Re: [PATCH v2 2/3] btrfs: use bio::remaining for async checksumming synchronization

[Qu Wenruo]

在 2026/9/3 15:53, Daniel Vacek 写道:
> We can use bio::remaining counter to sync the offloaded checksuming.
> As a result we can slim down the btrfs_bio structure by 24 bytes
> and simplify the code a bit.
>
> $ pahole | diff
> -	/* size: 328, cachelines: 6, members: 15 */
> +	/* size: 304, cachelines: 5, members: 14 */
>
> Moreover this will allow us enabling async checksumming with encryption
> where we need to checksum the bounce bio instead of our regular one
> embedded in btrfs_bio. And so we need to extend it's lifetime. This is
> the preffered way to do so.
>
> Signed-off-by: Daniel Vacek <neelx@xxxxxxxx>
> Reviewed-by: Qu Wenruo <wqu@xxxxxxxx>
> ---
>
> No change since v1
> ---
>   fs/btrfs/bio.c       | 4 ----
>   fs/btrfs/bio.h       | 4 ----
>   fs/btrfs/file-item.c | 6 ++----
>   3 files changed, 2 insertions(+), 12 deletions(-)
>
> diff --git a/fs/btrfs/bio.c b/fs/btrfs/bio.c
> index 19b4855969f5..771b7d598aee 100644
> --- a/fs/btrfs/bio.c
> +++ b/fs/btrfs/bio.c
> @@ -103,7 +103,6 @@ static struct btrfs_bio *btrfs_split_bio(struct btrfs_fs_info *fs_info,
>   	bbio->can_use_append = orig_bbio->can_use_append;
>   	bbio->is_scrub = orig_bbio->is_scrub;
>   	bbio->is_remap = orig_bbio->is_remap;
> -	bbio->async_csum = orig_bbio->async_csum;
>   
>   	atomic_inc(&orig_bbio->pending_ios);
>   	return bbio;
> @@ -114,9 +113,6 @@ void btrfs_bio_end_io(struct btrfs_bio *bbio, blk_status_t status)
>   	/* Make sure we're already in task context. */
>   	ASSERT(in_task());
>   
> -	if (bbio->async_csum)
> -		wait_for_completion(&bbio->csum_done);
> -
>   	bbio->bio.bi_status = status;
>   	if (bbio->bio.bi_pool == &btrfs_clone_bioset) {
>   		struct btrfs_bio *orig_bbio = bbio->private;
> diff --git a/fs/btrfs/bio.h b/fs/btrfs/bio.h
> index b7bd377a0162..bbf362b8668b 100644
> --- a/fs/btrfs/bio.h
> +++ b/fs/btrfs/bio.h
> @@ -58,7 +58,6 @@ struct btrfs_bio {
>   			struct btrfs_ordered_extent *ordered;
>   			struct btrfs_ordered_sum *sums;
>   			struct work_struct csum_work;
> -			struct completion csum_done;
>   			struct bvec_iter csum_saved_iter;
>   			u64 orig_physical;
>   			u64 orig_logical;
> @@ -93,9 +92,6 @@ struct btrfs_bio {
>   	/* Whether the bio is coming from copy_remapped_data_io(). */
>   	bool is_remap:1;
>   
> -	/* Whether the csum generation for data write is async. */
> -	bool async_csum:1;
> -
>   	/* Whether the bio is written using zone append. */
>   	bool can_use_append:1;
>   
> diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c
> index 4a7681557ec1..0fed4e0d32d5 100644
> --- a/fs/btrfs/file-item.c
> +++ b/fs/btrfs/file-item.c
> @@ -818,9 +818,8 @@ static void csum_one_bio_work(struct work_struct *work)
>   	struct btrfs_bio *bbio = container_of(work, struct btrfs_bio, csum_work);
>   
>   	ASSERT(btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE);
> -	ASSERT(bbio->async_csum == true);
>   	csum_one_bio(bbio);
> -	complete(&bbio->csum_done);
> +	bio_endio(&bbio->bio);
>   }
>   
>   /*
> @@ -854,8 +853,7 @@ int btrfs_csum_one_bio(struct btrfs_bio *bbio, bool async)
>   		csum_one_bio(bbio);
>   		return 0;
>   	}
> -	init_completion(&bbio->csum_done);
> -	bbio->async_csum = true;
> +	bio_inc_remaining(bio);

There seems to be a window where the bio can be finished before submission.

Firstly at btrfs_csum_one_bio() time, bio->bi_endio is not yet 
initialized, it's only properly assigned at btrfs_submit_bio().

Then we queue the csum generation work.

But by some bad timing, the bio submission is delayed, we can have the 
following sequence:

        Submission               |       Csum generation
---------------------------------+------------------------------------
btrfs_submit_chunk()             |
|- btrfs_csum_one_bio()          |
|  |- bio_inc_remaining()        |
|  |- schedule_work()            | csum_one_bio_work()
|                                | |- bio_endio()
|                                | Now the bio is finished, although
|                                | bi_end_io is NULL, nothing real
|                                | happened.
|- btrfs_submit_bio()


This can be even worse, if the bio_endio() is called when 
btrfs_submit_bio() has only partially setup the bio (e.g. bi_end_io() is 
set, but bi_bdev is not set)

Not mention now it changed the context where the endio function is called.

Previously csum_one_bio_work() will never call bi_end_io() function, but 
now it can.

The change has a much larger impact than I initially thought.
>   	INIT_WORK(&bbio->csum_work, csum_one_bio_work);
>   	schedule_work(&bbio->csum_work);
>   	return 0;