fix(image): clean partial mounts and refuse mounted export deletion

hydra/build.py

@@ -58,6 +58,21 @@ sys.exit(code)
 '''
 
 
+def rootfs_retry_guard_script():
+    """Refuse deletion of a still-mounted export, including nested mountpoints."""
+    return r"""mounts=$(findmnt --list --raw --noheadings --output TARGET)
+while IFS= read -r mountpoint; do
+  case "$mountpoint" in
+    "$root"|"$root/"*) echo 'Refusing to remove a mounted rootfs export' >&2; exit 1 ;;
+  esac
+done <<< "$mounts"
+if [ -e "$root" ] || [ -L "$root" ]; then
+  [ ! -L "$root" ] && [ -f "$root/.hydra-export-owned" ] || { echo 'Unowned rootfs output' >&2; exit 1; }
+  rm -rf -- "$root"
+fi
+"""
+
+
 def verify_hash(path, expected):
     digest = hashlib.sha256()
     with Path(path).open('rb') as stream:
@@ -421,12 +436,7 @@ fi'''
         snapshot = image.digest(original)
         assembler = self.assembly(snapshot)
         commit = self.run(['git', '-C', str(self.source_dir), 'rev-parse', 'HEAD'], capture=True).stdout.strip()
-        script = r"""root=/hydra-output/rootfs
+        script = 'root=/hydra-output/rootfs\n' + rootfs_retry_guard_script() + r"""mkdir -p "$root"
-if [ -e "$root" ] || [ -L "$root" ]; then
-  [ ! -L "$root" ] && [ -f "$root/.hydra-export-owned" ] || { echo 'Unowned rootfs output' >&2; exit 1; }
-  rm -rf -- "$root"
-fi
-mkdir -p "$root"
 touch "$root/.hydra-export-owned"
 tar --extract --preserve-permissions --numeric-owner --file /hydra-output/artifacts/builder-rootfs.tar --directory "$root"
 /hydra-source/image/configure-rootfs.sh "$root"

image/configure-rootfs.sh

@@ -7,10 +7,27 @@ fi
 root=$1
 python3 /hydra-source/hydra/image.py configure "$root" --source /hydra-source
 # This is a separate rootfs inside the assembly container, never the host root.
+mounted=()
+cleanup_mounts() {
+    status=$?
+    trap - EXIT
+    cleanup_failed=0
+    for ((index=${#mounted[@]}-1; index>=0; index--)); do
+        umount "${mounted[index]}" || cleanup_failed=1
+    done
+    if ((cleanup_failed)); then
+        echo 'configure-rootfs: temporary mount cleanup failed; refusing success' >&2
+        status=1
+    fi
+    exit "$status"
+}
+trap cleanup_mounts EXIT
 mount --bind /dev "$root/dev"
+mounted+=("$root/dev")
 mount -t proc proc "$root/proc"
+mounted+=("$root/proc")
 mount --bind /sys "$root/sys"
-trap 'umount "$root/sys" "$root/proc" "$root/dev"' EXIT
+mounted+=("$root/sys")
 for group in wheel kvm video render audio; do
     chroot "$root" getent group "$group" >/dev/null || chroot "$root" groupadd -r "$group"
 done

tests/test_image.py

@@ -97,6 +97,58 @@ class ImageTests(unittest.TestCase):
         with self.assertRaises(image.ImageError):
             image.kernel_version(self.root)
 
+    def test_mount_faults_clean_only_successful_bindings(self):
+        source = Path(__file__).resolve().parents[1]
+        tools = Path(self.tmp.name) / 'tools'
+        tools.mkdir()
+        for name in ('python3', 'chroot', 'ln', 'systemctl'):
+            script = tools / name
+            script.write_text('#!/bin/sh\nexit 0\n')
+            script.chmod(0o755)
+        mount = tools / 'mount'
+        mount.write_text('#!/bin/sh\nn=$(cat "$STATE/count" 2>/dev/null || echo 0)\n'
+                         'n=$((n+1)); echo "$n" > "$STATE/count"\n'
+                         'test "$n" != "$FAIL_MOUNT"\n')
+        mount.chmod(0o755)
+        unmount = tools / 'umount'
+        unmount.write_text('#!/bin/sh\nprintf "%s\\n" "$@" >> "$STATE/unmounts"\n'
+                           'test "$FAIL_UNMOUNT" != 1\n')
+        unmount.chmod(0o755)
+        for failed, cleanup_failed in ((1, 0), (2, 0), (3, 0), (0, 0), (0, 1)):
+            with self.subTest(failed=failed, cleanup_failed=cleanup_failed):
+                state = Path(self.tmp.name) / f'state-{failed}-{cleanup_failed}'
+                state.mkdir()
+                env = dict(os.environ, PATH=str(tools) + ':' + os.environ['PATH'],
+                           STATE=str(state), FAIL_MOUNT=str(failed), FAIL_UNMOUNT=str(cleanup_failed))
+                result = subprocess.run(['bash', str(source / 'image/configure-rootfs.sh'), '/hydra-output/rootfs'],
+                                        env=env, capture_output=True, text=True)
+                expected = ['/hydra-output/rootfs/dev', '/hydra-output/rootfs/proc', '/hydra-output/rootfs/sys']
+                expected = list(reversed(expected[:failed - 1 if failed else 3]))
+                actual = (state / 'unmounts').read_text().splitlines() if (state / 'unmounts').exists() else []
+                self.assertEqual(actual, expected)
+                self.assertEqual(result.returncode == 0, not failed and not cleanup_failed)
+
+    def test_retry_refuses_removing_owned_rootfs_with_stale_child_mount(self):
+        from hydra.build import rootfs_retry_guard_script
+        marker = self.file('.hydra-export-owned')
+        preserved = self.file('dev/guest-data')
+        tools = Path(self.tmp.name) / 'guard-tools'
+        tools.mkdir()
+        findmnt = tools / 'findmnt'
+        findmnt.write_text('#!/bin/sh\nprintf "%s\\n" / "$STALE_MOUNT"\n')
+        findmnt.chmod(0o755)
+        env = dict(os.environ, PATH=str(tools) + ':' + os.environ['PATH'], STALE_MOUNT=str(self.root / 'dev'))
+        script = 'set -eu\nroot=' + shlex.quote(str(self.root)) + '\n' + rootfs_retry_guard_script()
+        result = subprocess.run(['bash', '-c', script], env=env, capture_output=True, text=True)
+        self.assertNotEqual(result.returncode, 0)
+        self.assertTrue(marker.is_file())
+        self.assertTrue(preserved.is_file())
+        self.assertIn('mounted', result.stderr)
+        env['STALE_MOUNT'] = str(Path(self.tmp.name) / 'different-rootfs')
+        result = subprocess.run(['bash', '-c', script], env=env, capture_output=True, text=True)
+        self.assertEqual(result.returncode, 0, result.stderr)
+        self.assertFalse(self.root.exists())
+
     def test_export_sanitization_preserves_update_inputs(self):
         preserved = ['var/db/pkg/sys-apps/portage-3/USE', 'var/lib/portage/world',
                      'var/db/repos/gentoo/profiles/repo_name', 'usr/share/openpgp-keys/gentoo-release.asc',