fix wrong iput on d_inode introduced by e6bc45d65d
authorTörök Edwin <edwintorok@gmail.com>
Wed, 15 Jun 2011 21:06:14 +0000 (00:06 +0300)
committerAl Viro <viro@zeniv.linux.org.uk>
Thu, 16 Jun 2011 15:27:39 +0000 (11:27 -0400)
Git bisection shows that commit e6bc45d65df8599fdbae73be9cec4ceed274db53 causes
BUG_ONs under high I/O load:

kernel BUG at fs/inode.c:1368!
[ 2862.501007] Call Trace:
[ 2862.501007]  [<ffffffff811691d8>] d_kill+0xf8/0x140
[ 2862.501007]  [<ffffffff81169c19>] dput+0xc9/0x190
[ 2862.501007]  [<ffffffff8115577f>] fput+0x15f/0x210
[ 2862.501007]  [<ffffffff81152171>] filp_close+0x61/0x90
[ 2862.501007]  [<ffffffff81152251>] sys_close+0xb1/0x110
[ 2862.501007]  [<ffffffff814c14fb>] system_call_fastpath+0x16/0x1b

A reliable way to reproduce this bug is:
Login to KDE, run 'rsnapshot sync', and apt-get install openjdk-6-jdk,
and apt-get remove openjdk-6-jdk.

The buggy part of the patch is this:
struct inode *inode = NULL;
.....
-               if (nd.last.name[nd.last.len])
-                       goto slashes;
                inode = dentry->d_inode;
-               if (inode)
-                       ihold(inode);
+               if (nd.last.name[nd.last.len] || !inode)
+                       goto slashes;
+               ihold(inode)
...
if (inode)
iput(inode); /* truncate the inode here */

If nd.last.name[nd.last.len] is nonzero (and thus goto slashes branch is taken),
and dentry->d_inode is non-NULL, then this code now does an additional iput on
the inode, which is wrong.

Fix this by only setting the inode variable if nd.last.name[nd.last.len] is 0.

Reference: https://lkml.org/lkml/2011/6/15/50
Reported-by: Norbert Preining <preining@logic.at>
Reported-by: Török Edwin <edwintorok@gmail.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Török Edwin <edwintorok@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/namei.c

index 9802345df5e78c86396fa05f4c29c760e3a270a9..6301963b161f6cb5753bbd4723c547cd5b16a27e 100644 (file)
@@ -2713,8 +2713,10 @@ static long do_unlinkat(int dfd, const char __user *pathname)
        error = PTR_ERR(dentry);
        if (!IS_ERR(dentry)) {
                /* Why not before? Because we want correct error value */
+               if (nd.last.name[nd.last.len])
+                       goto slashes;
                inode = dentry->d_inode;
-               if (nd.last.name[nd.last.len] || !inode)
+               if (!inode)
                        goto slashes;
                ihold(inode);
                error = mnt_want_write(nd.path.mnt);