space: the diff's clock runs inside sudo, so zfs is killed too

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

PLAN.md

@@ -356,10 +356,12 @@ something long in one, switch away and back, refresh — it's all still there.
   snapshots, the scan records the newest snapshot of each dataset at start (`bases`), and the
   next one runs `zfs diff -H -F <base>` per dataset: only the directories it names are
   read, their ancestors kept but looked into, everything else kept wholesale without a
-  stat. The diffs run side by side and one is given four minutes (its process group is
-  killed, sudo and zfs both), after which that dataset is walked by mtime instead — on
-  `tank`, `immich` changes constantly and its diff ran past twenty minutes the first time,
-  longer than walking it. Elsewhere (ext4, or a dataset with no snapshot) a directory whose mtime — to the
+  stat. The diffs run side by side and one is given three minutes — `timeout -s KILL`
+  inside sudo, since Debian's sudo gives the command its own session and a process-group
+  kill from outside never reached zfs — after which that dataset is walked by mtime
+  instead. On `tank` a diff is metadata-bound like the walk: 8 s to 3 min per dataset for
+  a handful of changes, and `immich`, which changes constantly, ran past twenty minutes
+  the first time. A refresh of `tank` took 4 min 19 s against 7 min for the full scan. Elsewhere (ext4, or a dataset with no snapshot) a directory whose mtime — to the
   nanosecond — is unchanged keeps its list and only its kept files are stat'ed again, since
   a file can grow without its directory noticing. A **cancelled** scan keeps what it walked,
   marks the folders it never got to (`partial`, "not yet read" in the tree) and **Resume**

internal/space/manager.go

@@ -571,16 +571,19 @@ func newestSnapshot(ctx context.Context, dataset string) string {
 // zfsDiff lists what changed in a dataset since a snapshot: every changed, added, removed
 // or renamed path, with the directory whose listing that touches.
 func zfsDiff(ctx context.Context, snapshot, mount string) ([]change, error) {
-	// a diff slower than this is not worth waiting for: the walk by mtime is bounded
-	ctx, cancel := context.WithTimeout(ctx, 4*time.Minute)
+	// A diff slower than this is not worth waiting for: the walk by mtime is bounded. The
+	// clock runs inside sudo — `timeout` is zfs's parent and kills it as root — because sudo
+	// on Debian gives the command a session of its own, out of reach of our process group;
+	// the context is only a backstop.
+	const patience = 3 * time.Minute
+	ctx, cancel := context.WithTimeout(ctx, patience+30*time.Second)
 	defer cancel()
-	args := priv.Wrap("zfs", "diff", "-H", "-F", snapshot)
+	args := priv.Wrap("timeout", "-s", "KILL", strconv.Itoa(int(patience.Seconds())), "zfs", "diff", "-H", "-F", snapshot)
 	cmd := exec.CommandContext(ctx, args[0], args[1:]...)
-	// on timeout the whole process group goes, not just sudo: zfs itself would otherwise run
-	// on holding the pipe, and Wait would sit there with it
 	cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
 	cmd.Cancel = func() error { return syscall.Kill(-cmd.Process.Pid, syscall.SIGKILL) }
 	cmd.WaitDelay = 5 * time.Second
+	t0 := time.Now()
 	var stderr strings.Builder
 	cmd.Stderr = &stderr
 	out, err := cmd.StdoutPipe()
@@ -609,8 +612,8 @@ func zfsDiff(ctx context.Context, snapshot, mount string) ([]change, error) {
 	}
 	if err := cmd.Wait(); err != nil {
 		msg := strings.TrimSpace(stderr.String())
-		if errors.Is(ctx.Err(), context.DeadlineExceeded) {
-			msg = "gave up after 4 min"
+		if errors.Is(ctx.Err(), context.DeadlineExceeded) || time.Since(t0) >= patience {
+			msg = "gave up after " + patience.String()
 		}
 		if msg == "" {
 			msg = err.Error()