Add PROC_PDEATHSIG_SET to procctl interface.

[munro_ip9.org]

Summary

Allow processes to request the delivery of a signal upon death
of their parent process. This provides approximately the same
behaviour as prctl(PR_SET_PDEATHSIG, ...) on Linux.

As discussed on freebsd-hackers:

https://marc.info/?l=freebsd-hackers&m=152368119728365&w=3

Test Plan

ATF tests are included.

[kib]

It looks fine overall, please look at several minor comments I posted.

I will wait some more days for more opinions and suggestions before committing.

lib/libc/sys/procctl.2

402 ↗

(On Diff #41551)

I think it would be useful to note that zero acts same as the caller pid.

sys/kern/kern_procctl.c

577 ↗

(On Diff #41551)

I think it is unnatural to keep the actual code in the locking switch. Might be, put a switch with only PDEATHSIG commands above this switch.

[munro_ip9.org]

Updated documentation and moved PDEATHSIG handling into its own switch, as requested.

[mjg]

I only have cosmetic remarks. Definitely looks good enough to be shipped.

lib/libc/sys/procctl.2

547 ↗

(On Diff #41557)

A remark about the corner case of clearing the flag on creds-changing exec would be useful, but I don't know how to bolt it in here.

sys/kern/kern_procctl.c

568 ↗

(On Diff #41557)

both this and the case below should use the already declared p pointer to shorten the lines.

p = td->td_proc;
......

[munro_ip9.org]

Use 'p' variable instead of repeating td->td_proc, as requested.

[munro_ip9.org]

Thanks for the reviews and help!

lib/libc/sys/procctl.2

547 ↗

(On Diff #41557)

That is covered by the following man page hunk:

+The value is cleared for the children
+of fork() and when executing set-user-ID or set-group-ID binaries.

[munro_ip9.org]

Use 'p' instead of td->td_proc in one more place that I'd missed. (Thanks to mjg for prod via IRC.) Sorry for the churn.

[jilles]

The code itself looks good, but something could be improved in the man page and test.

lib/libc/sys/procctl.2

403–404 ↗

(On Diff #41562)

for child processes (since this is not just fork() but also vfork(), rfork() and pdfork())

406 ↗

(On Diff #41562)

.Vt int

419 ↗

(On Diff #41562)

.Vt int

523 ↗

(On Diff #41562)

.Fa id and .Fa id_type

tests/sys/kern/pdeathsig.c

191–192 ↗

(On Diff #41562)

This is nice and simple, but also slow. Consider mechanisms such as sigsuspend() and sigwait().

199 ↗

(On Diff #41562)

Exiting a child process should be done via _exit(0); (here and in some more locations). The code that executes after the test returns may not produce desirable effects when run in a child process.

[munro_ip9.org]

Used _exit(0) to exit the tests's subprocesses successfully. Use sigwait to wait for signals synchronously. Documentation tweaks.

[munro_ip9.org]

The "fork_no_inherit" and "exec_inherit" tests' child processes also need to call _exit(0), so here's a new version to do that. Thanks for the review jilles. I agree the tests look much better with those changes.

[kib]

sys/kern/kern_thread.c

116 ↗

(On Diff #41608)

This one must be changed to 0x318. On amd64 it seems that the new int member fits into some pad, but not on i386.

[munro_ip9.org]

Fixed assertion about struct layout on i386. Fixed typo in man page (s/pont/point/).

[kib]

I will commit this later today, giving some to for jilles to comment further.

[jilles]

I like it.

[munro_ip9.org]

Thank you!