Unlike mirror based profiles, RAID5/6 recovery needs to read out the whole full stripe.
And if we don't do proper protect, it can easily cause race condition. Introduce 2 new functions: lock_full_stripe() and unlock_full_stripe() for RAID5/6. Which stores a rb_tree of mutex for full stripes, so scrub callers can use them to lock a full stripe to avoid race. Signed-off-by: Qu Wenruo <quwen...@cn.fujitsu.com> --- fs/btrfs/ctree.h | 4 ++ fs/btrfs/extent-tree.c | 3 + fs/btrfs/scrub.c | 177 +++++++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 184 insertions(+) diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 50bcfb8..5896059 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -639,6 +639,10 @@ struct btrfs_block_group_cache { * Protected by free_space_lock. */ int needs_free_space; + + /* Scrub full stripe lock tree for RAID5/6 scrub */ + struct rb_root scrub_lock_root; + spinlock_t scrub_lock; }; /* delayed seq elem */ diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index e97302f..3e61604 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -130,6 +130,7 @@ void btrfs_put_block_group(struct btrfs_block_group_cache *cache) if (atomic_dec_and_test(&cache->count)) { WARN_ON(cache->pinned > 0); WARN_ON(cache->reserved > 0); + WARN_ON(!RB_EMPTY_ROOT(&cache->scrub_lock_root)); kfree(cache->free_space_ctl); kfree(cache); } @@ -9906,6 +9907,8 @@ btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, atomic_set(&cache->count, 1); spin_lock_init(&cache->lock); + spin_lock_init(&cache->scrub_lock); + cache->scrub_lock_root = RB_ROOT; init_rwsem(&cache->data_rwsem); INIT_LIST_HEAD(&cache->list); INIT_LIST_HEAD(&cache->cluster_list); diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c index 0d63d99..2b9555c 100644 --- a/fs/btrfs/scrub.c +++ b/fs/btrfs/scrub.c @@ -240,6 +240,13 @@ struct scrub_warning { struct btrfs_device *dev; }; +struct scrub_full_stripe_lock { + struct rb_node node; + u64 logical; + u64 refs; + struct mutex mutex; +}; + static void scrub_pending_bio_inc(struct scrub_ctx *sctx); static void scrub_pending_bio_dec(struct scrub_ctx *sctx); static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx); @@ -351,6 +358,176 @@ static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) } /* + * Caller must hold cache->scrub_root_lock. + * + * Return existing full stripe and increase it refs + * Or return NULL, and insert @fstripe_lock into the bg cache + */ +static struct scrub_full_stripe_lock * +add_scrub_lock(struct btrfs_block_group_cache *cache, + struct scrub_full_stripe_lock *fstripe_lock) +{ + struct rb_node **p; + struct rb_node *parent = NULL; + struct scrub_full_stripe_lock *entry; + + p = &cache->scrub_lock_root.rb_node; + while (*p) { + parent = *p; + entry = rb_entry(parent, struct scrub_full_stripe_lock, node); + if (fstripe_lock->logical < entry->logical) { + p = &(*p)->rb_left; + } else if (fstripe_lock->logical > entry->logical) { + p = &(*p)->rb_right; + } else { + entry->refs++; + return entry; + } + } + /* Insert new one */ + rb_link_node(&fstripe_lock->node, parent, p); + rb_insert_color(&fstripe_lock->node, &cache->scrub_lock_root); + + return NULL; +} + +static struct scrub_full_stripe_lock * +search_scrub_lock(struct btrfs_block_group_cache *cache, u64 bytenr) +{ + struct rb_node *node; + struct scrub_full_stripe_lock *entry; + + node = cache->scrub_lock_root.rb_node; + while (node) { + entry = rb_entry(node, struct scrub_full_stripe_lock, node); + if (bytenr < entry->logical) + node = node->rb_left; + else if (bytenr > entry->logical) + node = node->rb_right; + else + return entry; + } + return NULL; +} + +/* + * Helper to get full stripe logical from a normal bytenr. + * Thanks to the chaos of scrub structures, we need to get it all + * by ourselves, using btrfs_map_sblock(). + */ +static int get_full_stripe_logical(struct btrfs_fs_info *fs_info, u64 bytenr, + u64 *bytenr_ret) +{ + struct btrfs_bio *bbio = NULL; + u64 len; + int ret; + + /* Just use map_sblock() to get full stripe logical */ + ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, bytenr, + &len, &bbio, 0, 1); + if (ret || !bbio || !bbio->raid_map) + goto error; + *bytenr_ret = bbio->raid_map[0]; + btrfs_put_bbio(bbio); + return 0; +error: + btrfs_put_bbio(bbio); + if (ret) + return ret; + return -EIO; +} + +/* + * To lock a full stripe to avoid concurrency of recovery and read + * It's only used for profiles with parities(RAID5/6), for other profiles it + * does nothing + */ +static int lock_full_stripe(struct btrfs_fs_info *fs_info, u64 bytenr, + gfp_t gfp_flags) +{ + struct btrfs_block_group_cache *bg_cache; + struct scrub_full_stripe_lock *fstripe_lock; + struct scrub_full_stripe_lock *existing; + u64 fstripe_start; + int ret = 0; + + bg_cache = btrfs_lookup_block_group(fs_info, bytenr); + if (!bg_cache) + return -ENOENT; + + /* Mirror based profiles don't need full stripe lock */ + if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK)) + goto out; + + ret = get_full_stripe_logical(fs_info, bytenr, &fstripe_start); + if (ret < 0) + goto out; + + fstripe_lock = kmalloc(sizeof(*fstripe_lock), gfp_flags); + if (!fstripe_lock) { + ret = -ENOENT; + goto out; + } + + fstripe_lock->logical = fstripe_start; + fstripe_lock->refs = 1; + mutex_init(&fstripe_lock->mutex); + + /* Now insert the full stripe lock */ + spin_lock(&bg_cache->scrub_lock); + existing = add_scrub_lock(bg_cache, fstripe_lock); + if (existing) { + kfree(fstripe_lock); + fstripe_lock = existing; + } + spin_unlock(&bg_cache->scrub_lock); + mutex_lock(&fstripe_lock->mutex); + +out: + btrfs_put_block_group(bg_cache); + return ret; +} + +static int unlock_full_stripe(struct btrfs_fs_info *fs_info, u64 bytenr) +{ + struct btrfs_block_group_cache *bg_cache; + struct scrub_full_stripe_lock *fstripe_lock; + u64 fstripe_start; + int ret = 0; + + bg_cache = btrfs_lookup_block_group(fs_info, bytenr); + if (!bg_cache) + return -ENOENT; + if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK)) + goto out; + + ret = get_full_stripe_logical(fs_info, bytenr, &fstripe_start); + if (ret < 0) + goto out; + + spin_lock(&bg_cache->scrub_lock); + fstripe_lock = search_scrub_lock(bg_cache, fstripe_start); + /* This is a deadly problem, we hold a mutex but can't unlock it */ + if (WARN_ON(!fstripe_lock)) { + ret = -ENOENT; + goto unlock; + } + + mutex_unlock(&fstripe_lock->mutex); + if (!WARN_ON(fstripe_lock->refs == 0)) + fstripe_lock->refs--; + if (fstripe_lock->refs == 0) { + rb_erase(&fstripe_lock->node, &bg_cache->scrub_lock_root); + kfree(fstripe_lock); + } +unlock: + spin_unlock(&bg_cache->scrub_lock); +out: + btrfs_put_block_group(bg_cache); + return ret; +} + +/* * used for workers that require transaction commits (i.e., for the * NOCOW case) */ -- 2.10.2 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html