One process which would not die force me to power-cycle server and ALL UFS SUJ FSes are completely broken after that AGAIN!

Lev Serebryakov lev at FreeBSD.org
Sat Apr 12 13:10:37 UTC 2014


Hello, Freebsd-fs.

 On my 10-STABLE (r263965) system transmission-daemon stops to work, could not be
killed (waits forever in STOP state after "kill -KILL), kernel reports about
overfilled accept TCP queue for its socket (sonewconn: pcb 012345678FFFFFFF: Listen queue overflow).

 Try "shutdown -r now", process aborted due to process which would not die,
nothing could be done: system doesn't react on keyboard after that.

 Wait one hour (!). No result, only more "Listen queue overflow" messages on
console.

 Power-off. Power-on.

 All UFS2 filesystems can not be recovered with using of automated fsck, due
to journal/softupdate inconsistencies. I need to run "fsck -f" TWICE for
each of them (as first run ask to re-run fsck).

 Please note, they are filesystems on MBR slice + BSD label on simple SATA
disk attached to chipset port, no RAID, no "strange" GEOM modules, nothing
fancy. Plain and easy install -- MBR with one slice, BSD label, filesystems,
it's all.

 So, there are two questions:

(1) Does UFS2 SUJ works at all on STABLE system? Should it?!

(2) How could I avoid such situation, how could I reboot system WITHOUT such
disaster when one process refuse to die?

-- 
// Black Lion AKA Lev Serebryakov <lev at FreeBSD.org>



More information about the freebsd-stable mailing list