[Queuse]

What

Reject nonpositive SO_SNDBUF and SO_RCVBUF values with EINVAL before changing the host socket or cached guest size. Previously zero/negative sizes reached the host, producing host-specific success or errors. This follows FreeBSD 11.4 sosetopt; PS5 validation and host buffer-limit differences remain in debt.

Tested

Windows 11 x86-64, WinLibs GCC 15.2.0 UCRT, Release. guest_net_buffer_validation fails on unchanged main and passes here. IPv4/IPv6 TCP/UDP tests reject zero, negative values and INT_MIN through both sceNet and kernel setters, verify guest and native buffer sizes stay unchanged, and check later valid sets, invalid pointers/lengths and descriptor precedence. The three new focused network tests pass together (3/3). Conventions and test-registration audit pass. No Linux, PS5 or title run.

Maintenance validation (2026-10-22): merged main b7d1d47a5f into f555d351c8. Conventions and whitespace checks pass. The focused target builds on Windows x86-64, MSYS2 UCRT64 GCC 16.2.0, Ninja, Release. Runtime validation is blocked: the host lacks required AVX2, FMA, BMI1/BMI2, LZCNT and MOVBE. Both guest_net_boolean_readback and the unchanged guest_sce_net fixture abort in the unchanged main startup guard before assertions. A collaborator needs to run the Build workflow for the refreshed head on a compatible runner. Earlier runtime results remain tied to their stated revisions. Overlap: #6879 only adds a system header to libSceNet, and #7201 changes macOS resolver handling; neither changes these socket options. The current #8362 head merges cleanly with #9544 and #9542 after its author moved the debt entry. Order with #6044 and #5836 is awaiting author confirmation; no agreement is assumed. Retain #9523 before #9542 and #9525 before #9543.

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

[MarcZX10]

Reviewed the full diff against current main. The nonpositive guard runs after descriptor, level, pointer/length and option validation, before calling native setsockopt or updating the guest cached size; the test covers IPv4/IPv6, stream/datagram, sceNet and kernel paths, native/cached state preservation and valid follow-up calls. The overlap scanner found no current open-PR overlap. Hosted eligibility, conventions, intake, labels and report checks pass; product build jobs were skipped by path policy. Windows results and the focused 3/3 tests are reported by the author; I did not run a local build. Approved.

[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 5836	core/libs/prx/libSceNet/CMakeLists.txt core/libs/prx/libSceNet/Export.cpp docs/dev/debt/libSceNet.md
PR 6044	core/libs/prx/libSceNet/CMakeLists.txt
PR 6879	core/libs/prx/libSceNet/Export.cpp
PR 7201	core/libs/prx/libSceNet/Export.cpp
PR 8254	core/libs/prx/libSceNet/CMakeLists.txt core/libs/prx/libSceNet/Export.cpp docs/dev/debt/libSceNet.md
PR 9060	core/libs/prx/libSceNet/Export.cpp docs/dev/debt/libSceNet.md
PR 9651	docs/dev/debt/libSceNet.md

[jotitadev]

Linux run, since the description lists only Windows 11. On Linux the host accepts every nonpositive size, so the failure on main is a silent buffer change rather than an error.

Machine: Linux x86_64, i7-8700, Ubuntu 26.04 (kernel 7.0.0-38), GCC 15.2, Release, BUILD_TESTING=ON. Head 4e7559ae9 merged --no-commit onto main ab347d144 (clean).

• guest_net_buffer_validation passes, 3 of 3 runs. The other 20 libSceNet tests (guest_sce_net, net_dontwait_*, net_blocking_send, net_epoll_lifetime) pass with it.

• Red→green: dropping only the new value < 1 line from sceNetSetsockopt and keeping the test aborts at line 58 (sceNetSetsockopt(..., invalid) == 0x80410116): the call returns 0 instead of EINVAL.

• Why it returns 0 on Linux: the host setsockopt(SO_SNDBUF/SO_RCVBUF) never rejects a size. Measured on the same box with plain host sockets (IPv4 TCP/UDP, IPv6 TCP), after setting 32768 (read back as 65536):

  value	setsockopt	SO_SNDBUF read back	SO_RCVBUF read back
  0	ok	4608	2304
  -1, -4096, INT_MIN	ok	8388608	8388608

  Linux takes the int as unsigned and clamps it to wmem_max/rmem_max (4 MiB here, doubled), so without this PR a guest that passes -1 on Linux gets the largest host buffer, and 0 gets the smallest. The nativeBuffer() == originalNative checks in the test cover exactly that case, so the test guards it on Linux too.

Written with AI assistance (Claude Code). I ran the builds and tests above on my machine.

[MarcZX10]

I compared this against open #8362 on the exact shared debt file. The implementations/tests touch different functionality, but both branches add their debt entry at the same insertion point in docs/dev/debt/libSceNet.md; a pairwise three-way merge currently conflicts in that file. Please coordinate the merge order with #8362’s author and record the agreed order in the overlap note / Depends on if one PR must land first. The individual PR merges cleanly with current main, but its sync check is 1 ahead / 119 behind, so please also refresh it onto current main before merge.

[Queuse]

@MarcZX10 I rechecked the updated #8362 head against this branch after merging current main (b7d1d47a5). The pairwise three-way merge is clean, including the debt file. @MarrasFederico's moved entry removes the insertion conflict, so neither PR needs to wait for the other. Thanks for sorting that out.