2.6.39-rc5-git4: Reported regressions from 2.6.38

[Linus Torvalds]

On Sat, Apr 30, 2011 at 12:42 PM, Rafael J. Wysocki <rjw at sisk.pl> wrote:
>
> Bug-Entry       : http://bugzilla.kernel.org/show_bug.cgi?id=34012
> Subject         : 2.6.39-rc4+: oom-killer busy killing tasks
> Submitter       : Christian Kujau <lists at nerdbynature.de>
> Date            : 2011-04-22 1:57 (9 days old)
> Message-ID      : <alpine.DEB.2.01.1104211841510.18728 at trent.utfs.org>
> References      : http://marc.info/?l=linux-kernel&m=130343744622331&w=2

Judging by the extended debug info in:

   http://nerdbynature.de/bits/2.6.39-rc4/oom/
   (The -9 files are the current ones)

that thing shows a _lot_ of xfs inodes:

  xfs_inode         479187 479187   1120   14    4 : tunables    0
0    0 : slabdata  34329  34329      0

even though there aren't that many dentries (each inode should have at
least one dentry associated with it under normal circumstances):

  dentry             97896  97900    160   25    1 : tunables    0
0    0 : slabdata   3916   3916      0

and then later when the system gets low on memory, the dentries shrink:

  dentry              1017   3525    160   25    1 : tunables    0
0    0 : slabdata    141    141      0

but the XFS inodes do not:

  xfs_inode         557579 557579   1120   14    4 : tunables    0
0    0 : slabdata  41492  41492      0

so I suspect it's some kind of XFS inode leak.

                      Linus