On Sat, 2009-07-11 at 15:39 +0100, David Woodhouse wrote: > This is a preliminary attempt to add RAID5 and RAID6 support. > > So far it doesn't attempt to write or read the parity blocks -- it > just > lays the data blocks out as we want them, so it's effectively just a > complex and wasteful kind of RAID0. > > The next step is to make btrfs_map_bio() do the right thing: > - Satisfy read requests for mirrors #2 and #3 by recreating data from > RAID5 parity or RAID6 error correction stripe respectively. > - Write out parity and RAID6 blocks appropriately when data writes > happen. Actually, the next step is to tweak __btrfs_map_block() a bit more to let it return information about the whole stripe-set, so that btrfs_map_bio() _can_ do what we say above... So rather than just mapping the requested address as if it's RAID0, we (where appropriate) return information about the _entire_ disk set in the btrfs_multi_bio, with an auxiliary array giving the _logical_ offset corresponding to each physical stripe in the referenced set (with special values for the P and Q stripes). We do this for all writes, and for reads where mirror_num > 1 (i.e. when we're being asked to rebuild it from parity, rather than reading the original data blocks). git://, http://git.infradead.org/users/dwmw2/btrfs-raid56.git commit ed90c58ba7c60555af4b8c00a104c7d71f6db6d2 Author: David Woodhouse <David.Woodhouse@xxxxxxxxx> Date: Sun Jul 12 11:15:22 2009 +0100 Btrfs: Let btrfs_map_block() return full stripe information for RAID[56] ... in the cases where it's necessary -- which is for a write, or for a parity recovery attempt. We'll let btrfs_map_bio() do the rest. Signed-off-by: David Woodhouse <David.Woodhouse@xxxxxxxxx> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 3b231ef..55facd3 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -62,6 +62,11 @@ static int init_first_rw_device(struct btrfs_trans_handle *trans, struct btrfs_device *device); static int btrfs_relocate_sys_chunks(struct btrfs_root *root); +#define RAID5_P_STRIPE ((u64)-1) +#define RAID6_Q_STRIPE ((u64)-2) + +#define is_parity_stripe(x) ( ((x) == RAID5_P_STRIPE) || ((x) == RAID6_Q_STRIPE) ) + #define map_lookup_size(n) (sizeof(struct map_lookup) + \ (sizeof(struct btrfs_bio_stripe) * (n))) @@ -2614,7 +2619,8 @@ static int find_live_mirror(struct map_lookup *map, int first, int num, static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, u64 logical, u64 *length, struct btrfs_multi_bio **multi_ret, - int mirror_num, struct page *unplug_page) + int mirror_num, struct page *unplug_page, + u64 **raid_map_ret) { struct extent_map *em; struct map_lookup *map; @@ -2622,6 +2628,7 @@ static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, u64 offset; u64 stripe_offset; u64 stripe_nr; + u64 *raid_map = NULL; int stripes_allocated = 8; int stripes_required = 1; int stripe_index; @@ -2674,9 +2681,24 @@ again: max_errors = 1; } } - if (multi_ret && (rw & (1 << BIO_RW)) && - stripes_allocated < stripes_required) { - stripes_allocated = map->num_stripes; + if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) + && multi_ret && (rw & (1 << BIO_RW) || mirror_num > 1) && raid_map_ret) { + /* RAID[56] write or recovery. Return all stripes */ + stripes_required = map->num_stripes; + max_errors = nr_parity_stripes(map); + + /* Only allocate the map if we've already got a large enough multi_ret */ + if (stripes_allocated >= stripes_required) { + raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); + if (!raid_map) { + free_extent_map(em); + kfree(multi); + return -ENOMEM; + } + } + } + if (multi_ret && stripes_allocated < stripes_required) { + stripes_allocated = stripes_required; free_extent_map(em); kfree(multi); goto again; @@ -2749,18 +2771,43 @@ again: stripe_index = do_div(stripe_nr, nr_data_stripes(map)); - /* - * Mirror #0 or #1 means the original data block. - * Mirror #2 is RAID5 parity block. - * Mirror #3 is RAID6 Q block. - */ - if (mirror_num > 1) - stripe_index = nr_data_stripes(map) + mirror_num - 2; - - /* We distribute the parity blocks across stripes */ - tmp = stripe_nr + stripe_index; - stripe_index = do_div(tmp, map->num_stripes); - + if (unplug_page) { + stripe_index = 0; + num_stripes = map->num_stripes; + } else if (raid_map) { + int i, rot; + + /* Work out the disk rotation on this stripe-set */ + tmp = stripe_nr; + rot = do_div(tmp, map->num_stripes); + + /* Fill in the logical address of each stripe */ + tmp = stripe_nr * nr_data_stripes(map); + for (i = 0; i < nr_data_stripes(map); i++) + raid_map[(i+rot) % map->num_stripes] = + em->start + (tmp + i) * map->stripe_len; + + raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; + if (map->type & BTRFS_BLOCK_GROUP_RAID6) + raid_map[(i+rot+1) % map->num_stripes] = RAID6_Q_STRIPE; + + *length = map->stripe_len; + stripe_index = 0; + stripe_offset = 0; + num_stripes = map->num_stripes; + } else { + /* + * Mirror #0 or #1 means the original data block. + * Mirror #2 is RAID5 parity block. + * Mirror #3 is RAID6 Q block. + */ + if (mirror_num > 1) + stripe_index = nr_data_stripes(map) + mirror_num - 2; + + /* We distribute the parity blocks across stripes */ + tmp = stripe_nr + stripe_index; + stripe_index = do_div(tmp, map->num_stripes); + } } else { /* * after this do_div call, stripe_nr is the number of stripes @@ -2795,6 +2842,8 @@ again: multi->num_stripes = num_stripes; multi->max_errors = max_errors; } + if (raid_map_ret) + *raid_map_ret = raid_map; out: free_extent_map(em); return 0; @@ -2805,7 +2854,7 @@ int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, struct btrfs_multi_bio **multi_ret, int mirror_num) { return __btrfs_map_block(map_tree, rw, logical, length, multi_ret, - mirror_num, NULL); + mirror_num, NULL, NULL); } int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, @@ -2889,7 +2938,7 @@ int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree, { u64 length = PAGE_CACHE_SIZE; return __btrfs_map_block(map_tree, READ, logical, &length, - NULL, 0, page); + NULL, 0, page, NULL); } static void end_bio_multi_stripe(struct bio *bio, int err) -- David Woodhouse Open Source Technology Centre David.Woodhouse@xxxxxxxxx Intel Corporation -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
