btrfs: honor path->skip_locking in backref code
authorJosef Bacik <josef@toxicpanda.com>
Wed, 16 Jan 2019 16:00:57 +0000 (11:00 -0500)
committerDavid Sterba <dsterba@suse.com>
Mon, 25 Feb 2019 13:13:39 +0000 (14:13 +0100)
Qgroups will do the old roots lookup at delayed ref time, which could be
while walking down the extent root while running a delayed ref.  This
should be fine, except we specifically lock eb's in the backref walking
code irrespective of path->skip_locking, which deadlocks the system.
Fix up the backref code to honor path->skip_locking, nobody will be
modifying the commit_root when we're searching so it's completely safe
to do.

This happens since fb235dc06fac ("btrfs: qgroup: Move half of the qgroup
accounting time out of commit trans"), kernel may lockup with quota
enabled.

There is one backref trace triggered by snapshot dropping along with
write operation in the source subvolume.  The example can be reliably
reproduced:

  btrfs-cleaner   D    0  4062      2 0x80000000
  Call Trace:
   schedule+0x32/0x90
   btrfs_tree_read_lock+0x93/0x130 [btrfs]
   find_parent_nodes+0x29b/0x1170 [btrfs]
   btrfs_find_all_roots_safe+0xa8/0x120 [btrfs]
   btrfs_find_all_roots+0x57/0x70 [btrfs]
   btrfs_qgroup_trace_extent_post+0x37/0x70 [btrfs]
   btrfs_qgroup_trace_leaf_items+0x10b/0x140 [btrfs]
   btrfs_qgroup_trace_subtree+0xc8/0xe0 [btrfs]
   do_walk_down+0x541/0x5e3 [btrfs]
   walk_down_tree+0xab/0xe7 [btrfs]
   btrfs_drop_snapshot+0x356/0x71a [btrfs]
   btrfs_clean_one_deleted_snapshot+0xb8/0xf0 [btrfs]
   cleaner_kthread+0x12b/0x160 [btrfs]
   kthread+0x112/0x130
   ret_from_fork+0x27/0x50

When dropping snapshots with qgroup enabled, we will trigger backref
walk.

However such backref walk at that timing is pretty dangerous, as if one
of the parent nodes get WRITE locked by other thread, we could cause a
dead lock.

For example:

           FS 260     FS 261 (Dropped)
            node A        node B
           /      \      /      \
       node C      node D      node E
      /   \         /  \        /     \
  leaf F|leaf G|leaf H|leaf I|leaf J|leaf K

The lock sequence would be:

      Thread A (cleaner)             |       Thread B (other writer)
-----------------------------------------------------------------------
write_lock(B)                        |
write_lock(D)                        |
^^^ called by walk_down_tree()       |
                                     |       write_lock(A)
                                     |       write_lock(D) << Stall
read_lock(H) << for backref walk     |
read_lock(D) << lock owner is        |
                the same thread A    |
                so read lock is OK   |
read_lock(A) << Stall                |

So thread A hold write lock D, and needs read lock A to unlock.
While thread B holds write lock A, while needs lock D to unlock.

This will cause a deadlock.

This is not only limited to snapshot dropping case.  As the backref
walk, even only happens on commit trees, is breaking the normal top-down
locking order, makes it deadlock prone.

Fixes: fb235dc06fac ("btrfs: qgroup: Move half of the qgroup accounting time out of commit trans")
CC: stable@vger.kernel.org # 4.14+
Reported-and-tested-by: David Sterba <dsterba@suse.com>
Reported-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
[ rebase to latest branch and fix lock assert bug in btrfs/007 ]
Signed-off-by: Qu Wenruo <wqu@suse.com>
[ copy logs and deadlock analysis from Qu's patch ]
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/backref.c

index 136454dbb4afdde1330f07eae441392be18c0973..11459fe84a2965f795cfe6be3f91b9805fb72354 100644 (file)
@@ -712,7 +712,7 @@ out:
  * read tree blocks and add keys where required.
  */
 static int add_missing_keys(struct btrfs_fs_info *fs_info,
  * read tree blocks and add keys where required.
  */
 static int add_missing_keys(struct btrfs_fs_info *fs_info,
-                           struct preftrees *preftrees)
+                           struct preftrees *preftrees, bool lock)
 {
        struct prelim_ref *ref;
        struct extent_buffer *eb;
 {
        struct prelim_ref *ref;
        struct extent_buffer *eb;
@@ -737,12 +737,14 @@ static int add_missing_keys(struct btrfs_fs_info *fs_info,
                        free_extent_buffer(eb);
                        return -EIO;
                }
                        free_extent_buffer(eb);
                        return -EIO;
                }
-               btrfs_tree_read_lock(eb);
+               if (lock)
+                       btrfs_tree_read_lock(eb);
                if (btrfs_header_level(eb) == 0)
                        btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
                else
                        btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
                if (btrfs_header_level(eb) == 0)
                        btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
                else
                        btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
-               btrfs_tree_read_unlock(eb);
+               if (lock)
+                       btrfs_tree_read_unlock(eb);
                free_extent_buffer(eb);
                prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
                cond_resched();
                free_extent_buffer(eb);
                prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
                cond_resched();
@@ -1227,7 +1229,7 @@ again:
 
        btrfs_release_path(path);
 
 
        btrfs_release_path(path);
 
-       ret = add_missing_keys(fs_info, &preftrees);
+       ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
        if (ret)
                goto out;
 
        if (ret)
                goto out;
 
@@ -1288,11 +1290,15 @@ again:
                                        ret = -EIO;
                                        goto out;
                                }
                                        ret = -EIO;
                                        goto out;
                                }
-                               btrfs_tree_read_lock(eb);
-                               btrfs_set_lock_blocking_read(eb);
+
+                               if (!path->skip_locking) {
+                                       btrfs_tree_read_lock(eb);
+                                       btrfs_set_lock_blocking_read(eb);
+                               }
                                ret = find_extent_in_eb(eb, bytenr,
                                                        *extent_item_pos, &eie, ignore_offset);
                                ret = find_extent_in_eb(eb, bytenr,
                                                        *extent_item_pos, &eie, ignore_offset);
-                               btrfs_tree_read_unlock_blocking(eb);
+                               if (!path->skip_locking)
+                                       btrfs_tree_read_unlock_blocking(eb);
                                free_extent_buffer(eb);
                                if (ret < 0)
                                        goto out;
                                free_extent_buffer(eb);
                                if (ret < 0)
                                        goto out;