Btrfs: fix list transaction->pending_ordered corruption
authorFilipe Manana <fdmanana@suse.com>
Fri, 3 Jul 2015 19:30:34 +0000 (20:30 +0100)
committerFilipe Manana <fdmanana@suse.com>
Sat, 11 Jul 2015 21:35:05 +0000 (22:35 +0100)
When we call btrfs_commit_transaction(), we splice the list "ordered"
of our transaction handle into the transaction's "pending_ordered"
list, but we don't re-initialize the "ordered" list of our transaction
handle, this means it still points to the same elements it used to
before the splice. Then we check if the current transaction's state is
>= TRANS_STATE_COMMIT_START and if it is we end up calling
btrfs_end_transaction() which simply splices again the "ordered" list
of our handle into the transaction's "pending_ordered" list, leaving
multiple pointers to the same ordered extents which results in list
corruption when we are iterating, removing and freeing ordered extents
at btrfs_wait_pending_ordered(), resulting in access to dangling
pointers / use-after-free issues.
Similarly, btrfs_end_transaction() can end up in some cases calling
btrfs_commit_transaction(), and both did a list splice of the transaction
handle's "ordered" list into the transaction's "pending_ordered" without
re-initializing the handle's "ordered" list, resulting in exactly the
same problem.

This produces the following warning on a kernel with linked list
debugging enabled:

[109749.265416] ------------[ cut here ]------------
[109749.266410] WARNING: CPU: 7 PID: 324 at lib/list_debug.c:59 __list_del_entry+0x5a/0x98()
[109749.267969] list_del corruption. prev->next should be ffff8800ba087e20, but was fffffff8c1f7c35d
(...)
[109749.287505] Call Trace:
[109749.288135]  [<ffffffff8145f077>] dump_stack+0x4f/0x7b
[109749.298080]  [<ffffffff81095de5>] ? console_unlock+0x356/0x3a2
[109749.331605]  [<ffffffff8104b3b0>] warn_slowpath_common+0xa1/0xbb
[109749.334849]  [<ffffffff81260642>] ? __list_del_entry+0x5a/0x98
[109749.337093]  [<ffffffff8104b410>] warn_slowpath_fmt+0x46/0x48
[109749.337847]  [<ffffffff81260642>] __list_del_entry+0x5a/0x98
[109749.338678]  [<ffffffffa053e8bf>] btrfs_wait_pending_ordered+0x46/0xdb [btrfs]
[109749.340145]  [<ffffffffa058a65f>] ? __btrfs_run_delayed_items+0x149/0x163 [btrfs]
[109749.348313]  [<ffffffffa054077d>] btrfs_commit_transaction+0x36b/0xa10 [btrfs]
[109749.349745]  [<ffffffff81087310>] ? trace_hardirqs_on+0xd/0xf
[109749.350819]  [<ffffffffa055370d>] btrfs_sync_file+0x36f/0x3fc [btrfs]
[109749.351976]  [<ffffffff8118ec98>] vfs_fsync_range+0x8f/0x9e
[109749.360341]  [<ffffffff8118ecc3>] vfs_fsync+0x1c/0x1e
[109749.368828]  [<ffffffff8118ee1d>] do_fsync+0x34/0x4e
[109749.369790]  [<ffffffff8118f045>] SyS_fsync+0x10/0x14
[109749.370925]  [<ffffffff81465197>] system_call_fastpath+0x12/0x6f
[109749.382274] ---[ end trace 48e0d07f7c03d95a ]---

On a non-debug kernel this leads to invalid memory accesses, causing a
crash. Fix this by using list_splice_init() instead of list_splice() in
btrfs_commit_transaction() and btrfs_end_transaction().

Cc: stable@vger.kernel.org
Fixes: 50d9aa99bd35 ("Btrfs: make sure logged extents complete in the current transaction V3"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
fs/btrfs/transaction.c

index c0f18e7266b673c22357ebb033e40a4ec6e28b60..51e0f0d0053e52c965c4ebed464fc4a761523a67 100644 (file)
@@ -761,7 +761,7 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 
        if (!list_empty(&trans->ordered)) {
                spin_lock(&info->trans_lock);
-               list_splice(&trans->ordered, &cur_trans->pending_ordered);
+               list_splice_init(&trans->ordered, &cur_trans->pending_ordered);
                spin_unlock(&info->trans_lock);
        }
 
@@ -1866,7 +1866,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        }
 
        spin_lock(&root->fs_info->trans_lock);
-       list_splice(&trans->ordered, &cur_trans->pending_ordered);
+       list_splice_init(&trans->ordered, &cur_trans->pending_ordered);
        if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
                spin_unlock(&root->fs_info->trans_lock);
                atomic_inc(&cur_trans->use_count);