]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
btrfs: scrub: use scrub_simple_mirror() to handle RAID56 data stripe scrub
authorQu Wenruo <wqu@suse.com>
Fri, 11 Mar 2022 07:38:46 +0000 (15:38 +0800)
committerDavid Sterba <dsterba@suse.com>
Mon, 16 May 2022 15:17:30 +0000 (17:17 +0200)
Although RAID56 has complex repair mechanism, which involves reading the
whole full stripe, but inside one data stripe, it's in fact no different
than SINGLE/RAID1.

The point here is, for data stripe we just check the csum for each
extent we hit.  Only for csum mismatch case, our repair paths divide.

So we can still reuse scrub_simple_mirror() for RAID56 data stripes,
which saves quite some code.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/scrub.c

index cc2991ade2bd078069a4a7bdf700f039925079df..9efe7af1290deeee96521c709403e0d961070170 100644 (file)
@@ -3494,33 +3494,20 @@ static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
        struct btrfs_fs_info *fs_info = sctx->fs_info;
        struct btrfs_root *root;
        struct btrfs_root *csum_root;
-       struct btrfs_extent_item *extent;
        struct blk_plug plug;
        const u64 profile = map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
        const u64 chunk_logical = bg->start;
-       u64 flags;
        int ret;
-       int slot;
-       struct extent_buffer *l;
        u64 physical = map->stripes[stripe_index].physical;
+       const u64 physical_end = physical + dev_extent_len;
        u64 logical;
        u64 logic_end;
-       const u64 physical_end = physical + dev_extent_len;
-       u64 generation;
-       struct btrfs_key key;
+       /* The logical increment after finishing one stripe */
        u64 increment;
+       /* Offset inside the chunk */
        u64 offset;
-       u64 extent_logical;
-       u64 extent_physical;
-       /*
-        * Unlike chunk length, extent length should never go beyond
-        * BTRFS_MAX_EXTENT_SIZE, thus u32 is enough here.
-        */
-       u32 extent_len;
        u64 stripe_logical;
        u64 stripe_end;
-       struct btrfs_device *extent_dev;
-       int extent_mirror_num;
        int stop_loop = 0;
 
        path = btrfs_alloc_path();
@@ -3601,34 +3588,13 @@ static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
        get_raid56_logic_offset(physical, stripe_index, map, &offset, NULL);
        increment = map->stripe_len * nr_data_stripes(map);
 
+       /*
+        * Due to the rotation, for RAID56 it's better to iterate each stripe
+        * using their physical offset.
+        */
        while (physical < physical_end) {
-               /*
-                * canceled?
-                */
-               if (atomic_read(&fs_info->scrub_cancel_req) ||
-                   atomic_read(&sctx->cancel_req)) {
-                       ret = -ECANCELED;
-                       goto out;
-               }
-               /*
-                * check to see if we have to pause
-                */
-               if (atomic_read(&fs_info->scrub_pause_req)) {
-                       /* push queued extents */
-                       sctx->flush_all_writes = true;
-                       scrub_submit(sctx);
-                       mutex_lock(&sctx->wr_lock);
-                       scrub_wr_submit(sctx);
-                       mutex_unlock(&sctx->wr_lock);
-                       wait_event(sctx->list_wait,
-                                  atomic_read(&sctx->bios_in_flight) == 0);
-                       sctx->flush_all_writes = false;
-                       scrub_blocked_if_needed(fs_info);
-               }
-
-               ret = get_raid56_logic_offset(physical, stripe_index,
-                                             map, &logical,
-                                             &stripe_logical);
+               ret = get_raid56_logic_offset(physical, stripe_index, map,
+                                             &logical, &stripe_logical);
                logical += chunk_logical;
                if (ret) {
                        /* it is parity strip */
@@ -3639,194 +3605,23 @@ static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
                                                  stripe_end);
                        if (ret)
                                goto out;
-                       goto skip;
+                       goto next;
                }
 
-               if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
-                       key.type = BTRFS_METADATA_ITEM_KEY;
-               else
-                       key.type = BTRFS_EXTENT_ITEM_KEY;
-               key.objectid = logical;
-               key.offset = (u64)-1;
-
-               ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+               /*
+                * Now we're at a data stripe, scrub each extents in the range.
+                *
+                * At this stage, if we ignore the repair part, inside each data
+                * stripe it is no different than SINGLE profile.
+                * We can reuse scrub_simple_mirror() here, as the repair part
+                * is still based on @mirror_num.
+                */
+               ret = scrub_simple_mirror(sctx, root, csum_root, bg, map,
+                                         logical, map->stripe_len,
+                                         scrub_dev, physical, 1);
                if (ret < 0)
                        goto out;
-
-               if (ret > 0) {
-                       ret = btrfs_previous_extent_item(root, path, 0);
-                       if (ret < 0)
-                               goto out;
-                       if (ret > 0) {
-                               /* there's no smaller item, so stick with the
-                                * larger one */
-                               btrfs_release_path(path);
-                               ret = btrfs_search_slot(NULL, root, &key,
-                                                       path, 0, 0);
-                               if (ret < 0)
-                                       goto out;
-                       }
-               }
-
-               stop_loop = 0;
-               while (1) {
-                       u64 bytes;
-
-                       l = path->nodes[0];
-                       slot = path->slots[0];
-                       if (slot >= btrfs_header_nritems(l)) {
-                               ret = btrfs_next_leaf(root, path);
-                               if (ret == 0)
-                                       continue;
-                               if (ret < 0)
-                                       goto out;
-
-                               stop_loop = 1;
-                               break;
-                       }
-                       btrfs_item_key_to_cpu(l, &key, slot);
-
-                       if (key.type != BTRFS_EXTENT_ITEM_KEY &&
-                           key.type != BTRFS_METADATA_ITEM_KEY)
-                               goto next;
-
-                       if (key.type == BTRFS_METADATA_ITEM_KEY)
-                               bytes = fs_info->nodesize;
-                       else
-                               bytes = key.offset;
-
-                       if (key.objectid + bytes <= logical)
-                               goto next;
-
-                       if (key.objectid >= logical + map->stripe_len) {
-                               /* out of this device extent */
-                               if (key.objectid >= logic_end)
-                                       stop_loop = 1;
-                               break;
-                       }
-
-                       /*
-                        * If our block group was removed in the meanwhile, just
-                        * stop scrubbing since there is no point in continuing.
-                        * Continuing would prevent reusing its device extents
-                        * for new block groups for a long time.
-                        */
-                       spin_lock(&bg->lock);
-                       if (bg->removed) {
-                               spin_unlock(&bg->lock);
-                               ret = 0;
-                               goto out;
-                       }
-                       spin_unlock(&bg->lock);
-
-                       extent = btrfs_item_ptr(l, slot,
-                                               struct btrfs_extent_item);
-                       flags = btrfs_extent_flags(l, extent);
-                       generation = btrfs_extent_generation(l, extent);
-
-                       if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
-                           (key.objectid < logical ||
-                            key.objectid + bytes >
-                            logical + map->stripe_len)) {
-                               btrfs_err(fs_info,
-                                          "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
-                                      key.objectid, logical);
-                               spin_lock(&sctx->stat_lock);
-                               sctx->stat.uncorrectable_errors++;
-                               spin_unlock(&sctx->stat_lock);
-                               goto next;
-                       }
-
-again:
-                       extent_logical = key.objectid;
-                       ASSERT(bytes <= U32_MAX);
-                       extent_len = bytes;
-
-                       /*
-                        * trim extent to this stripe
-                        */
-                       if (extent_logical < logical) {
-                               extent_len -= logical - extent_logical;
-                               extent_logical = logical;
-                       }
-                       if (extent_logical + extent_len >
-                           logical + map->stripe_len) {
-                               extent_len = logical + map->stripe_len -
-                                            extent_logical;
-                       }
-
-                       extent_physical = extent_logical - logical + physical;
-                       extent_dev = scrub_dev;
-                       /* For RAID56 data stripes, mirror_num is fixed to 1 */
-                       extent_mirror_num = 1;
-                       if (sctx->is_dev_replace)
-                               scrub_remap_extent(fs_info, extent_logical,
-                                                  extent_len, &extent_physical,
-                                                  &extent_dev,
-                                                  &extent_mirror_num);
-
-                       if (flags & BTRFS_EXTENT_FLAG_DATA) {
-                               ret = btrfs_lookup_csums_range(csum_root,
-                                               extent_logical,
-                                               extent_logical + extent_len - 1,
-                                               &sctx->csum_list, 1);
-                               if (ret)
-                                       goto out;
-                       }
-
-                       ret = scrub_extent(sctx, map, extent_logical, extent_len,
-                                          extent_physical, extent_dev, flags,
-                                          generation, extent_mirror_num,
-                                          extent_logical - logical + physical);
-
-                       scrub_free_csums(sctx);
-
-                       if (ret)
-                               goto out;
-
-                       if (sctx->is_dev_replace)
-                               sync_replace_for_zoned(sctx);
-
-                       if (extent_logical + extent_len <
-                           key.objectid + bytes) {
-                               /*
-                                * loop until we find next data stripe
-                                * or we have finished all stripes.
-                                */
-loop:
-                               physical += map->stripe_len;
-                               ret = get_raid56_logic_offset(physical,
-                                               stripe_index, map,
-                                               &logical, &stripe_logical);
-                               logical += chunk_logical;
-
-                               if (ret && physical < physical_end) {
-                                       stripe_logical += chunk_logical;
-                                       stripe_end = stripe_logical +
-                                                       increment;
-                                       ret = scrub_raid56_parity(sctx,
-                                               map, scrub_dev,
-                                               stripe_logical,
-                                               stripe_end);
-                                       if (ret)
-                                               goto out;
-                                       goto loop;
-                               }
-                               if (logical < key.objectid + bytes) {
-                                       cond_resched();
-                                       goto again;
-                               }
-
-                               if (physical >= physical_end) {
-                                       stop_loop = 1;
-                                       break;
-                               }
-                       }
 next:
-                       path->slots[0]++;
-               }
-               btrfs_release_path(path);
-skip:
                logical += increment;
                physical += map->stripe_len;
                spin_lock(&sctx->stat_lock);