WASIX: support process groups and negative PIDs in kill() #7077

Problem

POSIX kill(pid, sig) with a negative pid addresses process groups:

• pid == -1: every process the caller may signal (broadcast); ESRCH only if there are none.
• pid < -1: every process in group |pid|; ESRCH if that group is empty.

WASIX doesn't support either:

• proc_signal takes an unsigned Pid, so wasix-libc's kill(-1234, sig) arrives as 4294966062 and is looked up as a plain PID. Since fix(wasix): report missing processes from proc_signal #7045, non-positive PIDs (interpreted as i32) are a silent no-op that returns success; before that, all failures were silent.
• There are no process groups in the runtime. wasix-libc fakes them per process: getpgrp() returns __wasilibc_pgrp (default 0), setpgid(0, x) stores x locally, and setpgid(other, …) and setsid() fail with EINVAL.

As a result, shell job control can't work: shells put each job in its own group with setpgid and signal it with kill(-pgid, sig) (e.g. forwarding Ctrl-C/SIGINT to a foreground job, or kill -- -$pgid). killpg() and Python's os.killpg are silently ignored as well.

Proposed direction

Move process groups (and sessions, if feasible) from wasix-libc into the runtime:

1. Each process in the control plane gets a PGID (and SID), inherited from the parent on spawn and fork; a new process is not its own group leader unless it asks.
2. New WASIX syscalls for setpgid, getpgid and setsid (and getsid), with POSIX errors (EPERM / ESRCH / EINVAL, EACCES after the child has exec'd).
3. wasix-libc implements setpgid/getpgid/getpgrp/setsid on top of them, dropping __wasilibc_pgrp.
4. proc_signal interprets the PID as signed: -1 broadcasts to every process in the instance except the caller itself (Linux additionally skips init); < -1 signals every process in group |pid|, returning ESRCH if it's empty; 0 is the caller's group (see WASIX: kill(0, sig) should signal the caller's process group #7078).
5. Ideally, terminal foreground process groups (tcsetpgrp/tcgetpgrp) follow, so the TTY can deliver SIGINT to the foreground group.

This is a new mechanism spanning the runtime, the WASIX ABI and wasix-libc, so the design should be agreed before implementation. Note that terminated processes are currently never removed from the control plane (#6787), which group lookups would also need to account for.

Acceptance

• setpgid/getpgid/setsid behave per POSIX across spawn and fork.
• kill(-pgid, sig) reaches every member of the group and only them; kill(-1, sig) reaches every process the caller may signal; empty groups give ESRCH.
• Guest tests under lib/wasix/tests/wasm_tests/signal/, plus a shell job-control smoke test (e.g. bash kill -- -$pgid).

Related: #7045, wasix-org/wasix-libc#144.

Filed by Claude Code (AI agent) on behalf of @Arshia001.