Re: [RFC] Removing b_end_io

[Theodore Y. Ts'o]

On Sun, Oct 25, 2020 at 04:44:38AM +0000, Matthew Wilcox wrote:
> @@ -3068,6 +3069,12 @@ static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
>  	}
>  
>  	submit_bio(bio);
> +}
> +
> +static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
> +			 enum rw_hint write_hint, struct writeback_control *wbc)
> +{
> +	__bh_submit(bh, op | op_flags, write_hint, wbc, end_bio_bh_io_sync);
>  	return 0;
>  }
>

I believe this will break use cases where the file system sets
bh->b_end_io and then calls submit_bh(), which then calls
submit_bh_wbc().  That's because with this change, calls to
submit_bh_wbc() --- include submit_bh() --- ignores bh->b_end_io and
results in end_bio_bh_io_sync getting used.

Filesystems that do this includes fs/ntfs, fs/resiserfs.

In this case, that can probably be fixed by changing submit_bh() to
pass in bh->b_end_io, or switching those users to use the new
bh_submit() function to prevent these breakages.

						- Ted