Patchwork [3/6] ext4: Do not clear EOFBLOCKS_FL too soon

login
register
mail settings
Submitter Dmitri Monakho
Date Oct. 20, 2011, 9:08 p.m.
Message ID <1319144939-28801-4-git-send-email-dmonakhov@openvz.org>
Download mbox | patch
Permalink /patch/120882/
State New
Headers show

Comments

Dmitri Monakho - Oct. 20, 2011, 9:08 p.m.
ext4_ext_insert_extent() may fail due to number of reasons (ENOSPC)
so let's update eof flag only after extent was successfully inserted.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
---
 fs/ext4/extents.c |   10 ++++++----
 1 files changed, 6 insertions(+), 4 deletions(-)
Theodore Ts'o - Oct. 25, 2011, 8:18 a.m.
On Fri, Oct 21, 2011 at 01:08:56AM +0400, Dmitry Monakhov wrote:
> ext4_ext_insert_extent() may fail due to number of reasons (ENOSPC)
> so let's update eof flag only after extent was successfully inserted.
> 
> Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>

The problem with this patch is that if the check_eofblocks_fl() fails,
the patch jumps to out: and doesn't undo the block allocation and
extent tree manipulation.

I suspect a better way of solving this problem is to keep the existing
order, but to save the state of eofblocks flag, and if
ext4_ext_insert_extent() fails, to restore the state of the eofblocks
flag before jumping to the exit routine.

					- Ted
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Dmitri Monakho - Oct. 25, 2011, 11:57 a.m.
On Tue, 25 Oct 2011 04:18:34 -0400, Ted Ts'o <tytso@mit.edu> wrote:
> On Fri, Oct 21, 2011 at 01:08:56AM +0400, Dmitry Monakhov wrote:
> > ext4_ext_insert_extent() may fail due to number of reasons (ENOSPC)
> > so let's update eof flag only after extent was successfully inserted.
> > 
> > Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
> 
> The problem with this patch is that if the check_eofblocks_fl() fails,
> the patch jumps to out: and doesn't undo the block allocation and
> extent tree manipulation.
check_eofblocks_fl may fail only due to ext4_journal_get_write_access
or ext4_mark_iloc_dirty, this means that something serious happen,
so it is very unlikely we can undo block allocation.
> 
> I suspect a better way of solving this problem is to keep the existing
> order, but to save the state of eofblocks flag, and if
> ext4_ext_insert_extent() fails, to restore the state of the eofblocks
> flag before jumping to the exit routine.
Yes. This is definitely looks better, will redo.
> 
> 					- Ted
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 315775e..d81085e 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -4115,10 +4115,7 @@  got_allocated_blocks:
 			map->m_flags |= EXT4_MAP_UNINIT;
 	}
 
-	err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
-	if (!err)
-		err = ext4_ext_insert_extent(handle, inode, path,
-					     &newex, flags);
+	err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
 	if (err && free_on_err) {
 		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
 			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
@@ -4130,6 +4127,11 @@  got_allocated_blocks:
 				 ext4_ext_get_actual_len(&newex), fb_flags);
 		goto out2;
 	}
+	err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
+	if (err) {
+		ext4_discard_preallocations(inode);
+		goto out2;
+	}
 
 	/* previous routine could use block we allocated */
 	newblock = ext4_ext_pblock(&newex);