Fix remote server socket path exceeding sun_path limit by kevinyang372 · Pull Request #11009 · warpdotdev/warp

kevinyang372 · 2026-05-15T14:40:52Z

Description

PR #10782 ("Version aware daemon socket") changed daemon socket filenames from server.sock (11 chars) to server-v0.2026.05.13.09.15.stable_01.sock (41 chars), adding ~30 bytes to the Unix domain socket path. This pushes users over the sun_path limit (107 bytes on Linux, 103 on macOS), causing UnixListener::bind to fail silently in the daemon. The proxy then times out after 10s, exits with code 1, and the client sees ResponseChannelClosed.

This accounts for the ~20% drop in remote server initialization success rate (90% → 70%). Sentry issue: WARP-CLIENT-BETA-STABLE-7M9M (11 occurrences across 9 users, all anonymous with 36-char UUID identity keys).

Fix (two changes):

Hash the version string in socket/PID filenames to an 8-hex-char suffix (server-a1b2c3d4.sock, 20 chars) instead of the full version string (41 chars).
Hash the identity key directory name to 8 hex chars instead of using the raw identity key (up to 36 chars for anonymous UUIDs). This is needed because longer channel base dirs like .warp-preview (+8 chars vs .warp) would otherwise narrow the headroom too much on macOS.

Both changes use std::hash::DefaultHasher for deterministic, fixed-length output. The existing cleanup_old_versions() logic automatically cleans up old long-form files.

Also adds an explicit sun_path length guard in the proxy that fails fast with a clear error message instead of silently timing out.

Worst-case path after fix: /home/{32-char-user}/.warp-preview/remote-server/{8-char-hash}/server-{8-char-hash}.sock = 97 bytes, well under both limits (103 macOS, 107 Linux).

Fixes WARP-CLIENT-BETA-STABLE-7M9M

Linked Issue

The linked issue is labeled ready-to-spec or ready-to-implement.

Testing

10 unit tests: version hash determinism, identity key hash (length, determinism, uniqueness), socket/PID name length bounds, and worst-case path length regression test against the preview channel (longest base dir)
Validated via Docker containers on linux/amd64, linux/arm64, and macOS/aarch64 with the actual 9 Sentry-reported identity keys
Cross-referenced Sentry events to confirm all affected users are anonymous (36-char UUID identity keys)
cargo fmt and cargo clippy pass
I have manually tested my changes locally with ./script/run

Agent Mode

Warp Agent Mode - This PR was created via Warp's AI Agent Mode

Co-Authored-By: Warp agent@warp.dev

kevinyang372 · 2026-05-15T14:41:06Z

Fix remote server socket path exceeding sun_path limit #11009 👈 (View in Graphite)
master

This stack of pull requests is managed by Graphite. Learn more about stacking.

oz-for-oss · 2026-05-15T14:41:06Z

@kevinyang372

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss

Overview

This PR shortens the versioned remote-server socket and PID filenames and adds an early proxy-side path length guard for Unix domain socket limits.

Concerns

The new server-{hash}.sock name is still too long for the same realistic identity/home inputs on longer channel directories such as .warp-preview, so Preview users can continue to exceed sun_path and fail before the daemon starts.

Verdict

Found: 0 critical, 1 important, 0 suggestions

Request changes

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

kevinyang372 · 2026-05-15T14:57:41Z

/oz-review

oz-for-oss · 2026-05-15T14:57:54Z

@kevinyang372

I'm re-reviewing this pull request in response to a review request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss

Overview

This PR shortens remote-server daemon socket/PID paths by hashing the version and identity components, then adds a proxy-side sun_path length guard.

Concerns

Hashing the identity directory changes the location used by remote_server_daemon_data_dir, which stores the remote daemon SQLite database, so existing per-identity state under the previous percent-encoded directory becomes unreachable without a migration.
The same new identity directory is only an 8-hex-character hash; because it scopes the daemon socket and data directory, collisions can make distinct identity keys share daemon/data state.

Security

The 32-bit identity hash is used for identity-scoped daemon/data isolation; a collision can cross-contaminate daemon state between distinct identities.

Verdict

Found: 0 critical, 1 important, 0 suggestions

Request changes

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss · 2026-05-15T15:02:21Z

-    encoded
+    let mut hasher = std::collections::hash_map::DefaultHasher::new();
+    identity_key.hash(&mut hasher);
+    format!("{:016x}", hasher.finish())[..8].to_string()


⚠️ [IMPORTANT] [SECURITY] Hashing the identity key into an 8-hex (32-bit) directory changes the existing data path and can collide across distinct identities; remote_server_daemon_data_dir stores the daemon SQLite DB under this directory, so this can orphan existing state or make two identities share daemon/data state. Use a stable digest with enough bits and migrate from the previous percent-encoded directory before switching.

We don't need migration here. SQLite DB is unused right now

cool, as long as we use the actual right identity key for sqlite directory i'm good

alokedesai · 2026-05-15T15:09:32Z

+        anyhow::bail!(
+            "daemon socket path is {path_len} bytes, which exceeds the \
+             platform sun_path limit of {SUN_PATH_MAX} bytes: {}",
+            socket_path.display()


In the case we hit this, what do we or the user do?

This is realistically never going to happen after this fix. It is just a safe guard so we could return clearer errors rather than hitting generic timeout

) ## Description PR warpdotdev#10782 ("Version aware daemon socket") changed daemon socket filenames from `server.sock` (11 chars) to `server-v0.2026.05.13.09.15.stable_01.sock` (41 chars), adding ~30 bytes to the Unix domain socket path. This pushes users over the `sun_path` limit (107 bytes on Linux, 103 on macOS), causing `UnixListener::bind` to fail silently in the daemon. The proxy then times out after 10s, exits with code 1, and the client sees `ResponseChannelClosed`. This accounts for the ~20% drop in remote server initialization success rate (90% → 70%). Sentry issue: WARP-CLIENT-BETA-STABLE-7M9M (11 occurrences across 9 users, all anonymous with 36-char UUID identity keys). **Fix (two changes):** 1. **Hash the version string** in socket/PID filenames to an 8-hex-char suffix (`server-a1b2c3d4.sock`, 20 chars) instead of the full version string (41 chars). 2. **Hash the identity key** directory name to 8 hex chars instead of using the raw identity key (up to 36 chars for anonymous UUIDs). This is needed because longer channel base dirs like `.warp-preview` (+8 chars vs `.warp`) would otherwise narrow the headroom too much on macOS. Both changes use `std::hash::DefaultHasher` for deterministic, fixed-length output. The existing `cleanup_old_versions()` logic automatically cleans up old long-form files. Also adds an explicit `sun_path` length guard in the proxy that fails fast with a clear error message instead of silently timing out. **Worst-case path after fix:** `/home/{32-char-user}/.warp-preview/remote-server/{8-char-hash}/server-{8-char-hash}.sock` = **97 bytes**, well under both limits (103 macOS, 107 Linux). Fixes WARP-CLIENT-BETA-STABLE-7M9M ## Linked Issue - [x] The linked issue is labeled `ready-to-spec` or `ready-to-implement`. ## Testing - 10 unit tests: version hash determinism, identity key hash (length, determinism, uniqueness), socket/PID name length bounds, and worst-case path length regression test against the preview channel (longest base dir) - Validated via Docker containers on linux/amd64, linux/arm64, and macOS/aarch64 with the actual 9 Sentry-reported identity keys - Cross-referenced Sentry events to confirm all affected users are anonymous (36-char UUID identity keys) - `cargo fmt` and `cargo clippy` pass - [x] I have manually tested my changes locally with `./script/run` ## Agent Mode - [x] Warp Agent Mode - This PR was created via Warp's AI Agent Mode Co-Authored-By: Warp <agent@warp.dev>

Fix socket path length

6fb3d4b

cla-bot Bot added the cla-signed label May 15, 2026

kevinyang372 changed the title ~~Fix socket path length~~ Fix remote server socket path exceeding sun_path limit May 15, 2026

oz-for-oss Bot reviewed May 15, 2026

View reviewed changes

Comment thread crates/remote_server/src/setup.rs

hash identity key too

47d774d

kevinyang372 requested a review from alokedesai May 15, 2026 15:00

oz-for-oss Bot reviewed May 15, 2026

View reviewed changes

alokedesai approved these changes May 15, 2026

View reviewed changes

comment

8c70cd6

kevinyang372 enabled auto-merge (squash) May 15, 2026 15:44

kevinyang372 merged commit 18baecd into master May 15, 2026
28 checks passed

kevinyang372 deleted the kevin/fix-socket-path-length branch May 15, 2026 15:50

Conversation

kevinyang372 commented May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Linked Issue

Testing

Agent Mode

Uh oh!

kevinyang372 commented May 15, 2026

Uh oh!

oz-for-oss Bot commented May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Verdict

Uh oh!

Uh oh!

kevinyang372 commented May 15, 2026

Uh oh!

oz-for-oss Bot commented May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Security

Verdict

Uh oh!

oz-for-oss Bot May 15, 2026

Choose a reason for hiding this comment

Uh oh!

kevinyang372 May 15, 2026

Choose a reason for hiding this comment

Uh oh!

alokedesai May 15, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

alokedesai May 15, 2026

Choose a reason for hiding this comment

Uh oh!

kevinyang372 May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

kevinyang372 commented May 15, 2026 •

edited

Loading

oz-for-oss Bot commented May 15, 2026 •

edited

Loading

oz-for-oss Bot commented May 15, 2026 •

edited

Loading

kevinyang372 May 15, 2026 •

edited

Loading