Re: [PATCH 2/2] Fix possible leakage of blocks in UDF

[Andrew Morton <akpm@linux-foundation.org>]

On Sat, 2 Jun 2007 22:57:07 +0400 Cyrill Gorcunov <gorcunov@gmail.com> wrote:

> [Andrew Morton - Sat, Jun 02, 2007 at 10:32:03AM -0700]
> | On Sat, 2 Jun 2007 18:06:19 +0400 Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> | 
> | > [Andrew Morton - Sat, Jun 02, 2007 at 12:06:45AM -0700]
> | > | On Sat, 2 Jun 2007 10:59:23 +0400 Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> | > | 
> | > | > [Andrew Morton - Fri, Jun 01, 2007 at 11:54:22PM -0700]
> | > | > | On Sat, 2 Jun 2007 10:34:03 +0400 Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> | > | > | 
> | > | > | > | That patch is DOA, methinks.
> | > | > | > | 
> | > | > | > 
> | > | > | > Andrew, what does it mean - "DOA"? Dead on arrival?
> | > | > | 
> | > | > | yes - I dropped it.
> | > | > | 
> | > | > 
> | > | > But that could lead to rejection of my code-style-conversion patch...
> | > | > Should I remake them?
> | > | 
> | > | Actually I've rebuilt those patches four times already.  People keep
> | > | changing stuff.
> | > | 
> | > | > Actually Jan was right, the current state of UDF (without his patches)
> | > | > could lead to lost blocks and his patch must be just fixed I think.
> | > | 
> | > | sure.
> | > | 
> | > 
> | > Andrew, you know I've been trying to reproduce Eric's lockup case almost
> | > two hour and still can't reach it. All manupulation I've done to UDF didn't
> | > lead to lockup. Moreover, I've added debug print for UDF module and here is
> | > the results (for single drop_inode call):
> | > 
> | > 	[12063.897000] UDF: udf_drop_inode:105 --> udf_drop_inode --> inode->i_count: 0
> | > 	[12063.897000] UDF: udf_drop_inode:107 udf_drop_inode -> discard_prealloc
> | > 	[12063.897000] UDF: udf_discard_prealloc:136 udf_discard_prealloc
> | > 	[12063.897000] UDF: udf_truncate_tail_extent:84 udf_truncate_tail_extent
> | > 	[12063.897000] UDF: udf_truncate_extents:194 udf_truncate_extents -->
> | > 	[12063.897000] UDF: extent_trunc:38 --->
> | > 	[12063.897000] UDF: extent_trunc:54 call to udf_write_aext
> | > 	[12063.897000] UDF: udf_write_aext:1843 udf_write_aext
> | > 	[12063.897000] UDF: udf_write_aext:1846 dont has epos->bh
> | > 	[12063.897000] UDF: udf_write_aext:1866 ICBTAG_FLAG_AD_LONG
> | > --->	[12063.897000] UDF: udf_write_aext:1893 ---> gotcha ---> call mark_inode_dirty
> | > --->	[12063.897000] UDF: extent_trunc:59 --> gotcha --> call mark_inode_dirty
> | > 	[12063.897000] UDF: extent_trunc:68 <---
> | > --->	[12063.897000] UDF: udf_truncate_extents:282 call mark_inode_dirty
> | > 	[12063.897000] UDF: udf_truncate_extents:330 udf_truncate_extents <--
> | > 	[12063.897000] UDF: udf_drop_inode:115 <-- udf_drop_inode <--
> | > 
> | > As you may see, mark_inode_dirty is called several time and no locking happened.
> | > Maybe I should use some test utils?
> | > 
> | 
> | Silly question: you _do_ have CONFIG_SMP=y, yes?
> | 
> Oh, no I don't :( So the problem is in kernel sync (as I thought)...
> damn... I have to rebuild my kernel... but hold on - my machine has only
> one CPU ;)

You should be able to run an SMP kernel on a single-CPU machine.

> | And did you enable lockdep?
> | 
> Yes
> 
> So the problem is 'cause of mark_inode_dirty may sleep? Right?
> So only thing to be checked is lock_kernel I think

No, the problem is that the patch caused the kernel to take inode_lock
within the newly-added drop_inode(), btu drop_inode() is already called
under inode_lock.

It has nothing to do with lock_kernel() and it has nothing to do with
sleeping.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/