[xXPBSXx]

What

setsockopt_nid_postfix and getsockopt_nid_postfix (Udp.cpp) answered ENOPROTOOPT for every level except SOL_SOCKET, IPPROTO_TCP and IPPROTO_IPV6. FreeBSD 11.4 hands such a level to the protocol, and ip_ctloutput (ip_output.c#L943-L944), ip6_ctloutput (ip6_output.c#L1459) and uipc_ctloutput (uipc_usrreq.c#L1228-L1229) fail every level they do not own with EINVAL. Now an unknown level fails with EINVAL. The levels a protocol owns keep ENOPROTOOPT for the options nothing here implements: IPPROTO_IP on an AF_INET socket and IPPROTO_UDP on a datagram socket (udp_ctloutput); level 0 on an AF_UNIX socket fails with EOPNOTSUPP for the options uipc_ctloutput does not handle; LOCAL_PEERCRED (get), LOCAL_CREDS and LOCAL_CONNWAIT, which it handles, throw as unimplemented. The Technical debt entry is not touched: main has since split the getsockopt, setsockopt entry into measured per-option entries, and its unmeasured sentence on the IPPROTO_IP level is now only exact for AF_INET sockets (this change answers EINVAL on AF_INET6 and EOPNOTSUPP/EINVAL on AF_UNIX); that is left for a separate note.
Fixes #8447
Overlap: git merge-tree of this head with each of #7490, #7475, #7485, #7090, #9515 and #9578 (every one on the merge base git finds for the pair) is clean, and so is this head against main 160f6592b5. #7511, which changed the same function bodies, was closed on 25 October; #8392, #7850 and #8312 are merged. pr_overlap was not run. #10320 (merged) changed the same two lines and routes level 0 to SetIpLevelOption/GetIpLevelOption; here level 0 reaches UnsupportedLevelError only on an AF_UNIX socket, so the PS5-measured AF_INET and AF_INET6 answers of #10320 are unchanged.

Tested

First measurement, Windows 10 (19045), MinGW-w64 GCC 15.2.0, Release, on upstream/main 02b5a8e + the first commit:

• guest_socket_options (the only test registered under that name): fails without the change to Udp.cpp (an unknown protocol level on a stream socket must fail with EINVAL (22), ip_ctloutput (errno 42), exit 0xc0000409) and passes with it.

• 40 socket and network tests (guest_socket_*, guest_sce_net, guest_udp, guest_tcp, guest_poll, guest_select*, guest_datagram_*, guest_recvmsg_memory) pass.

• python tools/check_conventions.py --base upstream/main: no finding.

• Rebased onto main b607bfdc8b (head 425ada08af; conflict only in Udp.cpp, one line in each of setsockopt_nid_postfix and getsockopt_nid_postfix, from fix(libkernel): keep IP_TTL, IP_TOS and the multicast options as measured on a PS5 #10320 routing level 0): Windows 10 19045, MinGW-w64 GCC 15.2.0, Release, both arms on the same base and rebuilt: guest_socket_options and guest_socket_stream_options pass 4 of 4 runs; with Udp.cpp restored to main, guest_socket_options fails 4 of 4 (exit code 0xc0000409) and guest_socket_stream_options passes 4 of 4. Linux (WSL2 Ubuntu 24.04, virtualised kernel, not bare-metal), GCC 13.3.0, Release, one run per arm: both pass; with Udp.cpp restored to main, guest_socket_options aborts and guest_socket_stream_options passes.

• Not tested: bare-metal Linux, macOS, a console, a title run. Not covered by the new test: level 17 on an AF_INET6 datagram socket and any console behaviour of the throwing options.

Checklist

• Based on current main; no other open PR implements the same functions (fix(libkernel): check socket address arguments before the descriptor #8392, merged, touched the same two functions for pointer and length checks, not the level)
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw; no silent stubs added
• No notes, investigation .md files or images added to the repository
• New third-party code is a submodule built from source (none added)
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: libkernel: setsockopt/getsockopt with an unknown level fails with ENOPROTOOPT where FreeBSD 11.4 fails with EINVAL #8447
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

AI-assisted (Claude Code).

🤖 Generated with Claude Code

[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	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7088	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7201	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7475	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 8605	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 9509	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 9713	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 9948	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 10621	core/libs/prx/libkernel/Socket/src/Udp.cpp

[xXPBSXx]

Rebased onto main 1896871ba7 (head 01b3b7e9a9, same commit). The conflict was only in Udp.cpp: main added ReceiveAnsweredHere next to where UnsupportedLevelError goes; both functions are kept. Windows 10 19045, WinLibs GCC 15.2.0 r7, Release: ctest -R '^guest_socket_options$' passes 3 of 3; the socket options, udp, tcp, socket_pair, poll and socket_address_length tests pass (6 of 6). check_conventions passes (one existing length warning on the long getsockopt debt entry); git merge-tree against main is clean. I did not re-run the test without the change on this head.

AI-assisted (Claude Code).

[harisibrar1]

Checked against main 522e32e and FreeBSD 11.4 sys/kern/uipc_usrreq.c, sys/netinet/udp_usrreq.c and sys/netinet/ip_output.c:

• Main still answers every level other than 0xffff/6/41 with ENOPROTOOPT (Udp.cpp:1148, 1160), and git merge-tree against main is clean. orezraey's #10320 changes the same two lines with PS5 13.60 measurements (gist d9efe0da…, pinet4): level 99 gives EINVAL, IPPROTO_IP on AF_INET6 gives EINVAL, and an unknown IP option gives ENOPROTOOPT. Those agree with this PR, so they may be worth citing. One difference: #10320's blanket EINVAL would also cover level 17 on a datagram socket, where udp_ctloutput (udp_usrreq.c:982) keeps ENOPROTOOPT as this PR does, so whichever lands second should keep that.
• AF_UNIX, level 0: uipc_ctloutput (uipc_usrreq.c:1237-1309) supports LOCAL_PEERCRED (1, get), LOCAL_CREDS (2) and LOCAL_CONNWAIT (4); only other options get EOPNOTSUPP. With this change (UnsupportedLevelError, Udp.cpp:729), setsockopt(unixfd, 0, 2, &one, 4) fails with EOPNOTSUPP where 11.4 returns 0. Could those three options throw (NotImplemented_nid_no_patch) and the rest return EOPNOTSUPP?
• Small test note: option 1 at IPPROTO_UDP is UDP_ENCAP (netinet/udp.h), which 11.4 handles on an IPSEC kernel (udp_usrreq.c:1004-1012). An option number with no case would test ENOPROTOOPT without depending on the kernel config.

Reviewed with AI assistance (Claude Code); the checks above were run locally on my machine.

[github-actions]

Progress report (acb900c → 77c7232)

📉 System libraries: 50.17% (-0.02%, -2 reverted functions)

⚠️ 2 went back to stubs

Library	Name
libkernel	getsockopt_nid_postfix
libkernel	setsockopt_nid_postfix

Changed libraries (implemented exports)

Library	Before	After	Change
libkernel	683/721 (94.73%)	681/721 (94.45%)	-0.28 pp

[xXPBSXx]

@harisibrar1 thanks for the review. Pushed b6ca1f3660 on top (no force):

• AF_UNIX level 0: getsockopt of LOCAL_PEERCRED, LOCAL_CREDS and LOCAL_CONNWAIT and setsockopt of LOCAL_CREDS and LOCAL_CONNWAIT now throw as unimplemented, as uipc_ctloutput answers them; every other option at level 0 fails with EOPNOTSUPP (45), setsockopt of LOCAL_PEERCRED included (11.4 has no set case for it). Any other level on an AF_UNIX socket fails with EINVAL.
• The IPPROTO_UDP test uses option 77, which has no case in udp_ctloutput, instead of UDP_ENCAP. New AF_UNIX cases cover the level, the unknown option and the throwing options.
• Level 17 on a datagram socket stays ENOPROTOOPT, as udp_ctloutput does; whichever of this PR and fix(libkernel): keep IP_TTL, IP_TOS and the multicast options as measured on a PS5 #10320 lands second should keep that. Thanks for pointing at fix(libkernel): keep IP_TTL, IP_TOS and the multicast options as measured on a PS5 #10320: its PS5 13.60 measurements (level 99 and IPPROTO_IP on AF_INET6 give EINVAL, an unknown IP option gives ENOPROTOOPT) agree with this PR; I have not measured on a console myself.
• Both runs on the same base main 379b1dca32, rebuilt each time. Windows 10 19045, WinLibs GCC 15.2.0: guest_socket_options and guest_socket_stream_options pass 3 of 3; with Udp.cpp as before this commit and the new tests, guest_socket_options aborts 3 of 3. Linux (WSL2 Ubuntu 24.04, virtualised kernel, not bare-metal): both pass 2 of 2; with the old Udp.cpp, guest_socket_options aborts.

AI-assisted (Claude Code).

[harisibrar1]

Re-checked head b6ca1f3 against main b607bfd:

• The AF_UNIX change matches uipc_ctloutput: on get, LOCAL_PEERCRED/LOCAL_CREDS/LOCAL_CONNWAIT now throw; on set, only LOCAL_CREDS/LOCAL_CONNWAIT throw (11.4's SOPT_SET switch has no LOCAL_PEERCRED case, so EOPNOTSUPP there is right), and every other level-0 option gets EOPNOTSUPP. Option 77 at IPPROTO_UDP avoids the IPSEC dependency. Thanks.
• #10320 has merged since, and it changed the same two lines: main now has if (level != 0xffff && level != 0 && level != 6 && level != 41) return Fail(22); in both setsockopt and getsockopt (3-way git merge-tree shows two conflicts in Udp.cpp, plus GuestSocketOptions.cpp changed on both sides). Two things to keep in mind when resolving:
  • Level 0 is now routed to SetIpLevelOption/GetIpLevelOption, which return EINVAL for anything that isn't IPv4 (SocketOptions.cpp:283, :320). So for the AF_UNIX handling to take effect, the socket.family == GuestUnix check has to run before that dispatch; otherwise every AF_UNIX level-0 call becomes EINVAL.
  • Main's blanket Fail(22) also covers IPPROTO_UDP (17) on a datagram socket, where your helper keeps udp_ctloutput's ENOPROTOOPT, so keeping UnsupportedLevelError for the non-zero levels preserves that.
• With #10320's PS5 13.60 measurements now on main (level 99 and IPPROTO_IP on AF_INET6 → EINVAL), the remaining parts of this PR are the AF_UNIX and IPPROTO_UDP cases.

Reviewed with AI assistance (Claude Code); the checks above were run locally on my machine.

[xXPBSXx]

Rebased onto main b607bfdc8b (head 425ada08af). The conflict was the level check in setsockopt_nid_postfix and getsockopt_nid_postfix from #10320, as you described. I kept level 0 on AF_INET and AF_INET6 going to SetIpLevelOption/GetIpLevelOption and let only AF_UNIX reach UnsupportedLevelError for level 0, so UnsupportedLevelError and the tests are unchanged. guest_socket_options and guest_socket_stream_options pass 4 of 4 on Windows and once on Linux (WSL2); with Udp.cpp restored to main on the same base guest_socket_options fails 4 of 4 on Windows and aborts on Linux, guest_socket_stream_options passes. The ownFamily term in UnsupportedLevelError (AF_INET, level 0) can no longer be reached, because #10320 answers that case; I left it so the patch stays as reviewed, and can remove it in a follow-up commit if you prefer.

AI-assisted (Claude Code).

[wachi35676]

Tested head 425ada08af merged onto main 49d1f9b07e, Windows 11 (WinLibs GCC 15.2.0, Release) and WSL2 Ubuntu 24.04 (GCC 13.3), BUILD_TESTING=ON.

• guest_socket_options, guest_socket_stream_options, guest_socket_error, guest_tcp, guest_udp, guest_socket_pair: 6/6 pass, 2 runs, on both hosts. With Udp.cpp from main, guest_socket_options fails on both and the other five pass.
• A probe of the cross-protocol levels gives EINVAL on both hosts, as FreeBSD 11.4's ip_ctloutput/ip6_ctloutput do: level 6 on a UDP/IPv4 socket, level 41 on TCP and UDP IPv4 sockets, level 17 on a TCP/IPv6 socket, and level 0 on a TCP/IPv6 socket.

One test note. SetFails(stream, 0, 1, &one, sizeof(one), ErrorNoProtocolOption) uses option 1, which is IP_OPTIONS. FreeBSD 11.4 handles it: ip_ctloutput passes the 4 bytes 01 00 00 00 to ip_pcbopts (ip_options.c:600), which reads a NOP, then an EOL, and returns 0. So the assertion pins an answer FreeBSD does not give, as harisibrar1 pointed out for UDP_ENCAP. The unknown IP option measured on the PS5 in #10320's pinet4.log is 0x7777 (ip_unknown ret=-1 errno=42). Using that number would test the same path and match the console.

AI-assisted: yes (Claude Code); results above were run on this machine.

[xXPBSXx]

Thank you for the runs and for the note on the test. You are right about option 1: at level 0 it is IP_OPTIONS, and in FreeBSD 11.4 ip_ctloutput hands the four bytes to ip_pcbopts, which reads a NOP, then an EOL, and returns 0 (sys/netinet/ip_output.c, case IP_OPTIONS; sys/netinet/ip_options.c:600). The check therefore asserted an answer FreeBSD does not give. The new commit 77c7232681 changes that line to option 0x7777, the number the IPPROTO_IP getsockopt check in the same file already uses, so it still tests the unhandled-option path. On Windows 10 (MinGW-w64 GCC 15.2, Release), the six socket tests guest_socket_error, guest_socket_options, guest_socket_stream_options, guest_udp, guest_tcp and guest_socket_pair pass in 3 of 3 runs with the new commit (ninja exit 0); I did not run Linux, and the change has no failing arm because it only moves the check to a number that returns ENOPROTOOPT either way.

AI-assisted (Claude Code).

[wachi35676]

Re-checked head 77c7232681 merged onto main 0f34ebb85 (clean). Windows 11 with WinLibs GCC 15.2.0, and WSL Ubuntu 24.04 with GCC 13.3.

• guest_socket_options, guest_socket_stream_options, guest_socket_error, guest_tcp, guest_udp and guest_socket_pair pass 6/6 on both hosts, 3 runs. This covers the Linux run you didn't do.
• With Udp.cpp from main, guest_socket_options still fails on Windows. Option 0x7777 at level 0 reaches ip_ctloutput's default ENOPROTOOPT in FreeBSD 11.4, so the check now matches both the source and the console log.

AI-assisted: yes (Claude Code); results above were run on this machine.