[MaxLikesCode]

What

dup2 from a guest socket now drops the kqueue and the knotes of the descriptor it replaces, as the pipe and file paths of dup2 already do.

GuestSockets::DuplicateTo, which the socket branch of dup2_nid_postfix (and so fcntl F_DUP2FD) uses, called EqueueDescriptorClosed only when the replaced descriptor was itself a guest socket. Two consequences, as #11429 reports:

• a kqueue replaced by a socket stayed registered, so kevent on the number that now names the socket still returned the old queue's events;
• knotes that another kqueue had on a replaced pipe or file stayed attached.

FreeBSD 11.4's kern_dup closes the replaced file through closefp, whose knote_fdclose drops every knote on that descriptor; kevent on a descriptor that is not a kqueue then fails with EBADF (kqueue_acquire), and EV_DELETE of a dropped knote with ENOENT (kqueue_register). DuplicateTo now notifies the event queues after every successful replacement. Merged #3764 already gives dup2 onto a kqueue these semantics for the other sources.

Tested

Windows 11 Pro 26200, WinLibs GCC 15.2.0 r7, full Release build with BUILD_TESTING=ON.

• guest_kqueue gains SocketReplacesDescriptor:
  • a kqueue with a pending one-shot timer is replaced by one end of a socketpair; kevent on that number must fail with EBADF;
  • another kqueue watches a readable pipe read end, which is replaced by the other end of the socketpair; it must then report nothing, and EV_DELETE of that knote must fail with ENOENT.
• With the new test and Udp.cpp from main, guest_kqueue aborts 3 of 3. A local diagnostic build showed both cases failing: kevent on the replaced kqueue returned 1 with the old timer event (ident 1, EVFILT_TIMER, the original udata), and EV_DELETE of the replaced pipe's knote returned 0. With the change it passes 3 of 3.
• Full suite on main 035ac96 with this change: 2472 of 2475 pass. The 3 failures fail the same way on main on this machine: guest_wide_formatting (fix(libc): convert the Windows time zone name from the ANSI code page to UTF-8 for tzname, %Z and tm_zone #10981, German code page, fixed by fix(libc): convert Windows time zone names to UTF-8 #10987), agc_host_import_cut_lifetime (agc_host_import_cut_lifetime fails on Windows since the merge of #9159 #10746) and agc_driver_depth_plane_partial_store (AGC: imported depth-plane partial-store test fails refresh-proof assertion on Windows (NVIDIA and AMD) #10664, fixed by fix(agc): collect upload generation after host import #10696).
• Rebased onto main bf77a9e: the full suite passes 2473 of 2476 with the same 3 failures, and guest_kqueue passes 3 of 3.
• python tools/check_conventions.py --base origin/main passes. python tools/pr_overlap.py --branch fix/dup2-socket-closes-kqueue could not compare 12 conflicting PRs; I checked their diffs by hand. feat(libkernel): report open, read and close notes on watched files #7121 also adds a dup2 case to GuestKqueue.cpp (close notes on a replaced vnode), in another function; none of them change DuplicateTo or this test.
• @B-Howell ran head 01679f8 (comment): guest_kqueue and guest_socket_descriptors pass 2 of 2 on Windows (MSYS2 UCRT64 GCC 16.2, Release) and on Linux (WSL Ubuntu, GCC 11, Release). On Linux, the same test executables with only libkernel.prx built from bf77a9e9cb's Udp.cpp make guest_kqueue abort, while guest_socket_descriptors still passes.
• @babyhuey ran head 01679f8 merged onto main 27ad724 with the test selection of CI's skipped Windows job (Windows 11, RTX 3080, WinLibs GCC 15.2, Release; comment): 1422 of the 2122 selected tests pass and none fail; the other 700 are skipped, as on main there.
• @Gonic0 ran head 01679f8 on macOS (Apple M5 Pro, macOS 27.2, Apple clang 21, x86-64 Release under Rosetta 2), on main 801d5b9 with the open macOS build PRs (comment): guest_kqueue passes 3 of 3 on that base and 3 of 3 with this change, and aborts 3 of 3 with this change's test but only Udp.cpp restored from the base.
• Rebased onto main 2ee8e97 at @MarcZX10's request (head 56c1a45, was 01679f8); the change is unchanged. guest_kqueue passes 3 of 3 and python tools/check_conventions.py --base origin/main passes. The full suite on the same change rebased onto main f2f759a, four commits earlier and none of them in libkernel, passes 2666 of 2667: only guest_wide_formatting fails, the same way as on main here (fix(libc): convert the Windows time zone name from the ANSI code page to UTF-8 for tzname, %Z and tm_zone #10981, fixed by fix(libc): convert Windows time zone names to UTF-8 #10987).
• Rebased again onto main 5454cd2 at @MarcZX10's request (head dcdf9f3, was 56c1a45), after main changed Udp.cpp in other functions; the change is unchanged. Full Release build: guest_kqueue passes 3 of 3, and the full suite passes 3016 of 3018. The 2 failures fail the same way on main on this machine: guest_wide_formatting (fix(libc): convert the Windows time zone name from the ANSI code page to UTF-8 for tzname, %Z and tm_zone #10981, German code page, fixed by fix(libc): convert Windows time zone names to UTF-8 #10987) and guest_freebsd_clocks (Tests: guest_freebsd_clocks aborts where the host zone is not UTC #12736, the host zone here is not UTC, fixed by test(libkernel): pin the zone before the kernel timezone check #12747). python tools/check_conventions.py --base origin/main passes.
• Not tested: bare-metal Linux, a console or a title. The change and the test are not platform-specific.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: Fixes libkernel: dup2 from a socket leaves the replaced kqueue alive #11429
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes. Written with Claude Code (Claude Opus 5.5).

🤖 Generated with Claude Code

[MarcZX10]

Reviewed head 01679f8dbe3856b1f70d8dc0d8ffd10453c00788 against current main 9656d366be903e570594e0726f9bd86f2c32ddb0. The replacement notification is correctly after successful descriptor duplication and is safe for unregistered descriptors: EqueueDescriptorClosed removes a kqueue at that descriptor number and removes descriptor notes from all owning queues. The two regression cases cover replacing a kqueue with a socket and replacing a pipe watched by another kqueue. The latest main commit after the reported bf77a9e9cb base does not touch the changed files. Local tools/check_conventions.py --base upstream/main --head upstream/pr-11445 and git diff --check upstream/main...upstream/pr-11445 pass.

The PR reports the focused guest_kqueue red/green test and full Windows suite after rebasing; I did not independently run that Windows suite. Hosted eligibility, conventions, intake, labels and report checks pass. Build and product jobs were skipped by path policy. I found no blocking issue.

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 10014	core/libs/tests/GuestKqueue.cpp L1627
PR 10016	core/libs/tests/GuestKqueue.cpp L1627

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 5021	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7088	core/libs/prx/libkernel/Socket/src/Udp.cpp core/libs/tests/GuestKqueue.cpp
PR 7121	core/libs/tests/GuestKqueue.cpp
PR 7507	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 8463	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 9843	core/libs/tests/GuestKqueue.cpp
PR 11896	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 12682	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 12688	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 12700	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 12965	core/libs/prx/libkernel/Socket/src/Udp.cpp

[B-Howell]

Independent component qualification of exact head 01679f8dbe3856b1f70d8dc0d8ffd10453c00788:

• Windows, MSYS2 UCRT64 GCC 16.2, Release: guest_kqueue and guest_socket_descriptors pass 2/2.
• Linux, WSL Ubuntu / GCC 11, Release: the same suites pass 2/2 (5.68 s and 0.06 s).
• Linux negative control: with the same PR test executables and dependencies, substituting only libkernel.prx built with baseline bf77a9e9cb's Udp.cpp makes guest_kqueue abort while guest_socket_descriptors still passes.
• Restoring the PR library passes 2/2 again. All 12 archived executable/library/source/configuration/dependency hashes match the original positive capture; only libkernel.prx changed for the negative control.

This qualifies these two component suites, not the full suite, a retail title, macOS or PS5 console behavior. The Linux negative control is a baseline-library substitution, not an unmodified-main full build.

I had independently reached the same implementation locally, but found this PR in the final overlap check and am contributing the qualification here rather than opening a duplicate.

AI-assisted: yes.

[MaxLikesCode]

Thanks @B-Howell for the Windows and Linux runs and the baseline-library control on Linux. I added them to the Tested section.

AI-assisted: yes (Claude Code)

[THEMANnt]

Reviewed 01679f8d against 9656d366. No actionable findings.

GuestSockets::DuplicateTo notifies event queues only after replacement succeeds and after releasing the socket lock. That clears a kqueue registered at the target and removes notes from every owning queue. The regression covers replacing a pending timer kqueue with a socket and replacing a pipe watched by another kqueue, checking EBADF, no event, and ENOENT.

Validation: The exact-head Windows Release build succeeded; focused CTest passed 3 tests, 0 skipped, 0 failed. A pinned current-main control passed the same two descriptor tests. The author also reports a baseline failure with the previous Udp.cpp; I did not independently rerun that negative control. Local conventions and git diff --check passed. Hosted intake, eligibility, labels, conventions, and report checks passed; hosted product jobs were skipped by path policy.

This is local Windows x64 validation. I did not test Linux, macOS, console firmware, or a title.

[babyhuey]

Windows run of this head 01679f8dbe3, merged onto main 27ad7244d22 (clean merge, local only), with the selection CI's skipped Windows product job uses: Windows 11, Ryzen 7 5800X, RTX 3080, WinLibs GCC 15.2, Release, BUILD_TESTING=ON. The tree was configured fresh after the merge, then select_tests.py select / ninja -f select.ninja selected / --target libs / select_tests.py ctest --parallel 4 --timeout 120.

• Selected 2122 tests: 1422 pass, 0 fail, 700 skip. No failures. 27ad7244d22 itself skips the same tests on this machine; one is machine-local (pad_sdl, because a controller is plugged in).

AI-assisted: yes (Claude Code)

[MaxLikesCode]

Thanks @babyhuey for running CI's Windows selection on this head. I added it to the Tested section.

AI-assisted: yes (Claude Code)

[MarcZX10]

Follow-up on branch currency: this PR is still based on 9656d366be, while upstream main is 801d5b9bc1. The exact changed paths (Udp.cpp and GuestKqueue.cpp) are unchanged on current main, and the submitted implementation/tests still look correct; however, the repository asks branches to stay up to date before merge. Please rebase onto current main and rerun the focused guest_kqueue and conventions checks. This supersedes the earlier approval until the current-main tree is checked.

[Gonic0]

macOS run of head 01679f8dbe, for the "Not tested: macOS" line: Apple M5 Pro, macOS 27.2, Apple clang 21, x86-64 Release under Rosetta 2.

Tree: main 801d5b9bc1 plus the open macOS build PRs (#7631, #11249, #11358, #10669, #10552, #7804, #11447, #11437, #11554), with this PR's diff (bf77a9e9cb..01679f8dbe) on top. The diff applies cleanly there.

	guest_kqueue
base, without this PR	passes, 3 of 3
with this PR	passes, 3 of 3
this PR with only Udp.cpp restored from the base	aborts, 3 of 3

Same pattern as your Windows run and B-Howell's Linux control.

AI-assisted: yes

[MaxLikesCode]

@MarcZX10 rebased onto main 2ee8e97 (head 56c1a45, was 01679f8); the change itself is unchanged. guest_kqueue passes 3 of 3 on Windows and tools/check_conventions.py --base origin/main passes. The full suite on the same change rebased onto main f2f759a, four commits earlier and none of them in libkernel, passes 2666 of 2667: the one failure, guest_wide_formatting, fails the same way on main here (#10981).

Thanks @Gonic0 for the macOS run with the Udp.cpp control. I added it and the rebase to the Tested section.

AI-assisted: yes (Claude Code)

[MarcZX10]

Thanks for rebasing this PR onto main 2ee8e9751a and reporting the focused Windows rerun. The change and regression tests still apply cleanly in a merge preview against current main, and the newer Udp.cpp edits are in separate functions from the DuplicateTo hunk.

The branch is still behind current main 4c7d275bef89, which has since changed Udp.cpp. Please rebase onto the current tip and rerun guest_kqueue plus conventions on that exact integrated head. The earlier Windows, Linux, and macOS reports remain useful evidence for the reviewed head, but do not cover this newer base.

I did not run a local build or test for this PR in this workspace.

[MaxLikesCode]

@MarcZX10 rebased onto main 5454cd2 (head dcdf9f3, was 56c1a45); the change itself is unchanged, and main's newer Udp.cpp edits are in other functions. On that exact head, full Release build on Windows: guest_kqueue passes 3 of 3 and tools/check_conventions.py --base origin/main passes. The full suite passes 3016 of 3018; the two failures, guest_wide_formatting (#10981) and guest_freebsd_clocks (#12736: this machine's zone is not UTC), fail the same way on main here. I added this to the Tested section.

AI-assisted: yes (Claude Code)