[nicolasfalesy]

What

Follow-up to #4391. With libSceNet sockets on ordinary descriptor numbers, a title that makes a libSceNet socket non-blocking the POSIX way, fcntl(fd, F_GETFL) then fcntl(fd, F_SETFL, flags | O_NONBLOCK), reaches File::Fcntl, whose classification throws for a socket:

what():  fcntl: descriptor 47 is neither a file, a directory, a pipe nor a character device

libSceNet now registers a status control with libkernel, the same way it registers its option setter (KernelSetSocketStatusControl_nid_no_patch). fcntl asks it before falling back to File::Fcntl, and for a libSceNet socket it behaves as fcntl already does for libkernel sockets: F_GETFL returns O_RDWR plus O_NONBLOCK when set, F_SETFL applies O_NONBLOCK through SO_NBIO (so sceNetGetsockopt sees the same state), flags outside the status mask fail with EOPNOTSUPP and other commands with EINVAL.

This branch contains #4391's commits; the change here is the last commit (1836f8e). It only makes sense on top of #4391: on main, libSceNet numbers can collide with host descriptors.

Tested

Linux (Arch, GCC 16.2.1, Release), on #4391's head e0ec0c8 plus this commit:

• New CheckKernelStatusFlags in guest_sce_net: get, set and clear O_NONBLOCK with fcntl on a libSceNet socket, checked against SO_NBIO, plus the EOPNOTSUPP and EINVAL cases. Passes 5/5. Without the libSceNet change it aborts with the fcntl: descriptor 7 is neither a file ... error above.
• guest_sce_net, guest_socket_descriptors, guest_poll, guest_select, guest_dup, guest_socket_pair: all pass.
• ProsperoLight (PPSA99002, catalog release, --to-intel), 3 runs of 60 s: no longer stops on fcntl; its next stop is poll: descriptor 47 is neither a socket nor a pipe, i.e. poll not yet knowing libSceNet sockets. That is a separate change and not part of this PR.
• tools/check_conventions.py --base upstream/main passes.
• Not tested on Windows.

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: fix(libkernel): give guest sockets ordinary descriptor numbers below FD_SETSIZE #4391
• Unblocks: none yet (PPSA99002 needs poll next)
• AI-assisted: yes

[VitzS7]

Heads-up from #4910, which also touches Equeue.cpp L596-605: once your lookup lands, a libSceNet socket reaches GuestSockets::Kind there as a descriptor that is not in the libkernel socket table, so a kevent read or write event on it would answer EINVAL instead of registering a knote. I noted on that PR that the native lookup you register is the natural source for the kind, and that I will rebase on whichever of the two lands first and ask it there rather than deriving the kind from the host descriptor alone. Nothing for you to change here.

[xXPBSXx]

Windows test of this PR.

Environment: Windows 10 (19045), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release, BUILD_TESTING=ON, -j2, libusb-cmake from a local checkout. Built from main c2565f014 with this PR and ten other open PRs (#4972, #4967, #4966, #4945, #4938, #4936, #4933, #4898, #4896, #4888) merged together; the merges had no conflict.

Result: guest_dup and guest_fstatfs (2 of 2; the branch is contained in #4933) pass.

Not checked: whether the test fails without the change in the library code; the PR was not tested alone, only in the combined merge; no title run; Linux not tested.

AI-assisted (Claude Code).

[YossiYad]

Linux title run with ProsperoLight (PPSA99002, catalog release): Bazzite 44, GCC 16.2, Release, RX 7700 XT (RADV), head 1836f8e93 merged into main 315260903.

• main alone: ProsperoLight aborts during start-up with fcntl: descriptor 48 is neither a file, a directory, a pipe nor a character device, the stop this PR describes.
• With this PR the fcntl stop is gone. The title gets through its UI start-up (EGL surface, GL programs, pad and audio) and then stops on the same socket in poll: poll: descriptor 48 is neither a socket nor a pipe. That is the stop fix(libkernel): poll libSceNet sockets with the host wait #4933 handles. With fix(libkernel): poll libSceNet sockets with the host wait #4933, which carries this commit, on main f74ec639c, ProsperoLight ran the full 60 s without a stop.
• guest_sce_net passes 3 of 3 (--repeat until-fail:3) on the merged tree.

Merging into current main conflicts only in core/libs/tests/GuestSceNet.cpp: main added CheckSocketError(); on the line where this PR adds CheckKernelStatusFlags();. Keeping both calls builds, and that's the tree the runs above used.

[adriadam10]

Rebased onto current main (#4391 is merged, so only the fcntl commit remains; one adjacency conflict in GuestSceNet.cpp's main() kept both calls). Checked: build and ctest for libSceNet/libkernel green, guest_sce_net fails without the change, flag values match FreeBSD (O_NONBLOCK 4, EOPNOTSUPP 45, EINVAL 22). Merging once the full build is green; thanks also to @YossiYad and @xXPBSXx for the runs.