[PATCH] Btrfs: fix regression in raid level conversion

[PATCH] Btrfs: fix regression in raid level conversion

[Chris Mason]

Commit 2f0810880f082fa8ba66ab2c33b02e4ff9770a5e changed
btrfs_set_block_group_ro to avoid trying to allocate new chunks with the
new raid profile during conversion.  This fixed failures when there was
no space on the drive to allocate a new chunk, but the metadata
reserves were sufficient to continue the conversion.

But this ended up causing a regression when the drive had plenty of
space to allocate new chunks, mostly because reduce_alloc_profile isn't
using the new raid profile.

Fixing btrfs_reduce_alloc_profile is a bigger patch.  For now, do a
partial revert of 2f0810880, and don't error out if we hit ENOSPC.

Signed-off-by: Chris Mason <clm@fb.com>
---
 fs/btrfs/extent-tree.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 45e3f08..a115599 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -8829,6 +8829,26 @@ again:
 		goto again;
 	}
 
+	/*
+	 * if we are changing raid levels, try to allocate a corresponding
+	 * block group with the new raid level.
+	 */
+	if (!(cache->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {
+		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);
+			/*
+			 * ENOSPC is allowed here, we may have enough space
+			 * already allocated at the new raid level to
+			 * carry on
+			 */
+			if (ret == -ENOSPC)
+				ret = 0;
+			if (ret < 0)
+				goto out;
+		}
+	}
 
 	ret = set_block_group_ro(cache, 0);
 	if (!ret)
-- 
2.4.0

Re: [PATCH] Btrfs: fix regression in raid level conversion

[David Sterba]

On Wed, May 20, 2015 at 10:13:11AM -0400, Chris Mason wrote:
> Commit 2f0810880f082fa8ba66ab2c33b02e4ff9770a5e changed
> btrfs_set_block_group_ro to avoid trying to allocate new chunks with the
> new raid profile during conversion.  This fixed failures when there was
> no space on the drive to allocate a new chunk, but the metadata
> reserves were sufficient to continue the conversion.
> 
> But this ended up causing a regression when the drive had plenty of
> space to allocate new chunks, mostly because reduce_alloc_profile isn't
> using the new raid profile.
> 
> Fixing btrfs_reduce_alloc_profile is a bigger patch.  For now, do a
> partial revert of 2f0810880, and don't error out if we hit ENOSPC.
> 
> Signed-off-by: Chris Mason <clm@fb.com>
> ---
>  fs/btrfs/extent-tree.c | 20 ++++++++++++++++++++
>  1 file changed, 20 insertions(+)
> 
> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> index 45e3f08..a115599 100644
> --- a/fs/btrfs/extent-tree.c
> +++ b/fs/btrfs/extent-tree.c
> @@ -8829,6 +8829,26 @@ again:
>  		goto again;
>  	}
>  
> +	/*
> +	 * if we are changing raid levels, try to allocate a corresponding
> +	 * block group with the new raid level.
> +	 */
> +	if (!(cache->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {

This prevents to switch the system chunk in all cases. What was the
reason to do it?

If I remove the check, then the conversions work in all combinations.
Eg.

* convert from ext (or use any single device fs)
* add second device
* balance data/metadata/system to raid1
* all fine

Re: [PATCH] Btrfs: fix regression in raid level conversion

[Chris Mason]

On 05/20/2015 01:02 PM, David Sterba wrote:
> On Wed, May 20, 2015 at 10:13:11AM -0400, Chris Mason wrote:
>> Commit 2f0810880f082fa8ba66ab2c33b02e4ff9770a5e changed
>> btrfs_set_block_group_ro to avoid trying to allocate new chunks with the
>> new raid profile during conversion.  This fixed failures when there was
>> no space on the drive to allocate a new chunk, but the metadata
>> reserves were sufficient to continue the conversion.
>>
>> But this ended up causing a regression when the drive had plenty of
>> space to allocate new chunks, mostly because reduce_alloc_profile isn't
>> using the new raid profile.
>>
>> Fixing btrfs_reduce_alloc_profile is a bigger patch.  For now, do a
>> partial revert of 2f0810880, and don't error out if we hit ENOSPC.
>>
>> Signed-off-by: Chris Mason <clm@fb.com>
>> ---
>>  fs/btrfs/extent-tree.c | 20 ++++++++++++++++++++
>>  1 file changed, 20 insertions(+)
>>
>> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
>> index 45e3f08..a115599 100644
>> --- a/fs/btrfs/extent-tree.c
>> +++ b/fs/btrfs/extent-tree.c
>> @@ -8829,6 +8829,26 @@ again:
>>  		goto again;
>>  	}
>>  
>> +	/*
>> +	 * if we are changing raid levels, try to allocate a corresponding
>> +	 * block group with the new raid level.
>> +	 */
>> +	if (!(cache->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {
> 
> This prevents to switch the system chunk in all cases. What was the
> reason to do it?

I thought the system chunk was being caught by check_system_chunk below,
but no, its using the wrong profile.  It did work when I tested, but I
ran it a few times in a row and got inconsistent results.

> 
> If I remove the check, then the conversions work in all combinations.
> Eg.

Thanks, I'm pushing out an updated patch without the check.

-chris