[KernelPanicNT]

What

__opendir2 no longer throws for flags outside DTF_HIDEW | DTF_NODUP | DTF_REWIND. It now handles every flag value the way FreeBSD 11.4 does.

Source of the behaviour. FreeBSD 11.4 libc, the release the libSceUsbd and libScePosixForWebKit debt entries already pin:

• __opendir2 returns NULL before it opens anything, with errno unchanged, when __DTF_READALL (0x8) or __DTF_SKIPREAD (0x10) is set. Any other value goes to __opendir_common.
• __opendir_common stores the flags in dd_flags and acts only on DTF_NODUP, and only when fstatfs reports unionfs or MNT_UNION. No guest filesystem reports either. DTF_REWIND is defined but never read by libc. DTF_HIDEW only makes readdir skip DT_WHT entries, and host filesystems never produce those. Bits above 0x10 are stored and never read. The flag values are in dirent.h.
• opendir is __opendir2(name, DTF_HIDEW|DTF_NODUP). So every value without 0x8 or 0x10 opens the same stream as the in-tree opendir_nid_postfix.
• The guest DIR is opaque: FreeBSD declares struct _dirdesc only in gen-private.h, and the AnyPS5 stream is a host record. That leaves no guest-visible dd_flags to fill.

Lines that change on main:

• Directory.cpp L26-L32, which throws for any other flag;
• the debt entry;
• GuestOpendir2.cpp L32-L42.

The device's answer was not measured, and the debt entry still says so.

Changes:

• libSceLibcInternal/Directory.cpp: __opendir2 returns null with errno untouched when 0x8 or 0x10 is set. Otherwise it calls opendir. The NotImplemented_nid_no_patch path is gone.
• guest_opendir2:
  • 0x0, 0x1, 0x20, 0x27, 0x100, 0x40000000 and 0xffffffe7 list the same entries as opendir;
  • 0x8, 0x10, 0x18, 0xf, 0x1f, 0x7fffffff and -1 return null and leave errno at a sentinel, both for an existing directory and for a null name;
  • a missing path with 0x20 fails with opendir's errno.
• docs/dev/debt/libSceLibcInternal.md: the entry is cut to what stays open, which is that the device's answer is not measured.

Tested

Linux (GitHub Codespace, native build of this branch on bb04a4f1ea):

• guest_opendir2 passes.
• Negative control: with Directory.cpp from main, guest_opendir2 aborts (flag 0x20 reaches NotImplemented_nid_no_patch).

Checks:

• python3 tools/check_conventions.py --base origin/main passes.
• tools/progress.py --compare gives libSceLibcInternal 72/75 to 73/75 (+1, __opendir2_nid_postfix).

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: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

[github-actions]

Progress report (bb04a4f → 03fc7bb)

📈 System libraries: 46.5% (+0.01%, +1 implemented functions)

✅ 1 implemented

Library	Name
libSceLibcInternal	__opendir2_nid_postfix

Changed libraries (implemented exports)

Library	Before	After	Change
libSceLibcInternal	72/75 (96.00%)	73/75 (97.33%)	+1.33 pp

[rooklz]

Reviewed head 03fc7bbca3 against main bb04a4f1ea, on macOS arm64. I only checked it on the host. I did not build or run anything on the console.

Gates

• tools/check_conventions.py --base upstream/main passes.
• git diff --check is clean, and git merge-tree against bb04a4f1ea is clean.
• pr_overlap merged 12 of 12 PRs. The only shared file is the debt file: #5021 and #10749 already conflict with main on it, so neither could be compared.

Source check (FreeBSD d106002967)

• __opendir2 (opendir.c L83-84) returns NULL with errno untouched for __DTF_READALL|__DTF_SKIPREAD before _open, so a null name never gets dereferenced. This matches the PR's null-name rows.
• __opendir_common stores dd_flags (L293) and branches only on DTF_NODUP, and only for unionfs/MNT_UNION (L312-317).
• In readdir.c, DTF_HIDEW is read only for DT_WHT (L85). The __DTF_* bits are internal state that readdir, rewinddir and telldir set and clear. Nothing reads bits above 0x10.
• opendir is __opendir2(name, DTF_HIDEW|DTF_NODUP) (L61).

So "every value without 0x8/0x10 is opendir" holds as long as there is no union mount and the DIR stays opaque.

Standalone harness
I used the verbatim PR body and main body over host opendir, with the PR's test rows ported 1:1, built with clang++ -O1 and ASan+UBSan:

• PR head: PASS.

• main: aborts at the first new flag (0x20 → NotImplemented). This matches the author's negative control.

• Mutants:

  • only 0x8 checked: KILLED (0x10 row)
  • only 0x10 checked: KILLED
  • sets errno=EINVAL: KILLED (errno sentinel row)
  • rejects all of ~0x7: KILLED (0x20/0x100 rows)
  • requires both bits: KILLED

  All 5 mutants were killed.

Not run

• The real guest_opendir2 target. Submodules are not initialised in my lane. The PR's windows check passed, and linux was skipped by CI.
• The device's answer for any flag. The debt line says this openly, which is the right call.

Verdict: APPROVE.

[github-actions]

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	docs/dev/debt/libSceLibcInternal.md
PR 11749	docs/dev/debt/libSceLibcInternal.md
PR 11750	docs/dev/debt/libSceLibcInternal.md
PR 11751	docs/dev/debt/libSceLibcInternal.md

[JJenkx]

Windows run of this head (03fc7bbca3), merged onto main bb04a4f1ea (clean merge, local only). Windows 11 Enterprise 10.0.22621, Ryzen 9 5950X, WinLibs GCC 15.2.0 UCRT r7, Ninja, Release, BUILD_TESTING=ON, no Vulkan driver visible.

• This PR: guest_opendir2 passes 3 of 3 (ctest --repeat until-fail:3).
• Negative control: with this PR's source files (core/libs/prx/libSceLibcInternal/Directory.cpp) put back to main's version and its tests kept, guest_opendir2 fails.

AI-assisted: yes (Claude Code).

[adriadam10]

Checked __opendir2 against FreeBSD 11.4.0 opendir.c (NULL before opening when __DTF_READALL/__DTF_SKIPREAD is set, everything else goes to the common path) and the DTF_* values in dirent.h; the new test rows pin both branches and the errno behaviour, and flag 0x20 threw on main; merged.