fix(security): decide source containment as text before any filesystem access

Every caller (validate-source, POST /settings, config.json at load) now
refuses a source through one check, sources.source_problem: the form
check first, then GalleryConfig.gallery_root_problem, which compares the
normalised spelling with the allowed folders before anything resolves,
stats or opens the path. A UNC share, the \\?\ and \\.\ namespaces, a
device or any outside folder is refused with no SMB request, no
executor stall and no existence or timing oracle; only then is the path
resolved and contained again, which still catches a symlink out. The
filesystem root and the home directory answer outside-allowed with the
same reason as any outside path, so no caller learns the home path.

Also:
- POST /settings runs the form checks and parses the roots off the loop;
  an unusable source is a coded 400 invalid-source naming its public
  path (as text) and its index.
- ComfyUI folders outside the base directory are path anchors, so their
  sources round-trip and never collide by folder name.
- A per-source scan sends each index with its public path; a stale or
  shifted index is 409 sources-changed and starts nothing.
- A source still watched after monitoring was switched off is
  "stopping".
- Roots are compared as folders, not spellings, so an unchanged legacy
  resend takes no source-change hold.
- Tests: no realpath/stat/lstat of an outside path (native and Windows
  forms with ntpath fakes), the alike answers with /, ~ and a link to ~,
  an event instead of a timer wait, and a teardown that drains jobs
  before the temporary database goes.
This commit is contained in:
Vito Sansevero
2026-10-23 12:08:43 -07:00
parent 07c2717422
commit 3f8c63f16e
8 changed files with 763 additions and 112 deletions
+42 -13
View File
@@ -21,8 +21,10 @@ def _not_found():
MAX_SCAN_ROOTS = 64
_SCOPE_ERROR = 'The scan scope must be "new" or "all"'
_ROOTS_ERROR = (
f"roots must be a non-empty list of up to {MAX_SCAN_ROOTS} source indexes"
f"roots must be a non-empty list of up to {MAX_SCAN_ROOTS} source indexes, "
"with paths naming the public path of each"
)
SOURCES_CHANGED_ERROR = "The sources changed since they were loaded. Reload them."
def _bad_request(message, code=None):
@@ -33,20 +35,29 @@ def _bad_request(message, code=None):
def _scan_roots(body):
"""``(roots, error)``: the sorted, deduplicated source indexes, or None.
"""``(roots, error)``: ``{index: public path}`` of the selected sources, or None.
Absent ``roots`` means every source. When present it is a non-empty list
of at most MAX_SCAN_ROOTS non-negative ints (never bools or floats).
of at most MAX_SCAN_ROOTS non-negative ints (never bools or floats), and
``paths`` the same-length list of the public path the caller read for
each (GET /settings), so a stale index never scans another folder.
"""
if "roots" not in body:
return None, None
raw = body["roots"]
paths = body.get("paths")
if not isinstance(raw, list) or not raw or len(raw) > MAX_SCAN_ROOTS:
return None, _ROOTS_ERROR
for value in raw:
if not isinstance(paths, list) or len(paths) != len(raw):
return None, _ROOTS_ERROR
selected = {}
for value, path in zip(raw, paths):
if isinstance(value, bool) or not isinstance(value, int) or value < 0:
return None, _ROOTS_ERROR
return sorted(set(raw)), None
if not isinstance(path, str) or selected.get(value, path) != path:
return None, _ROOTS_ERROR
selected[value] = path
return dict(sorted(selected.items())), None
async def _scan_request(request):
@@ -98,6 +109,13 @@ def _summary_event(job):
class JobRoutesMixin:
"""Mixin providing the /prompt_manager/jobs endpoints."""
def _roots_still_listed(self, selected, output_dirs):
"""Whether each ``{index: public path}`` still names that source."""
return all(
index < len(output_dirs) and self._public_path(output_dirs[index]) == path
for index, path in selected.items()
)
def _register_job_routes(self, routes):
@routes.get("/prompt_manager/jobs")
async def list_jobs_route(request):
@@ -167,24 +185,35 @@ class JobRoutesMixin:
The body's ``scope`` is "new" (the default: new and unresolved files)
or "all". ``roots``, when given, selects sources by their index in
the effective sources (``_get_all_output_dirs``); an index past them
is a 400 ``unknown-source`` and starts nothing. The job walks the
sources resolved here, so a later source change cannot move it.
the effective sources (``_get_all_output_dirs``), and ``paths`` names
the public path read for each. An index past them, or one whose
source now has another public path (the list changed since it was
read), is a 409 ``sources-changed`` and starts nothing. The job walks
the sources resolved here, so a later source change cannot move it.
"""
scan, error = await _scan_request(request)
if error:
return _bad_request(error)
output_dirs = None
roots = None
if scan["roots"] is not None:
output_dirs = await self._run_in_executor(self._get_all_output_dirs)
if scan["roots"][-1] >= len(output_dirs):
return _bad_request(
"That source is no longer configured. Reload the sources.",
code="unknown-source",
listed = await self._run_in_executor(
self._roots_still_listed, scan["roots"], output_dirs
)
if not listed:
return web.json_response(
{
"success": False,
"code": "sources-changed",
"error": SOURCES_CHANGED_ERROR,
},
status=409,
)
roots = list(scan["roots"])
try:
job = self._start_scan_job(
scan["files"], roots=scan["roots"], output_dirs=output_dirs
scan["files"], roots=roots, output_dirs=output_dirs
)
except JobBusy as busy:
return busy_response(busy)
+81 -19
View File
@@ -65,6 +65,45 @@ def _stat_identity(path):
return (stat.st_dev, stat.st_ino) if stat.st_ino else None
# Refusals that a reconnected drive or folder can lift.
_RECONNECTABLE = frozenset({"not-found", "not-directory"})
def _root_key(root):
"""The directory a configured root names (resolved), else its spelling."""
from ..config import GalleryConfig
try:
return GalleryConfig.resolve_gallery_root(root)
except (OSError, TypeError, ValueError, AttributeError):
return root
def _request_error(message):
return {"code": "invalid-request", "error": message}
def _invalid_source(index, entry, code, reason):
"""The 400 body for the source at ``index`` that cannot be used.
Names it by its public spelling (computed as text: a refused path is
never resolved), so the page can mark that row and say why.
"""
source = sources.public_spelling(entry)
advice = (
"Remove it or reconnect it, then save."
if code in _RECONNECTABLE
else "Remove it, then save."
)
return {
"code": "invalid-source",
"source": source,
"index": index,
"reason": reason,
"error": f"{source} can't be used: {reason}. {advice}",
}
def _section(config, name):
"""``config[name]`` as a fresh dict (a non-object section starts empty)."""
value = config.get(name)
@@ -255,9 +294,15 @@ class SettingsRoutesMixin:
{"success": False, "error": str(ve)}, status=400
)
new_roots, error = self._parse_gallery_roots(data)
# Off the loop: an accepted root is resolved (filesystem access).
new_roots, error = await self._run_in_executor(
self._parse_gallery_roots, data
)
if error:
return web.json_response({"success": False, "error": error}, status=400)
return web.json_response({"success": False, **error}, status=400)
roots_change = new_roots is not None and await self._run_in_executor(
self._roots_change, new_roots
)
published = {}
@@ -268,7 +313,7 @@ class SettingsRoutesMixin:
async with contextlib.AsyncExitStack() as stack:
hold = None
if self._roots_change(new_roots):
if roots_change:
try:
hold = await stack.enter_async_context(
self._jobs.hold(
@@ -344,12 +389,19 @@ class SettingsRoutesMixin:
@staticmethod
def _roots_change(new_roots):
"""Whether a save changes the sources (the legacy page resends them)."""
"""Whether a save changes the sources (blocking: it resolves them).
Compared as directories, not spellings: the legacy page resends the
roots resolved, while config.json may hold them as hand-written
(relative, another case), and an unchanged resend is no change.
"""
from ..config import GalleryConfig
return new_roots is not None and new_roots != list(
GalleryConfig.MONITORING_DIRECTORIES
)
if new_roots is None:
return False
return [_root_key(root) for root in new_roots] != [
_root_key(root) for root in GalleryConfig.MONITORING_DIRECTORIES
]
@staticmethod
def _observer_alive():
@@ -441,15 +493,21 @@ class SettingsRoutesMixin:
Accepts ``gallery_root_paths`` (list, preferred) or the legacy single
``gallery_root_path`` string. Blank entries are dropped.
Each entry passes the same checks as validate-source
(``sources.source_problem``): the form, then containment as text,
before anything is resolved. Blocking (an accepted root is resolved).
Returns:
(roots, error): ``roots`` is the validated list, or ``None`` when
the payload carries no gallery root key; ``error`` is a message
when validation failed (and ``roots`` is then ``None``).
the payload carries no gallery root key; ``error`` is the coded
body of the 400 when validation failed (``roots`` is then None).
An unusable source is ``invalid-source`` naming its public path
and its position in the list.
"""
if "gallery_root_paths" in data:
raw = data["gallery_root_paths"]
if not isinstance(raw, list):
return None, "gallery_root_paths must be a list"
return None, _request_error("gallery_root_paths must be a list")
elif "gallery_root_path" in data:
raw = [data["gallery_root_path"]]
else:
@@ -459,18 +517,15 @@ class SettingsRoutesMixin:
roots = []
seen = set()
for entry in raw:
for index, entry in enumerate(raw):
if not isinstance(entry, str):
return None, "Gallery root paths must be strings"
return None, _request_error("Gallery root paths must be strings")
entry = entry.strip()
if not entry:
continue
ok, reason = GalleryConfig.validate_gallery_root(entry)
if not ok:
return (
None,
f"Invalid gallery root '{os.path.basename(entry)}': {reason}",
)
code, reason = sources.source_problem(entry)
if code is not None:
return None, _invalid_source(index, entry, code, reason)
root = GalleryConfig.resolve_gallery_root(entry)
keys = {("path", _source_identity(root)), ("file", _stat_identity(root))}
keys.discard(("file", None))
@@ -481,11 +536,18 @@ class SettingsRoutesMixin:
return roots, None
def _apply_gallery_roots(self, roots):
"""Install validated gallery roots; returns True when they changed."""
"""Install validated gallery roots; returns True when they changed.
Another spelling of the same directories installs the resolved
spelling and is no change (nothing to restart for).
"""
from ..config import GalleryConfig
if roots == list(GalleryConfig.MONITORING_DIRECTORIES):
return False
if not self._roots_change(roots):
GalleryConfig.MONITORING_DIRECTORIES = roots
return False
GalleryConfig.MONITORING_DIRECTORIES = roots
self._cached_output_dir = None
self.reset_gallery_roots_cache()
+53 -23
View File
@@ -6,15 +6,21 @@
``\\\\?\\`` and ``\\\\.\\`` namespaces, a drive-relative path, an alternate
data stream, a device name or a part Windows would silently trim; on POSIX
no Windows drive or backslash form).
- ``source_problem`` is the one refusal every caller answers (validate-source
and the settings save): the form check, then
``GalleryConfig.gallery_root_problem``, which decides containment as text
before it touches the filesystem.
- ``describe_sources`` (blocking: it resolves paths) reports the configured
sources with their validity, the effective sources with their listing
index and watch state, and the image monitor's state. Every path in it is
a public path (``_public_path``), never an absolute one.
A source's watch state is ``watching`` when the monitor's observer is alive
and scheduled that directory; ``manual`` when monitoring is off or this
process has no monitor; else ``restart_pending`` (a source added since the
monitor started, monitoring switched on, or an observer that stopped).
A source's watch state is ``watching`` when the monitor's observer is alive,
scheduled that directory and monitoring is on; ``stopping`` when it still
watches it but monitoring was switched off (it stops at the next restart);
``manual`` when monitoring is off or this process has no monitor; else
``restart_pending`` (a source added since the monitor started, monitoring
switched on, or an observer that stopped).
"""
import logging
@@ -41,13 +47,7 @@ _COMFYUI_LABELS = (
("get_temp_directory", "ComfyUI temp"),
("get_user_directory", "ComfyUI user"),
)
# Codes validate-source folds into "outside-allowed" (see validate_source).
_OUTSIDE_CODES = frozenset({"filesystem-root", "home-directory", "outside-allowed"})
OUTSIDE_REASON = (
"Gallery root must be inside a ComfyUI directory (output, input, temp, user) "
"or one listed in PROMPT_MANAGER_EXTRA_GALLERY_ROOTS"
)
WATCH_STATES = ("watching", "manual", "restart_pending")
WATCH_STATES = ("watching", "stopping", "manual", "restart_pending")
def source_form_error(text, *, windows=None):
@@ -100,6 +100,45 @@ def _windows_form_error(text):
return None
def source_problem(text):
"""``(code, reason)`` that refuses ``text`` as a source, else ``(None, "")``.
The form check runs first and touches nothing; then containment is
decided as text before any filesystem access (see
``GalleryConfig.gallery_root_problem``). Every caller answers alike.
"""
from ..config import GalleryConfig
code = source_form_error(text)
if code is not None:
return code, FORM_REASONS[code]
return GalleryConfig.gallery_root_problem(text)
def public_spelling(text):
"""How a refused source is named in a response, computed as text only.
Relative to the path anchor it lies under (its public path), else its
last part: the caller sent the path, so naming it reveals nothing, and a
refused path is never resolved to name it.
"""
from ..config import GalleryConfig, _lexical_path, _lexically_inside
text = str(text).strip()
try:
lexical = os.path.normpath(text)
key = _lexical_path(text) # the same length as ``lexical``
if not os.path.isabs(text):
return lexical.replace("\\", "/")
for anchor in GalleryConfig.path_anchors():
if _lexically_inside(key, [anchor]):
rest = lexical[len(anchor.rstrip(os.sep)) :].lstrip(os.sep)
return rest.replace(os.sep, "/") or "."
return os.path.basename(lexical) or lexical
except (TypeError, ValueError):
return ""
def identity(path):
"""The comparison key of a directory: resolved and case-folded (Windows)."""
return os.path.normcase(os.path.realpath(str(path)))
@@ -162,9 +201,9 @@ def monitor_state(monitor):
def watch_state(path, *, present, alive, scheduled_keys, enabled):
"""``watching``, ``manual`` or ``restart_pending`` (see the module docstring)."""
"""One of ``WATCH_STATES`` (see the module docstring)."""
if alive and identity(path) in scheduled_keys:
return "watching"
return "watching" if enabled else "stopping"
if not present or not enabled:
return "manual"
return "restart_pending"
@@ -222,16 +261,7 @@ def validate_source(text):
from ..config import GalleryConfig
from . import _public_path
code = source_form_error(text)
reason = FORM_REASONS.get(code)
if code is None:
code, reason = GalleryConfig.gallery_root_problem(text)
if code in _OUTSIDE_CODES:
# One answer for every path outside the allowed folders: naming
# the filesystem root or the home directory would let a caller
# probe the server's layout. The detail goes to the log only.
logger.info(f"Source refused: {code}")
code, reason = "outside-allowed", OUTSIDE_REASON
code, reason = source_problem(text)
if code is not None:
return {
"valid": False,