Review of the previous head reproduced a check-then-use window in the
parent-chain walk: the link was judged by name (fstatat) and then opened
by name (openat), with nothing tying the two to the same entry. An
LD_PRELOAD shim that swaps the link for one pointing elsewhere the moment
the judgement returns made both earlier heads follow the swapped link,
including one owned by a foreign uid under root; the post-follow target
check only bounds where such a swap can land.
The link is now never opened by name after it has been judged. On Linux
an O_PATH|O_NOFOLLOW descriptor pins the inode, fstat on that descriptor
makes the ownership decision and readlinkat(fd, "") reads the target text
from the same inode; on other systems the link is stat'ed by name before
and after the readlinkat and both must be the same inode with the same
owner. The walk then opens the text from the parent directory, so a
relative text resolves against the link's own directory exactly as the
kernel resolves it, and the target check runs on what that opens. The
readlinkat form is used rather than /proc/self/fd so the walk does not
depend on /proc being mounted. What is still done by path is resolving
the text itself, which the target check bounds; the comment says so.
Test: relative, ../, absolute and chained link texts resolve from the
link's directory with the process CWD moved elsewhere. The race itself
is exercised out of tree with the harness from the review thread.
Refs #1722
Signed-off-by: TheStranjer <791672+TheStranjer@users.noreply.github.com>
Review of the previous head found the global rule unsound for one caller:
cbm_mkdir_p also creates <repo_path>/.codebase-memory inside the checkout
being indexed, and git creates symlinks owned by whoever cloned, so a
repository shipping `.codebase-memory -> ~/.ssh` would have been followed
by the artifact writer. "User-owned" says nothing about "user-intended"
once the path root is a checkout.
cbm_mkdir_p keeps the strict walk (root-owned links only). The new
cbm_mkdir_p_ex(path, mode, policy) with CBM_MKDIR_FOLLOW_OWNED also
follows a symlink owned by the invoking user, and only callers whose
paths are rooted in the user's own configuration opt in: the installer's
agent roots, skills, instruction files and hook directories, the
JSON-like config editor's parent creation, the cache root, <cache>/logs,
the UI config directory and the DB parent. The artifact directory inside
the checkout and the ~/.local/bin creations stay strict; the activation
transaction re-walks the latter with its own no-follow policy anyway.
What a follow lands on is now checked too: the opened target must be a
directory owned by root or the invoking user and not world-writable
unless sticky, so a link swapped between the fstatat and the openat
cannot steer the walk into a directory an attacker controls or one where
anyone can pre-plant entries. This applies to root-owned links as well;
macOS /private/tmp is root-owned and sticky and still passes. Refusal
restores the first openat's errno.
Tests: the plain walk refuses an own link that the opted-in walk
follows; targets that are world-writable without the sticky bit are
refused; as root, a foreign-owned target and a user-owned link are both
refused, each followed by restoring ownership and walking again; the
identical `.codebase-memory -> <user-owned dir>` link is refused by the
plain walk and followed once opted in; cbm_artifact_export refuses a
checked-in artifact-dir link and leaves its target empty; the installer
test gains the euid-0 leg with home, link and target owned by a non-root
uid. Under euid 0 the tests demote their links to a non-root owner, since
root-owned links are trusted by both walks.
Refs #1722
Signed-off-by: TheStranjer <791672+TheStranjer@users.noreply.github.com>
`install` failed every agent-config write under a config directory that
is a symlink the user themself created (`~/.config/opencode ->
/mnt/.../opencode/config`, the shape dotfile managers and configs kept on
another volume produce), reporting an opaque
`error: agent_config agent=OpenCode op=mcp_install ...` for each file.
cbm_mkdir_p walks the parent chain with O_NOFOLLOW and followed only
root-owned links. The refusal exists so the installer never writes through
a redirection created by someone with different authority: a link planted
by another account in a shared ancestor, or a user-owned link followed by
a privileged install. Neither applies to a link owned by the invoking
account. A symlink's owner is its creator, so such a link was made by the
user or by root, and following it cannot reach anything the user could not
already write directly. It is now followed. A link owned by any other
account stays refused, a privileged install (euid 0) still refuses
user-owned links, and the O_NOFOLLOW walk is unchanged so the
validate-then-write race stays closed. The rule rests only on POSIX
symlink ownership, so it holds on macOS and the BSDs as on Linux; it is
the rule the Linux kernel's fs.protected_symlinks applies, and the trust
the activation transaction already extends to ancestor directories. The
Windows reparse-point walk is a separate path and is unchanged.
The first openat's errno is preserved when a link is refused so the
caller's ENOENT check never reads fstatat's.
Tests: cbm_mkdir_p through an own symlink and through a dangling one;
the installer contract test now asserts own links are followed and, when
the suite runs as root, that a foreign-owned link is still refused.
Refs #1722
Signed-off-by: TheStranjer <791672+TheStranjer@users.noreply.github.com>
atoi and atol answer 0 for text they cannot read, and 0 is a real setting at
three places in this project. So a typo, a trailing unit such as "30s", or a
stray space silently chose a value nobody asked for, and nothing on screen
said the setting had been dropped.
src/mcp/index_supervisor.c CBM_INDEX_WORKER_TIMEOUT_S
src/cli/hook_augment.c CBM_HOOK_DEADLINE_MS
src/mcp/mcp.c CBM_INDEX_MAX_RESTARTS
CBM_HOOK_DEADLINE_MS was the worst of the three. atoi answered 0, 0 is below
HA_DEADLINE_MIN_MS, and the clamp then handed back 50 ms -- the SHORTEST
deadline the setting allows, for a setting whose only purpose is to give the
hook more room. The comment above that function records a hunt for hook runs
that never finished, 0 of 24 real sessions, which is the exact symptom a
silently-shortened deadline produces.
CBM_INDEX_MAX_RESTARTS lost twice. A typo kept the default of 100, and
CBM_INDEX_MAX_RESTARTS=0 -- which reads as "do not restart" to anybody who
sets it -- also kept 100. The setting did the opposite of the request.
CBM_INDEX_WORKER_TIMEOUT_S fell through to the 15-minute default, so a test
set to give up after 30 seconds hung for 15 minutes with nothing to explain
why.
The fix adds one helper rather than three copies of the same check:
bool cbm_env_long(const char *name, long *out);
It answers true only when the variable is set, is not empty, and reads
cleanly from its first character to its last. It holds no policy -- no
minimum, no maximum, no default -- because the three sites disagree on all
three, and a helper that guessed would be wrong at two of them. The shape is
the one src/main.c:1104 already uses: an end pointer, errno, and a check that
nothing was left over. It also refuses a leading blank, which strtol would
otherwise step over, so " 5" is a slip rather than the number 5.
Each site keeps its own rule:
worker timeout an unreadable value keeps the 15-minute default AND logs
the value it dropped
restart cap 0 now means no restarts; an unreadable value keeps 100
AND logs the value it dropped
hook deadline an unreadable value now yields HA_DEADLINE_DEFAULT_MS,
not the floor. This one stays silent on purpose: the file
includes no log header and writes no stderr, because its
output is hook protocol.
The restart-cap parse was lifted out of a very large function into a named
index_restart_cap(), so it can be read and reached on its own.
Four tests come with the change. The two that pin the user-visible behaviour
were seen failing before the fix:
FAIL tests/test_cli.c:402: ms == 50, expected HOOK_DEADLINE_DEFAULT == 2000
(with "unreadable value \"abc\" gave 50 ms" printed above it)
FAIL tests/test_cli.c:434: cbm_index_restart_cap_for_testing() == 100,
expected 0 == 0
After the fix, TEST_SUITES="platform cli mcp" reports 518 passed, 2 failed.
The full suite reports 7635 passed, 2 failed. Both failures are in
tests/test_cli.c (lines 1826 and 6802), print "error: one or more agent
cleanup operations failed", and reproduce on a clean tree without this
change -- they depend on the coding agents installed on the machine.
make -f Makefile.cbm lint-ci passes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Joshua Richter <jrichter5781@gmail.com>
Daemon lifecycle and Windows correctness, verified on a real Windows 11
ARM64 VM through the maintained test-infrastructure/vm drivers, plus the
macOS and Linux arm64 suites and the container lint gate.
Daemon lifecycle:
- daemon start/stop/status subcommands. `daemon start` launches a
PERMANENT daemon (spawn shape is byte-exact argv; survives idle
periods and session ends) and reports an already-active daemon
instead of failing. Permanence is honored at every stop latch:
last-committed-client disconnect, host initial-client window,
coordinator release, and application final-session close — a
permanent daemon also keeps admitting new sessions after its last
one closes.
- daemon stop refuses while sessions are active and lists the blocking
peers (pid/role) that must finish first; an idle daemon drains
through the activation-shutdown machinery with the ACK ordered after
connection interrupts. A second stop is idempotent. The wire ops are
no-cohort first-frame requests with peer fingerprint authentication,
so stop/status never conflict with an exact-build admission gate.
- One-shot CLI commands now execute through the daemon (index workers
keep their local supervised path). A cold CLI run that had to spawn a
temporary daemon prints a hint that `daemon start` removes the
per-command startup tax; a warm daemon is recycled silently.
- Hooks are connect-only fail-open: with no daemon present the hook
emits a visible, rate-limited notice (Claude-dialect systemMessage
plus stderr for other dialects) and always exits 0 — augmentation is
never allowed to block the caller's tool use.
- Version skew: a newer-build client automatically drains an
older-build permanent daemon (strict semantic-version triples only;
dev builds never auto-drain) and the build-conflict message names
`cbm daemon stop` as the manual escape hatch.
Windows IPC/runtime (real-VM verified):
- ipc(win): persistent pending overlapped ConnectNamedPipe. The accept
path used to destroy its listening pipe instance on every 20 ms poll
timeout; a client attaching in the teardown window was severed or left
on an orphaned pipe object whose HELLO no server handle could ever
read, absorbing the connect until the client's own timeout expired.
The pending connect now survives poll timeouts and nothing is
destroyed while a client could be attaching.
- ipc(win): drain-before-close for final responses. Closing a named-pipe
server handle can discard a just-sent response before the peer reads
it (POSIX stream sockets never lose buffered data on close). A bounded
cbm_daemon_ipc_connection_drain (read-until-peer-EOF; no-op on POSIX,
immediate on interrupted connections) now precedes close in
runtime_worker_finish and runtime_reject_inline, so hello-conflict,
capacity and disconnect acknowledgements reliably reach the peer.
- runtime: CLOSE_INTENT wire frame. A Windows named-pipe client has no
transport half-close, so close_begin now announces departure with an
explicit frame (ordered after APPLICATION_CANCEL, before the local
interrupt); the server releases the client's admission on receipt
instead of waiting for the handle to close. Admission-drop timing is
now identical to POSIX shutdown() semantics on every platform.
- runtime(win): client close cancellation. close_begin serializes with
request publication under the send lock, best-effort sends the active
token's APPLICATION_CANCEL frame, then interrupts local I/O; the
server cancels MCP/subprocess work promptly. Contract tests accept
both correct outcomes (interrupted transport or decoded CANCELLED).
- runtime: activation acknowledgement ordering. The activation ACK is
the requester's license to act on "snapshotted and draining", so every
connection interrupt is now initiated before the ACK is sent; a
session could previously get one more request serviced after the
requester observed the ACK.
- service(win): deadline-bounded private-file prepare. The conflict-log
prepare retry loop (100 x Sleep(2), which rounds up to the ~16 ms
timer granularity) burned ~1.6 s against permanently obstructed paths,
stalling hello rejections past the client's timeout. The retry budget
is now a 250 ms deadline; transient share collisions still retry.
- subprocess(win): cmd.exe /C payload encoder quotes metacharacters
correctly (root cause of the git-on-Windows failure cluster).
- watcher: SHA-256 buffer sizing (CBM_SZ_64 -> CBM_SZ_128) and a native
Windows stop/unwatch cancellation test with exact-image verification.
- httpd: send_all writes in bounded 64 KiB slices. A single giant
nonblocking send() on Windows is absorbed wholesale into AFD kernel
buffering regardless of SO_SNDBUF, so send deadlines and interrupts
could never engage against a slow peer (and the full payload was
pinned in nonpaged pool). Slicing restores a deterministic
backpressure point; a test hook pins SO_SNDBUF for the deadline and
interrupt tests.
- ui/http: shutdown lifecycle — interrupt checks, response-wide send
deadline, explicit connection states, refusal to free a server while
a listener-owned connection is active.
Windows long-path support:
- Central path-aware wide conversion (canonicalize via GetFullPathNameW
and prepend the extended-length prefix for absolute paths >=240) at
the compat chokepoints (cbm_fopen/compat_fs/mkstemp/mkdtemp), sqlite
store opens, and the daemon build-fingerprint/log paths. Deep managed
installs (a 64-hex generation directory routinely exceeds MAX_PATH)
now index, stage and activate correctly.
- activation transaction: its own file APIs and the component-walking
ancestry validators now operate in the extended-length namespace;
the launcher path is canonicalized (and prefixed when deep) once at
entry so every downstream exact-string comparison stays
form-consistent.
- Executable self-resolution uses the wide APIs (GetModuleFileNameW,
GetFileAttributesW) so non-ASCII install paths survive argv[0]
resolution.
Windows launcher install/uninstall transaction:
- FileRenameInfoEx names are NUL-terminated in an over-allocated
buffer. FileNameLength governs per the contract, but filter drivers
read FileName as NUL-terminated and appended adjacent heap bytes to
created names — a flaky, garbage-suffixed rename target. Both the CLI
and the launcher rename helpers are fixed.
- Uninstall retires state via rename-aside (.cbm ->
.cbm-retired-v1-<tag>-<pid>) with the retired tag shortened to 16 hex
chars so the bare rename target stays under the FileRenameInfoEx
NT-conversion ceiling at guard depths; 64 bits still uniquely
identify the generation.
- When the running launcher's mapped generation backings pin .cbm
against rename, the backings are relocated to activation-<pid>-N
.retired tombstones beside the install (a mapped image may be renamed,
never deleted; the launcher's liveness-guarded sweep reclaims stale
tombstones). Every relocation is recorded, and a FAILED uninstall
reverses the moves after restoring .cbm — via MoveFileExW with
extended-length paths on both arguments, since the deep generation
target is beyond the handle-based rename's bare-path reach — so a
restored install keeps its generation backings and stays runnable.
- After a committed uninstall the retired tree's backings are relocated
out so the tree is shallow enough for the detached cleanup's rd, and
the cleanup's working directory strips the extended-length prefix
(CreateProcessW lpCurrentDirectory silently ignores prefixed paths).
- Files created under Administrators-default-owner directories
(CopyFileW destinations, CREATE_NEW tombstones, probe directories)
are explicitly owner-stamped so the exact-owner validators hold on
runner images; guard fixtures stamp hand-built trees the same way.
Diagnostics, tests and infra:
- diagnostics: discovery is now an always-delivered JSON control record
(new cbm_log_control) that survives CBM_LOG_LEVEL suppression and
paths containing spaces; placement honors $TMPDIR with /tmp fallback
via a diagnostics-local helper; the soak parser reads the JSON record;
documented in docs/CONFIGURATION.md. Red-first coverage for suppressed
log levels, TMPDIR-with-spaces, and native Windows output-contract
assertions.
- tests(win): daemon_ipc/daemon_frontend fixtures now build endpoint
parents with production-shaped ancestry (LocalAppData on Windows, via
th_secure_runtime_parent_new) — the runtime ancestry validation
correctly refuses temp roots whose ancestors grant mutation rights to
Authenticated Users (C:/msys64/tmp, GitHub-runner work dirs) — and
drive the documented startup-owner publication flow before reading
generation-bound endpoint addresses. This turns the 26 Windows
failures previously visible in CI's full-test job green without
weakening any validation.
- tests(win): the launcher guard covers the full permanent-launcher
contract including failed-uninstall restore and immediate reinstall
after uninstall; new daemon lifecycle and reworked hook-augment
guards run the start/recycle/stop flow end to end.
- tests: CBM_SKIP_PERF is now actually consumed by the test runner
(it was set by CI but never read, so perf suites ran everywhere);
four throughput/bench suites are classified as perf, the heavy
store_arch suite moved to the slow-timeout tier, and two
wall-clock-sensitive assertions were rewritten as invariant checks
with coarse hang-detector backstops.
- build/test infra: build-dir safety contract, UI dev-proxy security
contract, soak daemon-recovery contract, path-safety helper, the
Windows VM worktree-sync contract wired into scripts/test.sh, and
vm/win.sh guards building its clean embedded-UI product in an
isolated BUILD_DIR so it cannot clobber the incremental test build.
provision-windows.sh now installs Node.js for the guards UI build.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
The CBM_WORKERS env tests called bare setenv/unsetenv, which are
POSIX-only and undeclared on the MSYS2 Windows test build. Switch to the
cbm_setenv/cbm_unsetenv compat wrappers (a no-op #define to setenv on
POSIX) and include foundation/compat.h. Fixes the test-windows build.
Inside a container, sysconf(_SC_NPROCESSORS_ONLN) and sysinfo() report
the host's CPU count and RAM, not the cgroup's effective quota — so
cbm_default_worker_count over-provisions workers and the mmap budget can
exceed the cgroup memory cap, driving OOMKills (#363).
detect_system_linux now reads the cgroup limits:
- cbm_detect_cgroup_cpus: cgroup v2 cpu.max, then v1 cpu.cfs_quota_us /
cpu.cfs_period_us → ceil(quota/period); -1 when unlimited.
- cbm_detect_cgroup_mem: cgroup v2 memory.max, then v1
memory.limit_in_bytes; 0 when unlimited or the v1 near-ULLONG_MAX
sentinel.
- Effective value is min(cgroup, host), guarding against mis-mounted
cgroups that report more than the host.
Helpers read via a bounded read_small_file (fopen "re", capped fread)
and are exposed through system_info_internal.h for unit tests that drive
them against a fake cgroup tree. The test teardown uses opendir/unlink/
rmdir (no shell spawn).
Distilled from #365 onto current main (unioned the test additions with
the CBM_WORKERS tests from #364, and replaced the test-cleanup
system("rm -rf") with a shell-free recursive remove). Closes#363
together with #364.
Adds a CBM_WORKERS env knob (clamped to [1, 256], invalid values warned
and ignored) that explicitly sets the parallel-indexing worker count.
In containers, sysconf(_SC_NPROCESSORS_ONLN) reports host CPUs rather
than the cgroup's effective quota, so a 1-vCPU pod on a 16-core node
otherwise spawns ~16 workers — the dominant OOMKill driver. When unset,
behaviour is unchanged.
From #364.
Complete C implementation of the indexing pipeline (src/), parallel worker
pool, graph buffer with merge support, SQLite writer, and 1893-test suite.
Linter setup: zero warnings from clang-tidy (all checks enabled, no NOLINT
suppressions), cppcheck, and clang-format. All issues fixed at source —
proper headers for external linkage, GROW_ARRAY macro restructured to
eliminate type parameter, null-deref paths guarded, named intermediates
for suspicious-argument checks.
News