Page MenuHomeFreeBSD

Rework jobc handling.

Authored by kib on Dec 31 2020, 7:10 PM.
Referenced Files
Unknown Object (File)
Thu, Jun 13, 8:32 AM
Unknown Object (File)
Fri, Jun 7, 11:22 AM
Unknown Object (File)
Sat, Jun 1, 4:52 PM
Unknown Object (File)
May 7 2024, 7:52 AM
Unknown Object (File)
May 1 2024, 3:50 AM
Unknown Object (File)
Apr 27 2024, 7:55 PM
Unknown Object (File)
Apr 25 2024, 8:39 AM
Unknown Object (File)
Apr 23 2024, 7:51 PM



Instead of trying to maintain pg_jobc counter on each process group update (and sometimes before), just calculate the counter when needed. Still, for the benefit of the signal delivery code, explicitly mark orphaned groups as such with the new process group flag.

Fix use after free when relocking p_pgrp in tty_wait_background (and probably other places). Change pgrp memory to type-stable zone, and ensure that pg_mtx is not destroyed.

Not sure how to present it in phab, there;a=shortlog;h=refs/heads/jobcdbg is the branch with individual commits.

Tested by: pho

Test Plan

Shells work fine.
I checked things like

# sh
# cat &
# ^D

to deliver expected signals to the stopped orphaned processes.

Also there is tools/regression/sigqueue/sigqtest2

Diff Detail

rG FreeBSD src repository
Lint Not Applicable
Tests Not Applicable

Event Timeline

kib requested review of this revision.Dec 31 2020, 7:10 PM
kib created this revision.

Fix LOR between proctree and pidhash bucket lock.
Do not forget un-orphaning groups when reparenting on exit.

FWIW, I upload patch series as a set of arc reviews. Originally this was tedious (using arc diff --create --head <N> <N-1> for each commit), but @markj has been working on a script that automates this (I used it to upload the series of zombproc-related debugger cleanups as a single command from a branch with multiple commits).

20210101 07:17:42 all (6/740):
panic: Lock (sx) proctree not locked @ kern/kern_proc.c:1172.
cpuid = 17
time = 1609481864
KDB: stack backtrace:
db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe01c7c16ab0
vpanic() at vpanic+0x181/frame 0xfffffe01c7c16b00
panic() at panic+0x43/frame 0xfffffe01c7c16b60
witness_assert() at witness_assert+0x23a/frame 0xfffffe01c7c16b90
fill_kinfo_proc() at fill_kinfo_proc+0x4e/frame 0xfffffe01c7c16be0
kern_proc_out() at kern_proc_out+0x62/frame 0xfffffe01c7c17370
sysctl_out_proc() at sysctl_out_proc+0x68/frame 0xfffffe01c7c17840
sysctl_kern_proc() at sysctl_kern_proc+0x87/frame 0xfffffe01c7c17890
sysctl_root_handler_locked() at sysctl_root_handler_locked+0x9c/frame 0xfffffe01c7c178e0
sysctl_root() at sysctl_root+0x20d/frame 0xfffffe01c7c17960
userland_sysctl() at userland_sysctl+0x180/frame 0xfffffe01c7c17a10
sys___sysctl() at sys___sysctl+0x5f/frame 0xfffffe01c7c17ac0
amd64_syscall() at amd64_syscall+0x147/frame 0xfffffe01c7c17bf0
fast_syscall_common() at fast_syscall_common+0xf8/frame 0xfffffe01c7c17bf0
--- syscall (202, FreeBSD ELF64, sys___sysctl), rip = 0x8003fb9ea, rsp = 0x7fffffffd258, rbp = 0x7fffffffd290 ---
KDB: enter: panic
[ thread pid 78475 tid 792240 ]
Stopped at      kdb_enter+0x37: movq    $0,0x10960c6(%rip)

One more place for proctree_lock.

kinfo_proc_out is also called from ELF coredump code

This revision is now accepted and ready to land.Jan 9 2021, 7:50 PM