Btrfs: fix race between enabling quotas and subvolume creation
authorFilipe Manana <fdmanana@suse.com>
Mon, 19 Nov 2018 16:20:34 +0000 (16:20 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 5 Dec 2018 18:32:06 +0000 (19:32 +0100)
commit 552f0329c75b3e1d7f9bb8c9e421d37403f192cd upstream.

We have a race between enabling quotas end subvolume creation that cause
subvolume creation to fail with -EINVAL, and the following diagram shows
how it happens:

              CPU 0                                          CPU 1

 btrfs_ioctl()
  btrfs_ioctl_quota_ctl()
   btrfs_quota_enable()
    mutex_lock(fs_info->qgroup_ioctl_lock)

                                                  btrfs_ioctl()
                                                   create_subvol()
                                                    btrfs_qgroup_inherit()
                                                     -> save fs_info->quota_root
                                                        into quota_root
                                                     -> stores a NULL value
                                                     -> tries to lock the mutex
                                                        qgroup_ioctl_lock
                                                        -> blocks waiting for
                                                           the task at CPU0

   -> sets BTRFS_FS_QUOTA_ENABLED in fs_info
   -> sets quota_root in fs_info->quota_root
      (non-NULL value)

   mutex_unlock(fs_info->qgroup_ioctl_lock)

                                                     -> checks quota enabled
                                                        flag is set
                                                     -> returns -EINVAL because
                                                        fs_info->quota_root was
                                                        NULL before it acquired
                                                        the mutex
                                                        qgroup_ioctl_lock
                                                   -> ioctl returns -EINVAL

Returning -EINVAL to user space will be confusing if all the arguments
passed to the subvolume creation ioctl were valid.

Fix it by grabbing the value from fs_info->quota_root after acquiring
the mutex.

CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/btrfs/qgroup.c

index b070401..ff43466 100644 (file)
@@ -2244,7 +2244,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
        int i;
        u64 *i_qgroups;
        struct btrfs_fs_info *fs_info = trans->fs_info;
-       struct btrfs_root *quota_root = fs_info->quota_root;
+       struct btrfs_root *quota_root;
        struct btrfs_qgroup *srcgroup;
        struct btrfs_qgroup *dstgroup;
        u32 level_size = 0;
@@ -2254,6 +2254,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
        if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
                goto out;
 
+       quota_root = fs_info->quota_root;
        if (!quota_root) {
                ret = -EINVAL;
                goto out;