btrfs: fix one bug that process may endlessly wait for ticket in wait_reserve_ticket()
authorWang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Wed, 31 Aug 2016 11:46:16 +0000 (19:46 +0800)
committerDavid Sterba <dsterba@suse.com>
Thu, 1 Sep 2016 15:23:24 +0000 (17:23 +0200)
If can_overcommit() in btrfs_calc_reclaim_metadata_size() returns true,
btrfs_async_reclaim_metadata_space() will not reclaim metadata space, just
return directly and also forget to wake up process which are waiting for
their tickets, so these processes will wait endlessly.

Fstests case generic/172 with mount option "-o compress=lzo" have revealed
this bug in my test machine. Here if we have tickets to handle, we must
handle them first.

Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/extent-tree.c

index 60d4ae7ce97464b8312b344ab35a41bef61d2439..64676a16d32a4117a27ce7623d42cdccffb0971a 100644 (file)
@@ -4901,11 +4901,6 @@ btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
        u64 expected;
        u64 to_reclaim = 0;
 
-       to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
-       if (can_overcommit(root, space_info, to_reclaim,
-                          BTRFS_RESERVE_FLUSH_ALL))
-               return 0;
-
        list_for_each_entry(ticket, &space_info->tickets, list)
                to_reclaim += ticket->bytes;
        list_for_each_entry(ticket, &space_info->priority_tickets, list)
@@ -4913,6 +4908,11 @@ btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
        if (to_reclaim)
                return to_reclaim;
 
+       to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
+       if (can_overcommit(root, space_info, to_reclaim,
+                          BTRFS_RESERVE_FLUSH_ALL))
+               return 0;
+
        used = space_info->bytes_used + space_info->bytes_reserved +
               space_info->bytes_pinned + space_info->bytes_readonly +
               space_info->bytes_may_use;