[v2,0/2] btrfs: zoned: activate new BG only from extent allocation context

Message

Naohiro Aota March 22, 2022, 9:11 a.m. UTC

In btrfs_make_block_group(), we activate the allocated block group,
expecting that the block group is soon used for the extent allocation.

With a lot of IOs, btrfs_async_reclaim_data_space() tries to prepare
for them by pre-allocating data block groups. That preallocation can
consume all the active zone counting. It is OK if they are soon
written and filled. However, that's not the case. As a result, an
allocation of non-data block groups fails due to the lack of an active
zone resource.

This series fixes the issue by activating a new block group only when
it's called from find_free_extent(). This series introduces
CHUNK_ALLOC_FORCE_FOR_EXTENT in btrfs_chunk_alloc_enum to distinguish
the context.

--
Changes
- v2
  - Fix a flipped condition

Naohiro Aota (2):
  btrfs: return allocated block group from do_chunk_alloc()
  btrfs: zoned: activate block group only for extent allocation

 fs/btrfs/block-group.c | 36 +++++++++++++++++++++++++++---------
 fs/btrfs/block-group.h |  1 +
 fs/btrfs/extent-tree.c |  2 +-
 3 files changed, 29 insertions(+), 10 deletions(-)

Comments

Johannes Thumshirn March 22, 2022, 4:03 p.m. UTC | #1

On 22/03/2022 10:11, Naohiro Aota wrote:
> In btrfs_make_block_group(), we activate the allocated block group,
> expecting that the block group is soon used for the extent allocation.
> 
> With a lot of IOs, btrfs_async_reclaim_data_space() tries to prepare
> for them by pre-allocating data block groups. That preallocation can
> consume all the active zone counting. It is OK if they are soon
> written and filled. However, that's not the case. As a result, an
> allocation of non-data block groups fails due to the lack of an active
> zone resource.
> 
> This series fixes the issue by activating a new block group only when
> it's called from find_free_extent(). This series introduces
> CHUNK_ALLOC_FORCE_FOR_EXTENT in btrfs_chunk_alloc_enum to distinguish
> the context.
> 
> --
> Changes
> - v2
>   - Fix a flipped condition
> 
> Naohiro Aota (2):
>   btrfs: return allocated block group from do_chunk_alloc()
>   btrfs: zoned: activate block group only for extent allocation
> 
>  fs/btrfs/block-group.c | 36 +++++++++++++++++++++++++++---------
>  fs/btrfs/block-group.h |  1 +
>  fs/btrfs/extent-tree.c |  2 +-
>  3 files changed, 29 insertions(+), 10 deletions(-)
> 


For the series,
Reviewed-by: Johannes Thumshirn <[e-mail]>
Tested-by: Johannes Thumshirn <[e-mail]>

David Sterba April 1, 2022, 2:22 p.m. UTC | #2

On Tue, Mar 22, 2022 at 06:11:32PM +0900, Naohiro Aota wrote:
> In btrfs_make_block_group(), we activate the allocated block group,
> expecting that the block group is soon used for the extent allocation.
> 
> With a lot of IOs, btrfs_async_reclaim_data_space() tries to prepare
> for them by pre-allocating data block groups. That preallocation can
> consume all the active zone counting. It is OK if they are soon
> written and filled. However, that's not the case. As a result, an
> allocation of non-data block groups fails due to the lack of an active
> zone resource.
> 
> This series fixes the issue by activating a new block group only when
> it's called from find_free_extent(). This series introduces
> CHUNK_ALLOC_FORCE_FOR_EXTENT in btrfs_chunk_alloc_enum to distinguish
> the context.
> 
> --
> Changes
> - v2
>   - Fix a flipped condition
> 
> Naohiro Aota (2):
>   btrfs: return allocated block group from do_chunk_alloc()
>   btrfs: zoned: activate block group only for extent allocation

Added to misc-next, thanks.