Re: Ext3 Errors...

[Jan Kara <jack@suse.cz>]

  Hi,

> Here is the output of /proc/slabinfo
> 
> slabinfo - version: 2.1
> # name            <active_objs> <num_objs> <objsize> <objperslab>
> <pagesperslab> : tunables <limit> <batchcount> <sharedfactor> : slabdata
> <active_slabs> <num_slabs> <sharedavail>
> jbd_4k                 5     10   4096    1    1 : tunables   24   12    8
> : slabdata      5     10      0
> ext3_inode_cache   22447  22696    968    4    1 : tunables   54   27    8
> : slabdata   5674   5674      0
> journal_head         740   1280     96   40    1 : tunables  120   60    8
> : slabdata     27     32    480
> buffer_head        56406  95386    104   37    1 : tunables  120   60    8
> : slabdata   2578   2578    480
  <snip>

  Thanks for the info. I should have said I'm interested about
/proc/slabinfo close to the state when the machine starts complaining
about OOM. Your current numbers look quite sensible so it's hard to see
if we really leak something or not...

> >> I have something really strange while running kernel 2.6.19.
> >>
> >> See the following lines:
> >>
> >> Dec  5 23:50:49 kernel: do_get_write_access: OOM for frozen_buffer
> >> Dec  5 23:50:49 kernel: ext3_free_blocks_sb: aborting transaction: Out
> >> of
> >> memory in __ext3_journal_get_undo_access
> >> Dec  5 23:50:49 kernel: EXT3-fs error (device md1) in
> >> ext3_free_blocks_sb:
> >> Out of memory
> >> Dec  5 23:50:49 kernel: EXT3-fs error (device md1) in
> >> ext3_reserve_inode_write: Readonly filesystem
> >> Dec  5 23:50:50 kernel: EXT3-fs error (device md1) in ext3_truncate: Out
> >> of memory
> >> Dec  5 23:50:51 kernel: EXT3-fs error (device md1) in
> >> ext3_reserve_inode_write: Readonly filesystem
> >> Dec  5 23:50:51 kernel: EXT3-fs error (device md1) in ext3_orphan_del:
> >> Readonly filesystem
> >> Dec  5 23:50:51 kernel: EXT3-fs error (device md1) in
> >> ext3_reserve_inode_write: Readonly filesystem
> >> Dec  5 23:50:51 kernel: EXT3-fs error (device md1) in ext3_delete_inode:
> >> Out of memory
> >>
> >> And three days later the same:
> >>
> >> Dec  8 08:24:29 kernel: do_get_write_access: OOM for frozen_buffer
> >> Dec  8 08:24:29 kernel: ext3_reserve_inode_write: aborting transaction:
> >> Out of memory in __ext3_journal_get_write_access
> >> Dec  8 08:24:29 kernel: EXT3-fs error (device md1) in
> >> ext3_reserve_inode_write: Out of memory
> >> Dec  8 08:24:32 kernel: EXT3-fs error (device md1) in ext3_dirty_inode:
> >> Out of memory
> >> Dec  8 08:24:32 kernel: EXT3-fs error (device md1) in ext3_new_blocks:
> >> Readonly filesystem
> >> Dec  8 08:24:32 kernel: EXT3-fs error (device md1) in
> >> ext3_reserve_inode_write: Readonly filesystem
> >> Dec  8 08:24:32 kernel: EXT3-fs error (device md1) in ext3_dirty_inode:
> >> Out of memory
> >> Dec  8 08:24:32 kernel: EXT3-fs error (device md1) in
> >> ext3_prepare_write:
> >> Out of memory
> >>
> >> Now the funny thing is, with kernel 2.6.18.3 I did not had these errors.
> >> Could it be my memory that is just going nuts, or something else? I have
> >> seen some other topics about the EXT3 corruption problems. Maybe this is
> >> also the same thing?
> >   Probably some error on the kernel's side. Maybe somebody (e.g. my new
> > code in jbd commit) forgets to release some buffers? What does your
> > /proc/slabinfo look like? Thanks for report.

								Honza