* [PATCH] libxfs: fix repair deadlock due to failed inode flushes.
@ 2019-02-05 3:24 Dave Chinner
2019-02-11 14:09 ` Brian Foster
0 siblings, 1 reply; 2+ messages in thread
From: Dave Chinner @ 2019-02-05 3:24 UTC (permalink / raw)
To: linux-xfs; +Cc: arekm
From: Dave Chinner <dchinner@redhat.com>
If inode_item_done() fails to flush an inode after we've grabbed a
reference to the underlying buffer during a transaction commit, we
fail to put the buffer and hence leak it. We then deadlock on the
next lookup ofthe inode buffer as it is still locked and no-one owns
it.
To fix it, put the buffer on error so that it gets unlocked and
can be recovered appropriately in a later phase of repair.
Reported-by: Arkadiusz Miskiewicz <arekm@maven.pl>
Fixes: d15188a1ec14 ("xfs: rework the inline directory verifiers")
Signed-off-by: Dave Chinner <dchinner@redhat.com>
---
libxfs/trans.c | 17 +++++++++++++----
1 file changed, 13 insertions(+), 4 deletions(-)
diff --git a/libxfs/trans.c b/libxfs/trans.c
index 46ff8b4ae798..10a35dd47b01 100644
--- a/libxfs/trans.c
+++ b/libxfs/trans.c
@@ -824,8 +824,10 @@ _("Transaction block reservation exceeded! %u > %u\n"),
/*
* Transaction commital code follows (i.e. write to disk in libxfs)
+ *
+ * XXX (dgc): should failure to flush the inode (e.g. due to uncorrected
+ * corruption) result in transaction commit failure w/ EFSCORRUPTED?
*/
-
static void
inode_item_done(
xfs_inode_log_item_t *iip)
@@ -856,17 +858,24 @@ inode_item_done(
return;
}
+ /*
+ * Flush the inode and disassociate it from the transaction regardless
+ * of whether the flush succeed or not. If we fail the flush, make sure
+ * we still release the buffer reference we currently hold.
+ */
bp->b_log_item = iip;
error = libxfs_iflush_int(ip, bp);
+ ip->i_transp = NULL; /* disassociate from transaction */
+ bp->b_log_item = NULL; /* remove log item */
+ bp->b_transp = NULL; /* remove xact ptr */
+
if (error) {
fprintf(stderr, _("%s: warning - iflush_int failed (%d)\n"),
progname, error);
+ libxfs_putbuf(bp);
return;
}
- ip->i_transp = NULL; /* disassociate from transaction */
- bp->b_log_item = NULL; /* remove log item */
- bp->b_transp = NULL; /* remove xact ptr */
libxfs_writebuf(bp, 0);
#ifdef XACT_DEBUG
fprintf(stderr, "flushing dirty inode %llu, buffer %p\n",
--
2.20.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH] libxfs: fix repair deadlock due to failed inode flushes.
2019-02-05 3:24 [PATCH] libxfs: fix repair deadlock due to failed inode flushes Dave Chinner
@ 2019-02-11 14:09 ` Brian Foster
0 siblings, 0 replies; 2+ messages in thread
From: Brian Foster @ 2019-02-11 14:09 UTC (permalink / raw)
To: Dave Chinner; +Cc: linux-xfs, arekm
On Tue, Feb 05, 2019 at 02:24:13PM +1100, Dave Chinner wrote:
> From: Dave Chinner <dchinner@redhat.com>
>
> If inode_item_done() fails to flush an inode after we've grabbed a
> reference to the underlying buffer during a transaction commit, we
> fail to put the buffer and hence leak it. We then deadlock on the
> next lookup ofthe inode buffer as it is still locked and no-one owns
> it.
>
> To fix it, put the buffer on error so that it gets unlocked and
> can be recovered appropriately in a later phase of repair.
>
> Reported-by: Arkadiusz Miskiewicz <arekm@maven.pl>
> Fixes: d15188a1ec14 ("xfs: rework the inline directory verifiers")
> Signed-off-by: Dave Chinner <dchinner@redhat.com>
> ---
Reviewed-by: Brian Foster <bfoster@redhat.com>
> libxfs/trans.c | 17 +++++++++++++----
> 1 file changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/libxfs/trans.c b/libxfs/trans.c
> index 46ff8b4ae798..10a35dd47b01 100644
> --- a/libxfs/trans.c
> +++ b/libxfs/trans.c
> @@ -824,8 +824,10 @@ _("Transaction block reservation exceeded! %u > %u\n"),
>
> /*
> * Transaction commital code follows (i.e. write to disk in libxfs)
> + *
> + * XXX (dgc): should failure to flush the inode (e.g. due to uncorrected
> + * corruption) result in transaction commit failure w/ EFSCORRUPTED?
> */
> -
> static void
> inode_item_done(
> xfs_inode_log_item_t *iip)
> @@ -856,17 +858,24 @@ inode_item_done(
> return;
> }
>
> + /*
> + * Flush the inode and disassociate it from the transaction regardless
> + * of whether the flush succeed or not. If we fail the flush, make sure
> + * we still release the buffer reference we currently hold.
> + */
> bp->b_log_item = iip;
> error = libxfs_iflush_int(ip, bp);
> + ip->i_transp = NULL; /* disassociate from transaction */
> + bp->b_log_item = NULL; /* remove log item */
> + bp->b_transp = NULL; /* remove xact ptr */
> +
> if (error) {
> fprintf(stderr, _("%s: warning - iflush_int failed (%d)\n"),
> progname, error);
> + libxfs_putbuf(bp);
> return;
> }
>
> - ip->i_transp = NULL; /* disassociate from transaction */
> - bp->b_log_item = NULL; /* remove log item */
> - bp->b_transp = NULL; /* remove xact ptr */
> libxfs_writebuf(bp, 0);
> #ifdef XACT_DEBUG
> fprintf(stderr, "flushing dirty inode %llu, buffer %p\n",
> --
> 2.20.1
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2019-02-11 14:09 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2019-02-05 3:24 [PATCH] libxfs: fix repair deadlock due to failed inode flushes Dave Chinner
2019-02-11 14:09 ` Brian Foster
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).