locks: fix a potential use-after-free problem when wakeup a waiter
authoryangerkun <yangerkun@huawei.com>
Wed, 4 Mar 2020 07:25:56 +0000 (15:25 +0800)
committerJeff Layton <jlayton@kernel.org>
Fri, 6 Mar 2020 16:54:13 +0000 (11:54 -0500)
'16306a61d3b7 ("fs/locks: always delete_block after waiting.")' add the
logic to check waiter->fl_blocker without blocked_lock_lock. And it will
trigger a UAF when we try to wakeup some waiter:

Thread 1 has create a write flock a on file, and now thread 2 try to
unlock and delete flock a, thread 3 try to add flock b on the same file.

Thread2                         Thread3
                                flock syscall(create flock b)
                        ...flock_lock_inode_wait
    flock_lock_inode(will insert
    our fl_blocked_member list
    to flock a's fl_blocked_requests)
   sleep
flock syscall(unlock)
...flock_lock_inode_wait
    locks_delete_lock_ctx
    ...__locks_wake_up_blocks
        __locks_delete_blocks(
b->fl_blocker = NULL)
...
                                   break by a signal
   locks_delete_block
    b->fl_blocker == NULL &&
    list_empty(&b->fl_blocked_requests)
                            success, return directly
 locks_free_lock b
wake_up(&b->fl_waiter)
trigger UAF

Fix it by remove this logic, and this patch may also fix CVE-2019-19769.

Cc: stable@vger.kernel.org
Fixes: 16306a61d3b7 ("fs/locks: always delete_block after waiting.")
Signed-off-by: yangerkun <yangerkun@huawei.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
fs/locks.c

index 44b6da0..426b55d 100644 (file)
@@ -753,20 +753,6 @@ int locks_delete_block(struct file_lock *waiter)
 {
        int status = -ENOENT;
 
-       /*
-        * If fl_blocker is NULL, it won't be set again as this thread
-        * "owns" the lock and is the only one that might try to claim
-        * the lock.  So it is safe to test fl_blocker locklessly.
-        * Also if fl_blocker is NULL, this waiter is not listed on
-        * fl_blocked_requests for some lock, so no other request can
-        * be added to the list of fl_blocked_requests for this
-        * request.  So if fl_blocker is NULL, it is safe to
-        * locklessly check if fl_blocked_requests is empty.  If both
-        * of these checks succeed, there is no need to take the lock.
-        */
-       if (waiter->fl_blocker == NULL &&
-           list_empty(&waiter->fl_blocked_requests))
-               return status;
        spin_lock(&blocked_lock_lock);
        if (waiter->fl_blocker)
                status = 0;