Make nested snapshots opt-in; fix snapshot leaks found in audit

Opt-in
------
Nested-dataset snapshot support changes how backups read their source data, so
it is now off by default and gated behind a marker file:

  install.sh --enable-nested-snapshots
  install.sh --disable-nested-snapshots

With neither flag install.sh preserves the current setting, so a routine
`git pull && bash install.sh` can never silently flip it. When disabled,
apply.sh skips the patch entirely and the stock guard remains. uninstall.sh
tears down staging mounts and removes the marker.

Snapshot lifecycle
------------------
zfs.snapshot.delete defaults to recursive=False and stock restic_backup() calls
it with no options. Stock is safe only because its validation means recursive is
never True in the field. Enabling nested datasets makes recursive snapshots real:
the parent then has one child snapshot per descendant dataset (160+ on an Apps
pool), so stock's delete would orphan every child on EVERY successful run.

The patch now owns the lifecycle end to end:

- delete_snapshot_tree() sweeps the parent and all children, and is idempotent
  against stock's finally winning the race once our mounts are released
- on a staging failure the tree is deleted here, because sync.py never completes
  `snapshot, local_path = await create_snapshot(...)` and so its finally deletes
  nothing at all
- the snapshot is recorded in a sidecar file before anything is mounted, so a
  middlewared restart mid-backup cannot orphan it
- a crashed run's snapshot tree is reclaimed on the next run instead of being
  overwritten and leaked

Silent-omission fix
-------------------
The dataset list is now enumerated AFTER the snapshot. Read beforehand it could
miss a dataset created in the gap, which the recursive snapshot would capture but
the staging plan would not -- silently omitting its data. Read afterwards, an
unsnapshotted dataset trips the staging check and fails the run loudly.

Also from the audit
-------------------
- plan_staging scopes by dataset name, so skipped-dataset warnings no longer
  include every mountpoint-less dataset on the box, which buried the ones that
  matter
- staging_root_for rejects "." / ".." components that would escape the staging
  base, and resolves STAGING_BASE at call time rather than freezing it into a
  default argument
- uninstall.sh no longer `rm -rf`s a tree that may still contain live bind
  mounts, and unmounts by path depth rather than string length
- apply_plan takes an injectable isdir; verify_staged drops an unused parameter
- pin the shellcheck action instead of tracking @master

61 tests, ruff and shellcheck clean.

.github/workflows/ci.yml

@@ -24,8 +24,10 @@ jobs:
           done < <(find . -name '*.sh' -not -path './.git/*')
           exit $fail
 
+      # Pinned to a release tag, not @master: a third-party action on a moving
+      # branch runs whatever that branch contains at the time CI fires.
       - name: shellcheck
-        uses: ludeeus/action-shellcheck@master
+        uses: ludeeus/action-shellcheck@2.0.0
         env:
           SHELLCHECK_OPTS: -S warning -e SC1091
 

.gitignore

@@ -4,6 +4,7 @@
 /apply.log.2
 /hook_status.json
 /disabled
+/nested_snapshots_enabled
 
 # Python
 __pycache__/

CHANGELOG.md

@@ -4,7 +4,13 @@
 
 ### Added
 
-- **`snapshot = true` now works on datasets that have child datasets.**
+- **`snapshot = true` now works on datasets that have child datasets** —
+  **opt-in, off by default** (`install.sh --enable-nested-snapshots` /
+  `--disable-nested-snapshots`). It changes how backups read their source data,
+  so it is never enabled implicitly; with neither flag `install.sh` preserves
+  the existing setting, so a `git pull && bash install.sh` cannot silently flip
+  it. When disabled, `apply.sh` skips the patch entirely and the stock guard
+  remains. `uninstall.sh` tears down any staging mounts and removes the marker.
   Stock TrueNAS refuses this with *"This option is only available for datasets
   that have no further nesting"*, which makes the snapshot option unusable for
   the single most common case on any box running Apps — every app is its own
@@ -46,6 +52,24 @@
     `snapshot.py`, then `sync.py`, and only then `crud.py`. A partial failure
     leaves the guard intact and the option merely unavailable — never
     "guard removed, traversal missing".
+  - **The patch owns the whole snapshot lifecycle.** `zfs.snapshot.delete`
+    defaults to `recursive=False` and stock `restic_backup()` calls it with no
+    options. Stock gets away with that only because its validation means
+    `recursive` is never True in the field — but enabling nested datasets makes
+    recursive snapshots real, so the parent now has one child snapshot per
+    descendant dataset (160+ on a typical Apps pool). Relying on stock's delete
+    would therefore orphan every child snapshot **on every successful run**.
+    This patch sweeps the parent *and* all children, is idempotent against
+    stock's `finally` winning the race, records the snapshot in a sidecar file
+    (so a middlewared restart mid-backup cannot orphan it), reclaims the tree
+    left by a crashed run, and deletes the tree when staging fails — where
+    sync.py's own `finally` would otherwise delete nothing at all, because its
+    `snapshot` local never gets assigned.
+  - **The dataset list is enumerated *after* the snapshot, never before.** A
+    list read beforehand can miss a dataset created in the gap: the recursive
+    snapshot would capture it but the staging plan would not, silently omitting
+    its data. Read afterwards, an unsnapshotted dataset trips the staging check
+    and fails the run loudly instead.
   - **Every injected block no-ops** if `_truecloud_nested` is absent.
   - Datasets that cannot contribute to a file tree (`mountpoint=none|legacy`,
     unmounted/locked, encrypted-and-locked) are skipped and **reported** —

README.md

@@ -89,6 +89,18 @@ support and the reason is logged to `apply.log` in your repo root.
 
 ## Nested-dataset snapshots
 
+> **Opt-in, and off by default.** This feature changes how backups read their
+> source data, so it is never enabled implicitly:
+>
+> ```bash
+> bash install.sh --enable-nested-snapshots    # turn it on
+> bash install.sh --disable-nested-snapshots   # turn it back off
+> ```
+>
+> With neither flag, `install.sh` leaves the current setting alone — so a
+> `git pull && bash install.sh` can never silently flip it. The B2/S3 provider
+> patch is unaffected either way.
+
 TrueCloud Backup's **Take Snapshot** option makes restic read from a frozen ZFS
 snapshot instead of live files. Without it the backup reads data *while apps are
 writing to it* — databases get captured mid-write, and an app that rewrites its

install.sh

@@ -23,6 +23,45 @@ VERSION="0.3.0"
 # The directory containing install.sh is the permanent install location.
 PATCH_DIR="$(cd "$(dirname "$0")" && pwd)"
 _HOOK_COMMENT='TrueCloud provider patch (S3/B2)'
+_NESTED_MARKER="$PATCH_DIR/nested_snapshots_enabled"
+
+# ── Options ───────────────────────────────────────────────────────────────────
+# Nested-dataset snapshot support is OPT-IN and off by default. It changes how
+# backups read their source data, so an unattended re-run (e.g. after a
+# `git pull`) must never flip it on or off by itself: with neither flag given,
+# whatever was chosen previously is preserved.
+_nested_choice=""
+
+usage() {
+    cat <<USAGE
+Usage: bash install.sh [options]
+
+Options:
+  --enable-nested-snapshots   Allow the "Take Snapshot" option on datasets that
+                              have child datasets (every pool running Apps).
+                              Stock TrueNAS refuses this; see README. Off by
+                              default because it changes how backups read data.
+  --disable-nested-snapshots  Turn it back off; the stock guard is restored.
+  -h, --help                  Show this help.
+
+With neither flag, the current setting is left unchanged.
+USAGE
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --enable-nested-snapshots)  _nested_choice="on" ;;
+        --disable-nested-snapshots) _nested_choice="off" ;;
+        -h|--help)                  usage; exit 0 ;;
+        *)
+            echo "ERROR: unknown option: $1" >&2
+            echo "" >&2
+            usage >&2
+            exit 1
+            ;;
+    esac
+    shift
+done
 
 if [ ! -f "$PATCH_DIR/patch/apply.sh" ]; then
     echo "ERROR: patch files not found at $PATCH_DIR/patch/" >&2
@@ -104,6 +143,42 @@ if [ -f "$PATCH_DIR/disabled" ]; then
     echo ""
 fi
 
+# ── Nested-dataset snapshot support (opt-in) ──────────────────────────────────
+
+case "$_nested_choice" in
+    on)
+        touch "$_NESTED_MARKER"
+        echo "Nested-dataset snapshots: ENABLED"
+        echo "  The \"Take Snapshot\" option will be allowed on datasets that have"
+        echo "  child datasets. Backups then read from a frozen, complete staging"
+        echo "  tree instead of live files."
+        echo ""
+        echo "  This changes how your backups read their source data. Verify that a"
+        echo "  backup completes AND that its restic snapshot actually contains"
+        echo "  child-dataset data before you rely on it."
+        ;;
+    off)
+        if [ -f "$_NESTED_MARKER" ]; then
+            rm -f "$_NESTED_MARKER"
+            echo "Nested-dataset snapshots: DISABLED (stock guard restored)."
+            echo "  Any task that already has snapshot=true on a nested dataset will"
+            echo "  fail validation on its next edit. Turn the option off on those"
+            echo "  tasks, or re-run with --enable-nested-snapshots."
+        else
+            echo "Nested-dataset snapshots: already disabled."
+        fi
+        ;;
+    *)
+        if [ -f "$_NESTED_MARKER" ]; then
+            echo "Nested-dataset snapshots: enabled (unchanged)."
+        else
+            echo "Nested-dataset snapshots: disabled (default)."
+            echo "  Enable with: bash install.sh --enable-nested-snapshots"
+        fi
+        ;;
+esac
+echo ""
+
 # ── Apply now ─────────────────────────────────────────────────────────────────
 
 echo "Applying patches ..."

patch/apply.sh

@@ -187,14 +187,25 @@ else
     _SYNC_PY="$_MW_DIR/plugins/cloud_backup/sync.py"
     _NESTED_SRC="$PATCH_DIR/patch/truecloud_nested.py"
 
+    # Nested-dataset snapshot support is OPT-IN. It changes how backups read
+    # their source data, so it is never enabled implicitly by a `git pull`.
+    # Enable:  bash install.sh --enable-nested-snapshots
+    # Disable: bash install.sh --disable-nested-snapshots
+    if [ -f "$PATCH_DIR/nested_snapshots_enabled" ]; then
+        _NESTED_ENABLED=1
+    else
+        _NESTED_ENABLED=0
+    fi
+
         # ── patch b2.py + restic.py + nested-snapshot + hook_status.json ────────
         # (single subprocess: PREINIT has a tight timeout budget)
         if "$PYTHON" - "$_B2_PY" "$_RESTIC_PY" "$PATCH_DIR/hook_status.json" \
-                      "$_CLOUD_DIR" "$_SYNC_PY" "$_NESTED_SRC" << 'PYEOF'
+                      "$_CLOUD_DIR" "$_SYNC_PY" "$_NESTED_SRC" "$_NESTED_ENABLED" << 'PYEOF'
 import json, os, shutil, sys, time
 
 b2_path, restic_path, status_path = sys.argv[1], sys.argv[2], sys.argv[3]
 cloud_dir, sync_path, nested_src = sys.argv[4], sys.argv[5], sys.argv[6]
+nested_enabled = sys.argv[7] == "1"
 
 B2_BLOCK = """
 # TRUECLOUD_PATCH — added by truenas-truecloud-patch/patch/apply.sh
@@ -277,25 +288,42 @@ if _tc_nested is not None:
     _tc_orig_create_snapshot = create_snapshot
 
     async def create_snapshot(middleware, path, name="cloud_task-onetime"):
-        # Determine nesting BEFORE delegating. We must never silently fall back
-        # to stock behaviour on a nested path: stock returns the parent's
-        # .zfs/snapshot/ path, where children are invisible, which is exactly
-        # the near-empty backup this feature exists to prevent. If we cannot
-        # tell, we raise -- loud failure beats a backup that lies.
-        datasets = await middleware.call("zfs.dataset.query", [["type", "=", "FILESYSTEM"]])
-        dataset, nested = get_dataset_recursive(datasets, path)
-
         # Stock takes the (already recursive) snapshot; we only replace the PATH.
         snapshot, snap_path = await _tc_orig_create_snapshot(middleware, path, name)
 
-        if not nested:
-            return snapshot, snap_path  # no children: stock behaviour, untouched
+        _logger = getattr(middleware, "logger", None)
+        try:
+            # Enumerate datasets AFTER the snapshot, never before. The snapshot is
+            # the point-in-time truth; a list read beforehand could miss a dataset
+            # created in the gap, which the recursive snapshot WOULD capture but
+            # our staging plan would not -- silently omitting it from the backup.
+            # Read afterwards, an unsnapshotted dataset instead trips the isdir()
+            # check in plan_staging and fails the run loudly. Loud beats silent.
+            datasets = await middleware.call(
+                "zfs.dataset.query", [["type", "=", "FILESYSTEM"]]
+            )
+            dataset, nested = get_dataset_recursive(datasets, path)
+
+            if not nested:
+                # No children: stock behaviour, untouched. Stock's `finally` owns
+                # the snapshot from here (its non-recursive delete is correct,
+                # because a non-nested snapshot has no children).
+                return snapshot, snap_path
+
+            staging_root = await _tc_nested.stage_nested(
+                middleware, path, snapshot,
+                dataset["name"], dataset["properties"]["mountpoint"]["value"],
+                name, datasets, logger=_logger,
+            )
+        except Exception:
+            # The snapshot exists, but this exception means sync.py never completes
+            # `snapshot, local_path = await create_snapshot(...)`, so its local
+            # `snapshot` stays None and its `finally` deletes NOTHING. Sweep the
+            # tree ourselves or leak the parent plus one snapshot per descendant
+            # dataset (160+ here) on every failed run.
+            await _tc_nested.delete_snapshot_tree(middleware, snapshot, logger=_logger)
+            raise
 
-        staging_root = await _tc_nested.stage_nested(
-            middleware, path, snapshot,
-            dataset["properties"]["mountpoint"]["value"], name,
-            logger=getattr(middleware, "logger", None),
-        )
         return snapshot, staging_root
 
     create_snapshot._truecloud_patched = True
@@ -401,29 +429,34 @@ else:
 # guard LAST. If anything fails partway, the guard is still in place and the
 # option stays unavailable -- we never expose "guard removed, traversal missing".
 nested_detail = ''
-try:
-    snapshot_py = os.path.join(cloud_dir, 'snapshot.py')
-    crud_py = os.path.join(cloud_dir, 'crud.py')
-    nested_dst = os.path.join(cloud_dir, '_truecloud_nested.py')
+if not nested_enabled:
+    nested_detail = 'disabled (opt-in; enable with: install.sh --enable-nested-snapshots)'
+    print('INFO: Nested-dataset snapshot support is disabled (opt-in feature).')
+    print('INFO: Enable with: bash install.sh --enable-nested-snapshots')
+else:
+    try:
+        snapshot_py = os.path.join(cloud_dir, 'snapshot.py')
+        crud_py = os.path.join(cloud_dir, 'crud.py')
+        nested_dst = os.path.join(cloud_dir, '_truecloud_nested.py')
 
-    missing = [p for p in (snapshot_py, crud_py, sync_path, nested_src) if not os.path.exists(p)]
-    if missing:
-        raise FileNotFoundError('missing: ' + ', '.join(missing))
+        missing = [p for p in (snapshot_py, crud_py, sync_path, nested_src) if not os.path.exists(p)]
+        if missing:
+            raise FileNotFoundError('missing: ' + ', '.join(missing))
 
-    shutil.copyfile(nested_src, nested_dst)   # 1. traversal implementation
-    patch_file(snapshot_py, SNAPSHOT_BLOCK)   # 2. build the staging tree
-    patch_file(sync_path, SYNC_BLOCK)         # 3. tear it down afterwards
-    patch_file(crud_py, CRUD_BLOCK)           # 4. ONLY NOW allow nested tasks
+        shutil.copyfile(nested_src, nested_dst)   # 1. traversal implementation
+        patch_file(snapshot_py, SNAPSHOT_BLOCK)   # 2. build the staging tree
+        patch_file(sync_path, SYNC_BLOCK)         # 3. tear it down afterwards
+        patch_file(crud_py, CRUD_BLOCK)           # 4. ONLY NOW allow nested tasks
 
-    nested_ok = True
-    nested_detail = 'nested-dataset snapshots enabled (staging tree)'
-    print(f'OK: Installed nested-snapshot support → {nested_dst}')
-    print(f'OK: Patched snapshot.py, sync.py, crud.py → {cloud_dir}')
-except Exception as e:
-    nested_detail = f'not applied: {e}'
-    print(f'WARNING: Failed to apply nested-snapshot patch: {e}')
-    print('WARNING: Stock nesting guard remains; snapshot option stays unavailable')
-    print('WARNING: for nested datasets. Existing backups are unaffected.')
+        nested_ok = True
+        nested_detail = 'nested-dataset snapshots enabled (staging tree)'
+        print(f'OK: Installed nested-snapshot support → {nested_dst}')
+        print(f'OK: Patched snapshot.py, sync.py, crud.py → {cloud_dir}')
+    except Exception as e:
+        nested_detail = f'not applied: {e}'
+        print(f'WARNING: Failed to apply nested-snapshot patch: {e}')
+        print('WARNING: Stock nesting guard remains; snapshot option stays unavailable')
+        print('WARNING: for nested datasets. Existing backups are unaffected.')
 
 patches = {
     'middlewared.rclone.remote.b2': {

patch/truecloud_nested.py

@@ -1,4 +1,4 @@
-"""Nested-dataset snapshot support for TrueCloud Backup / Cloud Sync.
+"""Nested-dataset snapshot support for TrueCloud Backup.
 
 Why this exists
 ---------------
@@ -12,8 +12,8 @@ already takes a **recursive** ZFS snapshot, but it then points the backup tool
 at the *parent* dataset's ``.zfs/snapshot/<snap>/`` directory -- and ZFS does
 not expose child datasets through a parent's snapshot directory::
 
-    /mnt/Tap/.zfs/snapshot/<snap>/apps/   ->  0 entries   (children invisible)
-    /mnt/Tap/apps/lidarr/config/.zfs/snapshot/<snap>/  ->  the real data
+    /mnt/Tap/.zfs/snapshot/<snap>/apps/                 ->  0 entries
+    /mnt/Tap/apps/lidarr/config/.zfs/snapshot/<snap>/   ->  the real data
 
 So without the guard, the backup tool would walk a near-empty tree, report
 SUCCESS, and upload almost nothing. A backup that lies about succeeding is the
@@ -32,19 +32,24 @@ Cardinal safety rule
 tree. Silently backing up an incomplete tree is precisely the failure this
 feature exists to prevent, and it would be worse than not having the feature.
 
-Notes
------
-* ZFS snapshots are immutable, so a plain ``mount --bind`` is inherently
-  read-only; no remount dance is needed.
-* Bind-mounting ``.zfs/snapshot/<snap>`` pins the snapshot, so ``zfs destroy``
-  of that snapshot returns EBUSY until we unmount. Stock ``restic_backup()``
-  deletes the snapshot in its ``finally``, which therefore logs one benign
-  "Error deleting snapshot ... busy" warning; :func:`cleanup_task` then unmounts
-  and deletes the snapshot for real. See ``patch/apply.sh``.
-* Staging roots live under a stable, per-task path so that the backup tool sees
-  the *same* path every run. Stock's ``.zfs/snapshot/<name>-<timestamp>/`` path
-  changes every run, which defeats restic's parent-snapshot detection; the
-  staging tree is an improvement on that.
+Snapshot lifecycle -- read this before changing anything
+--------------------------------------------------------
+``zfs.snapshot.delete`` defaults to ``recursive=False``, and stock
+``restic_backup()`` calls it with no options. Stock gets away with that because
+its validation means ``recursive`` is never actually True in the field. Enabling
+nested datasets makes recursive snapshots real, so the parent
+(``Tap@snap``) has one child snapshot per descendant dataset (160+ here).
+Deleting only the parent would orphan every child on **every successful run**.
+
+Therefore this module owns the whole lifecycle:
+
+* :func:`delete_snapshot_tree` sweeps the parent *and* every child snapshot, and
+  is idempotent -- it copes with stock's ``finally`` having already removed the
+  parent.
+* The snapshot name is recorded in a sidecar file next to the staging root, not
+  only in memory, so a middlewared restart mid-backup cannot orphan it.
+* Bind-mounting ``.zfs/snapshot/<snap>`` pins the snapshot, so stock's delete
+  fails with EBUSY and logs one benign warning; we unmount and then sweep.
 """
 
 from __future__ import annotations
@@ -54,21 +59,27 @@ import os
 import subprocess
 
 __all__ = [
-    "StagingError",
-    "STAGING_BASE",
     "ACTIVE",
-    "staging_root_for",
-    "plan_staging",
-    "current_mounts_under",
+    "STAGING_BASE",
+    "StagingError",
     "apply_plan",
-    "verify_staged",
+    "cleanup_task",
+    "current_mounts_under",
+    "delete_snapshot_tree",
+    "plan_staging",
+    "sidecar_for",
+    "snapshot_tree_names",
+    "stage_nested",
+    "staging_root_for",
     "teardown",
+    "verify_staged",
 ]
 
 #: Where staging trees are assembled. tmpfs; bind mounts consume no space.
 STAGING_BASE = "/run/truecloud-nested"
 
-#: staging_root -> zfs snapshot name ("pool/ds@snap"), for cleanup.
+#: staging_root -> zfs snapshot name. A cache; the sidecar file is the source of
+#: truth, so that a middlewared restart cannot orphan a snapshot.
 ACTIVE: dict[str, str] = {}
 
 
@@ -76,18 +87,76 @@ class StagingError(Exception):
     """Staging could not produce a complete tree. The backup must not proceed."""
 
 
-def staging_root_for(name: str, base: str = STAGING_BASE) -> str:
-    """Stable staging root for a task name (e.g. ``cloud_backup-5``)."""
-    safe = "".join(c if (c.isalnum() or c in "-_.") else "_" for c in name) or "task"
+# ── pure helpers ──────────────────────────────────────────────────────────────
+
+
+def staging_root_for(name: str, base: str | None = None) -> str:
+    """Stable staging root for a task name (e.g. ``cloud_backup-5``).
+
+    ``base`` defaults to :data:`STAGING_BASE` at CALL time, not at import time --
+    a ``base=STAGING_BASE`` default would freeze the value into the function
+    object and silently ignore any later override.
+    """
+    if base is None:
+        base = STAGING_BASE
+    safe = "".join(c if (c.isalnum() or c in "-_.") else "_" for c in name)
+    # A component of "." or ".." would escape STAGING_BASE once joined.
+    if not safe or safe.strip(".") == "":
+        safe = "task"
     return os.path.join(base, safe)
 
 
+def sidecar_for(staging_root: str) -> str:
+    """Path of the file recording which ZFS snapshot a staging tree pins."""
+    return staging_root + ".snapshot"
+
+
+def _write_sidecar(staging_root: str, snapshot: str) -> None:
+    """Record the pinned snapshot on disk. Blocking; call via run_in_thread."""
+    with contextlib.suppress(OSError):
+        os.makedirs(os.path.dirname(staging_root), exist_ok=True)
+        with open(sidecar_for(staging_root), "w", encoding="utf-8") as fh:
+            fh.write(snapshot)
+
+
+def _read_sidecar(staging_root: str) -> str | None:
+    """The snapshot a previous run recorded here, if any."""
+    try:
+        with open(sidecar_for(staging_root), encoding="utf-8") as fh:
+            return fh.read().strip() or None
+    except OSError:
+        return None
+
+
+def _remove_sidecar(staging_root: str) -> None:
+    with contextlib.suppress(OSError):
+        os.unlink(sidecar_for(staging_root))
+
+
 def _depth(path: str) -> int:
     return len([p for p in path.split("/") if p])
 
 
-def plan_staging(base_mountpoint, path, snapshot_name, datasets, staging_root,
-                 isdir=os.path.isdir):
+def snapshot_tree_names(snapshot: str, all_names) -> list[str]:
+    """Every snapshot produced by ``zfs snapshot -r <dataset>@<snap>``.
+
+    That is the parent plus one per descendant dataset, all sharing the same
+    name after the ``@``. Pure, so the sweep logic is testable without ZFS.
+    """
+    dataset, _, snapname = snapshot.partition("@")
+    if not snapname:
+        return []
+    parent = f"{dataset}@{snapname}"
+    prefix = dataset + "/"
+    suffix = "@" + snapname
+    return [
+        n for n in all_names
+        if n == parent or (n.startswith(prefix) and n.endswith(suffix))
+    ]
+
+
+def plan_staging(base_dataset, base_mountpoint, path, snapshot_name, datasets,
+                 staging_root, isdir=os.path.isdir):
     """Compute the bind-mount plan for staging a nested tree. Pure function.
 
     ``datasets`` is a list of dicts shaped like ``zfs.dataset.query`` results:
@@ -96,9 +165,12 @@ def plan_staging(base_mountpoint, path, snapshot_name, datasets, staging_root,
 
     Returns ``(mounts, skipped)`` where ``mounts`` is an ordered list of
     ``(source, target)`` pairs (parents before children) and ``skipped`` is a
-    list of ``(dataset_name, reason)``.
+    list of ``(dataset_name, reason)`` covering only datasets that are *in
+    scope* -- i.e. descendants of ``base_dataset``. Datasets elsewhere on the
+    system are ignored silently; reporting them would bury the ones that matter.
 
-    Raises StagingError if a descendant holds data we would silently omit.
+    Raises StagingError if an in-scope descendant holds data we would otherwise
+    silently omit.
     """
     def snapdir(mountpoint):
         return os.path.join(mountpoint, ".zfs", "snapshot", snapshot_name)
@@ -113,20 +185,30 @@ def plan_staging(base_mountpoint, path, snapshot_name, datasets, staging_root,
     mounts = [(root_src, staging_root)]
     skipped = []
 
-    prefix = path.rstrip("/") + "/"
+    ds_prefix = base_dataset.rstrip("/") + "/"
+    path_prefix = path.rstrip("/") + "/"
+
     for ds in datasets:
+        name = ds.get("name", "")
+        # Scope by DATASET NAME, not mountpoint: a dataset with no mountpoint
+        # cannot be scoped by path, and scoping by path first would drag in
+        # every mountpoint-less dataset on the box (all of Tank/.system/*, ...).
+        if not name.startswith(ds_prefix):
+            continue
+
         props = ds.get("properties", {})
         mp = props.get("mountpoint", {}).get("value", "")
-        name = ds.get("name", "?")
 
         if not mp or mp in ("none", "legacy", "-"):
             skipped.append((name, f"mountpoint is {mp or 'unset'}"))
             continue
-        if not mp.startswith(prefix):
-            continue  # not a descendant of the backup path
 
-        mounted = props.get("mounted", {}).get("value", "yes")
-        if mounted == "no":
+        if not mp.startswith(path_prefix):
+            # A descendant dataset mounted outside the backed-up path is
+            # genuinely not part of this tree. Not an omission.
+            continue
+
+        if props.get("mounted", {}).get("value", "yes") == "no":
             # An unmounted (e.g. locked/encrypted) dataset contributes nothing to
             # the live tree either, so skipping matches stock semantics -- but it
             # is a real gap and must be visible, never silent.
@@ -142,8 +224,7 @@ def plan_staging(base_mountpoint, path, snapshot_name, datasets, staging_root,
                 f"refusing to back up an incomplete tree"
             )
 
-        target = os.path.join(staging_root, os.path.relpath(mp, path))
-        mounts.append((src, target))
+        mounts.append((src, os.path.join(staging_root, os.path.relpath(mp, path))))
 
     # Parents before children, so each mountpoint exists before we mount onto it.
     mounts.sort(key=lambda m: _depth(m[1]))
@@ -168,11 +249,14 @@ def current_mounts_under(root, mounts_file="/proc/self/mounts"):
     return found
 
 
+# ── mount / unmount ───────────────────────────────────────────────────────────
+
+
 def _run(cmd):
     return subprocess.run(cmd, capture_output=True, text=True, check=False)
 
 
-def apply_plan(mounts, runner=_run):
+def apply_plan(mounts, runner=_run, isdir=os.path.isdir):
     """Execute the bind-mount plan. Blocking; call via ``run_in_thread``.
 
     Raises StagingError on the first failure, after rolling back what was
@@ -186,7 +270,7 @@ def apply_plan(mounts, runner=_run):
     try:
         os.makedirs(staging_root, exist_ok=True)
         for src, target in mounts:
-            if not os.path.isdir(target):
+            if not isdir(target):
                 # Child mountpoint dirs come from the parent snapshot, which is
                 # read-only -- we cannot mkdir them. Only the root is ours.
                 raise StagingError(f"staging target {target!r} does not exist")
@@ -206,7 +290,7 @@ def apply_plan(mounts, runner=_run):
     return staging_root
 
 
-def verify_staged(mounts, runner=_run, ismount=os.path.ismount, listdir=os.listdir):
+def verify_staged(mounts, ismount=os.path.ismount, listdir=os.listdir):
     """Assert the staged tree is real and complete. Raises StagingError if not.
 
     This is the anti-regression guard: it is what stops this feature from ever
@@ -251,15 +335,61 @@ def teardown(staging_root, runner=_run, mounts_file="/proc/self/mounts"):
 # ── async orchestration (middleware is duck-typed; no middlewared import) ─────
 
 
-async def stage_nested(middleware, path, snapshot, base_mountpoint, task_name, logger=None):
+async def delete_snapshot_tree(middleware, snapshot, logger=None):
+    """Delete the parent snapshot AND every child created by ``zfs snapshot -r``.
+
+    ``zfs.snapshot.delete`` is non-recursive by default and stock calls it with
+    no options, so relying on stock would orphan one snapshot per descendant
+    dataset on every run. Idempotent: tolerates the parent already being gone
+    (stock's ``finally`` may have won the race once our mounts were released).
+    """
+    dataset = snapshot.partition("@")[0]
+
+    try:
+        snaps = await middleware.call(
+            "zfs.snapshot.query", [["name", "^", dataset]], {"select": ["name"]}
+        )
+        # An empty result means the tree is already gone -- delete nothing, and
+        # do not fall back to the parent, which would only log a spurious
+        # "does not exist" warning on every clean run.
+        names = snapshot_tree_names(snapshot, [s["name"] for s in snaps])
+    except Exception as e:  # noqa: BLE001 - fall back to at least the parent
+        if logger:
+            logger.warning(
+                "truecloud-patch: could not enumerate snapshot tree for %s: %r",
+                snapshot, e,
+            )
+        names = [snapshot]
+
+    for name in names:
+        try:
+            await middleware.call("zfs.snapshot.delete", name)
+        except Exception as e:  # noqa: BLE001 - already gone is fine
+            if logger:
+                logger.warning(
+                    "truecloud-patch: could not delete snapshot %s: %r", name, e
+                )
+
+
+async def stage_nested(middleware, path, snapshot, base_dataset, base_mountpoint,
+                       task_name, datasets, logger=None):
     """Build a complete staging tree for `path` from the already-taken `snapshot`.
 
     `snapshot` is a full ZFS snapshot name ("Tap@cloud_backup-5-2026...").
+
+    `datasets` is the FILESYSTEM dataset list. **It MUST have been enumerated
+    AFTER `snapshot` was taken.** A list read beforehand can miss a dataset
+    created in the gap: the recursive snapshot would capture it, but the staging
+    plan would not, and its data would be silently omitted from the backup.
+    Enumerated afterwards, an unsnapshotted dataset instead trips the isdir()
+    check in plan_staging and fails the run loudly.
+
     Returns the staging root to hand to the backup tool.
 
     Raises StagingError if the tree cannot be staged completely -- the caller
     must let that propagate so the backup fails instead of silently uploading a
-    partial tree.
+    partial tree. The caller is responsible for deleting `snapshot` in that case
+    (see SNAPSHOT_BLOCK in apply.sh).
     """
     snapshot_name = snapshot.split("@", 1)[1]
     staging_root = staging_root_for(task_name)
@@ -267,23 +397,45 @@ async def stage_nested(middleware, path, snapshot, base_mountpoint, task_name, l
     # A previous run may have crashed mid-flight; never build on top of that.
     await middleware.run_in_thread(teardown, staging_root)
 
-    datasets = await middleware.call("zfs.dataset.query", [["type", "=", "FILESYSTEM"]])
+    # ...and if it left a sidecar behind, that snapshot tree is still on disk and
+    # nothing else will ever reclaim it. Sweep it before we overwrite the record,
+    # or a single crashed run orphans 160+ snapshots permanently.
+    stale = await middleware.run_in_thread(_read_sidecar, staging_root)
+    if stale and stale != snapshot:
+        if logger:
+            logger.warning(
+                "truecloud-patch: reclaiming snapshot tree from an earlier "
+                "interrupted run: %s", stale,
+            )
+        await delete_snapshot_tree(middleware, stale, logger=logger)
 
-    mounts, skipped = await middleware.run_in_thread(
-        plan_staging, base_mountpoint, path, snapshot_name, datasets, staging_root
-    )
-    if logger:
-        for name, reason in skipped:
-            logger.warning("truecloud-patch: not staging dataset %r: %s", name, reason)
+    # Record the snapshot BEFORE mounting anything, not after. middlewared can
+    # die at any point (this patch even schedules a restart at boot), and the
+    # sidecar is the only thing that survives it -- an in-process dict would take
+    # the sole record of a 160-snapshot tree with it. Writing it after apply_plan
+    # would leave exactly the crash window the sidecar exists to close.
+    await middleware.run_in_thread(_write_sidecar, staging_root, snapshot)
 
-    await middleware.run_in_thread(apply_plan, mounts)
     try:
+        mounts, skipped = await middleware.run_in_thread(
+            plan_staging, base_dataset, base_mountpoint, path, snapshot_name,
+            datasets, staging_root,
+        )
+        if logger:
+            for name, reason in skipped:
+                logger.warning(
+                    "truecloud-patch: not staging dataset %r: %s", name, reason
+                )
+
+        await middleware.run_in_thread(apply_plan, mounts)
         await middleware.run_in_thread(verify_staged, mounts)
     except Exception:
         await middleware.run_in_thread(teardown, staging_root)
+        await middleware.run_in_thread(_remove_sidecar, staging_root)
         raise
 
     ACTIVE[staging_root] = snapshot
+
     if logger:
         logger.info(
             "truecloud-patch: staged %d dataset(s) from %s at %s",
@@ -296,13 +448,14 @@ async def cleanup_task(middleware, task_name, logger=None):
     """Tear down a task's staging tree and delete the snapshot it pinned.
 
     Safe to call unconditionally: a no-op when the task was never staged.
-
-    Stock `restic_backup()` deletes the snapshot in its own `finally`, which
-    fails with EBUSY while our bind mounts pin it (it logs a warning and moves
-    on). We unmount here and then delete the snapshot for real.
     """
     staging_root = staging_root_for(task_name)
+    sidecar = sidecar_for(staging_root)
+
     snapshot = ACTIVE.pop(staging_root, None)
+    if snapshot is None:
+        # Sidecar survives a middlewared restart; ACTIVE does not.
+        snapshot = _read_sidecar(staging_root)
 
     if snapshot is None and not os.path.isdir(staging_root):
         return  # never staged; nothing to do
@@ -313,10 +466,7 @@ async def cleanup_task(middleware, task_name, logger=None):
             logger.warning("truecloud-patch: staging teardown: %s", err)
 
     if snapshot is not None:
-        try:
-            await middleware.call("zfs.snapshot.delete", snapshot)
-        except Exception as e:  # noqa: BLE001 - cleanup must never mask the real error
-            if logger:
-                logger.warning(
-                    "truecloud-patch: could not delete snapshot %s: %r", snapshot, e
-                )
+        await delete_snapshot_tree(middleware, snapshot, logger=logger)
+
+    with contextlib.suppress(OSError):
+        os.unlink(sidecar)

tests/test_apply_blocks.py

@@ -78,11 +78,55 @@ def test_nested_blocks_degrade_safely_without_the_module(name):
     assert "if _tc_nested is not None:" in block
 
 
+class TestSnapshotLeak:
+    """zfs.snapshot.delete is non-recursive and stock calls it with no options.
+
+    A recursive snapshot has one child per descendant dataset (160+ here), so
+    every path that creates one must also sweep the whole tree.
+    """
+
+    def test_staging_failure_deletes_the_snapshot_tree(self):
+        # On a staging failure, sync.py's `snapshot, local_path = await
+        # create_snapshot(...)` never completes, so its local `snapshot` stays
+        # None and its finally deletes nothing. We must sweep it ourselves.
+        block = extract_blocks()["SNAPSHOT_BLOCK"]
+        assert "except Exception:" in block
+        assert "delete_snapshot_tree" in block
+        assert "raise" in block
+
+    def test_sync_block_cleans_up_on_every_path(self):
+        block = extract_blocks()["SYNC_BLOCK"]
+        assert "finally:" in block
+        assert "cleanup_task" in block
+
+
 def test_crud_block_is_scoped_to_cloud_backup():
     # cloudsync has no staging teardown wired in, so its guard must stay.
     assert '!= "cloud_backup"' in extract_blocks()["CRUD_BLOCK"]
 
 
+class TestOptIn:
+    """Nested-snapshot support must be opt-in and must never self-enable."""
+
+    def test_heredoc_gates_on_the_opt_in_flag(self):
+        src = heredoc_source()
+        assert "nested_enabled = sys.argv[7]" in src
+        assert "if not nested_enabled:" in src
+
+    def test_apply_sh_reads_the_marker_file(self):
+        with open(APPLY_SH, encoding="utf-8") as fh:
+            sh = fh.read()
+        assert 'if [ -f "$PATCH_DIR/nested_snapshots_enabled" ]' in sh
+        assert '"$_NESTED_ENABLED"' in sh
+
+    def test_patching_is_skipped_entirely_when_disabled(self):
+        # The guard-relaxing crud.py patch must be inside the enabled branch.
+        src = heredoc_source()
+        gate = src.index("if not nested_enabled:")
+        crud = src.index("patch_file(crud_py, CRUD_BLOCK)")
+        assert gate < crud, "crud.py patch must sit inside the opt-in branch"
+
+
 def test_guard_is_relaxed_only_after_traversal_is_installed():
     # Ordering in apply.sh is a safety property: copy module -> patch snapshot.py
     # -> patch sync.py -> patch crud.py. crud.py (which unlocks the feature) must

tests/test_truecloud_nested.py

@@ -1,11 +1,18 @@
 """Tests for nested-dataset snapshot staging.
 
-The cardinal rule under test: a tree that cannot be staged completely must fail
-LOUDLY. A silently-incomplete backup is the exact failure that stock TrueNAS's
-"no further nesting" guard exists to prevent, and it is the one regression this
-feature must never introduce.
+Two rules are under test above all else:
+
+1. A tree that cannot be staged completely must fail LOUDLY. A silently
+   incomplete backup is the exact failure that stock TrueNAS's "no further
+   nesting" guard exists to prevent.
+
+2. Every snapshot we cause to exist must be cleaned up. ``zfs.snapshot.delete``
+   is non-recursive by default and stock calls it with no options, so a
+   recursive snapshot would otherwise orphan one snapshot per descendant dataset
+   on EVERY run.
 """
 
+import asyncio
 import os
 import sys
 
@@ -14,10 +21,15 @@ import pytest
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "patch"))
 
 from truecloud_nested import (  # noqa: E402
+    ACTIVE,
     StagingError,
     apply_plan,
+    cleanup_task,
     current_mounts_under,
+    delete_snapshot_tree,
     plan_staging,
+    sidecar_for,
+    snapshot_tree_names,
     staging_root_for,
     teardown,
     verify_staged,
@@ -47,22 +59,24 @@ DATASETS = [
     ds("Tap/apps/immich/pgdata", "/mnt/Tap/apps/immich/pgdata"),
 ]
 
-ALL_DIRS_EXIST = lambda _p: True  # noqa: E731
+
+def yes(_path):
+    return True
+
+
+def plan(datasets=DATASETS, base_dataset="Tap", base_mp="/mnt/Tap",
+         path="/mnt/Tap", isdir=yes):
+    return plan_staging(base_dataset, base_mp, path, SNAP, datasets, ROOT, isdir=isdir)
 
 
 class TestPlanStaging:
     def test_stages_every_descendant_dataset(self):
-        mounts, skipped = plan_staging(
-            "/mnt/Tap", "/mnt/Tap", SNAP, DATASETS, ROOT, isdir=ALL_DIRS_EXIST
-        )
+        mounts, skipped = plan()
         assert skipped == []
-
-        # Root + all 5 descendants. The base dataset itself is the root, not a
-        # descendant, so it must not be double-mounted.
-        assert len(mounts) == 6
+        assert len(mounts) == 6  # root + 5 descendants
         assert mounts[0] == (f"/mnt/Tap/.zfs/snapshot/{SNAP}", ROOT)
 
-        by_target = dict((t, s) for s, t in mounts)
+        by_target = {t: s for s, t in mounts}
         assert by_target[f"{ROOT}/apps"] == f"/mnt/Tap/apps/.zfs/snapshot/{SNAP}"
         assert by_target[f"{ROOT}/apps/immich/pgdata"] == (
             f"/mnt/Tap/apps/immich/pgdata/.zfs/snapshot/{SNAP}"
@@ -71,72 +85,282 @@ class TestPlanStaging:
     def test_parents_are_mounted_before_children(self):
         # A child's mountpoint dir only exists inside its parent's snapshot, so
         # mounting a child first would fail.
-        mounts, _ = plan_staging(
-            "/mnt/Tap", "/mnt/Tap", SNAP, DATASETS, ROOT, isdir=ALL_DIRS_EXIST
-        )
+        mounts, _ = plan()
         seen = set()
         for _src, target in mounts:
-            parent = os.path.dirname(target)
             if target != ROOT:
-                assert parent in seen or parent == ROOT, f"{target} mounted before {parent}"
+                assert os.path.dirname(target) in seen
             seen.add(target)
 
     def test_backup_path_below_dataset_root(self):
-        mounts, _ = plan_staging(
-            "/mnt/Tap", "/mnt/Tap/apps", SNAP, DATASETS, ROOT, isdir=ALL_DIRS_EXIST
-        )
-        # Root source is the *subdirectory* inside the base dataset's snapshot.
-        assert mounts[0] == (f"/mnt/Tap/.zfs/snapshot/{SNAP}/apps", ROOT)
+        mounts, _ = plan(base_dataset="Tap/apps", base_mp="/mnt/Tap/apps",
+                         path="/mnt/Tap/apps")
+        assert mounts[0] == (f"/mnt/Tap/apps/.zfs/snapshot/{SNAP}", ROOT)
         targets = [t for _s, t in mounts]
-        assert f"{ROOT}/lidarr" in targets       # relative to /mnt/Tap/apps
+        assert f"{ROOT}/lidarr" in targets
         assert f"{ROOT}/apps/lidarr" not in targets
 
     def test_base_dataset_is_not_a_descendant_of_itself(self):
-        mounts, _ = plan_staging(
-            "/mnt/Tap", "/mnt/Tap", SNAP, [ds("Tap", "/mnt/Tap")], ROOT, isdir=ALL_DIRS_EXIST
-        )
-        assert len(mounts) == 1  # just the root
+        mounts, _ = plan(datasets=[ds("Tap", "/mnt/Tap")])
+        assert len(mounts) == 1
 
 
-class TestSkipping:
-    @pytest.mark.parametrize("mp", ["none", "legacy", "-", ""])
-    def test_unmountable_mountpoints_are_skipped_and_reported(self, mp):
-        datasets = DATASETS + [ds("Tap/weird", mp)]
-        mounts, skipped = plan_staging(
-            "/mnt/Tap", "/mnt/Tap", SNAP, datasets, ROOT, isdir=ALL_DIRS_EXIST
-        )
+class TestScoping:
+    def test_unrelated_datasets_are_ignored_silently(self):
+        # Regression: scoping by mountpoint first dragged in every
+        # mountpoint-less dataset on the box (all of Tank/.system/*), burying the
+        # warnings that actually matter.
+        noisy = DATASETS + [
+            ds("Tank/.system", "none"),
+            ds("Tank/.system/cores", "legacy"),
+            ds("Tank/backups", "/mnt/Tank/backups"),
+        ]
+        mounts, skipped = plan(datasets=noisy)
         assert len(mounts) == 6
-        assert any(name == "Tap/weird" for name, _reason in skipped)
+        assert skipped == [], "datasets outside the base dataset must not be reported"
+
+    def test_in_scope_dataset_without_mountpoint_is_reported(self):
+        datasets = DATASETS + [ds("Tap/apps/weird", "none")]
+        _mounts, skipped = plan(datasets=datasets)
+        assert ("Tap/apps/weird", "mountpoint is none") in skipped
 
     def test_unmounted_dataset_is_skipped_but_never_silently(self):
-        # A locked/encrypted dataset contributes nothing to the live tree either,
-        # so skipping matches stock semantics -- but it MUST be reported.
         datasets = DATASETS + [ds("Tap/apps/vault", "/mnt/Tap/apps/vault", mounted="no")]
-        mounts, skipped = plan_staging(
-            "/mnt/Tap", "/mnt/Tap", SNAP, datasets, ROOT, isdir=ALL_DIRS_EXIST
-        )
+        mounts, skipped = plan(datasets=datasets)
         assert f"{ROOT}/apps/vault" not in [t for _s, t in mounts]
         assert ("Tap/apps/vault", "dataset is not mounted (locked/encrypted?)") in skipped
 
+    def test_descendant_mounted_outside_the_path_is_not_an_omission(self):
+        datasets = DATASETS + [ds("Tap/elsewhere", "/mnt/other")]
+        mounts, skipped = plan(datasets=datasets)
+        assert len(mounts) == 6
+        assert skipped == []
+
 
 class TestSilentOmissionGuard:
     """The whole point of the feature. These are the tests that matter."""
 
-    def test_missing_snapshot_on_descendant_raises(self):
-        # If the recursive snapshot somehow missed a dataset, staging it would
-        # silently omit its data. Refuse rather than upload an incomplete tree.
-        def isdir(path):
-            return "/mnt/Tap/apps/immich/pgdata/" not in path
+    @staticmethod
+    def _missing_pgdata(path):
+        return "/mnt/Tap/apps/immich/pgdata/" not in path
 
+    def test_missing_snapshot_on_descendant_raises(self):
         with pytest.raises(StagingError, match="incomplete tree"):
-            plan_staging("/mnt/Tap", "/mnt/Tap", SNAP, DATASETS, ROOT, isdir=isdir)
+            plan(isdir=self._missing_pgdata)
 
     def test_error_names_the_offending_dataset(self):
-        def isdir(path):
-            return "/mnt/Tap/apps/immich/pgdata/" not in path
-
         with pytest.raises(StagingError, match="Tap/apps/immich/pgdata"):
-            plan_staging("/mnt/Tap", "/mnt/Tap", SNAP, DATASETS, ROOT, isdir=isdir)
+            plan(isdir=self._missing_pgdata)
+
+
+class TestSnapshotTreeNames:
+    """zfs.snapshot.delete is non-recursive; we must sweep children ourselves."""
+
+    ALL = [
+        "Tap@cloud_backup-5-20260712030000",
+        "Tap/apps@cloud_backup-5-20260712030000",
+        "Tap/apps/lidarr/config@cloud_backup-5-20260712030000",
+        "Tap@auto-2026-07-12_03-00",            # unrelated periodic snapshot
+        "Tap/apps@cloud_backup-9-20260712030000",  # another task
+        "Tank/backups@cloud_backup-5-20260712030000",  # different pool
+    ]
+
+    def test_returns_parent_and_all_children(self):
+        got = snapshot_tree_names("Tap@cloud_backup-5-20260712030000", self.ALL)
+        assert set(got) == {
+            "Tap@cloud_backup-5-20260712030000",
+            "Tap/apps@cloud_backup-5-20260712030000",
+            "Tap/apps/lidarr/config@cloud_backup-5-20260712030000",
+        }
+
+    def test_never_touches_periodic_or_other_tasks_or_other_pools(self):
+        got = snapshot_tree_names("Tap@cloud_backup-5-20260712030000", self.ALL)
+        assert "Tap@auto-2026-07-12_03-00" not in got
+        assert "Tap/apps@cloud_backup-9-20260712030000" not in got
+        assert "Tank/backups@cloud_backup-5-20260712030000" not in got
+
+    def test_malformed_snapshot_name_yields_nothing(self):
+        assert snapshot_tree_names("Tap", self.ALL) == []
+
+
+class FakeMiddleware:
+    def __init__(self, snapshots=None):
+        self.snapshots = list(snapshots or [])
+        self.calls = []
+        self.logger = None
+
+    async def call(self, method, *args):
+        self.calls.append((method, args))
+        if method == "zfs.snapshot.query":
+            return [{"name": n} for n in self.snapshots]
+        if method == "zfs.snapshot.delete":
+            if args[0] not in self.snapshots:
+                raise RuntimeError("does not exist")
+            self.snapshots.remove(args[0])
+            return True
+        raise AssertionError(f"unexpected call {method}")
+
+    async def run_in_thread(self, fn, *args):
+        return fn(*args)
+
+
+class TestDeleteSnapshotTree:
+    def test_deletes_parent_and_every_child(self):
+        mw = FakeMiddleware([
+            "Tap@snap", "Tap/apps@snap", "Tap/apps/lidarr@snap", "Tap@keepme",
+        ])
+        asyncio.run(delete_snapshot_tree(mw, "Tap@snap"))
+        assert mw.snapshots == ["Tap@keepme"]
+
+    def test_is_idempotent_when_stock_already_removed_the_parent(self):
+        # Stock's finally can win the race once our mounts are released.
+        mw = FakeMiddleware(["Tap/apps@snap", "Tap/apps/lidarr@snap"])
+        asyncio.run(delete_snapshot_tree(mw, "Tap@snap"))
+        assert mw.snapshots == []
+
+    def test_survives_query_failure_by_deleting_at_least_the_parent(self):
+        class Broken(FakeMiddleware):
+            async def call(self, method, *args):
+                if method == "zfs.snapshot.query":
+                    raise RuntimeError("boom")
+                return await super().call(method, *args)
+
+        mw = Broken(["Tap@snap"])
+        asyncio.run(delete_snapshot_tree(mw, "Tap@snap"))
+        assert mw.snapshots == []
+
+    def test_attempts_no_delete_when_the_tree_is_already_gone(self):
+        # A successful query returning nothing means there is nothing to do.
+        # Falling back to the parent here would log a spurious "does not exist"
+        # warning on every clean run.
+        mw = FakeMiddleware(["Tap@unrelated"])
+        asyncio.run(delete_snapshot_tree(mw, "Tap@snap"))
+        assert [m for m, _a in mw.calls if m == "zfs.snapshot.delete"] == []
+        assert mw.snapshots == ["Tap@unrelated"]
+
+
+class TestStageNestedOrdering:
+    def setup_method(self):
+        ACTIVE.clear()
+
+    def test_sidecar_is_written_before_anything_is_mounted(self, tmp_path, monkeypatch):
+        # middlewared can die at any moment. If the snapshot were recorded only
+        # after apply_plan, a crash in that window would orphan a 160-snapshot
+        # tree -- the precise failure the sidecar exists to prevent.
+        import truecloud_nested as tn
+
+        monkeypatch.setattr(tn, "STAGING_BASE", str(tmp_path))
+        order = []
+
+        class Recorder(FakeMiddleware):
+            async def run_in_thread(self, fn, *args):
+                order.append(fn.__name__)
+                if fn.__name__ == "plan_staging":
+                    return ([("/src", str(tmp_path / "cloud_backup-5"))], [])
+                if fn.__name__ in ("apply_plan", "verify_staged", "teardown"):
+                    return [] if fn.__name__ == "teardown" else True
+                return fn(*args)
+
+        asyncio.run(tn.stage_nested(
+            Recorder(), "/mnt/Tap", "Tap@snap", "Tap", "/mnt/Tap",
+            "cloud_backup-5", DATASETS,
+        ))
+
+        assert order.index("_write_sidecar") < order.index("apply_plan")
+
+    def test_reclaims_the_snapshot_tree_left_by_a_crashed_run(self, tmp_path,
+                                                              monkeypatch):
+        # teardown() reclaims the crashed run's MOUNTS, but nothing else would
+        # ever reclaim its SNAPSHOTS -- and we are about to overwrite the only
+        # record of them. One crash would orphan 160+ snapshots permanently.
+        import truecloud_nested as tn
+
+        monkeypatch.setattr(tn, "STAGING_BASE", str(tmp_path))
+        root = tn.staging_root_for("cloud_backup-5")
+        os.makedirs(os.path.dirname(root), exist_ok=True)
+        with open(sidecar_for(root), "w", encoding="utf-8") as fh:
+            fh.write("Tap@old-crashed-run")
+
+        mw = FakeMiddleware(["Tap@old-crashed-run", "Tap/apps@old-crashed-run"])
+
+        class Stub(FakeMiddleware):
+            def __init__(self, inner):
+                super().__init__()
+                self.inner = inner
+
+            async def call(self, method, *args):
+                return await self.inner.call(method, *args)
+
+            async def run_in_thread(self, fn, *args):
+                if fn.__name__ == "plan_staging":
+                    return ([("/src", root)], [])
+                if fn.__name__ == "teardown":
+                    return []
+                if fn.__name__ in ("apply_plan", "verify_staged"):
+                    return True
+                return fn(*args)
+
+        asyncio.run(tn.stage_nested(
+            Stub(mw), "/mnt/Tap", "Tap@new", "Tap", "/mnt/Tap",
+            "cloud_backup-5", DATASETS,
+        ))
+
+        assert mw.snapshots == [], "the crashed run's snapshot tree must be reclaimed"
+
+    def test_sidecar_is_removed_when_staging_fails(self, tmp_path, monkeypatch):
+        import truecloud_nested as tn
+
+        monkeypatch.setattr(tn, "STAGING_BASE", str(tmp_path))
+        root = tn.staging_root_for("cloud_backup-5")
+
+        class Failing(FakeMiddleware):
+            async def run_in_thread(self, fn, *args):
+                if fn.__name__ == "plan_staging":
+                    raise StagingError("boom")
+                return fn(*args)
+
+        with pytest.raises(StagingError):
+            asyncio.run(tn.stage_nested(
+                Failing(), "/mnt/Tap", "Tap@snap", "Tap", "/mnt/Tap",
+                "cloud_backup-5", DATASETS,
+            ))
+
+        assert not os.path.exists(sidecar_for(root))
+
+
+class TestCleanupTask:
+    def setup_method(self):
+        ACTIVE.clear()
+
+    def test_recovers_snapshot_from_sidecar_after_middlewared_restart(self, tmp_path,
+                                                                      monkeypatch):
+        # ACTIVE is in-process; a restart wipes it. The sidecar is the source of
+        # truth, otherwise the snapshot tree is orphaned forever.
+        import truecloud_nested as tn
+
+        monkeypatch.setattr(tn, "STAGING_BASE", str(tmp_path))
+        root = tn.staging_root_for("cloud_backup-5", base=str(tmp_path))
+        os.makedirs(root, exist_ok=True)
+        with open(sidecar_for(root), "w", encoding="utf-8") as fh:
+            fh.write("Tap@snap")
+
+        ACTIVE.clear()  # simulate the restart
+        mw = FakeMiddleware(["Tap@snap", "Tap/apps@snap"])
+        monkeypatch.setattr(tn, "teardown", lambda *_a, **_k: [])
+
+        asyncio.run(cleanup_task(mw, "cloud_backup-5"))
+
+        assert mw.snapshots == []
+        assert not os.path.exists(sidecar_for(root))
+
+    def test_is_a_noop_when_never_staged(self, tmp_path, monkeypatch):
+        import truecloud_nested as tn
+
+        monkeypatch.setattr(tn, "STAGING_BASE", str(tmp_path / "nope"))
+        mw = FakeMiddleware(["Tap@snap"])
+        asyncio.run(cleanup_task(mw, "cloud_backup-5"))
+        assert mw.calls == []
+        assert mw.snapshots == ["Tap@snap"]
 
 
 class TestVerifyStaged:
@@ -144,22 +368,17 @@ class TestVerifyStaged:
 
     def test_passes_when_every_target_is_mounted_and_root_non_empty(self):
         mounts = [("/src", ROOT), ("/src/a", f"{ROOT}/a")]
-        assert verify_staged(
-            mounts, ismount=lambda p: True, listdir=lambda p: ["apps"]
-        )
+        assert verify_staged(mounts, ismount=lambda p: True, listdir=lambda p: ["apps"])
 
     def test_raises_when_a_target_is_not_actually_mounted(self):
         # This is the case that would produce a silently-empty backup.
         mounts = [("/src", ROOT), ("/src/a", f"{ROOT}/a")]
         with pytest.raises(StagingError, match="not a mountpoint"):
-            verify_staged(
-                mounts, ismount=lambda p: p == ROOT, listdir=lambda p: ["apps"]
-            )
+            verify_staged(mounts, ismount=lambda p: p == ROOT, listdir=lambda p: ["apps"])
 
     def test_raises_when_staging_root_is_empty(self):
-        mounts = [("/src", ROOT)]
         with pytest.raises(StagingError, match="empty"):
-            verify_staged(mounts, ismount=lambda p: True, listdir=lambda p: [])
+            verify_staged([("/src", ROOT)], ismount=lambda p: True, listdir=lambda p: [])
 
     def test_raises_on_empty_plan(self):
         with pytest.raises(StagingError):
@@ -185,25 +404,26 @@ class FakeRunner:
 
 
 class TestApplyPlanRollback:
-    def test_rolls_back_mounts_when_one_fails(self, tmp_path, monkeypatch):
+    def test_rolls_back_mounts_when_one_fails(self, tmp_path):
         # A half-built tree must never reach the backup tool.
         root = str(tmp_path / "root")
         mounts = [("/src", root), ("/src/a", root + "/a"), ("/src/b", root + "/b")]
-        monkeypatch.setattr(os.path, "isdir", lambda p: True)
 
         runner = FakeRunner(fail_on="/src/b")
         with pytest.raises(StagingError, match="bind-mount"):
-            apply_plan(mounts, runner=runner)
+            apply_plan(mounts, runner=runner, isdir=yes)
 
-        umounts = [c for c in runner.calls if c[0] == "umount"]
-        # Everything successfully mounted before the failure is unmounted again.
-        assert [c[-1] for c in umounts] == [root + "/a", root]
+        umounts = [c[-1] for c in runner.calls if c[0] == "umount"]
+        assert umounts == [root + "/a", root]
 
-    def test_raises_when_target_missing(self, tmp_path, monkeypatch):
+    def test_raises_when_target_missing(self, tmp_path):
         root = str(tmp_path / "root")
-        monkeypatch.setattr(os.path, "isdir", lambda p: p == root)
         with pytest.raises(StagingError, match="does not exist"):
-            apply_plan([("/src", root), ("/src/a", root + "/a")], runner=FakeRunner())
+            apply_plan(
+                [("/src", root), ("/src/a", root + "/a")],
+                runner=FakeRunner(),
+                isdir=lambda p: p == root,
+            )
 
 
 class TestTeardown:
@@ -226,7 +446,7 @@ class TestTeardown:
             f"{ROOT}/apps",
             ROOT,
         ]
-        assert "/somewhere/else" not in order  # never touch unrelated mounts
+        assert "/somewhere/else" not in order
 
     def test_is_idempotent_when_nothing_mounted(self, tmp_path):
         mounts_file = tmp_path / "mounts"
@@ -257,13 +477,12 @@ class TestTeardown:
 class TestCurrentMountsUnder:
     def test_matches_only_the_staging_subtree(self, tmp_path):
         mounts_file = tmp_path / "mounts"
-        # "/run/truecloud-nested/cloud_backup-50" must NOT match "cloud_backup-5".
+        # "cloud_backup-50" must NOT match "cloud_backup-5".
         mounts_file.write_text(
             f"tmpfs {ROOT} tmpfs rw 0 0\n"
             "tmpfs /run/truecloud-nested/cloud_backup-50 tmpfs rw 0 0\n"
         )
-        found = current_mounts_under(ROOT, mounts_file=str(mounts_file))
-        assert found == [ROOT]
+        assert current_mounts_under(ROOT, mounts_file=str(mounts_file)) == [ROOT]
 
 
 class TestStagingRootFor:
@@ -271,4 +490,10 @@ class TestStagingRootFor:
         assert staging_root_for("cloud_backup-5") == "/run/truecloud-nested/cloud_backup-5"
 
     def test_sanitises_path_separators(self):
-        assert "/" not in staging_root_for("evil/../../etc").rsplit("/", 1)[-1]
+        assert "/" not in staging_root_for("evil/name").rsplit("/", 1)[-1]
+
+    @pytest.mark.parametrize("name", ["..", ".", "...", "/", ""])
+    def test_dot_components_cannot_escape_the_staging_base(self, name):
+        # os.path.join(BASE, "..") normalises to /run — teardown would rmdir it.
+        root = staging_root_for(name)
+        assert os.path.normpath(root).startswith("/run/truecloud-nested/")

uninstall.sh

@@ -91,6 +91,52 @@ if [ "$_ov_found" -eq 0 ]; then
 fi
 echo ""
 
+# ── Unmount nested-snapshot staging trees ─────────────────────────────────────
+# These bind mounts pin their ZFS snapshots, so they must go before anything
+# tries to destroy those snapshots. Deepest first.
+
+echo "Unmounting nested-snapshot staging trees (if any) ..."
+_stage_found=0
+_stage_failed=0
+# Deepest FIRST, by path depth (slash count) — not string length, which would
+# let a long shallow path jump ahead of a short deep one and leave a child
+# mounted (and its ZFS snapshot pinned).
+while IFS= read -r _mp; do
+    [ -n "$_mp" ] || continue
+    if umount "$_mp" 2>/dev/null || umount -l "$_mp" 2>/dev/null; then
+        echo "  Unmounted: $_mp"
+    else
+        echo "  WARNING: Could not unmount $_mp"
+        _stage_failed=1
+    fi
+    _stage_found=1
+done < <(awk '$2 == "/run/truecloud-nested" || index($2, "/run/truecloud-nested/") == 1 {
+                 n = gsub(/\//, "/", $2); print n, $2
+             }' /proc/self/mounts 2>/dev/null | sort -rn | cut -d' ' -f2-)
+
+if [ "$_stage_found" -eq 0 ]; then
+    echo "  None active."
+fi
+
+# NEVER `rm -rf` here: if an unmount failed, that would recurse *through* a live
+# bind mount into the ZFS snapshot behind it. Remove empty directories only.
+if [ "$_stage_failed" -eq 0 ]; then
+    find /run/truecloud-nested -depth -type d -exec rmdir {} + 2>/dev/null || true
+    rm -f /run/truecloud-nested/*.snapshot 2>/dev/null || true
+    rmdir /run/truecloud-nested 2>/dev/null || true
+else
+    echo "  WARNING: staging mounts remain; leaving /run/truecloud-nested in place."
+    echo "           Unmount them manually, then remove the directory."
+fi
+
+# The opt-in marker lives in the repo dir; remove it so a later re-install
+# starts from the safe default (feature off).
+if [ -f "$PATCH_DIR/nested_snapshots_enabled" ]; then
+    rm -f "$PATCH_DIR/nested_snapshots_enabled"
+    echo "  Removed nested-snapshot opt-in marker."
+fi
+echo ""
+
 if [ "$_restore_failed" -eq 1 ]; then
     echo ""
     echo "ERROR: One or more UI bundle backups could not be restored." >&2