Skip to content

Commit

Permalink
Btrfs: relax the block group size limit for bitmaps
Browse files Browse the repository at this point in the history
Dave pointed out that xfstests 273 will tell you that it failed to load the
space cache for a block group when it remounts.  This is because we run out
of space writing out the block group cache.  This is ok and is working as it
should, but let's try to be a bit nicer.  This happens because the block
group was 100mb, but bitmap entries cover 128mb, so we were only getting
extent entries for this block group, which ended up being too many to fit in
the free space cache.  So relax the bitmap size requirements to block groups
that are at least half the size a bitmap will cover or larger, that way we
can still keep the amount of space used in the free space cache low enough
to be able to write it out.  With this patch I no longer fail to write out
the free space cache.  Thanks,

Reported-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
  • Loading branch information
Josef Bacik committed Feb 20, 2013
1 parent 3e39cea commit dde5740
Showing 1 changed file with 9 additions and 3 deletions.
12 changes: 9 additions & 3 deletions fs/btrfs/free-space-cache.c
Original file line number Diff line number Diff line change
Expand Up @@ -1356,6 +1356,8 @@ static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);

max_bitmaps = max(max_bitmaps, 1);

BUG_ON(ctl->total_bitmaps > max_bitmaps);

/*
Expand Down Expand Up @@ -1636,10 +1638,14 @@ static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
}

/*
* some block groups are so tiny they can't be enveloped by a bitmap, so
* don't even bother to create a bitmap for this
* The original block groups from mkfs can be really small, like 8
* megabytes, so don't bother with a bitmap for those entries. However
* some block groups can be smaller than what a bitmap would cover but
* are still large enough that they could overflow the 32k memory limit,
* so allow those block groups to still be allowed to have a bitmap
* entry.
*/
if (BITS_PER_BITMAP * ctl->unit > block_group->key.offset)
if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
return false;

return true;
Expand Down

0 comments on commit dde5740

Please sign in to comment.