Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
import pytest
|
|
|
|
|
|
|
|
|
|
from api.workspace import list_dir, read_file_content, safe_resolve_ws
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_safe_resolve_blocks_external_symlink_directory(tmp_path):
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "secret.txt").write_text("outside", encoding="utf-8")
|
|
|
|
|
(workspace / "escape").symlink_to(outside)
|
|
|
|
|
|
2026-06-21 08:55:34 +00:00
|
|
|
# The read/list gate still blocks navigation through the escape symlink.
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
with pytest.raises(ValueError, match="Path traversal blocked"):
|
|
|
|
|
safe_resolve_ws(workspace, "escape")
|
|
|
|
|
|
|
|
|
|
with pytest.raises(ValueError, match="Path traversal blocked"):
|
|
|
|
|
list_dir(workspace, "escape")
|
|
|
|
|
|
2026-06-21 08:55:34 +00:00
|
|
|
# The escape symlink is now emitted (display-only) with target_outside_workspace=True.
|
|
|
|
|
entries = {e["name"]: e for e in list_dir(workspace, ".")}
|
|
|
|
|
assert "escape" in entries
|
|
|
|
|
assert entries["escape"]["type"] == "symlink"
|
|
|
|
|
assert entries["escape"]["target_outside_workspace"] is True
|
2026-06-21 09:11:57 +00:00
|
|
|
# #4581 hardening: display-only escape rows are uniformly is_dir=False — the
|
|
|
|
|
# target's real dir/file nature is target-derived metadata we don't disclose
|
|
|
|
|
# (the row is non-navigable regardless).
|
|
|
|
|
assert entries["escape"]["is_dir"] is False
|
|
|
|
|
assert "target" not in entries["escape"]
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_read_file_blocks_external_symlink_file(tmp_path):
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "secret.txt").write_text("outside", encoding="utf-8")
|
|
|
|
|
(workspace / "secret-link.txt").symlink_to(outside / "secret.txt")
|
|
|
|
|
|
2026-06-21 08:55:34 +00:00
|
|
|
# The read gate still blocks reading through the escape symlink.
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
with pytest.raises(ValueError, match="Path traversal blocked"):
|
|
|
|
|
read_file_content(workspace, "secret-link.txt")
|
|
|
|
|
|
2026-06-21 08:55:34 +00:00
|
|
|
# The escape symlink is now emitted (display-only) with target_outside_workspace=True.
|
|
|
|
|
entries = {e["name"]: e for e in list_dir(workspace, ".")}
|
|
|
|
|
assert "secret-link.txt" in entries
|
|
|
|
|
assert entries["secret-link.txt"]["type"] == "symlink"
|
|
|
|
|
assert entries["secret-link.txt"]["target_outside_workspace"] is True
|
|
|
|
|
assert entries["secret-link.txt"]["is_dir"] is False
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_internal_symlink_still_resolves_within_workspace(tmp_path):
|
2026-06-05 19:58:11 +00:00
|
|
|
import api.workspace as w
|
|
|
|
|
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
nested = workspace / "nested"
|
|
|
|
|
nested.mkdir()
|
|
|
|
|
(nested / "inside.txt").write_text("inside", encoding="utf-8")
|
|
|
|
|
(workspace / "inside-link.txt").symlink_to(nested / "inside.txt")
|
|
|
|
|
|
|
|
|
|
resolved = safe_resolve_ws(workspace, "inside-link.txt")
|
|
|
|
|
|
|
|
|
|
assert resolved == (nested / "inside.txt").resolve()
|
|
|
|
|
assert read_file_content(workspace, "inside-link.txt")["content"] == "inside"
|
2026-06-05 19:58:11 +00:00
|
|
|
if not w._DIR_FD_OK:
|
|
|
|
|
pytest.skip("internal symlink listing is platform-dependent without dir_fd")
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
assert "inside-link.txt" in {entry["name"] for entry in list_dir(workspace, ".")}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# ── TOCTOU hardening (#3398): a path that passes safe_resolve_ws() but is then
|
|
|
|
|
# swapped to an external symlink before the open must not read/list/write
|
|
|
|
|
# outside the workspace. The read/list/write paths use a portable anchored
|
|
|
|
|
# openat-walk (openat + O_NOFOLLOW per component, dir_fd where supported). ──
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_read_file_toctou_swap_to_external_symlink_blocked(tmp_path, monkeypatch):
|
|
|
|
|
"""If the resolved path is swapped to an external symlink AFTER the
|
|
|
|
|
safe_resolve_ws() check, read_file_content must refuse, not follow the
|
|
|
|
|
symlink and leak external content."""
|
|
|
|
|
import api.workspace as w
|
2026-06-05 19:58:11 +00:00
|
|
|
if not w._DIR_FD_OK:
|
|
|
|
|
pytest.skip("TOCTOU symlink-swap hardening requires dir_fd support")
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
(workspace / "data.txt").write_text("LEGIT", encoding="utf-8")
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "secret.txt").write_text("SECRET-LEAK", encoding="utf-8")
|
|
|
|
|
|
|
|
|
|
real_resolve = w.safe_resolve_ws
|
|
|
|
|
|
|
|
|
|
def racing_resolve(root, rel):
|
|
|
|
|
p = real_resolve(root, rel)
|
|
|
|
|
if rel == "data.txt":
|
|
|
|
|
try:
|
|
|
|
|
p.unlink()
|
|
|
|
|
except OSError:
|
|
|
|
|
pass
|
|
|
|
|
p.symlink_to(outside / "secret.txt")
|
|
|
|
|
return p
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(w, "safe_resolve_ws", racing_resolve)
|
|
|
|
|
try:
|
|
|
|
|
result = w.read_file_content(workspace, "data.txt")
|
|
|
|
|
assert "SECRET" not in result["content"], "TOCTOU symlink swap leaked external content"
|
|
|
|
|
except (FileNotFoundError, ValueError):
|
|
|
|
|
pass # refused — the correct outcome
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_list_dir_toctou_swap_to_external_symlink_blocked(tmp_path, monkeypatch):
|
|
|
|
|
"""If a checked directory path is swapped to an external symlink after
|
|
|
|
|
safe_resolve_ws(), list_dir must refuse rather than enumerate the external
|
|
|
|
|
directory."""
|
|
|
|
|
import api.workspace as w
|
2026-06-05 19:58:11 +00:00
|
|
|
if not w._DIR_FD_OK:
|
|
|
|
|
pytest.skip("TOCTOU symlink-swap hardening requires dir_fd support")
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
(workspace / "sub").mkdir()
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "secret.txt").write_text("x", encoding="utf-8")
|
|
|
|
|
|
|
|
|
|
real_resolve = w.safe_resolve_ws
|
|
|
|
|
|
|
|
|
|
def racing_resolve(root, rel):
|
|
|
|
|
p = real_resolve(root, rel)
|
|
|
|
|
if rel == "sub":
|
|
|
|
|
try:
|
|
|
|
|
p.rmdir()
|
|
|
|
|
except OSError:
|
|
|
|
|
pass
|
|
|
|
|
p.symlink_to(outside)
|
|
|
|
|
return p
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(w, "safe_resolve_ws", racing_resolve)
|
|
|
|
|
try:
|
|
|
|
|
entries = w.list_dir(workspace, "sub")
|
|
|
|
|
names = {e["name"] for e in entries}
|
|
|
|
|
assert "secret.txt" not in names, "TOCTOU symlink swap leaked external dir listing"
|
|
|
|
|
except (FileNotFoundError, ValueError):
|
|
|
|
|
pass # refused — the correct outcome
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_anchored_create_blocks_symlinked_component(tmp_path):
|
|
|
|
|
"""open_anchored_create_fd must refuse to write through a symlinked path
|
|
|
|
|
component (the upload / archive-extraction write race), landing nothing
|
|
|
|
|
outside the workspace."""
|
2026-06-05 19:58:11 +00:00
|
|
|
import api.workspace as w
|
|
|
|
|
if not w._DIR_FD_OK:
|
|
|
|
|
pytest.skip("anchored symlink-component rejection requires dir_fd support")
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
from api.workspace import open_anchored_create_fd
|
|
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(workspace / "evil").symlink_to(outside) # symlinked intermediate dir
|
|
|
|
|
|
|
|
|
|
with pytest.raises((FileNotFoundError, ValueError, OSError)):
|
|
|
|
|
open_anchored_create_fd(workspace, (workspace / "evil" / "pwned.txt"))
|
|
|
|
|
assert not (outside / "pwned.txt").exists()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_anchored_create_no_fd_leak_on_rejection(tmp_path):
|
|
|
|
|
"""Repeated rejected anchored creates must not leak file descriptors."""
|
|
|
|
|
import os
|
|
|
|
|
|
|
|
|
|
from api.workspace import open_anchored_create_fd
|
|
|
|
|
|
|
|
|
|
if not os.path.isdir("/proc/self/fd"):
|
|
|
|
|
pytest.skip("fd-count check requires /proc/self/fd")
|
|
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(workspace / "evil").symlink_to(outside)
|
|
|
|
|
|
|
|
|
|
before = len(os.listdir("/proc/self/fd"))
|
|
|
|
|
for _ in range(200):
|
|
|
|
|
try:
|
|
|
|
|
open_anchored_create_fd(workspace, (workspace / "evil" / "x.txt"))
|
|
|
|
|
except Exception:
|
|
|
|
|
pass
|
|
|
|
|
after = len(os.listdir("/proc/self/fd"))
|
|
|
|
|
assert after <= before + 2, f"fd leak: before={before} after={after}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_anchored_create_nested_autocreates_dirs(tmp_path):
|
|
|
|
|
"""A normal (non-escaping) nested create works and lands under the workspace."""
|
|
|
|
|
import os
|
|
|
|
|
|
|
|
|
|
from api.workspace import open_anchored_create_fd
|
|
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
fd = open_anchored_create_fd(workspace, workspace / "a" / "b" / "file.txt")
|
|
|
|
|
os.write(fd, b"hello")
|
|
|
|
|
os.close(fd)
|
|
|
|
|
assert (workspace / "a" / "b" / "file.txt").read_text() == "hello"
|
|
|
|
|
|
|
|
|
|
|
2026-06-05 19:58:11 +00:00
|
|
|
def test_rename_anchored_reports_destination_traversal(tmp_path):
|
|
|
|
|
from api.workspace import rename_anchored
|
|
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
source = workspace / "inside.txt"
|
|
|
|
|
source.write_text("inside", encoding="utf-8")
|
|
|
|
|
dest = outside / "outside.txt"
|
|
|
|
|
|
|
|
|
|
with pytest.raises(ValueError) as exc_info:
|
|
|
|
|
rename_anchored(workspace, source, dest)
|
|
|
|
|
assert str(dest) in str(exc_info.value)
|
|
|
|
|
|
|
|
|
|
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
def test_list_read_create_work_on_no_dir_fd_fallback(tmp_path, monkeypatch):
|
|
|
|
|
"""The no-dir_fd portability fallback (Windows path) must still list, read,
|
|
|
|
|
and create within the workspace, and still hide/block external symlinks via
|
|
|
|
|
the static safe_resolve_ws guard — no fd-relative API that would brick on
|
|
|
|
|
platforms without os.supports_dir_fd."""
|
|
|
|
|
import os
|
|
|
|
|
|
|
|
|
|
import api.workspace as w
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(w, "_DIR_FD_OK", False)
|
|
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
(workspace / "a.txt").write_text("hi", encoding="utf-8")
|
|
|
|
|
(workspace / "sub").mkdir()
|
|
|
|
|
(workspace / "internal").symlink_to(workspace / "sub")
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "s.txt").write_text("x", encoding="utf-8")
|
|
|
|
|
(workspace / "escape").symlink_to(outside)
|
|
|
|
|
|
|
|
|
|
names = {e["name"] for e in w.list_dir(workspace, ".")}
|
|
|
|
|
assert "a.txt" in names
|
2026-06-05 19:58:11 +00:00
|
|
|
if w._DIR_FD_OK:
|
|
|
|
|
assert "internal" in names # legit internal symlink listed
|
2026-06-21 08:55:34 +00:00
|
|
|
assert "escape" in names # external symlink emitted (display-only)
|
|
|
|
|
escape_entry = next(e for e in w.list_dir(workspace, ".") if e["name"] == "escape")
|
|
|
|
|
assert escape_entry["target_outside_workspace"] is True
|
Release v0.51.221 — Release GO (stage-p3e — block all workspace symlink escapes + portable TOCTOU hardening [security]) (#3398) (#3451)
* [security] fix(workspace): block all symlink escapes from the selected workspace (#3398, @Hinotoi-agent)
Previously safe_resolve_ws allowed a symlink placed inside a workspace to resolve
to an external host path as long as it wasn't a system dir (/etc, /proc, etc).
But the workspace file API is reachable by LLM agent tool calls (read_file_content),
so an in-workspace symlink to ~/.ssh, ~/.hermes/auth.json (credentials), etc. was a
real read path. Now ANY symlink escape is blocked: safe_resolve_ws resolves and
requires the result stay under the workspace root; list_dir hides escaping symlinks
(they could never be opened anyway); internal symlinks resolving back under the
workspace still work. Updated the upload symlink-target test to accept the new
400 'Path traversal blocked' rejection (was 403) — the invariant (nothing lands
outside the workspace) is unchanged.
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
* docs(changelog): v0.51.221 release header for #3398 symlink-escape security fix
* [security] harden workspace file API against symlink-swap TOCTOU via portable anchored openat-walk (#3398 follow-up)
Codex review of #3398 flagged that safe_resolve_ws() validates a path but
list_dir/read_file_content/upload/extraction then re-open by pathname, leaving a
TOCTOU window: a symlink swapped in AFTER the check could still escape. (This
race pre-existed #3398; closing it here so the containment is complete.)
A first attempt used /proc/self/fd for the post-open containment check, but that
BRICKS workspace browsing on macOS/Windows (no /proc → every read/list rejected).
This version is portable:
- open_anchored_fd(): opens the (already symlink-resolved) target
component-by-component from the workspace root via openat (dir_fd) + O_NOFOLLOW.
Every component must be a real non-symlink entry, so a component swapped to a
symlink mid-flight is refused. No /proc dependency. Used by read_file_content
(read from the fd) and list_dir (enumerate via os.scandir(fd), per-entry
fstatat/readlinkat).
- open_anchored_create_fd(): same anchored walk for writes, creating missing
intermediate dirs with mkdir(dir_fd=) and the leaf with O_CREAT|O_EXCL|
O_NOFOLLOW. Used by the workspace upload write AND archive (zip+tar) member
writes, anchored against the TRUE workspace root (not the mutable extraction
dest_dir, closing Codex's root-swap finding). fd-leak-safe on rejection.
- Portability: gated on os.supports_dir_fd; platforms without it (Windows, where
symlink creation needs admin) fall back to a plain O_NOFOLLOW open/exclusive
create — no new race protection but no regression vs the prior path-based code.
Legit in-workspace symlinks still resolve and read/list fine (safe_resolve_ws
collapses them to a real in-workspace path, which the anchored walk then opens).
Verified: the swap-race leaks external content against the old path-based read
and is blocked here; macOS-class symlinked-root workspaces work; no fd leak over
300 rejected creates. Adds TOCTOU + anchored-create regression tests.
* [security] close 3 more #3398 TOCTOU gaps from Codex r3: root-swap, pre-create mkdir, Windows list_dir fallback
Codex round-3 review found three residual issues in the anchored openat-walk:
1. (CORE) The workspace ROOT itself could be swapped to a symlink after
resolve() but before the root os.open() — add _O_NOFOLLOW to the root open in
open_anchored_fd() and open_anchored_create_fd() (and make_anchored_dir()), so
a raced root symlink is refused. Verified: root-swap race now blocked.
2. (SILENT) Upload/extraction still did pathname Path.mkdir() AFTER the
containment check, so a raced symlink component could make the server create
dirs outside the workspace before the anchored file create rejected. Removed
the redundant member_path.parent.mkdir() calls (open_anchored_create_fd
already creates intermediates via anchored mkdirat) and replaced the two
base-dir mkdirs (upload target dir + archive extraction root) with a new
make_anchored_dir() that walks from the true workspace root via
openat+O_NOFOLLOW + mkdir(dir_fd=).
3. (CORE) list_dir() unconditionally used os.scandir(fd)/os.stat(dir_fd=)/
os.readlink(dir_fd=), which would brick workspace browsing on platforms
without os.supports_dir_fd (Windows). Split list_dir() into a _DIR_FD_OK
anchored branch and a path-based fallback branch (prior behaviour) sharing one
_process() entry builder. open_anchored_create_fd()'s Windows fallback now also
creates parent dirs.
Adds regression tests: no-dir_fd fallback (list+read+create+symlink filtering)
and the root-swap race. All prior TOCTOU + anchored-create tests still green.
* fix(workspace): portable symlink-loop filtering in list_dir via follow-stat ELOOP
CI on Python 3.13 caught test_mutual_symlink_loop_filtered failing: a mutual
symlink loop (a->b->a) was NOT filtered from the listing. Root cause: the new
readlink-based cycle detection relied on (target_resolved / raw_link).resolve()
RAISING on a loop, but Path.resolve() loop handling differs by Python version
(3.11 raises RuntimeError, 3.13 can return a path), so the loop slipped through
on 3.13.
Fix: compute a version-independent 'reachable' flag per symlink via
os.stat(..., follow_symlinks=True) — the syscall reliably returns ELOOP for
mutual/self loops and ENOENT for broken targets on every platform/version. A
symlink whose follow-stat raises can never be opened, so list_dir filters it.
Applied in both the dir_fd-anchored branch (fd-relative stat) and the Windows
path-based fallback branch. Mutual loop now filtered on all versions.
---------
Co-authored-by: nesquena-hermes <[email protected]>
Co-authored-by: Hinotoi-agent <Hinotoi-agent@users.noreply.github.com>
2026-06-02 23:56:30 +00:00
|
|
|
assert w.read_file_content(workspace, "a.txt")["content"] == "hi"
|
|
|
|
|
|
|
|
|
|
fd = w.open_anchored_create_fd(workspace, workspace / "new" / "f.txt")
|
|
|
|
|
os.write(fd, b"ok")
|
|
|
|
|
os.close(fd)
|
|
|
|
|
assert (workspace / "new" / "f.txt").read_text() == "ok"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_read_blocked_when_workspace_root_raced_to_symlink(tmp_path):
|
|
|
|
|
"""If the workspace root itself is swapped to an external symlink after
|
|
|
|
|
resolve() but before the anchored open, read_file_content must refuse
|
|
|
|
|
(O_NOFOLLOW on the root open), not follow it and leak external content."""
|
|
|
|
|
import os
|
|
|
|
|
import shutil
|
|
|
|
|
|
|
|
|
|
import api.workspace as w
|
|
|
|
|
|
|
|
|
|
if not w._DIR_FD_OK:
|
|
|
|
|
pytest.skip("anchored root-open race only applies on dir_fd platforms")
|
|
|
|
|
|
|
|
|
|
outside = tmp_path / "evil"
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "f.txt").write_text("SECRET-LEAK", encoding="utf-8")
|
|
|
|
|
wsroot = tmp_path / "wsroot"
|
|
|
|
|
wsroot.mkdir()
|
|
|
|
|
(wsroot / "f.txt").write_text("LEGIT", encoding="utf-8")
|
|
|
|
|
|
|
|
|
|
real_open = os.open
|
|
|
|
|
state = {"swapped": False}
|
|
|
|
|
|
|
|
|
|
def racing_open(path, *args, **kwargs):
|
|
|
|
|
if (not state["swapped"]) and "dir_fd" not in kwargs and str(path) == str(wsroot.resolve()):
|
|
|
|
|
state["swapped"] = True
|
|
|
|
|
shutil.rmtree(str(wsroot))
|
|
|
|
|
os.symlink(str(outside), str(wsroot))
|
|
|
|
|
return real_open(path, *args, **kwargs)
|
|
|
|
|
|
|
|
|
|
os.open = racing_open
|
|
|
|
|
try:
|
|
|
|
|
try:
|
|
|
|
|
result = w.read_file_content(wsroot, "f.txt")
|
|
|
|
|
assert "SECRET" not in result["content"], "root-swap race leaked external content"
|
|
|
|
|
except (FileNotFoundError, ValueError, NotADirectoryError, OSError):
|
|
|
|
|
pass # refused — correct
|
|
|
|
|
finally:
|
|
|
|
|
os.open = real_open
|
2026-06-21 08:55:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
# ── #4510: escape-target symlinks are now emitted as display-only rows ──────
|
|
|
|
|
# The escape filter was widened (not removed): symlinks whose resolved target
|
|
|
|
|
# sits outside the workspace root are now emitted with
|
|
|
|
|
# target_outside_workspace=True instead of being silently dropped. The
|
|
|
|
|
# read/list gate (safe_resolve_ws / open_anchored_fd) is unchanged and still
|
|
|
|
|
# blocks navigation through them. ──────────────────────────────────────────
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_list_dir_in_workspace_symlink_shape(tmp_path):
|
|
|
|
|
"""In-workspace symlinks emit type='symlink' with target_outside_workspace=False."""
|
|
|
|
|
import api.workspace as w
|
|
|
|
|
|
|
|
|
|
if not w._DIR_FD_OK:
|
|
|
|
|
pytest.skip("symlink listing requires dir_fd support")
|
|
|
|
|
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
(workspace / "data.txt").write_text("hello", encoding="utf-8")
|
|
|
|
|
(workspace / "link.txt").symlink_to(workspace / "data.txt")
|
|
|
|
|
|
|
|
|
|
entries = {e["name"]: e for e in w.list_dir(workspace, ".")}
|
|
|
|
|
assert "link.txt" in entries
|
|
|
|
|
assert entries["link.txt"]["type"] == "symlink"
|
|
|
|
|
assert entries["link.txt"]["target_outside_workspace"] is False
|
|
|
|
|
assert entries["link.txt"]["is_dir"] is False
|
|
|
|
|
assert entries["link.txt"]["target"] == str((workspace / "data.txt").resolve())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_list_dir_outside_workspace_symlink_emitted_with_flag(tmp_path):
|
|
|
|
|
"""Escape-target symlinks are emitted with target_outside_workspace=True."""
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "file.txt").write_text("external", encoding="utf-8")
|
|
|
|
|
(workspace / "ext-link.txt").symlink_to(outside / "file.txt")
|
|
|
|
|
|
|
|
|
|
entries = {e["name"]: e for e in list_dir(workspace, ".")}
|
|
|
|
|
assert "ext-link.txt" in entries
|
|
|
|
|
assert entries["ext-link.txt"]["type"] == "symlink"
|
|
|
|
|
assert entries["ext-link.txt"]["target_outside_workspace"] is True
|
|
|
|
|
assert entries["ext-link.txt"]["is_dir"] is False
|
2026-06-21 09:11:57 +00:00
|
|
|
# #4581 hardening: a display-only escape-target row must NOT disclose where it
|
|
|
|
|
# points — no resolved outside path, no target-derived size, no target-derived
|
|
|
|
|
# metadata. Only the link name/path + the display-only flag are emitted.
|
|
|
|
|
assert "target" not in entries["ext-link.txt"]
|
|
|
|
|
assert "size" not in entries["ext-link.txt"]
|
2026-06-21 08:55:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_list_dir_external_symlink_blocked_system_path_unchanged(tmp_path):
|
|
|
|
|
"""Symlinks to blocked system paths (/etc, /usr) are still filtered out."""
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
(workspace / "etc-link").symlink_to("/etc")
|
|
|
|
|
(workspace / "usr-link").symlink_to("/usr")
|
|
|
|
|
|
|
|
|
|
names = {e["name"] for e in list_dir(workspace, ".")}
|
|
|
|
|
assert "etc-link" not in names
|
|
|
|
|
assert "usr-link" not in names
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_list_dir_escape_symlink_read_still_blocked(tmp_path):
|
|
|
|
|
"""Listing shows the escape symlink (display-only) but read_file_content
|
|
|
|
|
on the same target still raises ValueError — proving the read gate is intact."""
|
|
|
|
|
workspace = tmp_path / "workspace"
|
|
|
|
|
outside = tmp_path / "outside"
|
|
|
|
|
workspace.mkdir()
|
|
|
|
|
outside.mkdir()
|
|
|
|
|
(outside / "secret.txt").write_text("secret", encoding="utf-8")
|
|
|
|
|
(workspace / "escape.txt").symlink_to(outside / "secret.txt")
|
|
|
|
|
|
|
|
|
|
# Listing emits the entry with target_outside_workspace=True
|
|
|
|
|
entries = {e["name"]: e for e in list_dir(workspace, ".")}
|
|
|
|
|
assert "escape.txt" in entries
|
|
|
|
|
assert entries["escape.txt"]["target_outside_workspace"] is True
|
|
|
|
|
|
|
|
|
|
# But reading through it is still blocked
|
|
|
|
|
with pytest.raises(ValueError, match="Path traversal blocked"):
|
|
|
|
|
read_file_content(workspace, "escape.txt")
|