Btrfs: fix leaking block group on balance
[safe/jmp/linux-2.6] / fs / btrfs / extent-tree.c
index d156385..1cc8924 100644 (file)
@@ -53,10 +53,6 @@ static int finish_current_insert(struct btrfs_trans_handle *trans, struct
                                 btrfs_root *extent_root, int all);
 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                               btrfs_root *extent_root, int all);
-static struct btrfs_block_group_cache *
-__btrfs_find_block_group(struct btrfs_root *root,
-                        struct btrfs_block_group_cache *hint,
-                        u64 search_start, int data, int owner);
 static int pin_down_bytes(struct btrfs_trans_handle *trans,
                          struct btrfs_root *root,
                          u64 bytenr, u64 num_bytes, int is_data);
@@ -142,6 +138,8 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
                        break;
                }
        }
+       if (ret)
+               atomic_inc(&ret->count);
        spin_unlock(&info->block_group_cache_lock);
 
        return ret;
@@ -189,6 +187,29 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group,
        return 0;
 }
 
+static int remove_sb_from_cache(struct btrfs_root *root,
+                               struct btrfs_block_group_cache *cache)
+{
+       u64 bytenr;
+       u64 *logical;
+       int stripe_len;
+       int i, nr, ret;
+
+       for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+               bytenr = btrfs_sb_offset(i);
+               ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
+                                      cache->key.objectid, bytenr, 0,
+                                      &logical, &nr, &stripe_len);
+               BUG_ON(ret);
+               while (nr--) {
+                       btrfs_remove_free_space(cache, logical[nr],
+                                               stripe_len);
+               }
+               kfree(logical);
+       }
+       return 0;
+}
+
 static int cache_block_group(struct btrfs_root *root,
                             struct btrfs_block_group_cache *block_group)
 {
@@ -197,9 +218,7 @@ static int cache_block_group(struct btrfs_root *root,
        struct btrfs_key key;
        struct extent_buffer *leaf;
        int slot;
-       u64 last = 0;
-       u64 first_free;
-       int found = 0;
+       u64 last = block_group->key.objectid;
 
        if (!block_group)
                return 0;
@@ -220,23 +239,13 @@ static int cache_block_group(struct btrfs_root *root,
         * skip the locking here
         */
        path->skip_locking = 1;
-       first_free = max_t(u64, block_group->key.objectid,
-                          BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
-       key.objectid = block_group->key.objectid;
+       key.objectid = max_t(u64, last, BTRFS_SUPER_INFO_OFFSET);
        key.offset = 0;
        btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
        ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
        if (ret < 0)
                goto err;
-       ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
-       if (ret < 0)
-               goto err;
-       if (ret == 0) {
-               leaf = path->nodes[0];
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
-               if (key.objectid + key.offset > first_free)
-                       first_free = key.objectid + key.offset;
-       }
+
        while(1) {
                leaf = path->nodes[0];
                slot = path->slots[0];
@@ -258,11 +267,6 @@ static int cache_block_group(struct btrfs_root *root,
                        break;
 
                if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
-                       if (!found) {
-                               last = first_free;
-                               found = 1;
-                       }
-
                        add_new_free_space(block_group, root->fs_info, last,
                                           key.objectid);
 
@@ -272,13 +276,11 @@ next:
                path->slots[0]++;
        }
 
-       if (!found)
-               last = first_free;
-
        add_new_free_space(block_group, root->fs_info, last,
                           block_group->key.objectid +
                           block_group->key.offset);
 
+       remove_sb_from_cache(root, block_group);
        block_group->cached = 1;
        ret = 0;
 err:
@@ -314,6 +316,12 @@ struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
        return cache;
 }
 
+static inline void put_block_group(struct btrfs_block_group_cache *cache)
+{
+       if (atomic_dec_and_test(&cache->count))
+               kfree(cache);
+}
+
 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
                                                  u64 flags)
 {
@@ -337,54 +345,16 @@ static u64 div_factor(u64 num, int factor)
        return num;
 }
 
-static struct btrfs_block_group_cache *
-__btrfs_find_block_group(struct btrfs_root *root,
-                        struct btrfs_block_group_cache *hint,
-                        u64 search_start, int data, int owner)
+u64 btrfs_find_block_group(struct btrfs_root *root,
+                          u64 search_start, u64 search_hint, int owner)
 {
        struct btrfs_block_group_cache *cache;
-       struct btrfs_block_group_cache *found_group = NULL;
-       struct btrfs_fs_info *info = root->fs_info;
        u64 used;
-       u64 last = 0;
-       u64 free_check;
+       u64 last = max(search_hint, search_start);
+       u64 group_start = 0;
        int full_search = 0;
-       int factor = 10;
+       int factor = 9;
        int wrapped = 0;
-
-       if (data & BTRFS_BLOCK_GROUP_METADATA)
-               factor = 9;
-
-       if (search_start) {
-               struct btrfs_block_group_cache *shint;
-               shint = btrfs_lookup_first_block_group(info, search_start);
-               if (shint && block_group_bits(shint, data)) {
-                       spin_lock(&shint->lock);
-                       used = btrfs_block_group_used(&shint->item);
-                       if (used + shint->pinned + shint->reserved <
-                           div_factor(shint->key.offset, factor)) {
-                               spin_unlock(&shint->lock);
-                               return shint;
-                       }
-                       spin_unlock(&shint->lock);
-               }
-       }
-       if (hint && block_group_bits(hint, data)) {
-               spin_lock(&hint->lock);
-               used = btrfs_block_group_used(&hint->item);
-               if (used + hint->pinned + hint->reserved <
-                   div_factor(hint->key.offset, factor)) {
-                       spin_unlock(&hint->lock);
-                       return hint;
-               }
-               spin_unlock(&hint->lock);
-               last = hint->key.objectid + hint->key.offset;
-       } else {
-               if (hint)
-                       last = max(hint->key.objectid, search_start);
-               else
-                       last = search_start;
-       }
 again:
        while (1) {
                cache = btrfs_lookup_first_block_group(root->fs_info, last);
@@ -395,16 +365,18 @@ again:
                last = cache->key.objectid + cache->key.offset;
                used = btrfs_block_group_used(&cache->item);
 
-               if (block_group_bits(cache, data)) {
-                       free_check = div_factor(cache->key.offset, factor);
+               if ((full_search || !cache->ro) &&
+                   block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
                        if (used + cache->pinned + cache->reserved <
-                           free_check) {
-                               found_group = cache;
+                           div_factor(cache->key.offset, factor)) {
+                               group_start = cache->key.objectid;
                                spin_unlock(&cache->lock);
+                               put_block_group(cache);
                                goto found;
                        }
                }
                spin_unlock(&cache->lock);
+               put_block_group(cache);
                cond_resched();
        }
        if (!wrapped) {
@@ -419,18 +391,7 @@ again:
                goto again;
        }
 found:
-       return found_group;
-}
-
-struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
-                                                struct btrfs_block_group_cache
-                                                *hint, u64 search_start,
-                                                int data, int owner)
-{
-
-       struct btrfs_block_group_cache *ret;
-       ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
-       return ret;
+       return group_start;
 }
 
 /* simple helper to search for an existing extent at a given offset */
@@ -1805,6 +1766,19 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
        return werr;
 }
 
+int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
+{
+       struct btrfs_block_group_cache *block_group;
+       int readonly = 0;
+
+       block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
+       if (!block_group || block_group->ro)
+               readonly = 1;
+       if (block_group)
+               put_block_group(block_group);
+       return readonly;
+}
+
 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
                             u64 total_bytes, u64 bytes_used,
                             struct btrfs_space_info **space_info)
@@ -1974,10 +1948,8 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                if (alloc) {
                        old_val += num_bytes;
                        cache->space_info->bytes_used += num_bytes;
-                       if (cache->ro) {
+                       if (cache->ro)
                                cache->space_info->bytes_readonly -= num_bytes;
-                               WARN_ON(1);
-                       }
                        btrfs_set_block_group_used(&cache->item, old_val);
                        spin_unlock(&cache->lock);
                        spin_unlock(&cache->space_info->lock);
@@ -1993,10 +1965,10 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                                int ret;
                                ret = btrfs_add_free_space(cache, bytenr,
                                                           num_bytes);
-                               if (ret)
-                                       return -1;
+                               WARN_ON(ret);
                        }
                }
+               put_block_group(cache);
                total -= num_bytes;
                bytenr += num_bytes;
        }
@@ -2006,12 +1978,16 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
 {
        struct btrfs_block_group_cache *cache;
+       u64 bytenr;
 
        cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
        if (!cache)
                return 0;
 
-       return cache->key.objectid;
+       bytenr = cache->key.objectid;
+       put_block_group(cache);
+
+       return bytenr;
 }
 
 int btrfs_update_pinned_extents(struct btrfs_root *root,
@@ -2053,6 +2029,7 @@ int btrfs_update_pinned_extents(struct btrfs_root *root,
                        if (cache->cached)
                                btrfs_add_free_space(cache, bytenr, len);
                }
+               put_block_group(cache);
                bytenr += len;
                num -= len;
        }
@@ -2083,6 +2060,7 @@ static int update_reserved_extents(struct btrfs_root *root,
                }
                spin_unlock(&cache->lock);
                spin_unlock(&cache->space_info->lock);
+               put_block_group(cache);
                bytenr += len;
                num -= len;
        }
@@ -2482,7 +2460,6 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                                mark_free = 1;
                        BUG_ON(ret < 0);
                }
-
                /* block accounting for super block */
                spin_lock_irq(&info->delalloc_lock);
                super_used = btrfs_super_bytes_used(&info->super_copy);
@@ -2502,6 +2479,11 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                                         mark_free);
                BUG_ON(ret);
 
+               if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+                       ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
+                       BUG_ON(ret);
+               }
+
 #ifdef BIO_RW_DISCARD
                /* Tell the block device(s) that the sectors can be discarded */
                ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
@@ -2718,6 +2700,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                        cache = btrfs_lookup_block_group(root->fs_info, bytenr);
                        BUG_ON(!cache);
                        btrfs_add_free_space(cache, bytenr, num_bytes);
+                       put_block_group(cache);
                        update_reserved_extents(root, bytenr, num_bytes, 0);
                        return 0;
                }
@@ -2922,6 +2905,8 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
                }
 new_group:
                mutex_unlock(&block_group->alloc_mutex);
+               put_block_group(block_group);
+               block_group = NULL;
 new_group_no_lock:
                /* don't try to compare new allocations against the
                 * last allocation any more
@@ -2991,6 +2976,8 @@ loop_check:
 
                block_group = list_entry(cur, struct btrfs_block_group_cache,
                                         list);
+               atomic_inc(&block_group->count);
+
                search_start = block_group->key.objectid;
                cur = cur->next;
        }
@@ -2998,7 +2985,7 @@ loop_check:
        /* we found what we needed */
        if (ins->objectid) {
                if (!(data & BTRFS_BLOCK_GROUP_DATA))
-                       trans->block_group = block_group;
+                       trans->block_group = block_group->key.objectid;
 
                if (last_ptr)
                        *last_ptr = ins->objectid + ins->offset;
@@ -3009,6 +2996,8 @@ loop_check:
                       loop, allowed_chunk_alloc);
                ret = -ENOSPC;
        }
+       if (block_group)
+               put_block_group(block_group);
 
        up_read(&space_info->groups_sem);
        return ret;
@@ -3118,6 +3107,7 @@ int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
                return -ENOSPC;
        }
        btrfs_add_free_space(cache, start, len);
+       put_block_group(cache);
        update_reserved_extents(root, start, len, 0);
        return 0;
 }
@@ -3282,6 +3272,7 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
        ret = btrfs_remove_free_space(block_group, ins->objectid,
                                      ins->offset);
        BUG_ON(ret);
+       put_block_group(block_group);
        ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
                                            ref_generation, owner, ins);
        return ret;
@@ -4081,7 +4072,8 @@ static int is_cowonly_root(u64 root_objectid)
            root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
            root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
            root_objectid == BTRFS_DEV_TREE_OBJECTID ||
-           root_objectid == BTRFS_TREE_LOG_OBJECTID)
+           root_objectid == BTRFS_TREE_LOG_OBJECTID ||
+           root_objectid == BTRFS_CSUM_TREE_OBJECTID)
                return 1;
        return 0;
 }
@@ -5491,8 +5483,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
        btrfs_set_inode_generation(leaf, item, 1);
        btrfs_set_inode_size(leaf, item, size);
        btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
-       btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
-                                         BTRFS_INODE_NOCOMPRESS);
+       btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
        btrfs_mark_buffer_dirty(leaf);
        btrfs_release_path(root, path);
 out:
@@ -5697,6 +5688,7 @@ next:
        WARN_ON(block_group->reserved > 0);
        WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
        spin_unlock(&block_group->lock);
+       put_block_group(block_group);
        ret = 0;
 out:
        btrfs_free_path(path);
@@ -5757,6 +5749,8 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
                down_write(&block_group->space_info->groups_sem);
                list_del(&block_group->list);
                up_write(&block_group->space_info->groups_sem);
+
+               WARN_ON(atomic_read(&block_group->count) != 1);
                kfree(block_group);
 
                spin_lock(&info->block_group_cache_lock);
@@ -5801,6 +5795,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                        break;
                }
 
+               atomic_set(&cache->count, 1);
                spin_lock_init(&cache->lock);
                mutex_init(&cache->alloc_mutex);
                mutex_init(&cache->cache_mutex);
@@ -5855,11 +5850,12 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 
        cache->key.objectid = chunk_offset;
        cache->key.offset = size;
+       cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+       atomic_set(&cache->count, 1);
        spin_lock_init(&cache->lock);
        mutex_init(&cache->alloc_mutex);
        mutex_init(&cache->cache_mutex);
        INIT_LIST_HEAD(&cache->list);
-       btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
 
        btrfs_set_block_group_used(&cache->item, bytes_used);
        btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
@@ -5920,10 +5916,8 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
        spin_unlock(&block_group->space_info->lock);
        block_group->space_info->full = 0;
 
-       /*
-       memset(shrink_block_group, 0, sizeof(*shrink_block_group));
-       kfree(shrink_block_group);
-       */
+       put_block_group(block_group);
+       put_block_group(block_group);
 
        ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
        if (ret > 0)