| Message ID | 0d54adfb390b3c9ffd8ff8f98db43de51df9c33e.1688368617.git.wqu@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | btrfs: scrub: introduce SCRUB_LOGICAL flag | expand |
On 2023/7/3 15:32, Qu Wenruo wrote: > Currently there is no way to read P/Q stripes of a RAID56 full stripe > directly. > > Normally caller should call btrfs_map_block() directly and fetch the > physical location directly of the P/Q stripes. > > But for the recent scrub rework, it's no longer that elegant as the full > scrub code is based on mirror_num based solution, thus this patch would > add two special mirror_num for this usages: > > - Mirror -1 for P stripes > - Mirror -2 for Q stripes > > Both mirrors only support read for now, and caller should make sure the > range start is stripe aligned. > > Signed-off-by: Qu Wenruo <wqu@suse.com> > --- > fs/btrfs/scrub.c | 2 +- > fs/btrfs/volumes.c | 30 +++++++++++++++++++++++++++++- > 2 files changed, 30 insertions(+), 2 deletions(-) > > diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c > index 1864856abb13..41c514db0793 100644 > --- a/fs/btrfs/scrub.c > +++ b/fs/btrfs/scrub.c > @@ -1601,7 +1601,7 @@ static void scrub_submit_initial_read(struct scrub_ctx *sctx, > int mirror = stripe->mirror_num; > > ASSERT(stripe->bg); > - ASSERT(stripe->mirror_num > 0); > + ASSERT(stripe->mirror_num); > ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state)); > > bbio = btrfs_bio_alloc(SCRUB_STRIPE_PAGES, REQ_OP_READ, fs_info, > diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c > index cd632b3f579f..c22007bd830b 100644 > --- a/fs/btrfs/volumes.c > +++ b/fs/btrfs/volumes.c > @@ -6251,10 +6251,12 @@ static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup * > * > * For RAID56 profile, mirror 1 means rebuild from P and > * the remaining data stripes. > + * And mirror -1 means read P stripes directly, -2 means > + * read Q stripes directly. > * > * For RAID6 profile, mirror > 2 means mark another > * data/P stripe error and rebuild from the remaining > - * stripes.. > + * stripes. > * > * @need_raid_map: (Used only for integrity checker) whether the map wants > * a full stripe map (including all data and P/Q stripes) > @@ -6297,6 +6299,12 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, > goto out_free_em; > } > > + /* Only allow mirror_num < 0 for RAID56. */ > + if (mirror_num < 0 && !(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) { > + free_extent_map(em); > + return -EINVAL; > + } > + > data_stripes = nr_data_stripes(map); > > map_offset = logical - em->start; > @@ -6382,6 +6390,26 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, > logical; > stripe_index = 0; > stripe_offset = 0; > + } else if (mirror_num < 0) { > + /* Only allow READ for direct P/Q operations. */ > + ASSERT(op == BTRFS_MAP_READ); > + /* > + * Caller should make sure the range start is stripe > + * aligned. > + */ > + ASSERT(stripe_offset == 0); > + > + /* > + * Stripes of @bioc is already sorted, stripes[0] is the > + * first data stripe and stripes[@data_stripes] is the > + * P stripe. > + * So we only need to update the @stripe_index to the > + * specified stripe, and set @stripe_nr/@stripe_offset > + * to 0, so we can return the beginning of the P/Q stripe. > + */ > + stripe_offset = 0; > + stripe_nr = 0; Oh, the stripe_nr calculation is wrong, it still needs to take full stripe number into consideration. Would update this patch in my branch. Thanks, Qu > + stripe_index = data_stripes + (-mirror_num - 1); > } else { > /* > * Mirror #0 or #1 means the original data block.
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c index 1864856abb13..41c514db0793 100644 --- a/fs/btrfs/scrub.c +++ b/fs/btrfs/scrub.c @@ -1601,7 +1601,7 @@ static void scrub_submit_initial_read(struct scrub_ctx *sctx, int mirror = stripe->mirror_num; ASSERT(stripe->bg); - ASSERT(stripe->mirror_num > 0); + ASSERT(stripe->mirror_num); ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state)); bbio = btrfs_bio_alloc(SCRUB_STRIPE_PAGES, REQ_OP_READ, fs_info, diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index cd632b3f579f..c22007bd830b 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -6251,10 +6251,12 @@ static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup * * * For RAID56 profile, mirror 1 means rebuild from P and * the remaining data stripes. + * And mirror -1 means read P stripes directly, -2 means + * read Q stripes directly. * * For RAID6 profile, mirror > 2 means mark another * data/P stripe error and rebuild from the remaining - * stripes.. + * stripes. * * @need_raid_map: (Used only for integrity checker) whether the map wants * a full stripe map (including all data and P/Q stripes) @@ -6297,6 +6299,12 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, goto out_free_em; } + /* Only allow mirror_num < 0 for RAID56. */ + if (mirror_num < 0 && !(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) { + free_extent_map(em); + return -EINVAL; + } + data_stripes = nr_data_stripes(map); map_offset = logical - em->start; @@ -6382,6 +6390,26 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, logical; stripe_index = 0; stripe_offset = 0; + } else if (mirror_num < 0) { + /* Only allow READ for direct P/Q operations. */ + ASSERT(op == BTRFS_MAP_READ); + /* + * Caller should make sure the range start is stripe + * aligned. + */ + ASSERT(stripe_offset == 0); + + /* + * Stripes of @bioc is already sorted, stripes[0] is the + * first data stripe and stripes[@data_stripes] is the + * P stripe. + * So we only need to update the @stripe_index to the + * specified stripe, and set @stripe_nr/@stripe_offset + * to 0, so we can return the beginning of the P/Q stripe. + */ + stripe_offset = 0; + stripe_nr = 0; + stripe_index = data_stripes + (-mirror_num - 1); } else { /* * Mirror #0 or #1 means the original data block.
Currently there is no way to read P/Q stripes of a RAID56 full stripe directly. Normally caller should call btrfs_map_block() directly and fetch the physical location directly of the P/Q stripes. But for the recent scrub rework, it's no longer that elegant as the full scrub code is based on mirror_num based solution, thus this patch would add two special mirror_num for this usages: - Mirror -1 for P stripes - Mirror -2 for Q stripes Both mirrors only support read for now, and caller should make sure the range start is stripe aligned. Signed-off-by: Qu Wenruo <wqu@suse.com> --- fs/btrfs/scrub.c | 2 +- fs/btrfs/volumes.c | 30 +++++++++++++++++++++++++++++- 2 files changed, 30 insertions(+), 2 deletions(-)