On Sat, 2007-01-13 at 00:02 +0300, Alex Tomas wrote: > interesting .. > > I thought VFS doesn't allow concurrent operations. > if unlink goes first, then link should wait on the > parent's i_mutex and then found no source name.
I don't think the VFS ever takes the source's parent's i_mutex. Unless the source and destination's parent is the same, in which case the i_mutex is taken after the source has already been looked up. > thanks, Alex > > >>>>> Eric Sandeen (ES) writes: > > ES> ) > ES> I've been looking at a case where many threads are opening, unlinking, > and > ES> hardlinking files on ext3 . At unmount time I see an oops, because the > superblock's > ES> orphan list points to a freed inode. > > ES> I did some tracing of the inodes, and it looks like this: > > ES> ext3_unlink():[/src/linux-2.6.18/fs/ext3/namei.c:2123] adding orphan > ES> i_state:0x7 cpu:1 i_count:2 i_nlink:0 > > ES> ext3_orphan_add():[/src/linux-2.6.18/fs/ext3/namei.c:1890] > ext3_orphan_add > ES> i_state:0x7 cpu:1 i_count:2 i_nlink:0 > > ES> iput():[/src/linux-2.6.18/fs/inode.c:1139] iput enter > ES> i_state:0x7 cpu:1 i_count:2 i_nlink:0 > > ES> ext3_link():[/src/linux-2.6.18/fs/ext3/namei.c:2202] ext3_link enter > ES> i_state:0x7 cpu:3 i_count:1 i_nlink:0 > > ES> ext3_inc_count():[/src/linux-2.6.18/fs/ext3/namei.c:1627] done > ES> i_state:0x7 cpu:3 i_count:1 i_nlink:1 > > ES> The unlink gets there first, finds i_count > 0 (in use) but nlink goes > to 0, so > ES> it puts it on the orphan inode list. Then link comes along, and bumps > the link > ES> back up to 1. So now we are on the orphan inode list, but we are not > unlinked. > > ES> Eventually when count goes to 0, and we still have 1 link, again no > action is > ES> taken to remove the inode from the orphan list, because it is still > linked (i.e. > ES> we don't go through ext3_delete()) > > ES> When this inode is eventually freed, the sb orphan list gets corrupted, > because > ES> we have freed it without first removing it from the orphan list. > > ES> I think the simple solution is to remove the inode from the orphan list > ES> when we bump the link back up from 0 to 1. I put that test in there > because > ES> there are other potential reasons that we might be on the list > (truncates, > ES> direct IO). > > ES> Comments? > > ES> Thanks, > ES> -Eric > > ES> p.s. ext3_inc_count and ext3_dec_count seem misnamed, have an unused > ES> arg, and are very infrequently called. I'll probably submit a patch > ES> to just put the single line of code into the caller, too. > > ES> --- > > ES> Remove inode from the orphan list in ext3_link() if we might have > ES> raced with ext3_unlink(), which potentially put it on the list. > ES> If we're on the list with nlink > 0, we'll never get cleaned up > ES> properly and eventually may corrupt the list. > > ES> Signed-off-by: Eric Sandeen <[EMAIL PROTECTED]> > > ES> Index: linux-2.6.19/fs/ext3/namei.c > ES> =================================================================== > ES> --- linux-2.6.19.orig/fs/ext3/namei.c > ES> +++ linux-2.6.19/fs/ext3/namei.c > ES> @@ -2204,6 +2204,9 @@ retry: > inode-> i_ctime = CURRENT_TIME_SEC; > ES> ext3_inc_count(handle, inode); > ES> atomic_inc(&inode->i_count); > ES> + /* did we race w/ unlink? */ > ES> + if (inode->i_nlink == 1) > ES> + ext3_orphan_del(handle, inode); > > ES> err = ext3_add_nondir(handle, dentry, inode); > ES> ext3_journal_stop(handle); -- David Kleikamp IBM Linux Technology Center - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
