2007-01-15 18:19:24

by Eric Sandeen

Subject: [PATCH] return ENOENT from ext3_link when racing with unlink

An update from the earlier thread,
[PATCH] [RFC] remove ext3 inode from orphan list when link and unlink race

I think this is better than the original idea of trying to handle the race;
I've seen that the orphan inode list can get corrupted, but there may well
be other implications of the race which haven't yet been exposed. I think
it's safer to simply return -ENOENT in this race window, and avoid other
potential problems. Anything wrong with this?

Thanks for the comments suggesting this approach in the prior thread.

Thanks,

-Eric

---

Return -ENOENT from ext[34]_link if we've raced with unlink and
i_nlink is 0. Doing otherwise has the potential to corrupt the
orphan inode list, because we'd wind up with an inode with a
non-zero link count on the list, and it will never get properly
cleaned up & removed from the orphan list before it is freed.

Signed-off-by: Eric Sandeen <[email protected]>

Index: linux-2.6.19/fs/ext3/namei.c
===================================================================
--- linux-2.6.19.orig/fs/ext3/namei.c
+++ linux-2.6.19/fs/ext3/namei.c
@@ -2191,6 +2191,8 @@ static int ext3_link (struct dentry * ol

if (inode->i_nlink >= EXT3_LINK_MAX)
return -EMLINK;
+ if (inode->i_nlink == 0)
+ return -ENOENT;

retry:
handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Index: linux-2.6.19/fs/ext4/namei.c
===================================================================
--- linux-2.6.19.orig/fs/ext4/namei.c
+++ linux-2.6.19/fs/ext4/namei.c
@@ -2189,6 +2189,8 @@ static int ext4_link (struct dentry * ol

if (inode->i_nlink >= EXT4_LINK_MAX)
return -EMLINK;
+ if (inode->i_nlink == 0)
+ return -ENOENT;

retry:
handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +

2007-01-16 22:02:43

by Peter Staubach

Subject: Re: [PATCH] return ENOENT from ext3_link when racing with unlink

Eric Sandeen wrote:
> An update from the earlier thread, [PATCH] [RFC] remove ext3 inode
> from orphan list when link and unlink race
>
> I think this is better than the original idea of trying to handle the
> race;
> I've seen that the orphan inode list can get corrupted, but there may
> well
> be other implications of the race which haven't yet been exposed. I
> think
> it's safer to simply return -ENOENT in this race window, and avoid other
> potential problems. Anything wrong with this?
>
> Thanks for the comments suggesting this approach in the prior thread.
>
> Thanks,
>
> -Eric
>
> ---
>
> Return -ENOENT from ext[34]_link if we've raced with unlink and
> i_nlink is 0. Doing otherwise has the potential to corrupt the
> orphan inode list, because we'd wind up with an inode with a
> non-zero link count on the list, and it will never get properly
> cleaned up & removed from the orphan list before it is freed.
>
> Signed-off-by: Eric Sandeen <[email protected]>
>
> Index: linux-2.6.19/fs/ext3/namei.c
> ===================================================================
> --- linux-2.6.19.orig/fs/ext3/namei.c
> +++ linux-2.6.19/fs/ext3/namei.c
> @@ -2191,6 +2191,8 @@ static int ext3_link (struct dentry * ol
>
> if (inode->i_nlink >= EXT3_LINK_MAX)
> return -EMLINK;
> + if (inode->i_nlink == 0)
> + return -ENOENT;
>
> retry:
> handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
> Index: linux-2.6.19/fs/ext4/namei.c
> ===================================================================
> --- linux-2.6.19.orig/fs/ext4/namei.c
> +++ linux-2.6.19/fs/ext4/namei.c
> @@ -2189,6 +2189,8 @@ static int ext4_link (struct dentry * ol
>
> if (inode->i_nlink >= EXT4_LINK_MAX)
> return -EMLINK;
> + if (inode->i_nlink == 0)
> + return -ENOENT;
>
> retry:
> handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
>

Just out of curosity, what keeps i_nlink from going to 0 immediately
after the new test is executed?

Thanx...

ps

2007-01-16 22:08:37

by Alex Tomas

Subject: Re: [PATCH] return ENOENT from ext3_link when racing with unlink

>>>>> Peter Staubach (PS) writes:


PS> Just out of curosity, what keeps i_nlink from going to 0 immediately
PS> after the new test is executed?

i_mutex in vfs_link() and vfs_unlink()

thanks, Alex

2007-01-16 22:17:20

by Peter Staubach

Subject: Re: [PATCH] return ENOENT from ext3_link when racing with unlink

Alex Tomas wrote:
>>>>>> Peter Staubach (PS) writes:
>>>>>>
>
>
> PS> Just out of curosity, what keeps i_nlink from going to 0 immediately
> PS> after the new test is executed?
>
> i_mutex in vfs_link() and vfs_unlink()
>

Ahhh... Okie doke, thanx!

ps