dont-release-reserved-space-for-previously-allocated-cluster

   1 ext4: don't release reserved space for previously allocated cluster
   2
   3 From: Eric Whitney <enwlinux@gmail.com>
   4
   5 When xfstests' auto group is run on a bigalloc filesystem with a
   6 4.0-rc3 kernel, e2fsck failures and kernel warnings occur for some
   7 tests. e2fsck reports incorrect iblocks values, and the warnings
   8 indicate that the space reserved for delayed allocation is being
   9 overdrawn at allocation time.
  10
  11 Some of these errors occur because the reserved space is incorrectly
  12 decreased by one cluster when ext4_ext_map_blocks satisfies an
  13 allocation request by mapping an unused portion of a previously
  14 allocated cluster.  Because a cluster's worth of reserved space was
  15 already released when it was first allocated, it should not be released
  16 again.
  17
  18 This patch appears to correct the e2fsck failure reported for
  19 generic/232 and the kernel warnings produced by ext4/001, generic/009,
  20 and generic/033.  Failures and warnings for some other tests remain to
  21 be addressed.
  22
  23 Signed-off-by: Eric Whitney <enwlinux@gmail.com>
  24 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
  25 ---
  26  fs/ext4/extents.c | 14 +-------------
  27  1 file changed, 1 insertion(+), 13 deletions(-)
  28
  29 diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
  30 index bed4308..554190e 100644
  31 --- a/fs/ext4/extents.c
  32 +++ b/fs/ext4/extents.c
  33 @@ -4535,19 +4535,7 @@ got_allocated_blocks:
  34                  */
  35                 reserved_clusters = get_reserved_cluster_alloc(inode,
  36                                                 map->m_lblk, allocated);
  37 -               if (map_from_cluster) {
  38 -                       if (reserved_clusters) {
  39 -                               /*
  40 -                                * We have clusters reserved for this range.
  41 -                                * But since we are not doing actual allocation
  42 -                                * and are simply using blocks from previously
  43 -                                * allocated cluster, we should release the
  44 -                                * reservation and not claim quota.
  45 -                                */
  46 -                               ext4_da_update_reserve_space(inode,
  47 -                                               reserved_clusters, 0);
  48 -                       }
  49 -               } else {
  50 +               if (!map_from_cluster) {
  51                         BUG_ON(allocated_clusters < reserved_clusters);
  52                         if (reserved_clusters < allocated_clusters) {
  53                                 struct ext4_inode_info *ei = EXT4_I(inode);
  54 --
  55 2.1.0
  56
  57