[Bugme-janitors] [Bug 9182] Critical memory leak (dirty pages)

[bugme-daemon at bugzilla.kernel.org]

http://bugzilla.kernel.org/show_bug.cgi?id=9182





------- Comment #56 from olel at ans.pl  2007-12-19 13:54 -------


On Wed, 19 Dec 2007, Linus Torvalds wrote:

>
>
> On Wed, 19 Dec 2007, Linus Torvalds wrote:
>>
>>> but I'd really like to understand how that page got marked dirty again,
>>> and why it seems to be related to "data=journal".
>>
>> That still holds. I'd really like to understand why/how this triggers.
>
> Hmm. "truncate_complete_page()" does:
>
>        cancel_dirty_page(page, PAGE_CACHE_SIZE);
>
>        if (PagePrivate(page))
>                do_invalidatepage(page, 0);
>
>        remove_from_page_cache(page);
>
> and yes, that "do_invalidatepage()" calls down to the filesystem layer
> (mapping->a_ops->invalidatepage), and yes, this all goes into the
> journalling code.
>
> So at a guess, the bug would go away if we just moved the
> "cancel_dirty_page()" to *after* the do_invalidatepage() case, although I
> wonder if we had some reason to do it in that order (ie maybe
> do_invalidatepage() likes to see the page being clean).
>
> Anyway, I think the fixups I added to __remove_from_page_cache() seem to
> continually become a better idea, considering that we let the filesystem
> mess around with the page in between, and if the filesystem messes with
> the dirty bits, it really means that the VM shouldn't just rely on it
> remaining clean.
>
> But I still want/hope-for a confirmation from Krzysztof that the patch
> actually fixes it for him too. At which point I'll just commit it without
> the stack dumping.

Just booted the system with 2.6.24-rc5+the debug/fixup patch. It took 2 
minutes to get this:

WARNING: at mm/filemap.c:132 __remove_from_page_cache()
Pid: 3734, comm: lmtp Not tainted 2.6.24-rc5 #1
  [<c014d772>] __remove_from_page_cache+0x87/0xe6
  [<c014d7f3>] remove_from_page_cache+0x22/0x2b
  [<c015327f>] truncate_complete_page+0x2b/0x3f
  [<c0153367>] truncate_inode_pages_range+0xd4/0x2d8
  [<c018b96e>] inotify_inode_is_dead+0x1a/0x70
  [<c018b96e>] inotify_inode_is_dead+0x1a/0x70
  [<c0245f52>] _atomic_dec_and_lock+0x2a/0x48
  [<c0153582>] truncate_inode_pages+0x17/0x1d
  [<c01a5b39>] ext3_delete_inode+0x13/0xbb
  [<c01a5b26>] ext3_delete_inode+0x0/0xbb
  [<c0178eda>] generic_delete_inode+0x5e/0xc6
  [<c0178604>] iput+0x60/0x62
  [<c0176779>] d_kill+0x2d/0x46
  [<c0176a94>] dput+0xdc/0xe4
  [<c01697c4>] __fput+0x113/0x13d
  [<c016727d>] filp_close+0x51/0x58
  [<c0168315>] sys_close+0x70/0xab
  [<c0103e92>] sysenter_past_esp+0x5f/0xa5
  =======================

WARNING: at mm/filemap.c:132 __remove_from_page_cache()
Pid: 3738, comm: smtp Not tainted 2.6.24-rc5 #1
  [<c014d772>] __remove_from_page_cache+0x87/0xe6
  [<c014d7f3>] remove_from_page_cache+0x22/0x2b
  [<c015327f>] truncate_complete_page+0x2b/0x3f
  [<c0153367>] truncate_inode_pages_range+0xd4/0x2d8
  [<c018b96e>] inotify_inode_is_dead+0x1a/0x70
  [<c018b96e>] inotify_inode_is_dead+0x1a/0x70
  [<c0245f52>] _atomic_dec_and_lock+0x2a/0x48
  [<c0153582>] truncate_inode_pages+0x17/0x1d
  [<c01a5b39>] ext3_delete_inode+0x13/0xbb
  [<c01a5b26>] ext3_delete_inode+0x0/0xbb
  [<c0178eda>] generic_delete_inode+0x5e/0xc6
  [<c0178604>] iput+0x60/0x62
  [<c0176779>] d_kill+0x2d/0x46
  [<c0176a94>] dput+0xdc/0xe4
  [<c01697c4>] __fput+0x113/0x13d
  [<c016727d>] filp_close+0x51/0x58
  [<c0168315>] sys_close+0x70/0xab
  [<c0103e92>] sysenter_past_esp+0x5f/0xa5
  =======================

ole at cougar:~$ dmesg |grep -c __remove_from_page_cache
10

ole at cougar:~$ uptime
  22:53:09 up 2 min,  1 user,  load average: 0.57, 0.37, 0.14


Best regards,

                                Krzysztof Oledzki


-- 
Configure bugmail: http://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.