You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(tailscale): take tailscale-rs from git to stop a UDP close panic - #1584
With the tailscale outbound, clash-rs aborts with a smoltcp panic:
panicked at smoltcp-0.13.1/src/iface/socket_set.rs:116:
handle does not refer to a valid socket
It happens whenever a UDP socket on the outbound is closed while a receive is still pending. Two common triggers:
a DNS nameserver routed through the outbound (nameserver-policy: "+.ts.net": "100.100.100.100#<tailscale proxy>"). tailscale-rs doesn't implement MagicDNS, so the query is never answered, times out, and the socket is dropped.
any UDP flow through the outbound that ends without a reply.
On iOS / Android, where clash-rs runs inside the VPN process, this takes the whole tunnel down.
Root cause, in tailscale-rs 0.6.1 netcore: a Recv that returns WouldBlock is parked in blocked_commands. When the socket is dropped, its Close removes the handle from the smoltcp SocketSet, but the parked Recv stays queued. The next pump_blocked_commands calls SocketSet::get_mut on the removed handle, which panics.
tailscale/tailscale-rs@9f7cec0 ("netcore: configure requests to cancel on drop", closestailscale/tailscale-rs#447) skips requests whose caller has gone away. It isn't released yet; crates.io is still at 0.6.1, and the git workspace is still versioned 0.6.1.
Changes
[patch.crates-io]: take tailscale from tailscale-rs git at 9f7cec0447.
Lockfile: only tailscale-rs crates change. tailscale-rs main has merged the ts_netstack_smoltcp* crates into one, so those entries collapse, and smoltcp goes from 0.13.1 to 0.14.0 (the 0.12 copy used elsewhere is untouched).
Remove the patch after the next tailscale-rs release.
Testing
Live tests on macOS (clash-rs binary, mixed inbound, against a real tailnet with subnet routers):
case
master
this PR
TCP to a domain whose lookup goes to 100.100.100.100#tailscale and times out
❌ panic
✅ times out, process keeps running
5 SOCKS5 UDP flows to a subnet host that never replies
❌ panic
✅
HTTP to a subnet host through a subnet router (10 requests)
✅
✅
UDP DNS query to a subnet host
✅
✅
Builds, all with --locked:
cargo check --workspace on macOS
cargo check -p clash-lib --no-default-features -F ring,shadowsocks,tuic,ssh,tun,zero_copy,tailscale for aarch64-apple-ios and for aarch64-apple-tvos (nightly, -Zbuild-std)
cargo test -p clash-lib --lib -F tailscale,tuic tailscale: 10 passed
-F tuic is needed because without it the wind-tuic dev-dependency fails to compile (unresolved import crate::quinn::outbound), the same problem noted in #1583. The tailscale live test (tailscale_live_auth_key_supports_real_tcp_and_udp_traffic) is Linux-only and wasn't run.
Risk: the pinned rev is 35 commits past v0.6.1, and it applies on every platform. The behaviour checked above is TCP/UDP through the netstack, subnet routes, and auth-key login.
Summary by CodeRabbit
Chores
Updated internal project configuration. This update does not add or remove user-facing features, change app behavior, or alter how existing features are accessed. No changes to the user interface or user workflows are included.
tailscale-rs 0.6.1 panics with "handle does not refer to a valid socket"
when a UDP socket is closed while a recv is still pending, e.g. after a
DNS query through the tailscale outbound times out, or when any UDP flow
without a reply ends. tailscale/tailscale-rs@9f7cec0 skips requests
whose caller has gone away. Patch `tailscale` to that rev; `ts_netmon`
moves from 2c67f76 to the same rev (2 commits later).
No actionable comments were generated in the recent review.
Walkthrough
The Cargo patch configuration now pins ts_netmon and tailscale to revision 9f7cec0447 of tailscale-rs. Comments identify an iOS PlatformMon re-export fix and handling for netstack requests whose caller was dropped.
Changes: tailscale-rs patch pins
Layer / File(s)
Summary
Update patch revisions Cargo.toml
The patch configuration updates the ts_netmon pin and adds a tailscale pin at revision 9f7cec0447. Comments describe changes at that revision.
No concrete merge-blocking issue is identified in the dependency pin update. Merge readiness remains subject to normal build and test checks.
Pre-merge checks: 5
Check name
Status
Explanation
Description Check
Check skipped - CodeRabbit’s high-level summary is enabled.
Title check
The title describes the dependency pin and its purpose: preventing a UDP close panic.
Linked Issues check
[#447] is closed and supplies historical context only. No active directly linked issue coding requirements apply.
Out of Scope Changes check
The Cargo.toml pins support the stated UDP close panic fix. The PR description reports related dependency lockfile updates. Cargo.lock is excluded by !**/*.lock, so its contents were not reviewed.
Docstring Coverage
No eligible functions found in the changed files. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentat…
Full details: Docstring Coverage
Explanation
No eligible functions found in the changed files. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentation-only edits and deletions within surviving functions are included; unchanged diff context and fully deleted functions are excluded. Analyzed 0 functions across 0 files. (1 skipped: 1 unsupported.)
Finishing Touches
Generate unit tests (beta)
Commit to this branch
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
File: No configuration file from this repository applied
Commit: Unavailable
Layers: defaults
Review profile: CHILL
Plan: Advanced
Run ID: 925426a0-b93a-41d0-880c-ff0bf5f83ef0
Commits
Reviewing files that changed from the base of the PR and between f10e4eb and ea0347b.
Files ignored due to path filters: 1
Cargo.lock is excluded by !**/*.lock
Files selected for processing: 1
Cargo.toml
Included review availability: This review used your included allowance. Your plan provides up to 4 included reviews per hour; 2 remain after this review.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
With the
tailscaleoutbound, clash-rs aborts with a smoltcp panic:It happens whenever a UDP socket on the outbound is closed while a receive is still pending. Two common triggers:
nameserver-policy: "+.ts.net": "100.100.100.100#<tailscale proxy>"). tailscale-rs doesn't implement MagicDNS, so the query is never answered, times out, and the socket is dropped.On iOS / Android, where clash-rs runs inside the VPN process, this takes the whole tunnel down.
Root cause, in tailscale-rs 0.6.1 netcore: a
Recvthat returnsWouldBlockis parked inblocked_commands. When the socket is dropped, itsCloseremoves the handle from the smoltcpSocketSet, but the parkedRecvstays queued. The nextpump_blocked_commandscallsSocketSet::get_muton the removed handle, which panics.tailscale/tailscale-rs@9f7cec0 ("netcore: configure requests to cancel on drop", closes tailscale/tailscale-rs#447) skips requests whose caller has gone away. It isn't released yet; crates.io is still at 0.6.1, and the git workspace is still versioned 0.6.1.
Changes
[patch.crates-io]: taketailscalefrom tailscale-rs git at9f7cec0447.ts_netmonmoves from2c67f76eab(added in fix(apple): build clash-lib for iOS/tvOS without patched dependencies #1583) to the same rev, which is 2 commits later. All tailscale-rs crates now come from one rev.ts_netstack_smoltcp*crates into one, so those entries collapse, and smoltcp goes from 0.13.1 to 0.14.0 (the 0.12 copy used elsewhere is untouched).Testing
Live tests on macOS (
clash-rsbinary, mixed inbound, against a real tailnet with subnet routers):100.100.100.100#tailscaleand times outBuilds, all with
--locked:cargo check --workspaceon macOScargo check -p clash-lib --no-default-features -F ring,shadowsocks,tuic,ssh,tun,zero_copy,tailscaleforaarch64-apple-iosand foraarch64-apple-tvos(nightly,-Zbuild-std)cargo test -p clash-lib --lib -F tailscale,tuic tailscale: 10 passedcargo clippy -p clash-lib --all-targets -F tailscale,tuic: clean-F tuicis needed because without it thewind-tuicdev-dependency fails to compile (unresolved import crate::quinn::outbound), the same problem noted in #1583. The tailscale live test (tailscale_live_auth_key_supports_real_tcp_and_udp_traffic) is Linux-only and wasn't run.Risk: the pinned rev is 35 commits past
v0.6.1, and it applies on every platform. The behaviour checked above is TCP/UDP through the netstack, subnet routes, and auth-key login.Summary by CodeRabbit