[PATCH v2] btrfs: zoned: don't wait for ordered extents in do_zone_finish()

[PATCH v2] btrfs: zoned: don't wait for ordered extents in do_zone_finish()

[Johannes Thumshirn]

In general, we want to ensure all writes in the block group to
complete before finishing the zone, as relocation does. However, the
current callers do not need the wait (or cannot wait) in do_zone_finish():

- Relocation already waits for reservations and ordered extents itself
  before calling btrfs_zone_finish().
- btrfs_delete_unused_bgs() checks that ->reserved is zero and sets the
  block group read-only while holding groups_sem, so there can be no
  in-flight data write.
- btrfs_zone_finish_one_bg() is called from the allocation path and
  cannot wait.

Dropping the wait is still safe. Once the block group is read-only,
->reserved can only drop, and for data it only drops after the ordered
extent's I/O has completed. So if ->reserved is zero, no data write is
left in flight. If it is non-zero, do_zone_finish() returns -EAGAIN,
which all callers handle.

This makes btrfs_zone_finish_one_bg() fail with -EAGAIN when it races
with an allocation in the selected block group.

Fixes: 393f646e34c1 ("btrfs: zoned: finish least available block group on data bg allocation")
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
Changes to v1:
- Reword changelog to undermine why this is safe (Naohiro) 

 fs/btrfs/zoned.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
index 1fa855a0c31a..35b422a53e2a 100644
--- a/fs/btrfs/zoned.c
+++ b/fs/btrfs/zoned.c
@@ -2604,9 +2604,10 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ
 
 		/* Ensure all writes in this block group finish */
 		btrfs_wait_block_group_reservations(block_group);
-		/* No need to wait for NOCOW writers. Zoned mode does not allow that */
-		btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group);
-		/* Wait for extent buffers to be written. */
+		/*
+		 * No need to wait for NOCOW writers, as zoned mode does not
+		 * allow that, only wait for extent buffers to be written.
+		 */
 		if (is_metadata)
 			wait_eb_writebacks(block_group);
 
-- 
2.56.0

Re: [PATCH v2] btrfs: zoned: don't wait for ordered extents in do_zone_finish()

[Naohiro Aota]

On Tue Oct 27, 2026 at 11:33 PM JST, Johannes Thumshirn wrote:
> In general, we want to ensure all writes in the block group to
> complete before finishing the zone, as relocation does. However, the
> current callers do not need the wait (or cannot wait) in do_zone_finish():
>
> - Relocation already waits for reservations and ordered extents itself
>   before calling btrfs_zone_finish().
> - btrfs_delete_unused_bgs() checks that ->reserved is zero and sets the
>   block group read-only while holding groups_sem, so there can be no
>   in-flight data write.
> - btrfs_zone_finish_one_bg() is called from the allocation path and
>   cannot wait.
>
> Dropping the wait is still safe. Once the block group is read-only,
> ->reserved can only drop, and for data it only drops after the ordered
> extent's I/O has completed. So if ->reserved is zero, no data write is
> left in flight. If it is non-zero, do_zone_finish() returns -EAGAIN,
> which all callers handle.
>
> This makes btrfs_zone_finish_one_bg() fail with -EAGAIN when it races
> with an allocation in the selected block group.

Oops, I meant to keep the pragraph mentioning the deadlock and the call
chain. I think we can get that part back on the commit.

Reviewed-by: Naohiro Aota <naohiro.aota@wdc.com>

>
> Fixes: 393f646e34c1 ("btrfs: zoned: finish least available block group on data bg allocation")
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
> Changes to v1:
> - Reword changelog to undermine why this is safe (Naohiro) 
>
>  fs/btrfs/zoned.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
> index 1fa855a0c31a..35b422a53e2a 100644
> --- a/fs/btrfs/zoned.c
> +++ b/fs/btrfs/zoned.c
> @@ -2604,9 +2604,10 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ
>  
>  		/* Ensure all writes in this block group finish */
>  		btrfs_wait_block_group_reservations(block_group);
> -		/* No need to wait for NOCOW writers. Zoned mode does not allow that */
> -		btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group);
> -		/* Wait for extent buffers to be written. */
> +		/*
> +		 * No need to wait for NOCOW writers, as zoned mode does not
> +		 * allow that, only wait for extent buffers to be written.
> +		 */
>  		if (is_metadata)
>  			wait_eb_writebacks(block_group);
>  

Re: [PATCH v2] btrfs: zoned: don't wait for ordered extents in do_zone_finish()

[Johannes Thumshirn]

On 10/27/26 4:18 PM, Naohiro Aota wrote:
> On Tue Oct 27, 2026 at 11:33 PM JST, Johannes Thumshirn wrote:
>> In general, we want to ensure all writes in the block group to
>> complete before finishing the zone, as relocation does. However, the
>> current callers do not need the wait (or cannot wait) in do_zone_finish():
>>
>> - Relocation already waits for reservations and ordered extents itself
>>    before calling btrfs_zone_finish().
>> - btrfs_delete_unused_bgs() checks that ->reserved is zero and sets the
>>    block group read-only while holding groups_sem, so there can be no
>>    in-flight data write.
>> - btrfs_zone_finish_one_bg() is called from the allocation path and
>>    cannot wait.
>>
>> Dropping the wait is still safe. Once the block group is read-only,
>> ->reserved can only drop, and for data it only drops after the ordered
>> extent's I/O has completed. So if ->reserved is zero, no data write is
>> left in flight. If it is non-zero, do_zone_finish() returns -EAGAIN,
>> which all callers handle.
>>
>> This makes btrfs_zone_finish_one_bg() fail with -EAGAIN when it races
>> with an allocation in the selected block group.
> Oops, I meant to keep the pragraph mentioning the deadlock and the call
> chain. I think we can get that part back on the commit.


OK will do that when committing.

>
> Reviewed-by: Naohiro Aota <naohiro.aota@wdc.com>
Thanks