diff mbox

Revert "btrfs: delete chunk allocation attemp when setting block group ro"

Message ID c99a895bc0facef23387ef7b44c98c0df41216fa.1429431096.git.osandov@osandov.com (mailing list archive)
State Accepted
Headers show

Commit Message

Omar Sandoval April 19, 2015, 8:17 a.m. UTC
This reverts commit 2f0810880f082fa8ba66ab2c33b02e4ff9770a5e.

This tried to fix the following test case:

	mkfs.ext4 -F /dev/sda
	btrfs-convert /dev/sda
	mount /dev/sda /mnt
	btrfs device add -f /dev/sdb /mnt
	btrfs balance start -v -dconvert=raid1 -mconvert=raid1 /mnt

Before the reverted commit, this test case failed with ENOSPC because
all chunks on the freshly converted filesystem were allocated, although
many were empty. The reverted commit removed an allocation attempt in
btrfs_set_block_group_ro(), but that fix wasn't right. After the
reverted commit, the balance succeeds, but the data/metadata profiles
aren't actually updated:

	# btrfs fi df /mnt
	Data, single: total=208.00MiB, used=49.48MiB
	System, single: total=32.00MiB, used=4.00KiB
	Metadata, single: total=208.00MiB, used=48.00KiB
	GlobalReserve, single: total=4.00MiB, used=0.00B

Indeed, several users reported that this commit caused a regression and
that converting the data and metadata profiles no longer works. This is
because the chunk allocation in question was where we actually allocated
the chunk with the new profile. Not seeing a more obvious fix, let's
just revert this. We can work around the ENOSPC in the original test
case by just issuing a balance to free up the unused block groups before
the conversion, anyways:

	mkfs.ext4 -F /dev/sda
	btrfs-convert /dev/sda
	mount /dev/sda /mnt
	btrfs balance start -v /mnt
	btrfs device add -f /dev/sdb /mnt
	btrfs balance start -v -dconvert=raid1 -mconvert=raid1 /mnt

Reported-by: Holger Hoffstätte <holger.hoffstaette@googlemail.com>
Cc: Shaohua Li <shli@fb.com>
Signed-off-by: Omar Sandoval <osandov@osandov.com>
---
Holger reported this awhile ago and I haven't seen a proper fix make it
to the ML. I messed around with it for awhile and came up empty, but I
figured I might as well send this in. Applies to v4.0.

 fs/btrfs/extent-tree.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

Comments

Omar Sandoval April 29, 2015, 6:24 p.m. UTC | #1
On Sun, Apr 19, 2015 at 01:17:05AM -0700, Omar Sandoval wrote:
> This reverts commit 2f0810880f082fa8ba66ab2c33b02e4ff9770a5e.
> 
> This tried to fix the following test case:
> 
> 	mkfs.ext4 -F /dev/sda
> 	btrfs-convert /dev/sda
> 	mount /dev/sda /mnt
> 	btrfs device add -f /dev/sdb /mnt
> 	btrfs balance start -v -dconvert=raid1 -mconvert=raid1 /mnt
> 
> Before the reverted commit, this test case failed with ENOSPC because
> all chunks on the freshly converted filesystem were allocated, although
> many were empty. The reverted commit removed an allocation attempt in
> btrfs_set_block_group_ro(), but that fix wasn't right. After the
> reverted commit, the balance succeeds, but the data/metadata profiles
> aren't actually updated:
> 
> 	# btrfs fi df /mnt
> 	Data, single: total=208.00MiB, used=49.48MiB
> 	System, single: total=32.00MiB, used=4.00KiB
> 	Metadata, single: total=208.00MiB, used=48.00KiB
> 	GlobalReserve, single: total=4.00MiB, used=0.00B
> 
> Indeed, several users reported that this commit caused a regression and
> that converting the data and metadata profiles no longer works. This is
> because the chunk allocation in question was where we actually allocated
> the chunk with the new profile. Not seeing a more obvious fix, let's
> just revert this. We can work around the ENOSPC in the original test
> case by just issuing a balance to free up the unused block groups before
> the conversion, anyways:
> 
> 	mkfs.ext4 -F /dev/sda
> 	btrfs-convert /dev/sda
> 	mount /dev/sda /mnt
> 	btrfs balance start -v /mnt
> 	btrfs device add -f /dev/sdb /mnt
> 	btrfs balance start -v -dconvert=raid1 -mconvert=raid1 /mnt
> 
> Reported-by: Holger Hoffstätte <holger.hoffstaette@googlemail.com>
> Cc: Shaohua Li <shli@fb.com>
> Signed-off-by: Omar Sandoval <osandov@osandov.com>
> ---
> Holger reported this awhile ago and I haven't seen a proper fix make it
> to the ML. I messed around with it for awhile and came up empty, but I
> figured I might as well send this in. Applies to v4.0.
> 
>  fs/btrfs/extent-tree.c | 13 ++++++++-----
>  1 file changed, 8 insertions(+), 5 deletions(-)
> 

Has another fix for this shown up yet? Currently 4.0 and up can't
convert between data/metadata profiles, which seems like a pretty major
regression.

Thanks,
Chris Mason April 29, 2015, 8:21 p.m. UTC | #2
On 04/29/2015 02:24 PM, Omar Sandoval wrote:
> On Sun, Apr 19, 2015 at 01:17:05AM -0700, Omar Sandoval wrote:

>> Holger reported this awhile ago and I haven't seen a proper fix make it
>> to the ML. I messed around with it for awhile and came up empty, but I
>> figured I might as well send this in. Applies to v4.0.
>>
>>  fs/btrfs/extent-tree.c | 13 ++++++++-----
>>  1 file changed, 8 insertions(+), 5 deletions(-)
>>
> 
> Has another fix for this shown up yet? Currently 4.0 and up can't
> convert between data/metadata profiles, which seems like a pretty major
> regression.
> 

I've got a fix in testing here, I'll get it out the door.

-chris


--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 8b353ad..6421852 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -8537,6 +8537,14 @@  int btrfs_set_block_group_ro(struct btrfs_root *root,
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
+	alloc_flags = update_block_group_flags(root, cache->flags);
+	if (alloc_flags != cache->flags) {
+		ret = do_chunk_alloc(trans, root, alloc_flags,
+				     CHUNK_ALLOC_FORCE);
+		if (ret < 0)
+			goto out;
+	}
+
 	ret = set_block_group_ro(cache, 0);
 	if (!ret)
 		goto out;
@@ -8547,11 +8555,6 @@  int btrfs_set_block_group_ro(struct btrfs_root *root,
 		goto out;
 	ret = set_block_group_ro(cache, 0);
 out:
-	if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
-		alloc_flags = update_block_group_flags(root, cache->flags);
-		check_system_chunk(trans, root, alloc_flags);
-	}
-
 	btrfs_end_transaction(trans, root);
 	return ret;
 }