configfs: fix race between dentry put and lookup
authorJunxiao Bi <junxiao.bi@oracle.com>
Thu, 21 Nov 2013 22:31:56 +0000 (14:31 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 22 Nov 2013 00:42:27 +0000 (16:42 -0800)
A race window in configfs, it starts from one dentry is UNHASHED and end
before configfs_d_iput is called.  In this window, if a lookup happen,
since the original dentry was UNHASHED, so a new dentry will be
allocated, and then in configfs_attach_attr(), sd->s_dentry will be
updated to the new dentry.  Then in configfs_d_iput(),
BUG_ON(sd->s_dentry != dentry) will be triggered and system panic.

sys_open:                     sys_close:
 ...                           fput
                                dput
                                 dentry_kill
                                  __d_drop <--- dentry unhashed here,
                                           but sd->dentry still point
                                           to this dentry.

 lookup_real
  configfs_lookup
   configfs_attach_attr---> update sd->s_dentry
                            to new allocated dentry here.

                                   d_kill
                                     configfs_d_iput <--- BUG_ON(sd->s_dentry != dentry)
                                                     triggered here.

To fix it, change configfs_d_iput to not update sd->s_dentry if
sd->s_count > 2, that means there are another dentry is using the sd
beside the one that is going to be put.  Use configfs_dirent_lock in
configfs_attach_attr to sync with configfs_d_iput.

With the following steps, you can reproduce the bug.

1. enable ocfs2, this will mount configfs at /sys/kernel/config and
   fill configure in it.

2. run the following script.
while [ 1 ]; do cat /sys/kernel/config/cluster/$your_cluster_name/idle_timeout_ms > /dev/null; done &
while [ 1 ]; do cat /sys/kernel/config/cluster/$your_cluster_name/idle_timeout_ms > /dev/null; done &

Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/configfs/dir.c

index 4522e0755773cdc1c5a2d293808b474a87d375ca..e081acbac2e756372340379fea7a69a84c9c3dd0 100644 (file)
@@ -56,10 +56,19 @@ static void configfs_d_iput(struct dentry * dentry,
        struct configfs_dirent *sd = dentry->d_fsdata;
 
        if (sd) {
-               BUG_ON(sd->s_dentry != dentry);
                /* Coordinate with configfs_readdir */
                spin_lock(&configfs_dirent_lock);
-               sd->s_dentry = NULL;
+               /* Coordinate with configfs_attach_attr where will increase
+                * sd->s_count and update sd->s_dentry to new allocated one.
+                * Only set sd->dentry to null when this dentry is the only
+                * sd owner.
+                * If not do so, configfs_d_iput may run just after
+                * configfs_attach_attr and set sd->s_dentry to null
+                * even it's still in use.
+                */
+               if (atomic_read(&sd->s_count) <= 2)
+                       sd->s_dentry = NULL;
+
                spin_unlock(&configfs_dirent_lock);
                configfs_put(sd);
        }
@@ -416,8 +425,11 @@ static int configfs_attach_attr(struct configfs_dirent * sd, struct dentry * den
        struct configfs_attribute * attr = sd->s_element;
        int error;
 
+       spin_lock(&configfs_dirent_lock);
        dentry->d_fsdata = configfs_get(sd);
        sd->s_dentry = dentry;
+       spin_unlock(&configfs_dirent_lock);
+
        error = configfs_create(dentry, (attr->ca_mode & S_IALLUGO) | S_IFREG,
                                configfs_init_file);
        if (error) {