aboutsummaryrefslogtreecommitdiff
path: root/fs/xfs
diff options
context:
space:
mode:
authorDavid Chinner <david@fromorbit.com>2008-10-30 18:32:43 +1100
committerLachlan McIlroy <lachlan@redback.melbourne.sgi.com>2008-10-30 18:32:43 +1100
commit6bfb3d065f4c498c17a3a07f3dc08cedff53aff4 (patch)
treec5c528c77e44584616a175e0dcc89713e7b76d0a /fs/xfs
parente0b8e8b65d578f5d5538465dff8392cf02e1cc5d (diff)
[XFS] Fix race when looking up reclaimable inodes
If we get a race looking up a reclaimable inode, we can end up with the winner proceeding to use the inode before it has been completely re-initialised. This is a Bad Thing. Fix the race by checking whether we are still initialising the inod eonce we have a reference to it, and if so wait for the initialisation to complete before continuing. While there, fix a leaked reference count in the same code when encountering an unlinked inode and we are not doing a lookup for a create operation. SGI-PV: 987246 SGI-Modid: xfs-linux-melb:xfs-kern:32429a Signed-off-by: David Chinner <david@fromorbit.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Diffstat (limited to 'fs/xfs')
-rw-r--r--fs/xfs/linux-2.6/xfs_linux.h1
-rw-r--r--fs/xfs/xfs_iget.c32
2 files changed, 23 insertions, 10 deletions
diff --git a/fs/xfs/linux-2.6/xfs_linux.h b/fs/xfs/linux-2.6/xfs_linux.h
index 214717650b2..77d6ddcaf54 100644
--- a/fs/xfs/linux-2.6/xfs_linux.h
+++ b/fs/xfs/linux-2.6/xfs_linux.h
@@ -77,6 +77,7 @@
#include <linux/spinlock.h>
#include <linux/random.h>
#include <linux/ctype.h>
+#include <linux/writeback.h>
#include <asm/page.h>
#include <asm/div64.h>
diff --git a/fs/xfs/xfs_iget.c b/fs/xfs/xfs_iget.c
index 837cae78153..bf4dc5eb4cf 100644
--- a/fs/xfs/xfs_iget.c
+++ b/fs/xfs/xfs_iget.c
@@ -52,7 +52,7 @@ xfs_iget_cache_hit(
int lock_flags) __releases(pag->pag_ici_lock)
{
struct xfs_mount *mp = ip->i_mount;
- int error = 0;
+ int error = EAGAIN;
/*
* If INEW is set this inode is being set up
@@ -60,7 +60,6 @@ xfs_iget_cache_hit(
* Pause and try again.
*/
if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) {
- error = EAGAIN;
XFS_STATS_INC(xs_ig_frecycle);
goto out_error;
}
@@ -73,7 +72,6 @@ xfs_iget_cache_hit(
* error immediately so we don't remove it from the reclaim
* list and potentially leak the inode.
*/
-
if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
error = ENOENT;
goto out_error;
@@ -91,27 +89,42 @@ xfs_iget_cache_hit(
error = ENOMEM;
goto out_error;
}
+
+ /*
+ * We must set the XFS_INEW flag before clearing the
+ * XFS_IRECLAIMABLE flag so that if a racing lookup does
+ * not find the XFS_IRECLAIMABLE above but has the igrab()
+ * below succeed we can safely check XFS_INEW to detect
+ * that this inode is still being initialised.
+ */
xfs_iflags_set(ip, XFS_INEW);
xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
/* clear the radix tree reclaim flag as well. */
__xfs_inode_clear_reclaim_tag(mp, pag, ip);
- read_unlock(&pag->pag_ici_lock);
} else if (!igrab(VFS_I(ip))) {
/* If the VFS inode is being torn down, pause and try again. */
- error = EAGAIN;
XFS_STATS_INC(xs_ig_frecycle);
goto out_error;
- } else {
- /* we've got a live one */
- read_unlock(&pag->pag_ici_lock);
+ } else if (xfs_iflags_test(ip, XFS_INEW)) {
+ /*
+ * We are racing with another cache hit that is
+ * currently recycling this inode out of the XFS_IRECLAIMABLE
+ * state. Wait for the initialisation to complete before
+ * continuing.
+ */
+ wait_on_inode(VFS_I(ip));
}
if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
error = ENOENT;
- goto out;
+ iput(VFS_I(ip));
+ goto out_error;
}
+ /* We've got a live one. */
+ read_unlock(&pag->pag_ici_lock);
+
if (lock_flags != 0)
xfs_ilock(ip, lock_flags);
@@ -122,7 +135,6 @@ xfs_iget_cache_hit(
out_error:
read_unlock(&pag->pag_ici_lock);
-out:
return error;
}