[7.0-R] Possible ufs livelock during coredump path?

[Xin LI]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Peter Holm wrote:
| Hi,
|
| It seems that we have a potential livelock during coredump on 7.0-R, the
| case was that two processes trying to coredump in the same time (e.g. if
| I configure kern.corefile=/var/tmp/%N.core and a lot of instances
| coredump in the same time), perhaps when paging involved with it.  Upon
| reboot, it would not recover but wait infinitely.  The box is running
| 7.0-R/i386, UP (Origin = "GenuineIntel"  Id = 0xf34  Stepping = 4).
|
| Is this an known issue?  This is my own server but I do not have my
| hands on it because it is in China, however I can provide some help if
| the experiment can be recovered with a power-cycle :)
|
|
|> AFAIK it is an old problem. I have some test where I had to disable core
|> dumps for the same reason. I seem to remember that the problem is related
|> to running out of VM?

For my case it does not seem to be ran out of VM (at least the system
did not printed out any messages, the log has a lot of kernel: pid 27223
(httpd), uid 80: exited on signal 11 (core dumped) but not the out of
swap one.

So, presumably we can reliably trigger this situation (or at least your
ones :)?

Cheers,
- --
Xin LI <delphij at delphij.net>	http://www.delphij.net/
FreeBSD - The Power to Serve!
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (FreeBSD)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkgVgMsACgkQi+vbBBjt66BnOwCeJLB5xoE27b3CN/x/VIL+0EAI
+c8AoJyYiqCi7tBeqZBx6cj/+gzBLmFn
=qZmb
-----END PGP SIGNATURE-----