Add basic support for System V shm

[Aransentin]

System V shared memory is kinda legacy cruft, having been supplanted by POSIX shared memory. Still, it's needed for some legacy applications; most notably X11 which uses it as a performant way of sharing pixel data with clients.

I did not add the IPC_STAT structs (e.g. shmid_ds), as they are even more useless and "broken" on Linux. The structs were specified wrongly on Linux a long time ago so they are incompatible with the standard, but not changed in the kernel due to not breaking backwards compatibility with ancient libcs.

[LemonBoy]

A basic test in os/linux/test.zig is needed, something that creates a key with shmget, fetch it with shmat and remove it with shmctl w/ IPC_RMID.

[Aransentin]

A basic test in os/linux/test.zig is needed

shmget(IPC_PRIVATE, 4096, IPC_CREAT|000) = 98328
shmat(98328, NULL, 0)                   = 0x7feb5b3c9000
shmdt(0x7feb5b3c9000)                   = 0
shmctl(98328, IPC_RMID, NULL)           = 0

[Aransentin]

Seems the CI server didn't like it. Hm. Hold on while I investigate.

[LemonBoy]

shmget(IPC_PRIVATE, 4096, IPC_CREAT|000) = 98328

You're passing 0o000 as permission parameter...

[Aransentin]

shmget(IPC_PRIVATE, 4096, IPC_CREAT|000) = 98328

You're passing 0o000 as permission parameter...

Yeah, I ran the test as root so I didn't notice. Fixed!

[LemonBoy]

Hmm, I just checked a few man pages and it seems the whole SHM api is well-supported across the board.
If you add the libc definitions in lib/std/c.zig the API will be available even when the libc is linked in (using the libc functions rather than our syscall stubs) and for BSDs too. The test can be moved to lib/std/os/test.zig using std.os.system to make it cross platform (I guess wasi and windows targets should be skipped).

[Aransentin]

Creating a proper wrapper in os.zig and not just having thin syscall wrappers is going to be moderately complicated; the same libc function maps to different syscalls (sys_ipc...) depending on architecture for non-recent kernels. I'll close the PR for now and get back to you with a proper implementation when I have time for it.

[LemonBoy]

Creating a proper wrapper in os.zig and not just having thin syscall wrappers is going to be moderately complicated;

No, a wrapper in os.zig is definitely the worst possible decision (see #6600).

he same libc function maps to different syscalls (sys_ipc...) depending on architecture for non-recent kernels.

Alright, I can't find what kernel deprecated sys_ipc on the "old" arches we support (mips/i386 iirc), you may want to use that for maximum compatibility as we already do with socketcall.

The libc wrappers don't do much beside calling the right syscall, except for shmctl that you didn't implement anyway.

[Aransentin]

@LemonBoy Where would you like the wrapper around linux/BSD shm (that chooses the correct syscall & translates errors) to be until the os.zig situation is resolved?