## Why
Bubblewrap discovery probes executables before confinement. Excluding only the command's current directory leaves candidates in other writable roots eligible to run outside the sandbox.
## What changed
- Filter canonical `PATH` candidates using the filesystem policy and effective user permissions, including writable ancestors, symlinked roots, and full-disk write access. Preserve protected system installations while rejecting replaceable paths.
- Select the launcher with the command's permissions before the proc-mount preflight, retaining the bundled bubblewrap fallback.
- Use the effective permission profile and policy working directory for startup warning probes.
## Testing
Add unit coverage for writable roots, symlinks, read-only carveouts, replaceable parents, and full-disk policies. Add Linux integration tests showing writable `bwrap` candidates are neither probed nor launched, including when the command runs in a workspace subdirectory or uses managed networking with full-disk write access.
GitOrigin-RevId: 188c9f45e727eaace062f5abb1205dc23fdf3c51
## Why
Bubblewrap consumes and closes the file descriptor for each `--ro-bind-data` mount. Reusing one descriptor across multiple file masks prevents the sandbox from starting.
## What changed
Open and preserve a separate `/dev/null` descriptor for each empty-file mask.
## Testing
Extend unit tests to verify distinct descriptors for multiple denied files and missing-path masks. Add integration coverage for exact-path and glob deny rules, checking that the sandbox starts, denied files remain unreadable and unwritable, and allowed files remain accessible.
GitOrigin-RevId: 85158c83699afd36bb7aca453b2652d777cf65e4
## Why
Concurrent Linux test spawns can inherit writable descriptors for executable fixtures, preventing those fixtures from being executed even after the writing test closes its descriptor.
## What changed
Replace direct executable fixture writes and copies across the Rust tests with `codex_utils_cargo_bin::write_executable` and `copy_executable`. These existing helpers complete writes and copies in separate processes on Linux.
Remove the empty temporary inode before creating fake bubblewrap executables, and stage tampered bubblewrap bytes in a separate data file before copying them into the executable fixture. Add the required test dependencies.
GitOrigin-RevId: c875c83088731a4933ef1574e8d79d8cecc12f54
## Why
Copying executable fixtures in the test process can expose writable file descriptors to sibling spawns. The shared `copy_executable` helper completes copies in a separate process on Linux before launching the fixtures.
## What changed
Use `copy_executable` for both `codex-linux-sandbox` and `bwrap` in `bazel_build_rejects_tampered_bundled_bwrap`, and add `codex-utils-cargo-bin` as a Linux test dependency.
GitOrigin-RevId: 05fb1381eeefb64568b525925c9e81863acc2b54
## Why
AWS profiles can select credential helpers that the application executes. Granting write access to a containing directory should therefore keep `.aws` protected by default.
## What changed
Add `.aws` to the protected metadata paths alongside `.git`, `.agents`, and `.codex`, including workspace roots and additional writable roots. Preserve explicit user rules that override the default protection.
## Testing
Extend policy and sandbox tests to cover `.aws` protection. Add macOS coverage for existing and absent `.aws` directories, deletion and replacement attempts, and symlink targets. Verify that Linux sandbox writes to protected configuration files fail while sibling files remain writable.
GitOrigin-RevId: 31ea287bd0dd9b13aff893783fba9ca3aed0c89d
## Why
Approved commands need broader write access while retaining explicit read denials. On Linux, binding the filesystem root writable can shadow standard devices, and additional root-metadata mounts can reopen denied symlink targets.
## What changed
- Add `FileSystemSandboxPolicy::for_approved_command` to grant root and root-metadata writes while retaining path and glob denials. Keep the original policy when denials cannot be resolved or deny the root.
- Restore standard devices after Linux writable root binds, then apply explicit deny masks. Avoid redundant root-alias binds and inherit root-metadata writes from the root mount so denied targets stay masked.
- Preserve literal deny paths alongside resolved targets so Linux can reject denials that cross writable symlinks.
- Advertise Linux device and approved-root restriction support through two opt-in exec-server capabilities, omitted from serialization when false.
## Testing
Add regression coverage for approval policy materialization across Unix, Windows drive, and UNC paths; Windows volume expansion; Linux device access and metadata symlink restrictions; and capability serialization compatibility.
GitOrigin-RevId: 40584473da98a15e606bb1a8934308f69528850f
## Why
When a container denies fresh `/proc` mounts, the existing sandbox fallback retains the caller's `/proc` while creating a new PID namespace. Process IDs inside the sandbox can then differ from those exposed by `/proc`, breaking process lookups.
## What changed
- Add `codex exec-server --linux-sandbox-pid-namespace=inherit` to reuse the caller's PID namespace and `/proc` for both process and filesystem helpers. Repository config and command environment variables cannot enable it.
- Keep `isolate` as the default, preserving the existing mount fallback and compatibility with older helpers. Inheritance requires an updated helper and omits `--unshare-pid` and `--as-pid-1`.
- Preserve other sandbox restrictions. With `:minimal`, bind the inherited `/proc` read-only with its container masks and apply explicit filesystem denials afterward.
Inheritance is intended for dedicated environments: it allows sandboxed commands to signal other same-UID processes, including the executor.
## Testing
Add unit and integration coverage for namespace flags, denied proc mounts, consistent process IDs, retained sandbox restrictions, and startup-only selection across process and filesystem helpers.
GitOrigin-RevId: dd13f2b9286d7edcf366b3a42a455f4d78daaf27
## Why
A `.git` pointer can resolve to a Git directory inside another writable root. Broader Seatbelt grants or later bubblewrap mounts can leave that directory writable despite its metadata protections.
## What changed
Include resolved Git directories in read-only exclusions for broader writable roots on Linux and macOS. Resolve symlink aliases when locating exclusion targets and retain explicit policy overrides.
## Testing
Add regression tests for Git directories reached directly or through symlink aliases in another writable root, and for linked worktree metadata under the macOS temporary directory. Verify that metadata writes are denied while ordinary files in the surrounding writable root remain writable.
GitOrigin-RevId: 6971f1cf35008d23eaeb0d02a9c61bee7cf83b12
## Why
Btrfs subvolumes can report an `st_dev` that differs from the device in `/proc/self/mountinfo`, causing daemon socket mount validation to reject valid mounts.
## What changed
Allow the mismatch only when the opened directory and its uniquely identified mount are both verified as Btrfs. Discover socket directory aliases using both device identities while retaining checks that reject unsafe directory and socket bind mounts.
## Testing
Add regression tests for aliases under both device numbers, unverified or inconsistent filesystems, unavailable or missing mount IDs, duplicate mount IDs, and unsafe bind mounts.
GitOrigin-RevId: 921d6465b7a97022573a42412606a26560cc086a
## Why
Nested writable roots must allow the Linux sandbox to start while preserving read-only metadata protections, including when a writable directory sits beneath a read-only `.codex` directory.
## What changed
Defer read-only subpath mounts shared with a deeper writable root until that root is bound. Keep symlink target remapping and writable-symlink validation for deferred paths so mount ordering does not bypass those checks.
## Testing
Add regression coverage for mounting `.git`, `.agents`, and `.codex` metadata exactly once after the nested writable bind. Add sandbox integration cases for ordinary and symlinked roots, using both parent and nested working directories, to verify startup, allowed writes, metadata protection, and temporary mountpoint cleanup.
GitOrigin-RevId: 45e439eaf537e10bfa8b839bad816b0401fb65aa
## Why
The Linux sandbox rejected bind mounts of ancestors of the app-server socket directory, even when the exposed socket paths could be masked. This prevented sandbox startup with otherwise usable temporary-directory mount layouts.
## What changed
- Collect the socket directory and its exposed ancestor-mount aliases, then mask each path after filesystem binds, including read-only subpaths and the synthetic-mount registry.
- Continue rejecting direct directory or socket aliases and nested mounts that cannot preserve socket isolation.
- Reject non-UTF-8 mask destinations to avoid masking a different path through lossy conversion.
## Testing
Extend mount-analysis tests for ancestor aliases, private temporary directories, and WSLg masks. Extend socket-isolation coverage to verify that ancestor and registry aliases hide daemon sockets while unrelated sockets remain reachable. Add a regression test for non-UTF-8 mask paths.
GitOrigin-RevId: 49a9a764fbe1fddacd99c3a78099fbffff53b944
## What changed
- Remove the private-desktop opt-out from elevated and unelevated Windows sandbox launches.
- Remove `windows.sandbox_private_desktop` and its managed requirement and API fields. Warn users to remove the obsolete setting.
- Require a private desktop name when launching through the Windows sandbox wrapper and command runner.
## Testing
Add coverage for the obsolete-setting migration warning and update wrapper tests to verify a live private desktop is passed and a missing desktop name is rejected.
GitOrigin-RevId: c7135f8d211aac8d812180691c2c1e433d77cde4
## Why
Namespace mounts can have roots such as `mnt:[inode]` or `net:[inode]` that are not filesystem paths. Treating every mount root as a path rejects these unrelated mounts and prevents sandbox startup.
## What changed
Parse mount roots as paths only for the daemon socket's filesystem. Continue validating every mount destination and checking mount ancestry, nested mounts, and socket aliases.
## Testing
Add regression cases for namespace mounts, invalid destinations, and socket aliases, both with and without a mount ID. Extend the socket-isolation integration fixture with an unrelated network namespace mount to exercise successful startup and rejection of a real socket alias.
GitOrigin-RevId: 9a1d2a69e5bd3f8d09b1de7cf998ee6d9c6bb980
## What changed
- Add `Platform` to `codex-utils-path-uri` with metadata parsing, native platform detection, and path convention mapping. Preserve missing or unrecognized metadata as `Unknown`.
- Replace `NetworkProxyExecutorOs` with the shared type and keep executor-specific socket path validation in the network proxy.
- Extract `effective_sandbox_mode` with explicit platform and Windows sandbox level inputs, preserving the native Windows fallback from `workspace-write` to `read-only` when the sandbox is disabled.
## Testing
Add unit tests for platform metadata, path conventions, native platform detection, and sandbox mode selection across platforms and Windows sandbox levels.
GitOrigin-RevId: 4fe0972e3a91040e35f2a6dfa5bcdf6c9a29be88
## Why
A controller and its executor can run different operating systems. Validating socket paths against the controller's OS can reject absolute paths that are valid on the executor, such as Windows paths on a Linux controller.
## What changed
- Thread `NetworkProxyExecutorOs` through network policy validation, proxy construction, and policy updates.
- Require allowed socket paths to be NUL-free and absolute for the executor OS, while preserving deny entries unchanged.
- Accept either Unix or Windows absolute syntax when executor metadata omits the OS, then validate against the executor's own OS at launch.
- Keep native path normalization and socket support checks at execution time.
## Testing
Add coverage for cross-platform absolute path syntax, invalid allow entries, preserved deny entries, and remote policy round trips that retain executor semantics through domain edits and proxy construction.
GitOrigin-RevId: 1ebc09cbce7138f60ec5fd62875591a3df52d067
## Why
Private `/tmp` bind mounts can leave hidden entries in `/proc/self/mountinfo`, causing the Linux sandbox to reject layouts that safely isolate daemon sockets. The proc-mount preflight also needs the main sandbox's WSL masks when checking aliases.
## What changed
- Identify the opened socket directory's mount using `fdinfo`, with a `statx` fallback, and follow its mount ancestry to distinguish hidden paths from exposed aliases.
- Accept safe private and stacked mounts while rejecting covered mounts, exposed aliases, and nested mounts. Retain conservative checks when no mount ID is available.
- Preserve WSL interop and WSLg masks in the proc-mount preflight.
- Heap-allocate large lifecycle futures in the TUI approval-gated MCP tool test to reduce Windows test-thread stack usage.
## Testing
Add mount-layout regression cases and a namespace integration test that verifies daemon sockets remain inaccessible under private `/tmp`, unrelated sockets remain reachable, and an exposed alias prevents startup. Add a preflight test for preserving WSL masks.
GitOrigin-RevId: 289b4446ba7776a7dd57ef5ff09e87bda19b069c
## Why
An omitted controller socket policy was treated as an explicit denial, preventing execution environments from supplying their own Unix socket grants.
## What changed
- Preserve omission of `dangerously_allow_all_unix_sockets` separately from `false`, and retain explicitly empty `unix_sockets` maps.
- Defer to attachment socket permissions when the controller supplies neither setting. Continue enforcing explicit restrictions, socket denials, and managed requirements.
- Resolve omitted values to `false` for ordinary execution and remote configuration, preserving the default for commands without attachment grants.
- Add debug logging for effective environment and remote execution network policies.
## Testing
Add regression coverage for omitted, explicit, finite, empty, and managed socket policies through remote launch configuration, including live policy replacement and serialization round trips. Adjust remote environment tests to tolerate child-completion ordering and box large cold-resume test futures to reduce Windows stack usage.
GitOrigin-RevId: 98a88d01cbb91aea1faaa03b3c414ef3be7af398
## Why
Network access and Unix-socket allowlists must not let commands with filesystem restrictions reach the privileged app-server RPC transport.
## What changed
- Bind Unix control sockets in a fixed, user-owned directory with mode `0700`, independent of environment settings, and expose the advertised paths as symlinks. Preserve existing parent permissions, reject unsafe parents, and serialize socket setup and publication.
- Mask the socket directory in Linux bubblewrap sandboxes after each bind that exposes it. Reject host mount aliases and nested mounts that compromise isolation.
- Deny access to the directory and outbound connections to its sockets in macOS Seatbelt policies, including when network access or Unix-socket allowlists grant broader access.
- Require bubblewrap for filesystem-restricted Linux execution. Users with `features.use_legacy_landlock` enabled must disable it for these policies.
## Testing
Add regression coverage for direct and symlink socket access, hardlink attempts, Linux host-process links and bind-mount aliases, and continued use of unrelated and sandbox-local sockets. Add transport coverage for parent permissions, concurrent restart after a stale symlink, and cleanup that preserves a replacement at the advertised path.
GitOrigin-RevId: 53372c27eea278d964f2cf68aed24323ef3b7082
## Why
WSLg's duplicate distro root can expose filesystem contents outside the sandbox's path masks.
## What changed
- Detect the duplicate root using filesystem identity, with mount metadata as a fallback, and hide it after applying filesystem grants and denials.
- Reject explicit grants, working directories, and executable paths that use the WSLg alias, directing users to the primary filesystem paths.
- Hide host procfs when a fresh procfs cannot be mounted so process roots cannot restore access to the masked view.
## Testing
Add regression tests for duplicate-root detection, mask ordering, existing ancestor masks, alias rejection, and unrestricted filesystem behavior. Add a WSLg runtime test covering masking with and without fresh procfs, continued access to an allowed file, and rejection of an executable using the alias.
GitOrigin-RevId: 9ea31724c00fb24e3a504b6658339e8dfea35343
## Why
Linux proxy-routed sandboxing denied standalone Unix sockets even when the effective network policy enabled `dangerously_allow_all_unix_sockets`.
## What changed
- Carry Unix socket permissions in `ManagedNetworkSandboxContext` and pass the prepared context through Linux sandbox launches with `--managed-network`.
- Allow `AF_UNIX` socket creation in proxy-routed mode when `dangerously_allow_all_unix_sockets` is enabled, while preserving network namespace isolation and restrictions on other socket families.
- Keep standalone Unix sockets denied by default and for path-only grants. Default missing fields in older serialized contexts to restrictive values.
## Testing
Add coverage for policy preparation and transport, legacy deserialization, and malformed policy rejection. Add a Linux integration test covering default denial, path-only denial, and explicit allow-all access, while checking that direct TCP access and `AF_NETLINK`/`AF_VSOCK` sockets remain blocked.
GitOrigin-RevId: 2695b945ad3e59fcb3faf7662d852a26650af16c
## What changed
- Add explicit MXC backend selection and carry its identity through exec-server process reporting and sandbox violation classification.
- Launch MXC through the Codex executable with the effective permission profile and command environment.
- Reject exec-server MXC requests when native MXC is unavailable or when they request a TTY, an `arg0` override, or managed networking. Reject private desktop isolation during MXC preparation.
- Allow an explicitly empty child environment and avoid exposing request payload values in launcher decode errors.
## Testing
Add coverage for sandbox selection and unsupported-request rejection, plus Windows RPC tests for stdin writes and temporary-directory permissions derived from the command environment. Native MXC tests skip when MXC is unavailable.
GitOrigin-RevId: 3626ff0f9ad7f9b812ce09b68c31ea9a5a9c72b1
## Why
With network access enabled, WSL interop can launch Windows processes outside the Linux filesystem sandbox, including using `wsl.exe` to reenter the distribution as root.
## What changed
- Mask WSL interop sockets in bubblewrap when filesystem write access is restricted, and hide inherited procfs when running without a fresh proc mount.
- Deny `AF_VSOCK` sockets and `io_uring` under restricted filesystem policies even when ordinary network access is allowed, closing the VM socket route to host services.
- Reject legacy Landlock configurations that combine restricted filesystem access with unrestricted networking when the WSL interop directory exists.
## Testing
Add unit coverage for mount masking and legacy policy rejection, plus WSL integration tests for Windows process launch, bind aliases, direct `/init` invocation, root reentry, and `--no-proc` isolation. WSL tests skip when host prerequisites are unavailable.
GitOrigin-RevId: 1e8b97ac43675807ee541ca614534d9b1febf410
## Why
Linux managed networking routed proxy connections through filesystem-backed Unix sockets, requiring an extra readable sandbox path and stale socket cleanup.
## What changed
- Transfer loopback TCP listeners from the sandbox namespace to host bridge processes over anonymous Unix socket pairs.
- Validate transferred descriptors, keep bootstrap channels private to each sandbox launch, and close them before running untrusted commands.
- Tie proxy bridge lifetime to the sandbox so cancellation closes active upstream connections.
## Testing
- Cover descriptor validation and interrupted handoffs.
- Verify concurrent sandboxes keep proxy endpoints isolated and do not inherit privileged sockets.
- Exercise managed networking end to end through `app-server`, confirming proxied access succeeds while direct host access remains blocked.
GitOrigin-RevId: 6561e09c83b927c604951bf2aa00ff532d179da5
## Why
The synthetic mount registry must remain private to the sandbox helper. A
writable bind overlapping its temporary directory could expose the registry,
and a pre-existing registry symlink could redirect bookkeeping into a writable
location.
## What changed
- Add the registry's read-only bind while constructing filesystem mounts, but
only when a writable bind exposes it and before applying deny masks.
- Reject a symlinked registry path before registering mounts or starting the
sandboxed command.
## Testing
Add coverage for symlink rejection and for denied temporary directories under
both full-root and minimal filesystem policies.
GitOrigin-RevId: 3adff28998f0ef8c8ec26a73e5a1183b4931c0e8
## Why
The Linux sandbox represents missing protected metadata paths such as `.git`
with empty read-only directories. Repository and project discovery treated any
`.git` directory as a checkout root, so this synthetic directory could hide a
real parent repository and its trust configuration.
## What changed
- Treat a `.git` directory as repository metadata only when it contains
`HEAD`, while continuing to recognize file-based `.git` entries.
- Protect missing `.git` paths with the same read-only synthetic mounts as
other workspace metadata without disrupting parent-repository discovery.
- Canonicalize and read-only bind the synthetic-mount registry into the
sandbox, and make protected-path cleanup handle read-only directory trees.
## Testing
Added coverage for repository, project-root, trust, synthetic-mount, and
Landlock behavior, including nested incomplete `.git` directories and a
redirected `TMPDIR`.
GitOrigin-RevId: 0724c54d56531143bb28011e4ca414cd3b0212be
## What changed
- Pass `--unshare-ipc` when building Bubblewrap arguments for both restricted-filesystem and full-filesystem sandbox paths.
- Update the Bubblewrap argument tests to cover the new namespace flag.
GitOrigin-RevId: 61fb1d5150d09626a3b6ee7a9c6a508108c52e80
## Why
System Bubblewrap installations that lack `--ro-bind-fd` cannot directly create
the descriptor-backed read-only mounts used by the Linux sandbox.
## What changed
- Detect `--ro-bind-fd` support when probing system Bubblewrap.
- On older versions, translate descriptor-backed mounts to `/proc/self/fd`
read-only binds and pass their descriptors and destinations to the trusted
inner sandbox stage for verification.
- Reject malformed, duplicate, mismatched, or symlink-substituted mounts, and
close inherited descriptors before running sandboxed code.
## Testing
Added unit coverage for capability detection, legacy argument translation,
invalid mount rejection, inode verification, and descriptor closure.
GitOrigin-RevId: dfc0a457b572cdd9aed093fb5abf57360e9724a2
## What changed
- Pass `--cap-drop ALL` in both bubblewrap launch modes.
- Verify in the inner sandbox stage that the effective and permitted Linux
capability sets are empty, and abort before executing the command otherwise.
- Cover the generated bubblewrap arguments, capability state of sandboxed
commands, and rejection of a capability-bearing inner-stage process.
GitOrigin-RevId: 9dbd651d306d92327e36c9b3f308619db7d80a67
## Why
Filesystem permission paths can use a convention that differs from the host
running Codex. Converting them immediately to native absolute paths can change
the meaning of ambiguous paths such as `/C:/secret` or Windows UNC paths.
## What changed
- Store literal filesystem permission paths as `PathUri` values through the
runtime policy and execution protocol.
- Keep legacy string-based serialization at explicit protocol boundaries and
reject conversions that cannot be represented losslessly.
- Encode native paths as opaque URIs when a normal file URI would imply the
wrong path convention.
## Testing
Added coverage for cross-platform and ambiguous path round trips, UNC path
variants, permission-profile serialization, and deny-policy enforcement.
GitOrigin-RevId: 5247713796d1f2bb4e02f94eb9fc82d4698060f0
## Why
Bazel builds did not embed the SHA-256 digest used to verify the bundled
`bwrap` binary at launch.
## What changed
- Strip the Bazel-built `bwrap`, generate its SHA-256 digest, and pass that
digest to `codex-linux-sandbox` through a compiler environment file.
- Report digest verification failures with exit code `8` instead of panicking.
- Add a Bazel integration test that accepts the original bundled binary and
rejects a tampered copy.
GitOrigin-RevId: 224eeb6ae3d1bfb100e048b296f75fd8ba109581
## What changed
Implement `From<AbsolutePathBuf>` for `FileSystemPath` and use the conversion
when constructing concrete filesystem permission paths.
GitOrigin-RevId: 244bcbd0c3f76cd87cc1240458fcac6c9ef0a8e3
## Why
Sandboxed descendants can outlive their immediate parent and must be collected by
PID 1 in the Bubblewrap namespace.
## What changed
- Launch `codex-linux-sandbox` with Bubblewrap's `--as-pid-1` option, and fall
back to the bundled Bubblewrap when the system version does not support it.
- Run the sandboxed command as a child, forward signals to it, reap other exited
descendants, and preserve the command's exit status.
- Verify proxy bridge parent identity when arming its parent-death signal.
## Testing
Added Linux sandbox coverage for the filtered namespace reaper, orphan
collection, and fallback from an incompatible system Bubblewrap.
GitOrigin-RevId: 379f08d6c2732ea0a4caeb61f93ae302e16d2458
## Why
Linux sandbox unreadable globs without a non-root directory prefix cannot be safely expanded because they would require scanning from `/`. Silently skipping these patterns leaves the requested deny-read rule unenforced.
## What changed
Return a fatal sandbox construction error when an unreadable glob cannot be split into a safe ripgrep search root. The error directs callers to use a pattern with a non-root directory prefix.
## Testing
Add a regression test confirming that `/**/*.env` rejects Bubblewrap command construction.
GitOrigin-RevId: 2f3cb398d39a9bc077d3a1b74e08dc506d239e60
## Why
Binding the full host filesystem into a network-isolated Bubblewrap sandbox also
inherits the host's device tree.
## What changed
- Overlay `/dev` with Bubblewrap's minimal device filesystem while preserving
the root bind's `nodev` behavior.
- Bind `/dev/shm` back into the sandbox when available so host and sandbox
processes can continue to share memory files.
## Testing
Add a managed-proxy integration test that verifies standard device nodes remain
usable, `/dev` is not the host directory, and `/dev/shm` writes are shared with
the host.
GitOrigin-RevId: 1d91fe44affd792d2777adae8f2cce6d85d14103
## Why
Managed proxy helpers can keep captured standard streams open after the sandboxed command exits. Proxy readiness can also fail when inherited standard descriptors are already closed, and zombie owners can leave stale socket directories behind.
## What changed
- Detach bridge and cleanup-worker standard I/O to `/dev/null`.
- Move readiness pipe descriptors above the standard descriptor range.
- Treat zombie processes as exited when cleaning proxy socket directories.
- Move helper lifecycle handling into a dedicated module.
## Testing
Add coverage for output release after command exit, readiness with closed standard descriptors, zombie detection, and stale socket cleanup.
GitOrigin-RevId: 9f4081cba1b73442f02a895473c0383a795fa30e
## What changed
- Forward remote managed-network policy requests to the controller-side decider when Guardian review is enabled.
- Bound callbacks using the configured permission-hook and Guardian review timeouts, restore trusted execution attribution, and recheck live baseline policy before requesting a decision.
- Cancel pending decisions when the process or execution scope ends, and propagate EOF across the sandbox proxy bridge with half-closes.
## Testing
- Cover callback enablement, strict allowlist behavior, live policy updates, trusted attribution, and execution-scope cancellation.
GitOrigin-RevId: 83b47657c8c4546dfe98eec8b272833e96a46b7e
## What changed
- Wrap `CodexErrorDetails` and an optional retry delay in `CodexErr`, allowing any mapped error to preserve server-provided retry timing.
- Generate the payload-free `CodexErrKind` classification alongside the error details and reuse it for analytics.
- Update error handling sites to inspect `CodexErr::details()` while preserving existing display, debug, protocol mapping, and retryability behavior.
## Testing
- Add coverage for legacy debug formatting, error-specific retryability, and retry-delay propagation through API error mapping.
GitOrigin-RevId: d3ab8a305f2a2ee21d0c0a8c8c388b06dda9c59a
## What changed
- Make the generated Linux proxy socket directory readable inside restricted
`bubblewrap` sandboxes so the proxy bridge can connect.
- Route `WS_PROXY` and `WSS_PROXY` through the Linux managed proxy bridge.
- Remove inherited proxy attribution tokens from unscoped executions while
continuing to replace them with the current token for scoped executions.
## Testing
- Exercise `WSS_PROXY` routing with a minimal filesystem policy.
- Cover attribution-token removal and replacement during environment setup.
GitOrigin-RevId: 2f14ddb26baab0147a786bc354d93eb5aad059f3
## Why
Default read-only protections for project metadata should apply when paths such
as `.git`, `.agents`, and `.codex` exist, without causing sandbox setup to
materialize missing paths as ACL targets.
## What changed
- Add an optional `missing_path_behavior` to filesystem sandbox entries and
mark default project-metadata protections with `skip`.
- Preserve the behavior through permission transforms and exec/MCP protocol
serialization while keeping existing path wire variants stable.
- Ignore skip-missing entries when projecting configuration or Windows sandbox
overrides, while retaining explicit metadata carveouts.
## Testing
- Cover protocol round trips for path and special-path entries.
- Verify default metadata protections and Windows explicit carveout handling.
GitOrigin-RevId: 6df13dadacdd131c44aab9f15a967c81051355c1
## What changed
- Run the bubblewrap `/proc` mount probe with the minimal read-only filesystem policy instead of the command's filesystem policy and working directory.
- Preserve the requested network namespace mode during the probe.
- Add coverage that the probe mounts a temporary root and `/proc` without binding the full filesystem.
GitOrigin-RevId: e3bc4352f510980aeebe81dedf3f9211947982ed
## Why
Special filesystem subpaths can contain relative or foreign-platform path
spellings. Treating this wire text as a host-native path can reinterpret it.
## What changed
- Store special filesystem subpaths as strings in the core permissions model.
- Represent app-server subpaths as `LegacyAppPathString` and convert them at
the protocol boundary without changing their spelling.
- Add `LegacyAppPathString::from_string` for callers that already own legacy
app-server path text.
GitOrigin-RevId: 2214f03aa997fd0e6eb2ab60d425304bb4bd6132
## Why
Managed-network commands within one Codex conversation share the same
HTTP and SOCKS proxy ingress. When several exec calls run concurrently,
the proxy sees the requested destination but cannot tell which exec
opened the connection.
For example:
```text
exec A: curl https://example.com/a ─┐
├─> conversation proxy ─> Guardian
exec B: curl https://example.com/b ─┘ host: example.com
trigger: unknown
```. Three parallel network execs reached Guardian without their
triggering call IDs or commands. Guardian denied the requests, but Codex
could not safely associate those outcomes with the individual tool
calls.
## What changes
Keep the shared proxy ingress and tag each connection at the existing
trusted Linux bridge:
```text
exec A ─> existing Linux bridge ─> [token A][proxy bytes] ─┐
├─> shared HTTP/SOCKS ingress
exec B ─> existing Linux bridge ─> [token B][proxy bytes] ─┘
│
token A ─> exec A ─────┤
token B ─> exec B ─────┘
```
The complete path is:
```text
active exec registration
│
├─ registers its UUID as a short-lived attribution token
├─ passes the token to the Linux sandbox helper
├─ helper removes the token before launching the user command
├─ existing host bridge prepends the token to each proxy connection
├─ shared proxy consumes the bounded attribution frame
└─ proxy attaches the matching execution-scoped state
├─ Guardian receives the exact call ID and command
└─ a denial finishes/cancels the matching tool call
```
Dropping the active or deferred exec registration removes the token.
Connections that were already accepted retain their resolved
attribution; new connections using an expired token fail closed.
## Before and after
Before, Guardian could receive only the network destination:
```json
{
"tool": "network_access",
"host": "www.17track.net",
"port": 443,
"protocol": "https"
}
```
After, the same request includes the action that caused it:
```json
{
"tool": "network_access",
"host": "www.17track.net",
"port": 443,
"protocol": "https",
"trigger": {
"callId": "exec-network-first",
"command": ["/bin/sh", "-c", "curl https://www.17track.net"]
}
}
```
## Listener accounting
This PR does **not** create proxy listeners per exec.
```text
Existing topology:
one conversation -> one HTTP listener + optional one SOCKS listener
Discarded per-exec approach:
one conversation -> existing listener pair
+ up to one additional listener pair per active exec
This PR:
one conversation -> existing listener pair only
+ one small token-map entry per active exec
```
The Linux sandbox already creates a trusted routing bridge for each
sandboxed command. This PR adds a short frame write to that bridge
rather than introducing another listener, task, or proxy process.
The existing conversation-scoped listener pair remains. Making a single
proxy service shared across multiple conversations would be a separate
multi-tenant architecture change involving per-conversation policy,
configuration, audit, and Guardian routing.
## Keeping the implementation small
The attribution is bound once, when the TCP connection enters the proxy.
The ingress installs an execution-scoped clone of the existing
`NetworkProxyState`, so the established HTTP, SOCKS, MITM, policy,
audit, and blocked-request paths continue using their existing state
lookup.
This avoids plumbing a new request-context type through every protocol
handler. Outside the two ingress wrappers, protocol-specific request
handling is unchanged.
## Security behavior
- Tokens are generated from the existing random execution registration
IDs.
- The trusted Linux helper consumes and removes the token before
executing user code.
- Attribution frames have a fixed magic prefix, bounded token length,
and bounded read timeout.
- Unknown or expired tokens close the connection.
- A token presented to a proxy for another environment closes the
connection.
- Existing unframed callers preserve the current conservative
attribution behavior.
## Platform scope
Exact bridge attribution is enabled on Linux. macOS and Windows retain
their current shared-proxy behavior.
## Test coverage
The concurrent end-to-end test starts two managed-network execs together
and synchronizes them so both are active before either connects. It then
inspects the two Guardian requests and compares the complete attribution
pairs:
```text
(exec-network-first, exact first command)
(exec-network-second, exact second command)
```
Focused proxy coverage verifies the bounded frame and that a registered
framed connection receives the matching execution and environment state.
## Scope
This fixes the Linux network-to-exec attribution path and records a
denial against the exact matching tool call. It intentionally does not
change:
- delivery of an entirely unattributed denial to the parent turn;
- how parallel denials count toward the Guardian circuit breaker;
- how the UI displays the rejection reason or completed-turn state.
Those remain separate concerns from attribution.
## Relationship to #29456 and #29668#29456 made the proxy environment and sandbox policy come from the same
prepared network context. This PR adds the execution token to that
prepared launch and consumes it at the shared ingress.
This follows #29668's shared-ingress framing direction, but completes
the production registration, Linux bridge, core call mapping, denial
mapping, and concurrent end-to-end path. It also keeps attribution in
the existing per-connection proxy state instead of introducing
request-context plumbing through every HTTP, SOCKS, and MITM handler.
This PR is intended to supersede #29668 for the Linux attribution fix.
---------
Co-authored-by: viyatb-oai <viyatb@openai.com>
Co-authored-by: Codex <noreply@openai.com>
## Why
Prepare network approval scoping to distinguish execution environments
without changing behavior yet.
## What changed
- Add optional environment IDs to network policy requests.
- Add optional network environment IDs to exec and sandbox request
structs.
- Thread default None values through existing construction points.
- Fix stale constructor call sites that caused the CI compile failures.
## Not included
- Per-environment proxy listeners.
- Network approval cache or prompt behavior changes.
- Ambiguous request attribution handling.
Those behavior changes moved to stacked follow-up #28899.
## Validation
- just fmt
- CI will run tests and clippy
The workspace denies `clippy::expect_used` in production. Although
`clippy.toml` allows `expect` in tests, Bazel Clippy compiles
integration-test helper code in a way that does not receive that
exemption, which encouraged verbose `unwrap_or_else(... panic!(...))`
and equivalent `match`/`let else` forms.
This allows `clippy::expect_used` once at each integration-test crate
root (including aggregated suites and test-support libraries), then
replaces manual panic-based Result and Option unwraps with
`expect`/`expect_err`. Standalone `tests/*.rs` files remain their own
crate roots. Intentional assertion and unexpected-variant panics remain
unchanged, and the production `expect_used = "deny"` lint remains in
place.
The cleanup is mechanical and net-negative in line count.
## Summary
- allow `socketpair(AF_UNIX, ...)` in the proxy-routed Linux seccomp
mode
- continue denying `socket(AF_UNIX, ...)` so user commands cannot create
pathname or abstract Unix sockets
- extend the managed-proxy integration test to verify both behaviors
## Root cause
`NetworkSeccompMode::ProxyRouted` treated anonymous Unix socket pairs
like externally addressable Unix sockets and returned `EPERM`. This
breaks tools that use socket pairs for local child-process IPC even
though a socket pair cannot connect outside the sandbox or bypass the
routed proxy.
`dangerously_allow_all_unix_sockets` controls Unix-socket requests
forwarded by the managed network proxy; it does not currently configure
the Linux seccomp filter. Socket pairs should not require that dangerous
setting because they are unnamed, process-local IPC.
Related but independent: #26553 fixes host proxy bridge socket path
length handling.
---------
Co-authored-by: Codex <noreply@openai.com>
## Summary
- avoid generating host proxy bridge Unix socket paths that exceed
Linux's `sockaddr_un.sun_path` limit
- fall back from a long `$CODEX_HOME/tmp` path to the system temp
directory, then `/tmp`
- add focused unit coverage for short and overlong parent paths
## Root cause
With a sufficiently long `CODEX_HOME`, the generated
`proxy-route-*.sock` path exceeds Linux's 107-byte pathname limit. The
host bridge child exits before writing its readiness byte, so the parent
reports the indirect error `failed to prepare host proxy routing bridge:
failed to fill whole buffer`.
## Validation
- reproduced the original error with a long `CODEX_HOME` using
`codex-cli 0.138.0-alpha.4`
- `cargo clippy -p codex-linux-sandbox --all-targets`
- `just fix -p codex-linux-sandbox`
- `just fmt`
The Linux-only unit test could not execute locally: the arm64 Docker
build was repeatedly OOM-killed by `rustc` while compiling an unrelated
`codex-app-server-protocol` dependency, before reaching the test.
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
#23813 switches the Windows sandbox runner path to `PermissionProfile`,
but it still left one runtime anchor for resolving symbolic
`:workspace_roots` entries. That is not enough once a turn has multiple
effective workspace roots: exact entries and deny globs under
`:workspace_roots` need to be materialized for every runtime root before
the command runner chooses token mode or builds ACL plans.
## What Changed
- Replaces the Windows runner/setup `permission_profile_cwd` plumbing
with `workspace_roots: Vec<AbsolutePathBuf>`.
- Resolves Windows-local `PermissionProfile` data with
`materialize_project_roots_with_workspace_roots(...)` instead of the
single-cwd helper.
- Threads `Config::effective_workspace_roots()` through core execution,
unified exec, TUI setup/read-grant flows, app-server setup, app-server
`command/exec`, and `debug sandbox` on Windows.
- Preserves those workspace roots through the zsh-fork escalation
executor instead of rebuilding them from `sandbox_policy_cwd`.
- Makes `ExecRequest::new(...)` and the remaining
`build_exec_request(...)` helper path take
`windows_sandbox_workspace_roots` explicitly so new call sites cannot
silently fall back to `vec![cwd]`.
- Clarifies the `debug sandbox` non-Windows comment: remaining
cwd-dependent resolution still uses `sandbox_policy_cwd`, while
`:workspace_roots` entries are already materialized from config roots.
- Updates elevated runner IPC `SpawnRequest` to send `workspace_roots`
and bumps the framed IPC protocol version to `3` for the payload shape
change.
- Adds Windows-local resolver coverage for expanding exact and glob
`:workspace_roots` entries across multiple roots, plus core helper
coverage proving explicit roots are preserved.
## Verification
- `cargo check -p codex-windows-sandbox -p codex-core -p codex-tui -p
codex-cli -p codex-app-server`
- `cargo test -p codex-windows-sandbox`
- `cargo test -p codex-core windows_sandbox`
- `cargo test -p codex-core unix_escalation`
- `cargo test -p codex-app-server windows_sandbox`
- `cargo test -p codex-tui windows_sandbox`
- `cargo test -p codex-cli debug_sandbox`
- `just test -p codex-core unified_exec`
- `just test -p codex-core
build_exec_request_preserves_windows_workspace_roots`
- `env -u CODEX_NETWORK_PROXY_ACTIVE -u
CODEX_NETWORK_ALLOW_LOCAL_BINDING just test -p codex-app-server --lib
command_exec`
- `just test -p codex-windows-sandbox`
- `just test -p codex-exec sandbox`
- `just fix -p codex-core -p codex-app-server -p codex-windows-sandbox`
A local macOS cross-check with `cargo check --target
x86_64-pc-windows-msvc ...` did not reach crate Rust code because native
dependencies require Windows SDK headers (`windows.h` / `assert.h`) in
this environment; Windows CI remains the real target validation.
Two local targeted filters compile but do not run assertions on macOS:
`env -u CODEX_NETWORK_PROXY_ACTIVE -u CODEX_NETWORK_ALLOW_LOCAL_BINDING
just test -p codex-app-server --lib command_exec_processor` matched zero
tests, and `just test -p codex-linux-sandbox landlock` matched zero
tests because the landlock suite is Linux-only.
## Why
`codex sandbox` previously required an OS subcommand like `linux`,
`macos`, or `windows`, even though the command can only run the sandbox
backend available on the current host. That made the CLI imply a
cross-OS choice that does not exist.
## What changed
- Collapse `codex sandbox <os>` into `codex sandbox [COMMAND]...` by
wiring the `sandbox` parser directly to the host-specific backend args
with `cfg`.
- Keep the existing backend runners for Seatbelt, Linux sandbox, and
Windows restricted token.
- Rename the public Windows debug sandbox runner to
`run_command_under_windows_sandbox` for clarity.
- Update the Rust sandbox docs and related README references to describe
host OS selection and avoid pointing readers at legacy `sandbox_mode`
config.
## Arg0 compatibility
The `codex-linux-sandbox` helper path is still handled before normal CLI
parsing. `arg0_dispatch()` checks whether the executable basename is
`codex-linux-sandbox` and directly calls
`codex_linux_sandbox::run_main()`, so removing the `sandbox linux`
parser branch does not affect the arg0 helper flow.
## Verification
- `cargo test -p codex-cli`
- `cargo test -p codex-arg0`
- `just fix -p codex-cli`
## Summary
The Linux sandbox should find bundled `bwrap` through the same
package-layout abstraction as the rest of the runtime, instead of
maintaining a separate standalone-specific lookup path.
This adds an `InstallContext` helper for bundled resources and updates
`codex-linux-sandbox` to ask the current install context for
`codex-resources/bwrap` before falling back to the old
executable-relative probes. The tests cover npm-style, standalone, and
canonical package layouts so `bwrap` lookup follows the package
structure introduced earlier in the stack.
## Test plan
- `cargo test -p codex-install-context`
- `cargo test -p codex-linux-sandbox --lib`
- `just fix -p codex-install-context -p codex-linux-sandbox`
- `just bazel-lock-check`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/23634).
* #23638
* #23637
* #23636
* #23635
* __->__ #23634
## Why
Filesystem permission profiles used `none` for deny-read entries, which
is less direct than the action the entry actually represents. This
change makes `deny` the canonical filesystem permission spelling while
preserving compatibility for older configs that still send `none`.
## What changed
- rename `FileSystemAccessMode::None` to `Deny`
- serialize and generate schemas with `deny` as the canonical value
- retain `none` only as a legacy input alias for temporary config
compatibility
- update filesystem glob diagnostics and regression coverage to use the
canonical spelling
- refresh config and app-server schema fixtures to match the new wire
shape
## Validation
- `cargo test -p codex-protocol`
- `cargo test -p codex-app-server-protocol`
- `cargo test -p codex-core config_toml_deserializes_permission_profiles
--lib`
- `cargo test -p codex-core
read_write_glob_patterns_still_reject_non_subpath_globs --lib`
Earlier in the session, a broad `cargo test -p codex-core` run reached
unrelated pre-existing failures in timing/snapshot/git-info tests under
this environment; the targeted surfaces touched by this PR passed
cleanly.
viyatb-oai