xfs: fix freeing of inodes not yet added to the inode cache
authorChristoph Hellwig <hch@lst.de>
Fri, 7 Aug 2009 17:38:34 +0000 (14:38 -0300)
committerChristoph Hellwig <hch@brick.lst.de>
Fri, 7 Aug 2009 17:38:34 +0000 (14:38 -0300)
When freeing an inode that lost race getting added to the inode cache we
must not call into ->destroy_inode, because that would delete the inode
that won the race from the inode cache radix tree.

This patch uses splits a new xfs_inode_free helper out of xfs_ireclaim
and uses that plus __destroy_inode to make sure we really only free
the memory allocted for the inode that lost the race, and not mess with
the inode cache state.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@sandeen.net>
Reported-by: Alex Samad <alex@samad.com.au>
Reported-by: Andrew Randrianasulu <randrik@mail.ru>
Reported-by: Stephane <sharnois@max-t.com>
Reported-by: Tommy <tommy@news-service.com>
Reported-by: Miah Gregory <mace@darksilence.net>
Reported-by: Gabriel Barazer <gabriel@oxeva.fr>
Reported-by: Leandro Lucarella <llucax@gmail.com>
Reported-by: Daniel Burr <dburr@fami.com.au>
Reported-by: Nickolay <newmail@spaces.ru>
Reported-by: Michael Guntsche <mike@it-loops.com>
Reported-by: Dan Carley <dan.carley+linuxkern-bugs@gmail.com>
Reported-by: Michael Ole Olsen <gnu@gmx.net>
Reported-by: Michael Weissenbacher <mw@dermichi.com>
Reported-by: Martin Spott <Martin.Spott@mgras.net>
Reported-by: Christian Kujau <lists@nerdbynature.de>
Tested-by: Michael Guntsche <mike@it-loops.com>
Tested-by: Dan Carley <dan.carley+linuxkern-bugs@gmail.com>
Tested-by: Christian Kujau <lists@nerdbynature.de>
fs/xfs/xfs_iget.c
fs/xfs/xfs_inode.h

index 719c85b155f440d34e7dc882a3e200cbb022a644..34ec86923f7e354044fbe9ffa27a77126ce354cf 100644 (file)
@@ -116,6 +116,71 @@ xfs_inode_alloc(
        return ip;
 }
 
+STATIC void
+xfs_inode_free(
+       struct xfs_inode        *ip)
+{
+       switch (ip->i_d.di_mode & S_IFMT) {
+       case S_IFREG:
+       case S_IFDIR:
+       case S_IFLNK:
+               xfs_idestroy_fork(ip, XFS_DATA_FORK);
+               break;
+       }
+
+       if (ip->i_afp)
+               xfs_idestroy_fork(ip, XFS_ATTR_FORK);
+
+#ifdef XFS_INODE_TRACE
+       ktrace_free(ip->i_trace);
+#endif
+#ifdef XFS_BMAP_TRACE
+       ktrace_free(ip->i_xtrace);
+#endif
+#ifdef XFS_BTREE_TRACE
+       ktrace_free(ip->i_btrace);
+#endif
+#ifdef XFS_RW_TRACE
+       ktrace_free(ip->i_rwtrace);
+#endif
+#ifdef XFS_ILOCK_TRACE
+       ktrace_free(ip->i_lock_trace);
+#endif
+#ifdef XFS_DIR2_TRACE
+       ktrace_free(ip->i_dir_trace);
+#endif
+
+       if (ip->i_itemp) {
+               /*
+                * Only if we are shutting down the fs will we see an
+                * inode still in the AIL. If it is there, we should remove
+                * it to prevent a use-after-free from occurring.
+                */
+               xfs_log_item_t  *lip = &ip->i_itemp->ili_item;
+               struct xfs_ail  *ailp = lip->li_ailp;
+
+               ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
+                                      XFS_FORCED_SHUTDOWN(ip->i_mount));
+               if (lip->li_flags & XFS_LI_IN_AIL) {
+                       spin_lock(&ailp->xa_lock);
+                       if (lip->li_flags & XFS_LI_IN_AIL)
+                               xfs_trans_ail_delete(ailp, lip);
+                       else
+                               spin_unlock(&ailp->xa_lock);
+               }
+               xfs_inode_item_destroy(ip);
+               ip->i_itemp = NULL;
+       }
+
+       /* asserts to verify all state is correct here */
+       ASSERT(atomic_read(&ip->i_iocount) == 0);
+       ASSERT(atomic_read(&ip->i_pincount) == 0);
+       ASSERT(!spin_is_locked(&ip->i_flags_lock));
+       ASSERT(completion_done(&ip->i_flush));
+
+       kmem_zone_free(xfs_inode_zone, ip);
+}
+
 /*
  * Check the validity of the inode we just found it the cache
  */
@@ -292,7 +357,8 @@ out_preload_end:
        if (lock_flags)
                xfs_iunlock(ip, lock_flags);
 out_destroy:
-       xfs_destroy_inode(ip);
+       __destroy_inode(VFS_I(ip));
+       xfs_inode_free(ip);
        return error;
 }
 
@@ -497,62 +563,7 @@ xfs_ireclaim(
        xfs_qm_dqdetach(ip);
        xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 
-       switch (ip->i_d.di_mode & S_IFMT) {
-       case S_IFREG:
-       case S_IFDIR:
-       case S_IFLNK:
-               xfs_idestroy_fork(ip, XFS_DATA_FORK);
-               break;
-       }
-
-       if (ip->i_afp)
-               xfs_idestroy_fork(ip, XFS_ATTR_FORK);
-
-#ifdef XFS_INODE_TRACE
-       ktrace_free(ip->i_trace);
-#endif
-#ifdef XFS_BMAP_TRACE
-       ktrace_free(ip->i_xtrace);
-#endif
-#ifdef XFS_BTREE_TRACE
-       ktrace_free(ip->i_btrace);
-#endif
-#ifdef XFS_RW_TRACE
-       ktrace_free(ip->i_rwtrace);
-#endif
-#ifdef XFS_ILOCK_TRACE
-       ktrace_free(ip->i_lock_trace);
-#endif
-#ifdef XFS_DIR2_TRACE
-       ktrace_free(ip->i_dir_trace);
-#endif
-       if (ip->i_itemp) {
-               /*
-                * Only if we are shutting down the fs will we see an
-                * inode still in the AIL. If it is there, we should remove
-                * it to prevent a use-after-free from occurring.
-                */
-               xfs_log_item_t  *lip = &ip->i_itemp->ili_item;
-               struct xfs_ail  *ailp = lip->li_ailp;
-
-               ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
-                                      XFS_FORCED_SHUTDOWN(ip->i_mount));
-               if (lip->li_flags & XFS_LI_IN_AIL) {
-                       spin_lock(&ailp->xa_lock);
-                       if (lip->li_flags & XFS_LI_IN_AIL)
-                               xfs_trans_ail_delete(ailp, lip);
-                       else
-                               spin_unlock(&ailp->xa_lock);
-               }
-               xfs_inode_item_destroy(ip);
-               ip->i_itemp = NULL;
-       }
-       /* asserts to verify all state is correct here */
-       ASSERT(atomic_read(&ip->i_iocount) == 0);
-       ASSERT(atomic_read(&ip->i_pincount) == 0);
-       ASSERT(!spin_is_locked(&ip->i_flags_lock));
-       ASSERT(completion_done(&ip->i_flush));
-       kmem_zone_free(xfs_inode_zone, ip);
+       xfs_inode_free(ip);
 }
 
 /*
index 1804f866a71d285b760db6bd5df1574676ecb898..65f24a3cc9922a174b8989c276691e01b1e56621 100644 (file)
@@ -309,23 +309,6 @@ static inline struct inode *VFS_I(struct xfs_inode *ip)
        return &ip->i_vnode;
 }
 
-/*
- * Get rid of a partially initialized inode.
- *
- * We have to go through destroy_inode to make sure allocations
- * from init_inode_always like the security data are undone.
- *
- * We mark the inode bad so that it takes the short cut in
- * the reclaim path instead of going through the flush path
- * which doesn't make sense for an inode that has never seen the
- * light of day.
- */
-static inline void xfs_destroy_inode(struct xfs_inode *ip)
-{
-       make_bad_inode(VFS_I(ip));
-       return destroy_inode(VFS_I(ip));
-}
-
 /*
  * i_flags helper functions
  */