
+	previousFiles, err := s.repos.Files.ListUnderPathInSnapshot(
-	for path, file := range knownFiles {
+		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():
@@ -1976,13 +1992,22 @@ func (s *Scanner) detectDeletedFilesFromMap(
 		default:
 		}
 
-		// Check if the file exists in our enumerated set
+		filePath := file.Path.String()
-		if _, exists := existingFiles[path]; !exists {
+
-			// File has been deleted
+		_, exists := existingFiles[filePath]
-			result.FilesDeleted++
+		if exists || s.excludedOrInExcludedDir(filePath, path) {
-			result.BytesDeleted += file.Size
+			continue
-			log.Debug("Detected deleted file", "path", path, "size", file.Size)
 		}
+
+		// 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 {
@@ -1993,6 +2018,36 @@ 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
@@ -2035,33 +2090,60 @@ func compileExcludePatterns(patterns []string) []compiledPattern {
 		pattern = strings.TrimSuffix(pattern, "/")
 
 		// A leading "**/" demands a "/" before the name and so misses an
-		// entry directly under the source path. An unanchored pattern
+		// entry directly under the source path. It means "at every depth",
-		// already matches at every depth, so drop the prefix: as in
+		// which is what an unanchored pattern already does, so drop the
-		// gitignore, "**/x" then also matches a top-level "x".
+		// prefix and match unanchored: as in gitignore, "**/x" and
-		if !anchored {
+		// "/**/x" then also match a top-level "x".
-			pattern = strings.TrimPrefix(pattern, "**/")
+		if rest, found := strings.CutPrefix(pattern, "**/"); found {
+			pattern = rest
+			anchored = false
 		}
 
-		// Compile the glob pattern
+		// Likewise "/**/" inside a pattern demands two slashes, so
-		// For patterns without path separators, we need to match them as components
+		// "a/**/*.log" misses "a/x.log". As in gitignore it stands for
-		// e.g., ".git" should match ".git" anywhere in the path
+		// zero or more directories, so also compile the pattern with it
-		g, err := glob.Compile(pattern, '/')
+		// made a single "/". The glob library's {a,b} is no help: with a
-		if err != nil {
+		// "**" inside it in mid-pattern, it misses paths it should match.
-			log.Warn("Invalid exclude pattern, skipping", "pattern", p, "error", err)
+		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)
 
-			continue
+				break
+			}
+
+			compiled = append(compiled, compiledPattern{
+				pattern:  g,
+				anchored: anchored,
+				original: p,
+			})
 		}
-
-		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/storage/file.go

@@ -120,9 +120,10 @@ func (f *FileStorer) Delete(_ context.Context, key string) error {
 	return nil
 }
 
-// List returns all keys with the given prefix. It fails when the
+// List returns all keys with the given prefix, reading through symlinked
-// destination directory is missing; a missing prefix under it is an
+// directories. It fails when the destination directory is missing, or
-// empty listing.
+// 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 +134,7 @@ func (f *FileStorer) List(ctx context.Context, prefix string) ([]string, error)
 
 	basePath := f.fullPath(prefix)
 
-	// Check if the prefix exists
+	exists, err := f.prefixExists(prefix)
-	exists, err := afero.Exists(f.fs, basePath)
 	if err != nil {
 		return nil, fmt.Errorf("checking path: %w", err)
 	}
@@ -143,7 +143,7 @@ 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 +176,9 @@ 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 +194,7 @@ func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan Objec
 
 		basePath := f.fullPath(prefix)
 
-		// Check if the prefix exists
+		exists, err := f.prefixExists(prefix)
-		exists, err := afero.Exists(f.fs, basePath)
 		if err != nil {
 			ch <- ObjectInfo{Err: fmt.Errorf("checking path: %w", err)}
 
@@ -204,7 +205,7 @@ 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():
@@ -243,11 +244,12 @@ 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 +258,7 @@ 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
 		}
@@ -352,6 +354,54 @@ 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 +63,232 @@ 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)
+			}
+		})
+	}
+}

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()
 

internal/storage/rclone_test.go

@@ -11,6 +11,7 @@ import (
 	"testing"
 
 	"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 +391,41 @@ 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)
+	}
+}

internal/storage/url.go

@@ -30,6 +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.
@@ -90,8 +96,9 @@ func parseFileURL(rawPath string) (*URL, error) {
 }
 
 // parseS3URL parses an s3://bucket/prefix URL. It rejects credentials in
-// the userinfo and any query parameter other than endpoint, region and
+// the userinfo or in the endpoint parameter's userinfo, a query that
-// ssl, so a credential-bearing URL is never stored or echoed.
+// 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 {
@@ -107,13 +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,
@@ -125,9 +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)
@@ -137,12 +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

@@ -150,6 +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,
@@ -203,6 +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

@@ -104,7 +104,8 @@ 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.
+// 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) {
@@ -119,7 +120,7 @@ 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 newCloseOnCancelReader(ctx, rc), nil

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
+// each blob once no file still to be restored in this run needs it.
-// written. LRU never fires under that configuration; it is kept as a
+// LRU never fires under that configuration; it is kept as a safety net
-// safety net for callers that don't manage eviction themselves.
+// 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/deep_verify_test.go

@@ -6,6 +6,7 @@ import (
 	"crypto/rand"
 	"io"
 	"os"
+	"path"
 	"path/filepath"
 	"testing"
 
@@ -142,7 +143,8 @@ func TestDeepVerifyJSONCountsBlobs(t *testing.T) {
 
 // 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.
+// 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()
@@ -168,11 +170,14 @@ func TestDeepVerifyJSONCountsWrongSizeBlob(t *testing.T) {
 	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 with the storage error.
+// 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()
@@ -187,6 +192,8 @@ func TestDeepVerifyStopsOnBlobThatCannotBeChecked(t *testing.T) {
 	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())
 }
 

internal/vaultik/fault_injection_test.go

@@ -661,17 +661,18 @@ func faultScannerFactory(
 ) snapshot.ScannerFactory {
 	return func(params snapshot.ScannerParams) *snapshot.Scanner {
 		return snapshot.NewScanner(snapshot.ScannerConfig{
-			FS:               params.Fs,
+			FS:                 params.Fs,
-			Storage:          storer,
+			Storage:            storer,
-			ChunkSize:        faultChunkSize,
+			ChunkSize:          faultChunkSize,
-			MaxBlobSize:      faultMaxBlobSize,
+			MaxBlobSize:        faultMaxBlobSize,
-			CompressionLevel: cfg.CompressionLevel,
+			CompressionLevel:   cfg.CompressionLevel,
-			AgeRecipients:    cfg.AgeRecipients,
+			AgeRecipients:      cfg.AgeRecipients,
-			Repositories:     repos,
+			Repositories:       repos,
-			EnableProgress:   params.EnableProgress,
+			EnableProgress:     params.EnableProgress,
-			UI:               params.UI,
+			UI:                 params.UI,
-			Exclude:          params.Exclude,
+			Exclude:            params.Exclude,
-			SkipErrors:       params.SkipErrors,
+			SkipErrors:         params.SkipErrors,
+			PreviousSnapshotID: params.PreviousSnapshotID,
 		})
 	}
 }

internal/vaultik/nuke_remote_test.go

@@ -18,7 +18,8 @@ import (
 // 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()
@@ -38,6 +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 {
@@ -100,6 +104,34 @@ func TestNukeRemoteFailsWhenADeleteFails(t *testing.T) {
 	}
 }
 
+// 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:

internal/vaultik/prune.go

@@ -22,6 +22,11 @@ var errNukeRequiresForce = errors.New(
 // 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.
@@ -35,10 +40,11 @@ const metadataDirName = "metadata"
 // (blobs/<aa>/<bb>/<hash>).
 const blobKeyParts = 4
 
-// NukeRemote deletes every snapshot's metadata and every blob from remote
+// NukeRemote deletes every object under metadata/ and blobs/ in remote
-// storage, along with any object an upload cut off part-way left under a
+// storage, whatever its name, along with any object an upload cut off
-// temporary `.partial` name. After this returns successfully the bucket
+// part-way left under a temporary `.partial` name. It returns
-// prefix is empty and the next backup starts from scratch.
+// successfully only when a listing afterwards finds nothing left there,
+// and the next backup then starts from scratch.
 //
 // Refuses to run unless force is true. The caller is responsible for
 // confirming with the user.
@@ -68,7 +74,8 @@ func (v *Vaultik) NukeRemote(force bool) error {
 }
 
 // emptyDestinationStore does the deleting for NukeRemote and stops at
-// the first step that fails.
+// 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.")
 
@@ -77,25 +84,62 @@ func (v *Vaultik) emptyDestinationStore() 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})
+	var left []string
-	if err != nil {
+
-		return fmt.Errorf("pruning blobs: %w", err)
+	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
+	for _, key := range left {
-	// steps above never delete them.
+		v.UI.Warningf("Still on the backup destination store: %s", key)
-	for _, prefix := range []string{"metadata/", "blobs/"} {
+	}
-		err = v.Storage.DeletePartialUploads(v.ctx, prefix)
+
-		if err != nil {
+	if len(left) > 0 {
-			return fmt.Errorf("deleting partial uploads: %w", err)
+		return fmt.Errorf("%d %w", len(left), errObjectsLeft)
-		}
 	}
 
 	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 {
+			log.Error("Failed to delete object", "key", key, "error", err)
+		}
+	}
+
+	// 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)
+	}
+
+	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
 //
 //nolint:tagliatelle // snake_case is the established JSON output format
@@ -160,7 +204,7 @@ 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
 	}
@@ -168,12 +212,17 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 	log.Info("Found unreferenced blobs",
 		"count", len(unreferencedBlobs), "total_size", ubytes(totalSize))
 
-	if !opts.JSON {
+	if opts.Force && !opts.JSON {
-		v.stdoutf("Found %d unreferenced blob(s) totaling %s\n",
+		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
@@ -200,7 +249,7 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 			return err
 		}
 	} else {
-		v.stdoutf("\nDeleted %d blob(s) totaling %s\n",
+		v.UI.Infof("Deleted %d blob(s) totaling %s.",
 			result.BlobsDeleted, ubytes(result.BytesFreed))
 	}
 

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/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"
 )
 
@@ -58,13 +59,16 @@ 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"
 )
 
@@ -86,6 +87,7 @@ func setupPurgeTest(
 		Stdout:       stdout,
 		Stderr:       stderr,
 		Stdin:        stdin,
+		UI:           ui.NewWithColor(stdout, false),
 	}
 	v.SetContext(ctx)
 
@@ -279,7 +281,8 @@ 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 +1,148 @@
+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)
+
+		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"
 )
@@ -350,7 +351,7 @@ 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
@@ -431,6 +432,40 @@ 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()
@@ -504,7 +539,7 @@ 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" +
@@ -150,16 +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))
@@ -185,6 +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),
@@ -199,21 +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) " +
+		v.warnIfOwnershipNotPreserved(files, opts.TargetDir)
-			"requires root. Re-run as root (e.g. with sudo) if you need " +
-			"original UID/GID preserved.")
 	}
 
 	if result.FilesFailed > 0 {
@@ -236,7 +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
@@ -299,12 +359,6 @@ 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
@@ -316,6 +370,12 @@ func (v *Vaultik) restoreAllFiles(
 		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 {
@@ -403,10 +463,8 @@ func (v *Vaultik) runRestoreLoop(
 			continue
 		}
 
-		// Record the file as restored so the sweeper can free blobs
+		// Drop the file from the plan's indexes so future picks ignore
-		// once all referencing files are done, and drop it from the
+		// it and the sweeper can free blobs no pending file needs.
-		// plan's indexes so future picks ignore it.
-		session.sweeper.fileRestored(fileID.String())
 		plan.finishFile(fileID)
 
 		processed++
@@ -840,27 +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
+		files, err := repos.Files.ListUnderPath(ctx, filepath.Clean(filter))
-		filter = filepath.Clean(filter)
-
-		files, err := repos.Files.ListUnderPath(ctx, 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 {
@@ -871,7 +936,7 @@ func (v *Vaultik) getFilesToRestore(
 		}
 	}
 
-	return result, nil
+	return result, unmatched, nil
 }
 
 // buildChunkToBlobMap creates a mapping from chunk hash to blob information
@@ -936,8 +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
+	// the call entirely as non-root; warnIfOwnershipNotPreserved warns
-	// of the restore explaining that ownership was not preserved.
+	// 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.

internal/vaultik/restore_interrupt_test.go

@@ -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

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
+// files of srcBytes pack into 3 blobs. The index database is closed
-// so the restore half runs from remote bytes only.
+// 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
+	// An average chunk size of four times srcBytes makes the minimum
-	// becomes one chunk. maxBlobSize = 5 MiB packs exactly 5 chunks
+	// chunk srcBytes, so each source file becomes one chunk. A blob limit
-	// per blob, yielding 3 blobs from 15 source files.
+	// of five times srcBytes packs exactly 5 chunks per blob, yielding 3
-	chunkSize := int64(4 * 1024 * 1024)
+	// blobs from 15 source files.
-	maxBlobSize := int64(5 * 1024 * 1024)
+	chunkSize := 4 * srcBytes
+	maxBlobSize := 5 * srcBytes
 
 	storer, err := storage.NewFileStorer(storeDir)
 	require.NoError(t, err)
@@ -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_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_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{}
@@ -237,6 +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

@@ -231,7 +231,8 @@ func (f trailingByteFile) Close() error {
 }
 
 // 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()
@@ -247,6 +248,8 @@ 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

internal/vaultik/restore_sweeper.go

@@ -1,46 +1,30 @@
 package vaultik
 
-import (
+// restoreSweeper frees cached blobs once no file still to be restored in
-	"context"
+// this run needs them. It works as follows:
-	"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:
 //
-//  1. Callers add a file's ID to an in-memory restored set via
+//  1. After each chunk is restored, chunkRestored accumulates a running
-//     fileRestored once the file is fully written to disk.
-//  2. 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
+//     the cache and deletes each blob that the restore plan no longer
-//     reference any chunk in that blob, then compares that list against
+//     needs.
-//     the in-memory restored set. If any referencing file is missing
-//     from the set the blob is kept; otherwise the cache entry is
-//     deleted.
 //
-// All DB reads happen against the snapshot's temporary metadata DB,
+// The plan drops a file once it is restored or has failed, and holds
-// which is local, indexed, and not under contention — the queries are
+// only the files this run restores, so files outside the requested paths
-// cheap and run at most once per blob per sweep interval.
+// never keep a blob.
 type restoreSweeper struct {
-	ctx        context.Context //nolint:containedctx // ctx bound at construction by design
+	plan       *restorePlan
-	repos      *database.Repositories
 	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,
+	plan *restorePlan,
-	repos *database.Repositories,
 	cache *blobDiskCache,
 	threshold int64,
 ) *restoreSweeper {
@@ -49,21 +33,12 @@ func newRestoreSweeper(
 	}
 
 	return &restoreSweeper{
-		ctx:       ctx,
+		plan:      plan,
-		repos:     repos,
 		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
+// sweep deletes every cached blob that no pending file references.
-// 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.
 func (s *restoreSweeper) sweep() {
 	for _, blobHash := range s.cache.Keys() {
-		needed, err := s.blobStillNeeded(blobHash)
+		if !s.plan.pendingFileUsesBlob(blobHash) {
-		if err != nil {
-			log.Debug("sweeper referencing-files query failed",
-				"blob_hash", blobHash[:16], "error", err)
-
-			continue
-		}
-
-		if !needed {
 			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_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 := filepath.Join(tempDir, "remote")
+	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_unmatched_path_test.go

@@ -0,0 +1,202 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"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/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"
+)
+
+// These tests cover https://git.eeqj.de/sneak/vaultik/issues/309: a path
+// given to restore that matches nothing in the snapshot is named in a
+// warning and makes the restore fail, after the paths that do match are
+// restored.
+
+func TestRestoreNamesUnmatchedPathAndRestoresTheMatchingOne(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "src")
+	storeDir := filepath.Join(tempDir, "remote")
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	ctx := context.Background()
+	cfg := faultTestConfig()
+	files := writeFaultSourceTree(t, fs, dataDir)
+
+	store, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+
+	id := fullFaultBackup(ctx, t, fs, store, cfg,
+		database.NewRepositories(db), dataDir, dbPath, "unmatched")
+	require.NoError(t, db.Close())
+
+	matched := filepath.Join(dataDir, "small.txt")
+	unmatched := filepath.Join(dataDir, "typo.txt")
+
+	var out bytes.Buffer
+
+	v := newReaderVaultik(ctx, cfg, store, nil, fs)
+	v.UI = ui.NewWithColor(&out, false)
+
+	err = v.Restore(&vaultik.RestoreOptions{
+		SnapshotID: id,
+		TargetDir:  restoreDir,
+		Paths:      []string{matched, unmatched},
+	})
+
+	require.Error(t, err, "a path that matches nothing must fail the restore")
+	assert.Contains(t, out.String(), unmatched,
+		"the path that matches nothing must be named")
+	assert.NotContains(t, out.String(), matched,
+		"the path that matches must not be reported")
+
+	got, err := afero.ReadFile(fs, filepath.Join(restoreDir, matched))
+	require.NoError(t, err, "the path that matches must still be restored")
+	assert.Equal(t, files[matched], got)
+}
+
+// When no requested path matches, nothing is restored and each path is
+// named. One path is relative. The other goes through the symlink that
+// is the snapshot's configured path, while the entries are stored under
+// the directory the symlink points to.
+func TestRestoreNamesEveryPathWhenNothingMatches(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	const snapshotName = "linked"
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "src")
+	linkDir := filepath.Join(tempDir, "srclink")
+	storeDir := filepath.Join(tempDir, "remote")
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	ctx := context.Background()
+
+	writeFaultSourceTree(t, fs, dataDir)
+	require.NoError(t, os.Symlink(dataDir, linkDir))
+
+	cfg := faultTestConfig()
+	cfg.IndexPath = dbPath
+	cfg.ChunkSize = config.Size(faultChunkSize)
+	cfg.Snapshots = map[string]config.SnapshotConfig{
+		snapshotName: {Paths: []string{linkDir}},
+	}
+
+	store, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+
+	repos := database.NewRepositories(db)
+	v := newBackupVaultik(ctx, cfg, store, repos, db, fs)
+
+	require.NoError(t, backUp(v, snapshotName))
+
+	id := localSnapshotID(ctx, t, repos, snapshotName)
+	require.NoError(t, db.Close())
+
+	throughLink := filepath.Join(linkDir, "sub")
+	relative := strings.TrimPrefix(filepath.Join(dataDir, "small.txt"), "/")
+
+	var out bytes.Buffer
+
+	reader := newReaderVaultik(ctx, cfg, store, nil, fs)
+	reader.UI = ui.NewWithColor(&out, false)
+
+	err = reader.Restore(&vaultik.RestoreOptions{
+		SnapshotID: id,
+		TargetDir:  restoreDir,
+		Paths:      []string{throughLink, relative},
+	})
+
+	require.Error(t, err, "a restore that matches nothing must fail")
+	assert.Contains(t, out.String(), throughLink,
+		"the path through the symlink must be named")
+	assert.Contains(t, out.String(), relative,
+		"the relative path must be named")
+
+	_, err = fs.Stat(restoreDir)
+	assert.True(t, os.IsNotExist(err), "nothing must be restored")
+}
+
+// Without --skip-errors a damaged blob stops the restore part way. The
+// path that matches nothing must still be named.
+func TestRestoreNamesUnmatchedPathWhenADamagedBlobStopsIt(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "src")
+	storeDir := filepath.Join(tempDir, "remote")
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	ctx := context.Background()
+	cfg := faultTestConfig()
+
+	writeFaultSourceTree(t, fs, dataDir)
+
+	inner, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+
+	id := fullFaultBackup(ctx, t, fs, inner, cfg,
+		database.NewRepositories(db), dataDir, dbPath, "damaged")
+	require.NoError(t, db.Close())
+
+	damaged := faultstore.New(inner)
+	damaged.OnGet = func(key string) faultstore.GetFault {
+		if strings.HasPrefix(key, "blobs/") {
+			return faultstore.GetCorrupt
+		}
+
+		return faultstore.GetNormal
+	}
+
+	matched := filepath.Join(dataDir, "small.txt")
+	unmatched := filepath.Join(dataDir, "typo.txt")
+
+	var out bytes.Buffer
+
+	v := newReaderVaultik(ctx, cfg, damaged, nil, fs)
+	v.UI = ui.NewWithColor(&out, false)
+
+	err = v.Restore(&vaultik.RestoreOptions{
+		SnapshotID: id,
+		TargetDir:  restoreDir,
+		Paths:      []string{matched, unmatched},
+	})
+
+	require.Error(t, err, "a damaged blob must stop the restore")
+	assert.Contains(t, err.Error(), "blob",
+		"the restore must stop on the damaged blob")
+	assert.Contains(t, out.String(), unmatched,
+		"the path that matches nothing must be named")
+}

internal/vaultik/shallow_verify_test.go

@@ -27,7 +27,7 @@ import (
 // then grows one stored blob so its size no longer matches the manifest.
 // Shallow verify must then fail, count the grown blob as a size mismatch,
 // and drop it from the verified count rather than continuing to report it
-// as checked.
+// as checked, and its error must not repeat the cli layer's prefix.
 func TestShallowVerifyDetectsWrongBlobSize(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -96,6 +96,12 @@ func TestShallowVerifyDetectsWrongBlobSize(t *testing.T) {
 	require.Equal(t, 1, bad.Mismatched)
 	require.Equal(t, healthy.BlobCount-1, bad.Verified,
 		"the wrong-sized blob must not be counted as verified")
+
+	// The cli layer prints this error after its own "Verification failed:".
+	err = newVerifier(io.Discard).VerifySnapshotWithOptions(
+		snapshotID, &vaultik.VerifyOptions{})
+	require.EqualError(t, err,
+		"problems found in storage: 1 blobs have the wrong size")
 }
 
 // TestShallowVerifyDetectsMissingDatabase backs up a real snapshot,

internal/vaultik/snapshot.go

@@ -23,6 +23,7 @@ var (
 	errSnapshotNotInConfig  = errors.New("snapshot not found in config")
 	errNoSnapshotsInConfig  = errors.New("no snapshots configured")
 	errSnapshotVerifyFailed = errors.New("verification failed")
+	errProblemsInStorage    = errors.New("problems found in storage")
 	errRemoveAllNeedsForce  = errors.New("--all requires --force")
 	errSnapshotNotFound     = errors.New("snapshot not found")
 	errSnapshotRemoveFailed = errors.New("snapshot(s) could not be removed")
@@ -243,12 +244,18 @@ func (v *Vaultik) createNamedSnapshot(
 		return err
 	}
 
+	previousSnapshotID, err := v.previousSnapshotID(snapName)
+	if err != nil {
+		return err
+	}
+
 	scanner := v.ScannerFactory(snapshot.ScannerParams{
-		EnableProgress: !opts.Cron,
+		EnableProgress:     !opts.Cron,
-		UI:             v.UI,
+		UI:                 v.UI,
-		Fs:             v.Fs,
+		Fs:                 v.Fs,
-		Exclude:        v.Config.GetExcludes(snapName),
+		Exclude:            v.Config.GetExcludes(snapName),
-		SkipErrors:     opts.SkipErrors,
+		SkipErrors:         opts.SkipErrors,
+		PreviousSnapshotID: previousSnapshotID,
 	})
 
 	snapshotID, err := v.SnapshotManager.CreateSnapshotWithName(
@@ -283,6 +290,24 @@ func (v *Vaultik) createNamedSnapshot(
 	return nil
 }
 
+// previousSnapshotID returns the ID of the most recent completed snapshot
+// named snapName in the local index, or "" when there is none.
+func (v *Vaultik) previousSnapshotID(snapName string) (string, error) {
+	snapshots, err := v.Repositories.Snapshots.ListRecent(v.ctx, listRecentLimit)
+	if err != nil {
+		return "", fmt.Errorf("listing snapshots: %w", err)
+	}
+
+	for _, s := range snapshots {
+		if s.CompletedAt != nil &&
+			parseSnapshotName(s.ID.String(), s.Hostname.String()) == snapName {
+			return s.ID.String(), nil
+		}
+	}
+
+	return "", nil
+}
+
 // resolveSnapshotPaths resolves source directories to absolute paths
 // with symlink resolution.
 func (v *Vaultik) resolveSnapshotPaths(snapName string) ([]string, error) {
@@ -574,7 +599,7 @@ func (v *Vaultik) PurgeSnapshotsWithOptions(opts *SnapshotPurgeOptions) error {
 
 	if len(toDelete) == 0 {
 		if !opts.Quiet {
-			v.printlnStdout("No snapshots to delete")
+			v.UI.Infof("No snapshots to delete.")
 		}
 
 		return nil
@@ -612,7 +637,7 @@ func selectSnapshotsToPurge(
 	case opts.OlderThan != "":
 		duration, err := parseDuration(opts.OlderThan)
 		if err != nil {
-			return nil, fmt.Errorf("invalid duration: %w", err)
+			return nil, err
 		}
 
 		cutoff := time.Now().UTC().Add(-duration)
@@ -631,19 +656,19 @@ func selectSnapshotsToPurge(
 func (v *Vaultik) confirmAndExecutePurge(
 	toDelete []SnapshotInfo, force, quiet bool,
 ) error {
-	if !quiet {
+	// Confirm unless --force is used. The list, the prompt and its answer
+	// bypass the UI: they are an interactive exchange the operator must
+	// see, so --quiet does not silence them.
+	if !force {
 		v.stdoutf("The following snapshots will be deleted:\n\n")
 
 		for _, snap := range toDelete {
 			v.stdoutf("  %s (%s, %s)\n",
 				snap.ID,
-				snap.Timestamp.Format("2006-01-02 15:04:05"),
+				snap.Timestamp.Format(time.DateTime),
 				ubytes(snap.CompressedSize))
 		}
-	}
 
-	// Confirm unless --force is used
-	if !force {
 		v.stdoutf("\nDelete %d snapshot(s)? [y/N] ", len(toDelete))
 
 		var confirm string
@@ -661,7 +686,16 @@ func (v *Vaultik) confirmAndExecutePurge(
 			return nil
 		}
 	} else if !quiet {
-		v.stdoutf("\nDeleting %d snapshot(s) (--force specified)\n", len(toDelete))
+		v.UI.Infof("The following snapshots will be deleted:")
+
+		for _, snap := range toDelete {
+			v.UI.Detailf("%s (%s, %s)",
+				v.UI.Snapshot(snap.ID.String()),
+				snap.Timestamp.Format(time.DateTime),
+				ubytes(snap.CompressedSize))
+		}
+
+		v.UI.Infof("Deleting %d snapshot(s) (--force specified).", len(toDelete))
 	}
 
 	// Delete snapshots (both local and remote)
@@ -669,10 +703,11 @@ func (v *Vaultik) confirmAndExecutePurge(
 		snapshotID := snap.ID.String()
 		log.Info("Deleting snapshot", "id", snapshotID)
 
+		// Stop before the remote delete, so a snapshot still in the local
+		// index keeps its metadata on the destination store.
 		err := v.deleteSnapshotFromLocalDB(snapshotID)
 		if err != nil {
-			log.Error("Failed to delete from local database",
+			return fmt.Errorf("deleting snapshot %s from local index: %w", snapshotID, err)
-				"snapshot_id", snapshotID, "error", err)
 		}
 
 		_, err = v.deleteRemoteSnapshotByKey(snapshot.RemoteSnapshotKey(snapshotID))
@@ -693,9 +728,9 @@ func (v *Vaultik) confirmAndExecutePurge(
 	}
 
 	if !quiet {
-		v.stdoutf("Deleted %d snapshot(s)\n", len(toDelete))
+		v.UI.Infof("Deleted %d snapshot(s).", len(toDelete))
-		v.printlnStdout(
+		v.UI.Noticef("Run '%s' to clean up unreferenced remote blobs.",
-			"\nNote: Run 'vaultik prune' to clean up unreferenced remote blobs.")
+			pruneCommandHint)
 	}
 
 	return nil
@@ -931,8 +966,10 @@ func (v *Vaultik) formatVerifyResult(
 		v.printVerifySummary(result, failure)
 	}
 
+	// Not errSnapshotVerifyFailed: the cli layer prints this after its own
+	// "Verification failed:".
 	if failure != "" {
-		return fmt.Errorf("%w: %s", errSnapshotVerifyFailed, failure)
+		return fmt.Errorf("%w: %s", errProblemsInStorage, failure)
 	}
 
 	return nil
@@ -1068,7 +1105,7 @@ func (v *Vaultik) CleanupLocalSnapshots(opts *PruneOptions) error {
 			log.Info("Removing stale local snapshot record", "snapshot_id", id)
 
 			if !opts.JSON {
-				v.stdoutf("Removing stale local record: %s\n", id)
+				v.UI.Infof("Removing stale local record: %s.", v.UI.Snapshot(id))
 			}
 
 			err = v.deleteSnapshotFromLocalDB(id)
@@ -1091,9 +1128,9 @@ func (v *Vaultik) CleanupLocalSnapshots(opts *PruneOptions) error {
 	}
 
 	if removed == 0 {
-		v.printlnStdout("No stale local snapshots found.")
+		v.UI.Infof("No stale local snapshots found.")
 	} else {
-		v.stdoutf("Removed %d stale local snapshot record(s).\n", removed)
+		v.UI.Infof("Removed %d stale local snapshot record(s).", removed)
 	}
 
 	return nil
@@ -1218,9 +1255,11 @@ const snapshotRemoveCommandHint = "vaultik snapshot remove"
 // destination store. Blobs are NOT touched: removing a snapshot's
 // blobs requires reading every remaining remote manifest (the remote
 // may hold snapshots this host doesn't know about), which is what
-// `vaultik prune` is for — the command prints the prune invocation to
+// `vaultik prune` is for — when metadata was removed from the
-// run as a follow-up. When the remote is unreachable the command still
+// destination store, the text output names the prune invocation to run
-// completes the local-DB removal and warns.
+// as a follow-up. When removing the metadata fails, a snapshot in the
+// local index is still removed from it, with a warning; a snapshot
+// only on the destination store is an error.
 func (v *Vaultik) RemoveSnapshot(
 	snapshotID string, opts *RemoveOptions,
 ) (*RemoveResult, error) {
@@ -1255,8 +1294,10 @@ func (v *Vaultik) RemoveSnapshot(
 		}
 	}
 
-	if !opts.LocalOnly {
+	result.RemoteRemoved, err = v.removeSnapshotRemote(
-		result.RemoteRemoved = v.removeSnapshotRemote(snapshotID, opts)
+		snapshotID, inLocalIndex, opts)
+	if err != nil {
+		return result, err
 	}
 
 	if v.SnapshotManager != nil {
@@ -1272,14 +1313,15 @@ func (v *Vaultik) RemoveSnapshot(
 	}
 
 	if inLocalIndex {
-		v.stdoutf("Removed snapshot '%s' from local database\n", snapshotID)
+		v.UI.Infof("Removed snapshot '%s' from local database.",
+			v.UI.Snapshot(snapshotID))
 	}
 
 	if result.RemoteRemoved {
-		v.printlnStdout("Removed snapshot metadata from remote storage")
+		v.UI.Infof("Removed snapshot metadata from remote storage.")
-		v.stdoutf("\nNote: The removed snapshot's blobs remain on the remote. "+
+		v.UI.Noticef("The removed snapshot's blobs remain on the remote. "+
 			"Run '%s' to delete any blobs no longer referenced by any "+
-			"remaining remote snapshot.\n", pruneCommandHint)
+			"remaining remote snapshot.", pruneCommandHint)
 	}
 
 	return result, nil
@@ -1335,21 +1377,23 @@ func (v *Vaultik) removeSnapshotDryRun(
 		return result, v.outputRemoveJSON(result)
 	}
 
-	v.stdoutf("Would remove snapshot: %s\n", snapshotID)
+	v.UI.Infof("Would remove snapshot: %s.", v.UI.Snapshot(snapshotID))
 
 	if !opts.LocalOnly {
-		v.printlnStdout("Would also remove snapshot metadata from " +
+		v.UI.Infof("Would also remove snapshot metadata from " +
-			"remote storage (blobs untouched)")
+			"remote storage (blobs untouched).")
 	}
 
-	v.printlnStdout("[Dry run - no changes made]")
+	v.UI.Noticef("Dry run: no changes made.")
 
 	return result, nil
 }
 
 // confirmRemoveSnapshot prompts for confirmation before removing a
 // snapshot. Returns false (treating EOF/read errors as "no") unless the
-// user answers "y".
+// user answers "y". The prompt and its answer bypass the UI: they are an
+// interactive exchange the operator must see, so --quiet does not
+// silence them.
 func (v *Vaultik) confirmRemoveSnapshot(snapshotID string, opts *RemoveOptions) bool {
 	if opts.LocalOnly {
 		v.stdoutf("Remove snapshot '%s' from local database "+
@@ -1378,19 +1422,33 @@ func (v *Vaultik) confirmRemoveSnapshot(snapshotID string, opts *RemoveOptions)
 }
 
 // removeSnapshotRemote strips the snapshot's metadata from the
-// destination store, warning and proceeding on failure: the local-DB
+// destination store. It returns true only when the delete finished
-// removal has already happened, so the user is told the remote half
+// without an error and removed at least one object; under LocalOnly it
-// didn't finish and to run `vaultik snapshot remove` for the snapshot
+// does nothing. When the snapshot is only on the destination store, a
-// again once the destination store is reachable (`vaultik prune` never
+// failure is returned as an error, because that removal is all the
-// removes snapshot metadata). Returns true when the remote removal
+// command does. When the snapshot was in the local index, its removal
-// succeeded and deleted something.
+// there has already happened, so a failure is only a warning telling the
-func (v *Vaultik) removeSnapshotRemote(snapshotID string, opts *RemoveOptions) bool {
+// user to run `vaultik snapshot remove` for the snapshot again once the
+// destination store is reachable (`vaultik prune` never removes snapshot
+// metadata).
+func (v *Vaultik) removeSnapshotRemote(
+	snapshotID string, inLocalIndex bool, opts *RemoveOptions,
+) (bool, error) {
+	if opts.LocalOnly {
+		return false, nil
+	}
+
 	log.Info("Removing snapshot metadata from remote storage",
 		"snapshot_id", snapshotID)
 
 	remoteKey := snapshot.RemoteSnapshotKey(snapshotID)
 
 	deleted, err := v.deleteRemoteSnapshotByKey(remoteKey)
+	if err != nil && !inLocalIndex {
+		return false, fmt.Errorf(
+			"removing snapshot metadata from destination store: %w", err)
+	}
+
 	if err != nil {
 		log.Warn("Could not remove snapshot metadata from remote storage; "+
 			"run '"+snapshotRemoveCommandHint+"' with the snapshot's ID "+
@@ -1405,10 +1463,10 @@ func (v *Vaultik) removeSnapshotRemote(snapshotID string, opts *RemoveOptions) b
 				err, snapshotRemoveCommandHint, snapshotID)
 		}
 
-		return false
+		return false, nil
 	}
 
-	return deleted > 0
+	return deleted > 0, nil
 }
 
 // RemoveAllSnapshots removes every snapshot known to the local
@@ -1455,7 +1513,7 @@ func (v *Vaultik) RemoveAllSnapshots(opts *RemoveOptions) (*RemoveResult, error)
 
 	if len(localSnaps) == 0 && len(orphanRemoteKeys) == 0 {
 		if !opts.JSON {
-			v.printlnStdout("No snapshots found")
+			v.UI.Infof("No snapshots found.")
 		}
 
 		return &RemoveResult{}, nil
@@ -1560,27 +1618,27 @@ func (v *Vaultik) handleRemoveAllDryRun(
 	}
 
 	if !opts.JSON {
-		v.stdoutf("Would remove %d local snapshot(s):\n", len(localSnaps))
+		v.UI.Infof("Would remove %d local snapshot(s):", len(localSnaps))
 
 		for _, id := range localSnaps {
-			v.stdoutf("  %s\n", id)
+			v.UI.Detailf("%s", v.UI.Snapshot(id))
 		}
 
 		if !opts.LocalOnly {
 			if len(orphanRemoteKeys) > 0 {
-				v.stdoutf("Would also remove %d orphan remote snapshot key(s):\n",
+				v.UI.Infof("Would also remove %d orphan remote snapshot key(s):",
 					len(orphanRemoteKeys))
 
 				for _, key := range orphanRemoteKeys {
-					v.stdoutf("  %s\n", key)
+					v.UI.Detailf("%s", key)
 				}
 			} else {
-				v.printlnStdout("Would also remove snapshot metadata from " +
+				v.UI.Infof("Would also remove snapshot metadata from " +
-					"remote storage (blobs untouched)")
+					"remote storage (blobs untouched).")
 			}
 		}
 
-		v.printlnStdout("[Dry run - no changes made]")
+		v.UI.Noticef("Dry run: no changes made.")
 	}
 
 	if opts.JSON {
@@ -1629,10 +1687,10 @@ func (v *Vaultik) executeRemoveAll(
 		return result, v.outputRemoveJSON(result)
 	}
 
-	v.stdoutf("Removed %d snapshot(s)\n", len(result.SnapshotsRemoved))
+	v.UI.Infof("Removed %d snapshot(s).", len(result.SnapshotsRemoved))
 
 	if !opts.LocalOnly && result.RemoteRemoved {
-		v.printlnStdout("Removed snapshot metadata from remote storage")
+		v.UI.Infof("Removed snapshot metadata from remote storage.")
 	}
 
 	return result, nil
@@ -1889,13 +1947,17 @@ func (v *Vaultik) PruneDatabase() (*PruneResult, error) {
 	snapshotsRemain := countDiff(snapshotCountBefore, &result.SnapshotsDeleted)
 
 	v.UI.Completef("Pruned local index database.")
-	v.UI.Detailf("Incomplete snapshots: %s removed (%s remain).",
+	v.UI.Detailf("Incomplete snapshots: %s removed "+
+		"(total snapshots left in the local index: %s).",
 		countText(&result.SnapshotsDeleted), countText(snapshotsRemain))
-	v.UI.Detailf("Orphaned files: %s removed (%s remain).",
+	v.UI.Detailf("Orphaned files: %s removed "+
+		"(total files left in the local index: %s).",
 		countText(result.FilesDeleted), countText(fileCountAfter))
-	v.UI.Detailf("Orphaned chunks: %s removed (%s remain).",
+	v.UI.Detailf("Orphaned chunks: %s removed "+
+		"(total chunks left in the local index: %s).",
 		countText(result.ChunksDeleted), countText(chunkCountAfter))
-	v.UI.Detailf("Orphaned blobs: %s removed (%s remain).",
+	v.UI.Detailf("Orphaned blobs: %s removed "+
+		"(total blobs left in the local index: %s).",
 		countText(result.BlobsDeleted), countText(blobCountAfter))
 
 	return result, nil

internal/vaultik/snapshot_deleted_count_test.go

@@ -0,0 +1,249 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"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/ui"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// These tests cover https://git.eeqj.de/sneak/vaultik/issues/304: a
+// backup counts a file as deleted only when the previous snapshot of
+// the same name had it and it is gone, and never counts an excluded
+// file.
+
+// deletedCountEnv is a backup setup over the source directory src whose
+// user-facing output is kept in out. Each test configures its snapshots
+// with setSnapshot.
+type deletedCountEnv struct {
+	v   *vaultik.Vaultik
+	out *bytes.Buffer
+	src string
+}
+
+// newDeletedCountEnv writes files, keyed by path relative to src.
+func newDeletedCountEnv(t *testing.T, files map[string][]byte) *deletedCountEnv {
+	t.Helper()
+
+	ctx := context.Background()
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	srcDir := filepath.Join(tempDir, "src")
+
+	for rel, content := range files {
+		path := filepath.Join(srcDir, rel)
+		require.NoError(t, fs.MkdirAll(filepath.Dir(path), 0o755))
+		require.NoError(t, afero.WriteFile(fs, path, content, 0o644))
+	}
+
+	cfg := faultTestConfig()
+	cfg.IndexPath = filepath.Join(tempDir, "index.sqlite")
+	cfg.Snapshots = map[string]config.SnapshotConfig{}
+
+	store, err := storage.NewFileStorer(filepath.Join(tempDir, "remote"))
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, cfg.IndexPath)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	repos := database.NewRepositories(db)
+	out := &bytes.Buffer{}
+	v := newBackupVaultik(ctx, cfg, store, repos, db, fs)
+	v.UI = ui.NewWithColor(out, false)
+
+	return &deletedCountEnv{v: v, out: out, src: srcDir}
+}
+
+// setSnapshot configures the snapshot name to back up the path rel
+// beneath src, with the given excludes.
+func (e *deletedCountEnv) setSnapshot(name, rel string, exclude ...string) {
+	e.v.Config.Snapshots[name] = config.SnapshotConfig{
+		Paths:   []string{filepath.Join(e.src, rel)},
+		Exclude: exclude,
+	}
+}
+
+// backUp runs a backup of the named snapshot and returns its output.
+func (e *deletedCountEnv) backUp(t *testing.T, name string) string {
+	t.Helper()
+
+	e.out.Reset()
+	require.NoError(t, e.v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		Snapshots: []string{name},
+	}))
+
+	return e.out.String()
+}
+
+// A file this snapshot excludes is not counted as deleted: neither one
+// that another snapshot backs up, nor one that the previous snapshot of
+// this name had before the exclude was added.
+func TestBackupDoesNotCountExcludedFileAsDeleted(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newDeletedCountEnv(t, map[string][]byte{
+		"top.txt":       []byte("top"),
+		"cache/tmp.bin": []byte("cache"),
+		"photos/a.jpg":  []byte("photo a"),
+		"photos/b.jpg":  []byte("photo b"),
+		"photos/c.jpg":  []byte("photo c"),
+	})
+	env.setSnapshot("photos", "photos")
+	env.setSnapshot("home", "", "/photos")
+
+	env.backUp(t, "photos")
+	assert.Contains(t, env.backUp(t, "home"), filesLine(2, 2, 0))
+
+	env.setSnapshot("home", "", "/photos", "/cache")
+	assert.Contains(t, env.backUp(t, "home"), filesLine(1, 0, 1))
+}
+
+// A file inside a directory that an exclude added after a backup
+// matches is not counted as deleted, though the pattern matches only the
+// directory and the directory is gone from disk.
+func TestBackupDoesNotCountFileInExcludedDirectoryAsDeleted(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newDeletedCountEnv(t, map[string][]byte{
+		"top.txt":                         []byte("top"),
+		"user/.local/share/Trash/old.txt": []byte("old"),
+	})
+	env.setSnapshot("home", "")
+
+	assert.Contains(t, env.backUp(t, "home"), filesLine(2, 2, 0))
+
+	require.NoError(t, os.RemoveAll(
+		filepath.Join(env.src, "user/.local/share/Trash")))
+
+	env.setSnapshot("home", "", "*/.local/share/Trash")
+	assert.Contains(t, env.backUp(t, "home"), filesLine(1, 0, 1))
+}
+
+// A file whose directory was replaced by a regular file of the same name
+// is counted as deleted.
+func TestBackupCountsFileInDirectoryReplacedByFileAsDeleted(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newDeletedCountEnv(t, map[string][]byte{
+		"dir/inside.txt": []byte("inside"),
+	})
+	env.setSnapshot("home", "")
+
+	assert.Contains(t, env.backUp(t, "home"), filesLine(1, 1, 0))
+
+	dir := filepath.Join(env.src, "dir")
+	require.NoError(t, os.RemoveAll(dir))
+	require.NoError(t, os.WriteFile(dir, []byte("now a file"), 0o600))
+
+	assert.Contains(t, env.backUp(t, "home"),
+		"Files: 1 examined, 1 backed up, 0 unchanged, 1 deleted.")
+}
+
+// Under --skip-errors a file inside a directory that the walk cannot
+// open is not counted as deleted: it is still on disk.
+func TestBackupDoesNotCountFileInUnopenableDirectoryAsDeleted(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newDeletedCountEnv(t, map[string][]byte{
+		"open.txt":          []byte("open"),
+		"locked/inside.txt": []byte("inside"),
+	})
+	env.setSnapshot("work", "")
+
+	assert.Contains(t, env.backUp(t, "work"), filesLine(2, 2, 0))
+
+	// A directory's mode does not lock out root, who runs the tests, so
+	// opening the directory is made to fail instead.
+	env.v.Fs = &openFailFs{
+		path: filepath.Join(env.src, "locked"),
+		err:  os.ErrPermission,
+	}
+
+	env.out.Reset()
+	require.NoError(t, env.v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		SkipErrors: true,
+		Snapshots:  []string{"work"},
+	}))
+	assert.Contains(t, env.out.String(), filesLine(1, 0, 1))
+}
+
+// A backup counts only files that the previous snapshot of its own name
+// had, not one that a later snapshot of another name over the same path
+// had and that is since removed.
+func TestBackupIgnoresSnapshotOfAnotherName(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newDeletedCountEnv(t, map[string][]byte{
+		"kept.txt": []byte("kept"),
+	})
+	env.setSnapshot("home", "")
+	env.setSnapshot("other", "")
+
+	env.backUp(t, "home")
+
+	// Snapshot start times are kept in whole seconds: wait so that other's
+	// snapshot is strictly the most recent.
+	time.Sleep(time.Second)
+
+	extra := filepath.Join(env.src, "extra.txt")
+	require.NoError(t, os.WriteFile(extra, []byte("extra"), 0o600))
+	env.backUp(t, "other")
+	require.NoError(t, os.Remove(extra))
+
+	assert.Contains(t, env.backUp(t, "home"), filesLine(1, 0, 1))
+}
+
+// A file removed from disk is counted as deleted by the next backup of
+// the snapshot that had it, and not by any later backup of that
+// snapshot, nor by the first backup of another name over the same path.
+func TestBackupCountsDeletedFileOnce(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	aContent := bytes.Repeat([]byte("a"), 1000)
+	bContent := bytes.Repeat([]byte("b"), 2000)
+	keptContent := bytes.Repeat([]byte("k"), 500)
+
+	env := newDeletedCountEnv(t, map[string][]byte{
+		"a.txt":    aContent,
+		"b.txt":    bContent,
+		"kept.txt": keptContent,
+	})
+	env.setSnapshot("home", "")
+	env.setSnapshot("other", "")
+
+	assert.Contains(t, env.backUp(t, "home"), filesLine(3, 3, 0))
+
+	require.NoError(t, os.Remove(filepath.Join(env.src, "a.txt")))
+	require.NoError(t, os.Remove(filepath.Join(env.src, "b.txt")))
+
+	summary := env.backUp(t, "home")
+	assert.Contains(t, summary,
+		"Files: 1 examined, 0 backed up, 1 unchanged, 2 deleted.")
+	assert.Contains(t, summary,
+		fmt.Sprintf("Data: %s total (%s backed up), %s deleted.",
+			env.v.UI.Size(int64(len(keptContent))), env.v.UI.Size(0),
+			env.v.UI.Size(int64(len(aContent)+len(bContent)))))
+
+	assert.Contains(t, env.backUp(t, "home"), filesLine(1, 0, 1))
+	assert.Contains(t, env.backUp(t, "other"), filesLine(1, 0, 1))
+}

internal/vaultik/snapshot_list.go

@@ -61,7 +61,9 @@ const tabPadding = 3
 // attempt is made to recover or fabricate the human ID.
 //
 // Snapshots in the local index with no counterpart on the destination
-// store are reported as drift below the table.
+// store are reported as drift below the table, unless the local index
+// is bound to a different destination than the configured one: that is
+// reported instead.
 //
 // If remote listing fails (unmounted volume, permission denied,
 // network), we degrade to local-only with a warning. List never fails
@@ -103,7 +105,26 @@ func (v *Vaultik) ListSnapshots(jsonOutput bool) error {
 		return snapshots[i].Timestamp.After(snapshots[j].Timestamp)
 	})
 
+	stored, configured, err := v.storageBinding()
+	if err != nil {
+		return err
+	}
+
+	// A local index bound to another destination describes that one, so
+	// its records are expected to be missing here, and prune refuses to
+	// run. The binding is reported instead of advising prune.
+	boundElsewhere := stored != "" && stored != configured
+
 	if jsonOutput {
+		// stdout holds only the document, so this goes to the logger.
+		if boundElsewhere {
+			log.Warn("Local index is bound to a different backup "+
+				"destination; to proceed, revert storage_url (or the s3.* "+
+				"endpoint, bucket and prefix) to the bound destination, "+
+				"or run 'vaultik database delete'",
+				"bound_to", stored, "configured", configured)
+		}
+
 		if remoteErr == nil {
 			v.reportJSONListingLimits(listing)
 		}
@@ -125,8 +146,13 @@ func (v *Vaultik) ListSnapshots(jsonOutput bool) error {
 		}
 	}
 
+	if boundElsewhere {
+		v.UI.Warningf("%v\n%s", errStorageBindingMismatch,
+			buildBindingMismatchMessage(stored, configured))
+	}
+
 	if remoteErr == nil {
-		v.reportListDrift(snapshots, listing)
+		v.reportListDrift(snapshots, listing, boundElsewhere)
 	}
 
 	return nil
@@ -432,7 +458,9 @@ func (v *Vaultik) snapshotInfoFromLocal(ls *database.Snapshot) SnapshotInfo {
 // reportListDrift prints the reconciliation notes the merged table
 // cannot express on its own: local records with no counterpart on the
 // destination store, plus counts of remote snapshots that were
-// unreadable or omitted.
+// unreadable or omitted. Local records are not reported when
+// boundElsewhere is set: ListSnapshots has already said why they are
+// missing.
 //
 // This is what remains of the old reportRemoteDrift, and it no longer
 // touches the destination store. Its remote-only half collapsed into
@@ -441,6 +469,7 @@ func (v *Vaultik) snapshotInfoFromLocal(ls *database.Snapshot) SnapshotInfo {
 // so `snapshot list` lists the destination exactly once per invocation.
 func (v *Vaultik) reportListDrift(
 	snapshots []SnapshotInfo, listing *remoteSnapshotListing,
+	boundElsewhere bool,
 ) {
 	var localOnly []string
 
@@ -450,7 +479,7 @@ func (v *Vaultik) reportListDrift(
 		}
 	}
 
-	if len(localOnly) > 0 {
+	if len(localOnly) > 0 && !boundElsewhere {
 		v.UI.Warningf("%d local snapshot record(s) not found in backup "+
 			"destination store:", len(localOnly))
 

internal/vaultik/snapshot_list_test.go

@@ -421,6 +421,74 @@ func TestListSnapshots_LocalOnlyReportedAsDrift(t *testing.T) {
 	assert.NotContains(t, out, "snapshot cleanup")
 }
 
+// TestListSnapshots_BoundElsewhereDoesNotAdvisePrune covers a changed
+// storage_url: the local index is still bound to the old destination,
+// so its records are missing from the new one. prune refuses to run in
+// that state, so the listing must give the choices prune gives instead
+// of advising it.
+func TestListSnapshots_BoundElsewhereDoesNotAdvisePrune(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	const (
+		oldURL = "file:///mnt/backups/old"
+		newURL = "file:///mnt/backups/new"
+	)
+
+	env := newListEnv(t)
+	env.v.Config.StorageURL = newURL
+	require.NoError(t, env.v.Repositories.LocalMeta.Set(context.Background(),
+		database.LocalMetaKeyStorageURL, oldURL))
+	env.addLocal(t, listLocalID, time.Date(2026, 3, 1, 10, 0, 0, 0, time.UTC))
+
+	err := env.v.ListSnapshots(false)
+	require.NoError(t, err)
+
+	out := env.stdout.String()
+
+	assert.Contains(t, out, listLocalID)
+	assert.Contains(t, out,
+		"local index is bound to a different backup destination")
+	assert.Contains(t, out, oldURL)
+	assert.Contains(t, out, newURL)
+	assert.Contains(t, out, "revert storage_url")
+	assert.Contains(t, out, "vaultik database delete")
+	assert.NotContains(t, out, "vaultik prune")
+}
+
+// TestListSnapshots_JSONReportsBoundElsewhere is the --json side of the
+// test above. Every local row comes out with remote_present false, which
+// a script reads as stale local records unless it is told, on stderr,
+// that the local index is bound to a different destination.
+//
+//nolint:paralleltest // captureProcessStderr replaces os.Stderr
+func TestListSnapshots_JSONReportsBoundElsewhere(t *testing.T) {
+	const (
+		oldURL = "file:///mnt/backups/old"
+		newURL = "file:///mnt/backups/new"
+	)
+
+	env := newListEnv(t)
+	env.v.Config.StorageURL = newURL
+	require.NoError(t, env.v.Repositories.LocalMeta.Set(context.Background(),
+		database.LocalMetaKeyStorageURL, oldURL))
+	env.addLocal(t, listLocalID, time.Date(2026, 3, 1, 10, 0, 0, 0, time.UTC))
+
+	stderr := captureProcessStderr(t, func() {
+		require.NoError(t, env.v.ListSnapshots(true))
+	})
+
+	rows := decodeListJSON(t, env.stdout.String())
+	require.Len(t, rows, 1)
+	assert.Equal(t, listLocalID, rows[0].ID)
+
+	assert.Contains(t, stderr, "bound to a different backup destination")
+	assert.Contains(t, stderr, oldURL)
+	assert.Contains(t, stderr, newURL)
+	assert.Contains(t, stderr, "revert storage_url")
+	assert.Contains(t, stderr, "vaultik database delete")
+}
+
 // TestListSnapshots_UnreachableRemoteDegrades covers the promise in the
 // doc comment: an unreachable destination is a warning plus local-only
 // output, never a failure.

internal/vaultik/storage_bind.go

@@ -45,17 +45,9 @@ func (v *Vaultik) EnsureStorageBinding() error {
 		return nil
 	}
 
-	configured := v.Config.StorageURL
+	stored, configured, err := v.storageBinding()
-	if configured == "" {
-		// The s3.* form: bind on its endpoint, bucket and prefix,
-		// written the way storage_url would name the same destination.
-		s3 := v.Config.S3
-		configured = "s3://" + s3.Bucket + "/" + s3.Prefix + "?endpoint=" + s3.Endpoint
-	}
-
-	stored, err := v.Repositories.LocalMeta.Get(v.ctx, database.LocalMetaKeyStorageURL)
 	if err != nil {
-		return fmt.Errorf("reading local storage binding: %w", err)
+		return err
 	}
 
 	if stored == "" {
@@ -78,6 +70,27 @@ func (v *Vaultik) EnsureStorageBinding() error {
 		errStorageBindingMismatch, buildBindingMismatchMessage(stored, configured))
 }
 
+// storageBinding returns the destination the local index is bound to
+// ("" when no binding is recorded yet), then the configured destination
+// written the same way. It records nothing, so a read-only command can
+// call it.
+func (v *Vaultik) storageBinding() (string, string, error) {
+	configured := v.Config.StorageURL
+	if configured == "" {
+		// The s3.* form: bind on its endpoint, bucket and prefix,
+		// written the way storage_url would name the same destination.
+		s3 := v.Config.S3
+		configured = "s3://" + s3.Bucket + "/" + s3.Prefix + "?endpoint=" + s3.Endpoint
+	}
+
+	stored, err := v.Repositories.LocalMeta.Get(v.ctx, database.LocalMetaKeyStorageURL)
+	if err != nil {
+		return "", "", fmt.Errorf("reading local storage binding: %w", err)
+	}
+
+	return stored, configured, nil
+}
+
 // buildBindingMismatchMessage assembles the multi-line explanation
 // shown when the local index is bound to a different destination than
 // the currently-configured one (the first line lives in the

internal/vaultik/symlinked_destination_test.go

@@ -0,0 +1,147 @@
+package vaultik_test
+
+import (
+	"context"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestPruneDeletesNothingThroughSymlinkedDirectory checks that prune keeps
+// every blob and the local snapshot record when the metadata/ or blobs/
+// directory of a file:// destination is a symlink to a directory elsewhere
+// (https://git.eeqj.de/sneak/vaultik/issues/296). A backup writes through
+// such a link, so prune has to list through it too.
+func TestPruneDeletesNothingThroughSymlinkedDirectory(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	for _, dir := range []string{"metadata", "blobs"} {
+		t.Run(dir, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+			storeDir := filepath.Join(t.TempDir(), "store")
+			v, repos, _ := backUpToFileDestination(ctx, t, storeDir)
+			blobFiles := blobFilesIn(t, storeDir)
+			moveBehindSymlink(t, storeDir, dir)
+
+			require.NoError(t, v.Prune(&vaultik.PruneOptions{Force: true}))
+
+			// When blobs/ is the link, these paths run through it.
+			for _, blobFile := range blobFiles {
+				assert.FileExists(t, blobFile, "prune must delete no blob")
+			}
+
+			snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+			require.NoError(t, err)
+			assert.Len(t, snapshots, 1, "prune must delete no local snapshot record")
+		})
+	}
+}
+
+// TestPruneFailsOnSymlinkedDirectoryWithMissingTarget checks that prune
+// fails, deleting no blob and keeping the local snapshot record, when the
+// metadata/ or blobs/ directory of a file:// destination is a symlink
+// whose target is missing, as a link to a volume that is not mounted is.
+func TestPruneFailsOnSymlinkedDirectoryWithMissingTarget(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	for _, dir := range []string{"metadata", "blobs"} {
+		t.Run(dir, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+			storeDir := filepath.Join(t.TempDir(), "store")
+			v, repos, _ := backUpToFileDestination(ctx, t, storeDir)
+			blobFiles := blobFilesIn(t, storeDir)
+			targetPath := moveBehindSymlink(t, storeDir, dir)
+
+			// Unmount the volume the link points to, and mount it again
+			// once prune has run.
+			unmountedPath := targetPath + "-unmounted"
+			require.NoError(t, os.Rename(targetPath, unmountedPath))
+
+			err := v.Prune(&vaultik.PruneOptions{Force: true})
+			require.ErrorIs(t, err, fs.ErrNotExist)
+
+			require.NoError(t, os.Rename(unmountedPath, targetPath))
+
+			// When blobs/ is the link, these paths run through it.
+			for _, blobFile := range blobFiles {
+				assert.FileExists(t, blobFile, "prune must delete no blob")
+			}
+
+			snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+			require.NoError(t, err)
+			assert.Len(t, snapshots, 1, "prune must delete no local snapshot record")
+		})
+	}
+}
+
+// TestRemoveSnapshotWarnsOnSymlinkedMetadataWithMissingTarget checks that
+// snapshot remove warns that the metadata could not be removed from the
+// destination when metadata/ is a symlink whose target is missing, as it
+// does when the destination directory is missing, instead of finding no
+// metadata to remove.
+func TestRemoveSnapshotWarnsOnSymlinkedMetadataWithMissingTarget(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, repos, out := backUpToFileDestination(ctx, t, storeDir)
+	targetPath := moveBehindSymlink(t, storeDir, "metadata")
+	require.NoError(t, os.Rename(targetPath, targetPath+"-unmounted"))
+
+	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.NotContains(t, out.String(),
+		"Removed snapshot metadata from remote storage")
+}
+
+// blobFilesIn returns the paths of the blob files under storeDir/blobs, and
+// fails the test when there are none.
+func blobFilesIn(t *testing.T, storeDir string) []string {
+	t.Helper()
+
+	var blobFiles []string
+
+	err := filepath.WalkDir(filepath.Join(storeDir, "blobs"),
+		func(path string, entry fs.DirEntry, err error) error {
+			if err == nil && !entry.IsDir() {
+				blobFiles = append(blobFiles, path)
+			}
+
+			return err
+		})
+	require.NoError(t, err)
+	require.NotEmpty(t, blobFiles)
+
+	return blobFiles
+}
+
+// moveBehindSymlink moves storeDir/dir to a directory elsewhere, puts a
+// symlink to it in its place, and returns where it went.
+func moveBehindSymlink(t *testing.T, storeDir, dir string) string {
+	t.Helper()
+
+	linkPath := filepath.Join(storeDir, dir)
+	targetPath := filepath.Join(t.TempDir(), dir)
+	require.NoError(t, os.Rename(linkPath, targetPath))
+	require.NoError(t, os.Symlink(targetPath, linkPath))
+
+	return targetPath
+}

internal/vaultik/verify.go

@@ -680,7 +680,7 @@ func (v *Vaultik) verifyManifestAgainstDatabase(
 func (v *Vaultik) verifyBlobExistenceFromDB(
 	blobs []snapshot.BlobInfo, result *VerifyResult,
 ) error {
-	log.Info("Verifying blob existence in S3", "blob_count", len(blobs))
+	log.Info("Verifying blob existence in storage", "blob_count", len(blobs))
 
 	for i, blob := range blobs {
 		// The hash is read from the snapshot database, which is not
@@ -700,7 +700,7 @@ func (v *Vaultik) verifyBlobExistenceFromDB(
 			result.Missing++
 			result.MissingSize += blob.CompressedSize
 		case err != nil:
-			return fmt.Errorf("checking blob %s in storage: %w",
+			return fmt.Errorf("could not check blob %s in storage: %w",
 				blob.Hash, err)
 		case stat.Size != blob.CompressedSize:
 			log.Warn("Blob has the wrong size in storage",