On Fri, Oct 02, 2020 at 03:36:32AM +0900, Naohiro Aota wrote:
> This commit enables zone append writing for zoned btrfs. Three parts are
> necessary to enable it. First, it modifies bio to use REQ_OP_ZONE_APPEND in
> btrfs_submit_bio_hook() and adjust the bi_sector to point the beginning of
> the zone.
>
> Second, it records returned physical address (and disk/partno) to
> the ordered extent in end_bio_extent_writepage().
That sounds fishy ...
> Finally, it rewrites logical addresses of the extent mapping and checksum
> data according to the physical address (using __btrfs_rmap_block). If the
> returned address match to the originaly allocated address, we can skip the
> rewriting process.
>
> [Johannes] fixed bvec handling
> Signed-off-by: Naohiro Aota <naohiro.aota@xxxxxxx>
> --- a/fs/btrfs/ordered-data.h
> +++ b/fs/btrfs/ordered-data.h
> @@ -127,6 +127,10 @@ struct btrfs_ordered_extent {
> struct completion completion;
> struct btrfs_work flush_work;
> struct list_head work_list;
> +
> + u64 physical;
> + struct gendisk *disk;
> + u8 partno;
btrfs_ordered_extent is on the logical layer, disk/partno is physical.
This needs proper explanation why it has to be done that way and that
there's not really anything better than layering violation. Why
__btrfs_rmap_block takes bdev is unexplained.
