X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=fs%2Fbtrfs%2Fextent-tree.c;h=cd86bffbdc9fac6efa8b2afb5e7fca596c79fb8c;hb=4008c04a07c73ec3cb1be4c1391d2159a8f75d6d;hp=29380467b6717b00da27ae651060856893d7c3ac;hpb=30c43e2444c16afe3b2130f40ad273541bf3dc36;p=safe%2Fjmp%2Flinux-2.6 diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 2938046..cd86bff 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -19,6 +19,8 @@ #include #include #include +#include +#include "compat.h" #include "hash.h" #include "crc32c.h" #include "ctree.h" @@ -42,34 +44,21 @@ struct pending_extent_op { u64 generation; u64 orig_generation; int level; + struct list_head list; + int del; }; -static int finish_current_insert(struct btrfs_trans_handle *trans, struct - btrfs_root *extent_root); -static int del_pending_extents(struct btrfs_trans_handle *trans, struct - btrfs_root *extent_root); -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); - -void maybe_lock_mutex(struct btrfs_root *root) -{ - if (root != root->fs_info->extent_root && - root != root->fs_info->chunk_root && - root != root->fs_info->dev_root) { - mutex_lock(&root->fs_info->alloc_mutex); - } -} - -void maybe_unlock_mutex(struct btrfs_root *root) -{ - if (root != root->fs_info->extent_root && - root != root->fs_info->chunk_root && - root != root->fs_info->dev_root) { - mutex_unlock(&root->fs_info->alloc_mutex); - } -} +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 int pin_down_bytes(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, int is_data); +static int update_block_group(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, int alloc, + int mark_free); static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) { @@ -80,7 +69,7 @@ static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) * this adds the block group to the fs_info rb tree for the block group * cache */ -int btrfs_add_block_group_cache(struct btrfs_fs_info *info, +static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, struct btrfs_block_group_cache *block_group) { struct rb_node **p; @@ -148,6 +137,8 @@ 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; @@ -164,6 +155,7 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group, u64 extent_start, extent_end, size; int ret; + mutex_lock(&info->pinned_mutex); while (start < end) { ret = find_first_extent_bit(&info->pinned_extents, start, &extent_start, &extent_end, @@ -175,7 +167,8 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group, start = extent_end + 1; } else if (extent_start > start && extent_start < end) { size = extent_start - start; - ret = btrfs_add_free_space(block_group, start, size); + ret = btrfs_add_free_space(block_group, start, + size); BUG_ON(ret); start = extent_end + 1; } else { @@ -188,10 +181,34 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group, ret = btrfs_add_free_space(block_group, start, size); BUG_ON(ret); } + mutex_unlock(&info->pinned_mutex); 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) { @@ -200,9 +217,7 @@ 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; if (!block_group) return 0; @@ -223,24 +238,15 @@ 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; + last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); + key.objectid = last; 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) { + + while (1) { leaf = path->nodes[0]; slot = path->slots[0]; if (slot >= btrfs_header_nritems(leaf)) { @@ -261,11 +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); @@ -275,13 +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: @@ -292,9 +291,8 @@ err: /* * return the block group that starts at or after bytenr */ -struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct - btrfs_fs_info *info, - u64 bytenr) +static struct btrfs_block_group_cache * +btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) { struct btrfs_block_group_cache *cache; @@ -306,9 +304,9 @@ struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct /* * return the block group that contains teh given bytenr */ -struct btrfs_block_group_cache *btrfs_lookup_block_group(struct - btrfs_fs_info *info, - u64 bytenr) +struct btrfs_block_group_cache *btrfs_lookup_block_group( + struct btrfs_fs_info *info, + u64 bytenr) { struct btrfs_block_group_cache *cache; @@ -317,121 +315,43 @@ struct btrfs_block_group_cache *btrfs_lookup_block_group(struct return cache; } -static int noinline find_free_space(struct btrfs_root *root, - struct btrfs_block_group_cache **cache_ret, - u64 *start_ret, u64 num, int data) -{ - int ret; - struct btrfs_block_group_cache *cache = *cache_ret; - struct btrfs_free_space *info = NULL; - u64 last; - u64 search_start = *start_ret; - - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); - if (!cache) - goto out; - - last = max(search_start, cache->key.objectid); - -again: - ret = cache_block_group(root, cache); - if (ret) - goto out; - - if (cache->ro || !block_group_bits(cache, data)) - goto new_group; - - info = btrfs_find_free_space(cache, last, num); - if (info) { - *start_ret = info->offset; - return 0; - } - -new_group: - last = cache->key.objectid + cache->key.offset; - - cache = btrfs_lookup_first_block_group(root->fs_info, last); - if (!cache) - goto out; - - *cache_ret = cache; - goto again; - -out: - return -ENOSPC; -} - -static u64 div_factor(u64 num, int factor) +static inline void put_block_group(struct btrfs_block_group_cache *cache) { - if (factor == 10) - return num; - num *= factor; - do_div(num, 10); - return num; + if (atomic_dec_and_test(&cache->count)) + kfree(cache); } static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, u64 flags) { struct list_head *head = &info->space_info; - struct list_head *cur; struct btrfs_space_info *found; - list_for_each(cur, head) { - found = list_entry(cur, struct btrfs_space_info, list); + list_for_each_entry(found, head, list) { if (found->flags == flags) return found; } return NULL; } -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 u64 div_factor(u64 num, int factor) +{ + if (factor == 10) + return num; + num *= factor; + do_div(num, 10); + return num; +} + +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) && !shint->ro) { - 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 && !hint->ro && 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); @@ -442,16 +362,18 @@ again: last = cache->key.objectid + cache->key.offset; used = btrfs_block_group_used(&cache->item); - if (!cache->ro && 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) { @@ -466,18 +388,7 @@ 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 */ @@ -489,13 +400,11 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) path = btrfs_alloc_path(); BUG_ON(!path); - maybe_lock_mutex(root); key.objectid = start; key.offset = len; btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, 0, 0); - maybe_unlock_mutex(root); btrfs_free_path(path); return ret; } @@ -525,31 +434,28 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) * - Objectid of the subvolume root * - Generation number of the tree holding the reference * - objectid of the file holding the reference - * - offset in the file corresponding to the key holding the reference * - number of references holding by parent node (alway 1 for tree blocks) * * Btree leaf may hold multiple references to a file extent. In most cases, * these references are from same file and the corresponding offsets inside - * the file are close together. So inode objectid and offset in file are - * just hints, they provide hints about where in the btree the references - * can be found and when we can stop searching. + * the file are close together. * * When a file extent is allocated the fields are filled in: - * (root_key.objectid, trans->transid, inode objectid, offset in file, 1) + * (root_key.objectid, trans->transid, inode objectid, 1) * * When a leaf is cow'd new references are added for every file extent found * in the leaf. It looks similar to the create case, but trans->transid will * be different when the block is cow'd. * - * (root_key.objectid, trans->transid, inode objectid, offset in file, + * (root_key.objectid, trans->transid, inode objectid, * number of references in the leaf) * - * Because inode objectid and offset in file are just hints, they are not - * used when backrefs are deleted. When a file extent is removed either - * during snapshot deletion or file truncation, we find the corresponding - * back back reference and check the following fields. + * When a file extent is removed either during snapshot deletion or + * file truncation, we find the corresponding back reference and check + * the following fields: * - * (btrfs_header_owner(leaf), btrfs_header_generation(leaf)) + * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), + * inode objectid) * * Btree extents can be referenced by: * @@ -558,21 +464,21 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) * * When a tree block is created, back references are inserted: * - * (root->root_key.objectid, trans->transid, level, 0, 1) + * (root->root_key.objectid, trans->transid, level, 1) * * When a tree block is cow'd, new back references are added for all the * blocks it points to. If the tree block isn't in reference counted root, * the old back references are removed. These new back references are of * the form (trans->transid will have increased since creation): * - * (root->root_key.objectid, trans->transid, level, 0, 1) + * (root->root_key.objectid, trans->transid, level, 1) * * When a backref is in deleting, the following fields are checked: * * if backref was for a tree root: - * (btrfs_header_owner(itself), btrfs_header_generation(itself)) + * (btrfs_header_owner(itself), btrfs_header_generation(itself), level) * else - * (btrfs_header_owner(parent), btrfs_header_generation(parent)) + * (btrfs_header_owner(parent), btrfs_header_generation(parent), level) * * Back Reference Key composing: * @@ -582,15 +488,17 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) * to the key objectid. */ -static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, +static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans, struct btrfs_root *root, - struct btrfs_path *path, u64 bytenr, - u64 parent, u64 ref_root, - u64 ref_generation, int del) + struct btrfs_path *path, + u64 bytenr, u64 parent, + u64 ref_root, u64 ref_generation, + u64 owner_objectid, int del) { struct btrfs_key key; struct btrfs_extent_ref *ref; struct extent_buffer *leaf; + u64 ref_objectid; int ret; key.objectid = bytenr; @@ -607,8 +515,11 @@ static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, leaf = path->nodes[0]; ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); + ref_objectid = btrfs_ref_objectid(leaf, ref); if (btrfs_ref_root(leaf, ref) != ref_root || - btrfs_ref_generation(leaf, ref) != ref_generation) { + btrfs_ref_generation(leaf, ref) != ref_generation || + (ref_objectid != owner_objectid && + ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { ret = -EIO; WARN_ON(1); goto out; @@ -618,12 +529,262 @@ out: return ret; } -static int noinline insert_extent_backref(struct btrfs_trans_handle *trans, +/* + * updates all the backrefs that are pending on update_list for the + * extent_root + */ +static noinline int update_backrefs(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, + struct btrfs_path *path, + struct list_head *update_list) +{ + struct btrfs_key key; + struct btrfs_extent_ref *ref; + struct btrfs_fs_info *info = extent_root->fs_info; + struct pending_extent_op *op; + struct extent_buffer *leaf; + int ret = 0; + struct list_head *cur = update_list->next; + u64 ref_objectid; + u64 ref_root = extent_root->root_key.objectid; + + op = list_entry(cur, struct pending_extent_op, list); + +search: + key.objectid = op->bytenr; + key.type = BTRFS_EXTENT_REF_KEY; + key.offset = op->orig_parent; + + ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1); + BUG_ON(ret); + + leaf = path->nodes[0]; + +loop: + ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); + + ref_objectid = btrfs_ref_objectid(leaf, ref); + + if (btrfs_ref_root(leaf, ref) != ref_root || + btrfs_ref_generation(leaf, ref) != op->orig_generation || + (ref_objectid != op->level && + ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { + printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, " + "root %llu, owner %u\n", + (unsigned long long)op->bytenr, + (unsigned long long)op->orig_parent, + (unsigned long long)ref_root, op->level); + btrfs_print_leaf(extent_root, leaf); + BUG(); + } + + key.objectid = op->bytenr; + key.offset = op->parent; + key.type = BTRFS_EXTENT_REF_KEY; + ret = btrfs_set_item_key_safe(trans, extent_root, path, &key); + BUG_ON(ret); + ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); + btrfs_set_ref_generation(leaf, ref, op->generation); + + cur = cur->next; + + list_del_init(&op->list); + unlock_extent(&info->extent_ins, op->bytenr, + op->bytenr + op->num_bytes - 1, GFP_NOFS); + kfree(op); + + if (cur == update_list) { + btrfs_mark_buffer_dirty(path->nodes[0]); + btrfs_release_path(extent_root, path); + goto out; + } + + op = list_entry(cur, struct pending_extent_op, list); + + path->slots[0]++; + while (path->slots[0] < btrfs_header_nritems(leaf)) { + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + if (key.objectid == op->bytenr && + key.type == BTRFS_EXTENT_REF_KEY) + goto loop; + path->slots[0]++; + } + + btrfs_mark_buffer_dirty(path->nodes[0]); + btrfs_release_path(extent_root, path); + goto search; + +out: + return 0; +} + +static noinline int insert_extents(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, + struct btrfs_path *path, + struct list_head *insert_list, int nr) +{ + struct btrfs_key *keys; + u32 *data_size; + struct pending_extent_op *op; + struct extent_buffer *leaf; + struct list_head *cur = insert_list->next; + struct btrfs_fs_info *info = extent_root->fs_info; + u64 ref_root = extent_root->root_key.objectid; + int i = 0, last = 0, ret; + int total = nr * 2; + + if (!nr) + return 0; + + keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS); + if (!keys) + return -ENOMEM; + + data_size = kzalloc(total * sizeof(u32), GFP_NOFS); + if (!data_size) { + kfree(keys); + return -ENOMEM; + } + + list_for_each_entry(op, insert_list, list) { + keys[i].objectid = op->bytenr; + keys[i].offset = op->num_bytes; + keys[i].type = BTRFS_EXTENT_ITEM_KEY; + data_size[i] = sizeof(struct btrfs_extent_item); + i++; + + keys[i].objectid = op->bytenr; + keys[i].offset = op->parent; + keys[i].type = BTRFS_EXTENT_REF_KEY; + data_size[i] = sizeof(struct btrfs_extent_ref); + i++; + } + + op = list_entry(cur, struct pending_extent_op, list); + i = 0; + while (i < total) { + int c; + ret = btrfs_insert_some_items(trans, extent_root, path, + keys+i, data_size+i, total-i); + BUG_ON(ret < 0); + + if (last && ret > 1) + BUG(); + + leaf = path->nodes[0]; + for (c = 0; c < ret; c++) { + int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY; + + /* + * if the first item we inserted was a backref, then + * the EXTENT_ITEM will be the odd c's, else it will + * be the even c's + */ + if ((ref_first && (c % 2)) || + (!ref_first && !(c % 2))) { + struct btrfs_extent_item *itm; + + itm = btrfs_item_ptr(leaf, path->slots[0] + c, + struct btrfs_extent_item); + btrfs_set_extent_refs(path->nodes[0], itm, 1); + op->del++; + } else { + struct btrfs_extent_ref *ref; + + ref = btrfs_item_ptr(leaf, path->slots[0] + c, + struct btrfs_extent_ref); + btrfs_set_ref_root(leaf, ref, ref_root); + btrfs_set_ref_generation(leaf, ref, + op->generation); + btrfs_set_ref_objectid(leaf, ref, op->level); + btrfs_set_ref_num_refs(leaf, ref, 1); + op->del++; + } + + /* + * using del to see when its ok to free up the + * pending_extent_op. In the case where we insert the + * last item on the list in order to help do batching + * we need to not free the extent op until we actually + * insert the extent_item + */ + if (op->del == 2) { + unlock_extent(&info->extent_ins, op->bytenr, + op->bytenr + op->num_bytes - 1, + GFP_NOFS); + cur = cur->next; + list_del_init(&op->list); + kfree(op); + if (cur != insert_list) + op = list_entry(cur, + struct pending_extent_op, + list); + } + } + btrfs_mark_buffer_dirty(leaf); + btrfs_release_path(extent_root, path); + + /* + * Ok backref's and items usually go right next to eachother, + * but if we could only insert 1 item that means that we + * inserted on the end of a leaf, and we have no idea what may + * be on the next leaf so we just play it safe. In order to + * try and help this case we insert the last thing on our + * insert list so hopefully it will end up being the last + * thing on the leaf and everything else will be before it, + * which will let us insert a whole bunch of items at the same + * time. + */ + if (ret == 1 && !last && (i + ret < total)) { + /* + * last: where we will pick up the next time around + * i: our current key to insert, will be total - 1 + * cur: the current op we are screwing with + * op: duh + */ + last = i + ret; + i = total - 1; + cur = insert_list->prev; + op = list_entry(cur, struct pending_extent_op, list); + } else if (last) { + /* + * ok we successfully inserted the last item on the + * list, lets reset everything + * + * i: our current key to insert, so where we left off + * last time + * last: done with this + * cur: the op we are messing with + * op: duh + * total: since we inserted the last key, we need to + * decrement total so we dont overflow + */ + i = last; + last = 0; + total--; + if (i < total) { + cur = insert_list->next; + op = list_entry(cur, struct pending_extent_op, + list); + } + } else { + i += ret; + } + + cond_resched(); + } + ret = 0; + kfree(keys); + kfree(data_size); + return ret; +} + +static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, u64 bytenr, u64 parent, u64 ref_root, u64 ref_generation, - u64 owner_objectid, u64 owner_offset) + u64 owner_objectid) { struct btrfs_key key; struct extent_buffer *leaf; @@ -643,7 +804,6 @@ static int noinline insert_extent_backref(struct btrfs_trans_handle *trans, btrfs_set_ref_root(leaf, ref, ref_root); btrfs_set_ref_generation(leaf, ref, ref_generation); btrfs_set_ref_objectid(leaf, ref, owner_objectid); - btrfs_set_ref_offset(leaf, ref, owner_offset); btrfs_set_ref_num_refs(leaf, ref, 1); } else if (ret == -EEXIST) { u64 existing_owner; @@ -663,14 +823,10 @@ static int noinline insert_extent_backref(struct btrfs_trans_handle *trans, btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); existing_owner = btrfs_ref_objectid(leaf, ref); - if (existing_owner == owner_objectid && - btrfs_ref_offset(leaf, ref) > owner_offset) { - btrfs_set_ref_offset(leaf, ref, owner_offset); - } else if (existing_owner != owner_objectid && - existing_owner != BTRFS_MULTIPLE_OBJECTIDS) { + if (existing_owner != owner_objectid && + existing_owner != BTRFS_MULTIPLE_OBJECTIDS) { btrfs_set_ref_objectid(leaf, ref, BTRFS_MULTIPLE_OBJECTIDS); - btrfs_set_ref_offset(leaf, ref, 0); } ret = 0; } else { @@ -682,7 +838,7 @@ out: return ret; } -static int noinline remove_extent_backref(struct btrfs_trans_handle *trans, +static noinline int remove_extent_backref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path) { @@ -706,12 +862,297 @@ static int noinline remove_extent_backref(struct btrfs_trans_handle *trans, return ret; } +#ifdef BIO_RW_DISCARD +static void btrfs_issue_discard(struct block_device *bdev, + u64 start, u64 len) +{ + blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL); +} +#endif + +static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, + u64 num_bytes) +{ +#ifdef BIO_RW_DISCARD + int ret; + u64 map_length = num_bytes; + struct btrfs_multi_bio *multi = NULL; + + /* Tell the block device(s) that the sectors can be discarded */ + ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, + bytenr, &map_length, &multi, 0); + if (!ret) { + struct btrfs_bio_stripe *stripe = multi->stripes; + int i; + + if (map_length > num_bytes) + map_length = num_bytes; + + for (i = 0; i < multi->num_stripes; i++, stripe++) { + btrfs_issue_discard(stripe->dev->bdev, + stripe->physical, + map_length); + } + kfree(multi); + } + + return ret; +#else + return 0; +#endif +} + +static noinline int free_extents(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, + struct list_head *del_list) +{ + struct btrfs_fs_info *info = extent_root->fs_info; + struct btrfs_path *path; + struct btrfs_key key, found_key; + struct extent_buffer *leaf; + struct list_head *cur; + struct pending_extent_op *op; + struct btrfs_extent_item *ei; + int ret, num_to_del, extent_slot = 0, found_extent = 0; + u32 refs; + u64 bytes_freed = 0; + + path = btrfs_alloc_path(); + if (!path) + return -ENOMEM; + path->reada = 1; + +search: + /* search for the backref for the current ref we want to delete */ + cur = del_list->next; + op = list_entry(cur, struct pending_extent_op, list); + ret = lookup_extent_backref(trans, extent_root, path, op->bytenr, + op->orig_parent, + extent_root->root_key.objectid, + op->orig_generation, op->level, 1); + if (ret) { + printk(KERN_ERR "btrfs unable to find backref byte nr %llu " + "root %llu gen %llu owner %u\n", + (unsigned long long)op->bytenr, + (unsigned long long)extent_root->root_key.objectid, + (unsigned long long)op->orig_generation, op->level); + btrfs_print_leaf(extent_root, path->nodes[0]); + WARN_ON(1); + goto out; + } + + extent_slot = path->slots[0]; + num_to_del = 1; + found_extent = 0; + + /* + * if we aren't the first item on the leaf we can move back one and see + * if our ref is right next to our extent item + */ + if (likely(extent_slot)) { + extent_slot--; + btrfs_item_key_to_cpu(path->nodes[0], &found_key, + extent_slot); + if (found_key.objectid == op->bytenr && + found_key.type == BTRFS_EXTENT_ITEM_KEY && + found_key.offset == op->num_bytes) { + num_to_del++; + found_extent = 1; + } + } + + /* + * if we didn't find the extent we need to delete the backref and then + * search for the extent item key so we can update its ref count + */ + if (!found_extent) { + key.objectid = op->bytenr; + key.type = BTRFS_EXTENT_ITEM_KEY; + key.offset = op->num_bytes; + + ret = remove_extent_backref(trans, extent_root, path); + BUG_ON(ret); + btrfs_release_path(extent_root, path); + ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); + BUG_ON(ret); + extent_slot = path->slots[0]; + } + + /* this is where we update the ref count for the extent */ + leaf = path->nodes[0]; + ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item); + refs = btrfs_extent_refs(leaf, ei); + BUG_ON(refs == 0); + refs--; + btrfs_set_extent_refs(leaf, ei, refs); + + btrfs_mark_buffer_dirty(leaf); + + /* + * This extent needs deleting. The reason cur_slot is extent_slot + + * num_to_del is because extent_slot points to the slot where the extent + * is, and if the backref was not right next to the extent we will be + * deleting at least 1 item, and will want to start searching at the + * slot directly next to extent_slot. However if we did find the + * backref next to the extent item them we will be deleting at least 2 + * items and will want to start searching directly after the ref slot + */ + if (!refs) { + struct list_head *pos, *n, *end; + int cur_slot = extent_slot+num_to_del; + u64 super_used; + u64 root_used; + + path->slots[0] = extent_slot; + bytes_freed = op->num_bytes; + + mutex_lock(&info->pinned_mutex); + ret = pin_down_bytes(trans, extent_root, op->bytenr, + op->num_bytes, op->level >= + BTRFS_FIRST_FREE_OBJECTID); + mutex_unlock(&info->pinned_mutex); + BUG_ON(ret < 0); + op->del = ret; + + /* + * we need to see if we can delete multiple things at once, so + * start looping through the list of extents we are wanting to + * delete and see if their extent/backref's are right next to + * eachother and the extents only have 1 ref + */ + for (pos = cur->next; pos != del_list; pos = pos->next) { + struct pending_extent_op *tmp; + + tmp = list_entry(pos, struct pending_extent_op, list); + + /* we only want to delete extent+ref at this stage */ + if (cur_slot >= btrfs_header_nritems(leaf) - 1) + break; + + btrfs_item_key_to_cpu(leaf, &found_key, cur_slot); + if (found_key.objectid != tmp->bytenr || + found_key.type != BTRFS_EXTENT_ITEM_KEY || + found_key.offset != tmp->num_bytes) + break; + + /* check to make sure this extent only has one ref */ + ei = btrfs_item_ptr(leaf, cur_slot, + struct btrfs_extent_item); + if (btrfs_extent_refs(leaf, ei) != 1) + break; + + btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1); + if (found_key.objectid != tmp->bytenr || + found_key.type != BTRFS_EXTENT_REF_KEY || + found_key.offset != tmp->orig_parent) + break; + + /* + * the ref is right next to the extent, we can set the + * ref count to 0 since we will delete them both now + */ + btrfs_set_extent_refs(leaf, ei, 0); + + /* pin down the bytes for this extent */ + mutex_lock(&info->pinned_mutex); + ret = pin_down_bytes(trans, extent_root, tmp->bytenr, + tmp->num_bytes, tmp->level >= + BTRFS_FIRST_FREE_OBJECTID); + mutex_unlock(&info->pinned_mutex); + BUG_ON(ret < 0); + + /* + * use the del field to tell if we need to go ahead and + * free up the extent when we delete the item or not. + */ + tmp->del = ret; + bytes_freed += tmp->num_bytes; + + num_to_del += 2; + cur_slot += 2; + } + end = pos; + + /* update the free space counters */ + spin_lock(&info->delalloc_lock); + super_used = btrfs_super_bytes_used(&info->super_copy); + btrfs_set_super_bytes_used(&info->super_copy, + super_used - bytes_freed); + + root_used = btrfs_root_used(&extent_root->root_item); + btrfs_set_root_used(&extent_root->root_item, + root_used - bytes_freed); + spin_unlock(&info->delalloc_lock); + + /* delete the items */ + ret = btrfs_del_items(trans, extent_root, path, + path->slots[0], num_to_del); + BUG_ON(ret); + + /* + * loop through the extents we deleted and do the cleanup work + * on them + */ + for (pos = cur, n = pos->next; pos != end; + pos = n, n = pos->next) { + struct pending_extent_op *tmp; + tmp = list_entry(pos, struct pending_extent_op, list); + + /* + * remember tmp->del tells us wether or not we pinned + * down the extent + */ + ret = update_block_group(trans, extent_root, + tmp->bytenr, tmp->num_bytes, 0, + tmp->del); + BUG_ON(ret); + + list_del_init(&tmp->list); + unlock_extent(&info->extent_ins, tmp->bytenr, + tmp->bytenr + tmp->num_bytes - 1, + GFP_NOFS); + kfree(tmp); + } + } else if (refs && found_extent) { + /* + * the ref and extent were right next to eachother, but the + * extent still has a ref, so just free the backref and keep + * going + */ + ret = remove_extent_backref(trans, extent_root, path); + BUG_ON(ret); + + list_del_init(&op->list); + unlock_extent(&info->extent_ins, op->bytenr, + op->bytenr + op->num_bytes - 1, GFP_NOFS); + kfree(op); + } else { + /* + * the extent has multiple refs and the backref we were looking + * for was not right next to it, so just unlock and go next, + * we're good to go + */ + list_del_init(&op->list); + unlock_extent(&info->extent_ins, op->bytenr, + op->bytenr + op->num_bytes - 1, GFP_NOFS); + kfree(op); + } + + btrfs_release_path(extent_root, path); + if (!list_empty(del_list)) + goto search; + +out: + btrfs_free_path(path); + return ret; +} + static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 orig_parent, u64 parent, u64 orig_root, u64 ref_root, u64 orig_generation, u64 ref_generation, - u64 owner_objectid, u64 owner_offset) + u64 owner_objectid) { int ret; struct btrfs_root *extent_root = root->fs_info->extent_root; @@ -723,8 +1164,9 @@ static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL); num_bytes = btrfs_level_size(root, (int)owner_objectid); + mutex_lock(&root->fs_info->extent_ins_mutex); if (test_range_bit(&root->fs_info->extent_ins, bytenr, - bytenr + num_bytes - 1, EXTENT_LOCKED, 0)) { + bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { u64 priv; ret = get_state_private(&root->fs_info->extent_ins, bytenr, &priv); @@ -733,6 +1175,7 @@ static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, (unsigned long)priv; BUG_ON(extent_op->parent != orig_parent); BUG_ON(extent_op->generation != orig_generation); + extent_op->parent = parent; extent_op->generation = ref_generation; } else { @@ -747,13 +1190,16 @@ static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, extent_op->generation = ref_generation; extent_op->orig_generation = orig_generation; extent_op->level = (int)owner_objectid; + INIT_LIST_HEAD(&extent_op->list); + extent_op->del = 0; set_extent_bits(&root->fs_info->extent_ins, bytenr, bytenr + num_bytes - 1, - EXTENT_LOCKED, GFP_NOFS); + EXTENT_WRITEBACK, GFP_NOFS); set_state_private(&root->fs_info->extent_ins, bytenr, (unsigned long)extent_op); } + mutex_unlock(&root->fs_info->extent_ins_mutex); return 0; } @@ -762,7 +1208,7 @@ static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, return -ENOMEM; ret = lookup_extent_backref(trans, extent_root, path, bytenr, orig_parent, orig_root, - orig_generation, 1); + orig_generation, owner_objectid, 1); if (ret) goto out; ret = remove_extent_backref(trans, extent_root, path); @@ -770,10 +1216,10 @@ static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, goto out; ret = insert_extent_backref(trans, extent_root, path, bytenr, parent, ref_root, ref_generation, - owner_objectid, owner_offset); + owner_objectid); BUG_ON(ret); - finish_current_insert(trans, extent_root); - del_pending_extents(trans, extent_root); + finish_current_insert(trans, extent_root, 0); + del_pending_extents(trans, extent_root, 0); out: btrfs_free_path(path); return ret; @@ -783,18 +1229,16 @@ int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 orig_parent, u64 parent, u64 ref_root, u64 ref_generation, - u64 owner_objectid, u64 owner_offset) + u64 owner_objectid) { int ret; if (ref_root == BTRFS_TREE_LOG_OBJECTID && owner_objectid < BTRFS_FIRST_FREE_OBJECTID) return 0; - maybe_lock_mutex(root); ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, parent, ref_root, ref_root, ref_generation, ref_generation, - owner_objectid, owner_offset); - maybe_unlock_mutex(root); + owner_objectid); return ret; } @@ -803,7 +1247,7 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, u64 orig_parent, u64 parent, u64 orig_root, u64 ref_root, u64 orig_generation, u64 ref_generation, - u64 owner_objectid, u64 owner_offset) + u64 owner_objectid) { struct btrfs_path *path; int ret; @@ -831,7 +1275,13 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, l = path->nodes[0]; btrfs_item_key_to_cpu(l, &key, path->slots[0]); - BUG_ON(key.objectid != bytenr); + if (key.objectid != bytenr) { + btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]); + printk(KERN_ERR "btrfs wanted %llu found %llu\n", + (unsigned long long)bytenr, + (unsigned long long)key.objectid); + BUG(); + } BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); @@ -845,10 +1295,10 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, ret = insert_extent_backref(trans, root->fs_info->extent_root, path, bytenr, parent, ref_root, ref_generation, - owner_objectid, owner_offset); + owner_objectid); BUG_ON(ret); - finish_current_insert(trans, root->fs_info->extent_root); - del_pending_extents(trans, root->fs_info->extent_root); + finish_current_insert(trans, root->fs_info->extent_root, 0); + del_pending_extents(trans, root->fs_info->extent_root, 0); btrfs_free_path(path); return 0; @@ -858,25 +1308,40 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root, u64 ref_generation, - u64 owner_objectid, u64 owner_offset) + u64 owner_objectid) { int ret; if (ref_root == BTRFS_TREE_LOG_OBJECTID && owner_objectid < BTRFS_FIRST_FREE_OBJECTID) return 0; - maybe_lock_mutex(root); ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, 0, ref_root, 0, ref_generation, - owner_objectid, owner_offset); - maybe_unlock_mutex(root); + owner_objectid); return ret; } int btrfs_extent_post_op(struct btrfs_trans_handle *trans, struct btrfs_root *root) { - finish_current_insert(trans, root->fs_info->extent_root); - del_pending_extents(trans, root->fs_info->extent_root); + u64 start; + u64 end; + int ret; + + while(1) { + finish_current_insert(trans, root->fs_info->extent_root, 1); + del_pending_extents(trans, root->fs_info->extent_root, 1); + + /* is there more work to do? */ + ret = find_first_extent_bit(&root->fs_info->pending_del, + 0, &start, &end, EXTENT_WRITEBACK); + if (!ret) + continue; + ret = find_first_extent_bit(&root->fs_info->extent_ins, + 0, &start, &end, EXTENT_WRITEBACK); + if (!ret) + continue; + break; + } return 0; } @@ -902,7 +1367,8 @@ int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, goto out; if (ret != 0) { btrfs_print_leaf(root, path->nodes[0]); - printk("failed to find block number %Lu\n", bytenr); + printk(KERN_INFO "btrfs failed to find block number %llu\n", + (unsigned long long)bytenr); BUG(); } l = path->nodes[0]; @@ -913,9 +1379,8 @@ out: return 0; } -static int get_reference_status(struct btrfs_root *root, u64 bytenr, - u64 parent_gen, u64 ref_objectid, - u64 *min_generation, u32 *ref_count) +int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, + struct btrfs_root *root, u64 objectid, u64 bytenr) { struct btrfs_root *extent_root = root->fs_info->extent_root; struct btrfs_path *path; @@ -923,8 +1388,8 @@ static int get_reference_status(struct btrfs_root *root, u64 bytenr, struct btrfs_extent_ref *ref_item; struct btrfs_key key; struct btrfs_key found_key; - u64 root_objectid = root->root_key.objectid; - u64 ref_generation; + u64 ref_root; + u64 last_snapshot; u32 nritems; int ret; @@ -933,12 +1398,13 @@ static int get_reference_status(struct btrfs_root *root, u64 bytenr, key.type = BTRFS_EXTENT_ITEM_KEY; path = btrfs_alloc_path(); - mutex_lock(&root->fs_info->alloc_mutex); ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); if (ret < 0) goto out; BUG_ON(ret == 0); - if (ret < 0 || path->slots[0] == 0) + + ret = -ENOENT; + if (path->slots[0] == 0) goto out; path->slots[0]--; @@ -946,14 +1412,10 @@ static int get_reference_status(struct btrfs_root *root, u64 bytenr, btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); if (found_key.objectid != bytenr || - found_key.type != BTRFS_EXTENT_ITEM_KEY) { - ret = 1; + found_key.type != BTRFS_EXTENT_ITEM_KEY) goto out; - } - - *ref_count = 0; - *min_generation = (u64)-1; + last_snapshot = btrfs_root_last_snapshot(&root->root_item); while (1) { leaf = path->nodes[0]; nritems = btrfs_header_nritems(leaf); @@ -976,115 +1438,23 @@ static int get_reference_status(struct btrfs_root *root, u64 bytenr, ref_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); - ref_generation = btrfs_ref_generation(leaf, ref_item); - /* - * For (parent_gen > 0 && parent_gen > ref_generation): - * - * we reach here through the oldest root, therefore - * all other reference from same snapshot should have - * a larger generation. - */ - if ((root_objectid != btrfs_ref_root(leaf, ref_item)) || - (parent_gen > 0 && parent_gen > ref_generation) || - (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID && - ref_objectid != btrfs_ref_objectid(leaf, ref_item))) { - *ref_count = 2; - break; - } - - *ref_count = 1; - if (*min_generation > ref_generation) - *min_generation = ref_generation; - - path->slots[0]++; - } - ret = 0; -out: - mutex_unlock(&root->fs_info->alloc_mutex); - btrfs_free_path(path); - return ret; -} - -int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - struct btrfs_key *key, u64 bytenr) -{ - struct btrfs_root *old_root; - struct btrfs_path *path = NULL; - struct extent_buffer *eb; - struct btrfs_file_extent_item *item; - u64 ref_generation; - u64 min_generation; - u64 extent_start; - u32 ref_count; - int level; - int ret; - - BUG_ON(trans == NULL); - BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY); - ret = get_reference_status(root, bytenr, 0, key->objectid, - &min_generation, &ref_count); - if (ret) - return ret; - - if (ref_count != 1) - return 1; - - old_root = root->dirty_root->root; - ref_generation = old_root->root_key.offset; - - /* all references are created in running transaction */ - if (min_generation > ref_generation) { - ret = 0; - goto out; - } - - path = btrfs_alloc_path(); - if (!path) { - ret = -ENOMEM; - goto out; - } - - path->skip_locking = 1; - /* if no item found, the extent is referenced by other snapshot */ - ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0); - if (ret) - goto out; - - eb = path->nodes[0]; - item = btrfs_item_ptr(eb, path->slots[0], - struct btrfs_file_extent_item); - if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG || - btrfs_file_extent_disk_bytenr(eb, item) != bytenr) { - ret = 1; - goto out; - } - - for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) { - if (level >= 0) { - eb = path->nodes[level]; - if (!eb) - continue; - extent_start = eb->start; - } else - extent_start = bytenr; - - ret = get_reference_status(root, extent_start, ref_generation, - 0, &min_generation, &ref_count); - if (ret) + ref_root = btrfs_ref_root(leaf, ref_item); + if ((ref_root != root->root_key.objectid && + ref_root != BTRFS_TREE_LOG_OBJECTID) || + objectid != btrfs_ref_objectid(leaf, ref_item)) { + ret = 1; goto out; - - if (ref_count != 1) { + } + if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) { ret = 1; goto out; } - if (level >= 0) - ref_generation = btrfs_header_generation(eb); + + path->slots[0]++; } ret = 0; out: - if (path) - btrfs_free_path(path); + btrfs_free_path(path); return ret; } @@ -1154,6 +1524,14 @@ int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, } ret = btrfs_add_leaf_ref(root, ref, shared); + if (ret == -EEXIST && shared) { + struct btrfs_leaf_ref *old; + old = btrfs_lookup_leaf_ref(root, ref->bytenr); + BUG_ON(!old); + btrfs_remove_leaf_ref(root, old); + btrfs_free_leaf_ref(root, old); + ret = btrfs_add_leaf_ref(root, ref, shared); + } WARN_ON(ret); btrfs_free_leaf_ref(root, ref); } @@ -1161,15 +1539,55 @@ out: return ret; } -int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, - struct extent_buffer *orig_buf, struct extent_buffer *buf, - u32 *nr_extents) +/* when a block goes through cow, we update the reference counts of + * everything that block points to. The internal pointers of the block + * can be in just about any order, and it is likely to have clusters of + * things that are close together and clusters of things that are not. + * + * To help reduce the seeks that come with updating all of these reference + * counts, sort them by byte number before actual updates are done. + * + * struct refsort is used to match byte number to slot in the btree block. + * we sort based on the byte number and then use the slot to actually + * find the item. + * + * struct refsort is smaller than strcut btrfs_item and smaller than + * struct btrfs_key_ptr. Since we're currently limited to the page size + * for a btree block, there's no way for a kmalloc of refsorts for a + * single node to be bigger than a page. + */ +struct refsort { + u64 bytenr; + u32 slot; +}; + +/* + * for passing into sort() + */ +static int refsort_cmp(const void *a_void, const void *b_void) +{ + const struct refsort *a = a_void; + const struct refsort *b = b_void; + + if (a->bytenr < b->bytenr) + return -1; + if (a->bytenr > b->bytenr) + return 1; + return 0; +} + + +noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct extent_buffer *orig_buf, + struct extent_buffer *buf, u32 *nr_extents) { u64 bytenr; u64 ref_root; u64 orig_root; u64 ref_generation; u64 orig_generation; + struct refsort *sorted; u32 nritems; u32 nr_file_extents = 0; struct btrfs_key key; @@ -1178,8 +1596,10 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, int level; int ret = 0; int faili = 0; + int refi = 0; + int slot; int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, - u64, u64, u64, u64, u64, u64, u64, u64, u64); + u64, u64, u64, u64, u64, u64, u64, u64); ref_root = btrfs_header_owner(buf); ref_generation = btrfs_header_generation(buf); @@ -1189,6 +1609,9 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, nritems = btrfs_header_nritems(buf); level = btrfs_header_level(buf); + sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS); + BUG_ON(!sorted); + if (root->ref_cows) { process_func = __btrfs_inc_extent_ref; } else { @@ -1201,6 +1624,11 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, process_func = __btrfs_update_extent_ref; } + /* + * we make two passes through the items. In the first pass we + * only record the byte number and slot. Then we sort based on + * byte number and do the actual work based on the sorted results + */ for (i = 0; i < nritems; i++) { cond_resched(); if (level == 0) { @@ -1216,38 +1644,60 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, if (bytenr == 0) continue; - nr_file_extents++; + nr_file_extents++; + sorted[refi].bytenr = bytenr; + sorted[refi].slot = i; + refi++; + } else { + bytenr = btrfs_node_blockptr(buf, i); + sorted[refi].bytenr = bytenr; + sorted[refi].slot = i; + refi++; + } + } + /* + * if refi == 0, we didn't actually put anything into the sorted + * array and we're done + */ + if (refi == 0) + goto out; + + sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); + + for (i = 0; i < refi; i++) { + cond_resched(); + slot = sorted[i].slot; + bytenr = sorted[i].bytenr; + + if (level == 0) { + btrfs_item_key_to_cpu(buf, &key, slot); - maybe_lock_mutex(root); ret = process_func(trans, root, bytenr, orig_buf->start, buf->start, orig_root, ref_root, orig_generation, ref_generation, - key.objectid, key.offset); - maybe_unlock_mutex(root); + key.objectid); if (ret) { - faili = i; + faili = slot; WARN_ON(1); goto fail; } } else { - bytenr = btrfs_node_blockptr(buf, i); - maybe_lock_mutex(root); ret = process_func(trans, root, bytenr, orig_buf->start, buf->start, orig_root, ref_root, orig_generation, ref_generation, - level - 1, 0); - maybe_unlock_mutex(root); + level - 1); if (ret) { - faili = i; + faili = slot; WARN_ON(1); goto fail; } } } out: + kfree(sorted); if (nr_extents) { if (level == 0) *nr_extents = nr_file_extents; @@ -1256,6 +1706,7 @@ out: } return 0; fail: + kfree(sorted); WARN_ON(1); return ret; } @@ -1309,24 +1760,20 @@ int btrfs_update_ref(struct btrfs_trans_handle *trans, bytenr = btrfs_file_extent_disk_bytenr(buf, fi); if (bytenr == 0) continue; - maybe_lock_mutex(root); ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_buf->start, buf->start, orig_root, ref_root, orig_generation, ref_generation, - key.objectid, key.offset); - maybe_unlock_mutex(root); + key.objectid); if (ret) goto fail; } else { bytenr = btrfs_node_blockptr(buf, slot); - maybe_lock_mutex(root); ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_buf->start, buf->start, orig_root, ref_root, orig_generation, ref_generation, - level - 1, 0); - maybe_unlock_mutex(root); + level - 1); if (ret) goto fail; } @@ -1359,8 +1806,8 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans, btrfs_mark_buffer_dirty(leaf); btrfs_release_path(extent_root, path); fail: - finish_current_insert(trans, extent_root); - pending_ret = del_pending_extents(trans, extent_root); + finish_current_insert(trans, extent_root, 0); + pending_ret = del_pending_extents(trans, extent_root, 0); if (ret) return ret; if (pending_ret) @@ -1383,8 +1830,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; - mutex_lock(&root->fs_info->alloc_mutex); - while(1) { + while (1) { cache = NULL; spin_lock(&root->fs_info->block_group_cache_lock); for (n = rb_first(&root->fs_info->block_group_cache_tree); @@ -1417,10 +1863,22 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, } } btrfs_free_path(path); - mutex_unlock(&root->fs_info->alloc_mutex); 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) @@ -1429,24 +1887,28 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags, found = __find_space_info(info, flags); if (found) { + spin_lock(&found->lock); found->total_bytes += total_bytes; found->bytes_used += bytes_used; found->full = 0; + spin_unlock(&found->lock); *space_info = found; return 0; } - found = kmalloc(sizeof(*found), GFP_NOFS); + found = kzalloc(sizeof(*found), GFP_NOFS); if (!found) return -ENOMEM; list_add(&found->list, &info->space_info); INIT_LIST_HEAD(&found->block_groups); + init_rwsem(&found->groups_sem); spin_lock_init(&found->lock); found->flags = flags; found->total_bytes = total_bytes; found->bytes_used = bytes_used; found->bytes_pinned = 0; found->bytes_reserved = 0; + found->bytes_readonly = 0; found->full = 0; found->force_alloc = 0; *space_info = found; @@ -1469,9 +1931,22 @@ static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) } } -static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) +static void set_block_group_readonly(struct btrfs_block_group_cache *cache) { - u64 num_devices = root->fs_info->fs_devices->num_devices; + spin_lock(&cache->space_info->lock); + spin_lock(&cache->lock); + if (!cache->ro) { + cache->space_info->bytes_readonly += cache->key.offset - + btrfs_block_group_used(&cache->item); + cache->ro = 1; + } + spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); +} + +u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) +{ + u64 num_devices = root->fs_info->fs_devices->rw_devices; if (num_devices == 1) flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); @@ -1503,11 +1978,11 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, { struct btrfs_space_info *space_info; u64 thresh; - u64 start; - u64 num_bytes; - int ret = 0, waited = 0; + int ret = 0; + + mutex_lock(&extent_root->fs_info->chunk_mutex); - flags = reduce_alloc_profile(extent_root, flags); + flags = btrfs_reduce_alloc_profile(extent_root, flags); space_info = __find_space_info(extent_root->fs_info, flags); if (!space_info) { @@ -1517,46 +1992,31 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, } BUG_ON(!space_info); + spin_lock(&space_info->lock); if (space_info->force_alloc) { force = 1; space_info->force_alloc = 0; } - if (space_info->full) + if (space_info->full) { + spin_unlock(&space_info->lock); goto out; + } - thresh = div_factor(space_info->total_bytes, 6); + thresh = space_info->total_bytes - space_info->bytes_readonly; + thresh = div_factor(thresh, 6); if (!force && (space_info->bytes_used + space_info->bytes_pinned + - space_info->bytes_reserved + alloc_bytes) < thresh) + space_info->bytes_reserved + alloc_bytes) < thresh) { + spin_unlock(&space_info->lock); goto out; - - while (!mutex_trylock(&extent_root->fs_info->chunk_mutex)) { - if (!force) - goto out; - mutex_unlock(&extent_root->fs_info->alloc_mutex); - cond_resched(); - mutex_lock(&extent_root->fs_info->alloc_mutex); - waited = 1; } + spin_unlock(&space_info->lock); - if (waited && space_info->full) - goto out_unlock; - - ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); - if (ret == -ENOSPC) { -printk("space info full %Lu\n", flags); + ret = btrfs_alloc_chunk(trans, extent_root, flags); + if (ret) space_info->full = 1; - goto out_unlock; - } - BUG_ON(ret); - - ret = btrfs_make_block_group(trans, extent_root, 0, flags, - BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); - BUG_ON(ret); - -out_unlock: - mutex_unlock(&extent_root->fs_info->chunk_mutex); out: + mutex_unlock(&extent_root->fs_info->chunk_mutex); return ret; } @@ -1571,15 +2031,14 @@ static int update_block_group(struct btrfs_trans_handle *trans, u64 old_val; u64 byte_in_group; - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); - while(total) { + while (total) { cache = btrfs_lookup_block_group(info, bytenr); - if (!cache) { + if (!cache) return -1; - } byte_in_group = bytenr - cache->key.objectid; WARN_ON(byte_in_group > cache->key.offset); + spin_lock(&cache->space_info->lock); spin_lock(&cache->lock); cache->dirty = 1; old_val = btrfs_block_group_used(&cache->item); @@ -1587,21 +2046,32 @@ 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) + cache->space_info->bytes_readonly -= num_bytes; btrfs_set_block_group_used(&cache->item, old_val); spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); } else { old_val -= num_bytes; cache->space_info->bytes_used -= num_bytes; + if (cache->ro) + cache->space_info->bytes_readonly += num_bytes; btrfs_set_block_group_used(&cache->item, old_val); spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); if (mark_free) { int ret; + + ret = btrfs_discard_extent(root, bytenr, + num_bytes); + WARN_ON(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; } @@ -1611,12 +2081,16 @@ 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, @@ -1626,7 +2100,7 @@ int btrfs_update_pinned_extents(struct btrfs_root *root, struct btrfs_block_group_cache *cache; struct btrfs_fs_info *fs_info = root->fs_info; - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); + WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex)); if (pin) { set_extent_dirty(&fs_info->pinned_extents, bytenr, bytenr + num - 1, GFP_NOFS); @@ -1640,18 +2114,25 @@ int btrfs_update_pinned_extents(struct btrfs_root *root, len = min(num, cache->key.offset - (bytenr - cache->key.objectid)); if (pin) { + spin_lock(&cache->space_info->lock); spin_lock(&cache->lock); cache->pinned += len; cache->space_info->bytes_pinned += len; spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); fs_info->total_pinned += len; } else { + spin_lock(&cache->space_info->lock); spin_lock(&cache->lock); cache->pinned -= len; cache->space_info->bytes_pinned -= len; spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); fs_info->total_pinned -= len; + if (cache->cached) + btrfs_add_free_space(cache, bytenr, len); } + put_block_group(cache); bytenr += len; num -= len; } @@ -1665,23 +2146,24 @@ static int update_reserved_extents(struct btrfs_root *root, struct btrfs_block_group_cache *cache; struct btrfs_fs_info *fs_info = root->fs_info; - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); while (num > 0) { cache = btrfs_lookup_block_group(fs_info, bytenr); BUG_ON(!cache); len = min(num, cache->key.offset - (bytenr - cache->key.objectid)); + + spin_lock(&cache->space_info->lock); + spin_lock(&cache->lock); if (reserve) { - spin_lock(&cache->lock); cache->reserved += len; cache->space_info->bytes_reserved += len; - spin_unlock(&cache->lock); } else { - spin_lock(&cache->lock); cache->reserved -= len; cache->space_info->bytes_reserved -= len; - spin_unlock(&cache->lock); } + spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); + put_block_group(cache); bytenr += len; num -= len; } @@ -1696,7 +2178,8 @@ int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; int ret; - while(1) { + mutex_lock(&root->fs_info->pinned_mutex); + while (1) { ret = find_first_extent_bit(pinned_extents, last, &start, &end, EXTENT_DIRTY); if (ret) @@ -1704,6 +2187,7 @@ int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) set_extent_dirty(copy, start, end, GFP_NOFS); last = end + 1; } + mutex_unlock(&root->fs_info->pinned_mutex); return 0; } @@ -1714,108 +2198,212 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, u64 start; u64 end; int ret; - struct btrfs_block_group_cache *cache; - mutex_lock(&root->fs_info->alloc_mutex); - while(1) { + mutex_lock(&root->fs_info->pinned_mutex); + while (1) { ret = find_first_extent_bit(unpin, 0, &start, &end, EXTENT_DIRTY); if (ret) break; + + ret = btrfs_discard_extent(root, start, end + 1 - start); + btrfs_update_pinned_extents(root, start, end + 1 - start, 0); clear_extent_dirty(unpin, start, end, GFP_NOFS); - cache = btrfs_lookup_block_group(root->fs_info, start); - if (cache->cached) - btrfs_add_free_space(cache, start, end - start + 1); + if (need_resched()) { - mutex_unlock(&root->fs_info->alloc_mutex); + mutex_unlock(&root->fs_info->pinned_mutex); cond_resched(); - mutex_lock(&root->fs_info->alloc_mutex); + mutex_lock(&root->fs_info->pinned_mutex); } } - mutex_unlock(&root->fs_info->alloc_mutex); - return 0; + mutex_unlock(&root->fs_info->pinned_mutex); + return ret; } static int finish_current_insert(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root) + struct btrfs_root *extent_root, int all) { u64 start; u64 end; u64 priv; + u64 search = 0; struct btrfs_fs_info *info = extent_root->fs_info; struct btrfs_path *path; - struct btrfs_extent_ref *ref; - struct pending_extent_op *extent_op; - struct btrfs_key key; - struct btrfs_extent_item extent_item; + struct pending_extent_op *extent_op, *tmp; + struct list_head insert_list, update_list; int ret; - int err = 0; + int num_inserts = 0, max_inserts, restart = 0; - WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); - btrfs_set_stack_extent_refs(&extent_item, 1); path = btrfs_alloc_path(); + INIT_LIST_HEAD(&insert_list); + INIT_LIST_HEAD(&update_list); - while(1) { - ret = find_first_extent_bit(&info->extent_ins, 0, &start, - &end, EXTENT_LOCKED); - if (ret) + max_inserts = extent_root->leafsize / + (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) + + sizeof(struct btrfs_extent_ref) + + sizeof(struct btrfs_extent_item)); +again: + mutex_lock(&info->extent_ins_mutex); + while (1) { + ret = find_first_extent_bit(&info->extent_ins, search, &start, + &end, EXTENT_WRITEBACK); + if (ret) { + if (restart && !num_inserts && + list_empty(&update_list)) { + restart = 0; + search = 0; + continue; + } break; + } + + ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS); + if (!ret) { + if (all) + restart = 1; + search = end + 1; + if (need_resched()) { + mutex_unlock(&info->extent_ins_mutex); + cond_resched(); + mutex_lock(&info->extent_ins_mutex); + } + continue; + } ret = get_state_private(&info->extent_ins, start, &priv); BUG_ON(ret); - extent_op = (struct pending_extent_op *)(unsigned long)priv; + extent_op = (struct pending_extent_op *)(unsigned long) priv; if (extent_op->type == PENDING_EXTENT_INSERT) { - key.objectid = start; - key.offset = end + 1 - start; - key.type = BTRFS_EXTENT_ITEM_KEY; - err = btrfs_insert_item(trans, extent_root, &key, - &extent_item, sizeof(extent_item)); - BUG_ON(err); - - clear_extent_bits(&info->extent_ins, start, end, - EXTENT_LOCKED, GFP_NOFS); - - err = insert_extent_backref(trans, extent_root, path, - start, extent_op->parent, - extent_root->root_key.objectid, - extent_op->generation, - extent_op->level, 0); - BUG_ON(err); + num_inserts++; + list_add_tail(&extent_op->list, &insert_list); + search = end + 1; + if (num_inserts == max_inserts) { + restart = 1; + break; + } } else if (extent_op->type == PENDING_BACKREF_UPDATE) { - err = lookup_extent_backref(trans, extent_root, path, - start, extent_op->orig_parent, - extent_root->root_key.objectid, - extent_op->orig_generation, 0); - BUG_ON(err); - - clear_extent_bits(&info->extent_ins, start, end, - EXTENT_LOCKED, GFP_NOFS); - - key.objectid = start; - key.offset = extent_op->parent; - key.type = BTRFS_EXTENT_REF_KEY; - err = btrfs_set_item_key_safe(trans, extent_root, path, - &key); - BUG_ON(err); - ref = btrfs_item_ptr(path->nodes[0], path->slots[0], - struct btrfs_extent_ref); - btrfs_set_ref_generation(path->nodes[0], ref, - extent_op->generation); - btrfs_mark_buffer_dirty(path->nodes[0]); - btrfs_release_path(extent_root, path); + list_add_tail(&extent_op->list, &update_list); + search = end + 1; } else { - BUG_ON(1); + BUG(); + } + } + + /* + * process the update list, clear the writeback bit for it, and if + * somebody marked this thing for deletion then just unlock it and be + * done, the free_extents will handle it + */ + list_for_each_entry_safe(extent_op, tmp, &update_list, list) { + clear_extent_bits(&info->extent_ins, extent_op->bytenr, + extent_op->bytenr + extent_op->num_bytes - 1, + EXTENT_WRITEBACK, GFP_NOFS); + if (extent_op->del) { + list_del_init(&extent_op->list); + unlock_extent(&info->extent_ins, extent_op->bytenr, + extent_op->bytenr + extent_op->num_bytes + - 1, GFP_NOFS); + kfree(extent_op); } - kfree(extent_op); + } + mutex_unlock(&info->extent_ins_mutex); - if (need_resched()) { - mutex_unlock(&extent_root->fs_info->alloc_mutex); - cond_resched(); - mutex_lock(&extent_root->fs_info->alloc_mutex); + /* + * still have things left on the update list, go ahead an update + * everything + */ + if (!list_empty(&update_list)) { + ret = update_backrefs(trans, extent_root, path, &update_list); + BUG_ON(ret); + + /* we may have COW'ed new blocks, so lets start over */ + if (all) + restart = 1; + } + + /* + * if no inserts need to be done, but we skipped some extents and we + * need to make sure everything is cleaned then reset everything and + * go back to the beginning + */ + if (!num_inserts && restart) { + search = 0; + restart = 0; + INIT_LIST_HEAD(&update_list); + INIT_LIST_HEAD(&insert_list); + goto again; + } else if (!num_inserts) { + goto out; + } + + /* + * process the insert extents list. Again if we are deleting this + * extent, then just unlock it, pin down the bytes if need be, and be + * done with it. Saves us from having to actually insert the extent + * into the tree and then subsequently come along and delete it + */ + mutex_lock(&info->extent_ins_mutex); + list_for_each_entry_safe(extent_op, tmp, &insert_list, list) { + clear_extent_bits(&info->extent_ins, extent_op->bytenr, + extent_op->bytenr + extent_op->num_bytes - 1, + EXTENT_WRITEBACK, GFP_NOFS); + if (extent_op->del) { + u64 used; + list_del_init(&extent_op->list); + unlock_extent(&info->extent_ins, extent_op->bytenr, + extent_op->bytenr + extent_op->num_bytes + - 1, GFP_NOFS); + + mutex_lock(&extent_root->fs_info->pinned_mutex); + ret = pin_down_bytes(trans, extent_root, + extent_op->bytenr, + extent_op->num_bytes, 0); + mutex_unlock(&extent_root->fs_info->pinned_mutex); + + spin_lock(&info->delalloc_lock); + used = btrfs_super_bytes_used(&info->super_copy); + btrfs_set_super_bytes_used(&info->super_copy, + used - extent_op->num_bytes); + used = btrfs_root_used(&extent_root->root_item); + btrfs_set_root_used(&extent_root->root_item, + used - extent_op->num_bytes); + spin_unlock(&info->delalloc_lock); + + ret = update_block_group(trans, extent_root, + extent_op->bytenr, + extent_op->num_bytes, + 0, ret > 0); + BUG_ON(ret); + kfree(extent_op); + num_inserts--; } } + mutex_unlock(&info->extent_ins_mutex); + + ret = insert_extents(trans, extent_root, path, &insert_list, + num_inserts); + BUG_ON(ret); + + /* + * if restart is set for whatever reason we need to go back and start + * searching through the pending list again. + * + * We just inserted some extents, which could have resulted in new + * blocks being allocated, which would result in new blocks needing + * updates, so if all is set we _must_ restart to get the updated + * blocks. + */ + if (restart || all) { + INIT_LIST_HEAD(&insert_list); + INIT_LIST_HEAD(&update_list); + search = 0; + restart = 0; + num_inserts = 0; + goto again; + } +out: btrfs_free_path(path); return 0; } @@ -1827,7 +2415,6 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans, int err = 0; struct extent_buffer *buf; - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); if (is_data) goto pinit; @@ -1870,8 +2457,7 @@ static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner_objectid, u64 owner_offset, - int pin, int mark_free) + u64 owner_objectid, int pin, int mark_free) { struct btrfs_path *path; struct btrfs_key key; @@ -1885,7 +2471,6 @@ static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_extent_item *ei; u32 refs; - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); key.objectid = bytenr; btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); key.offset = num_bytes; @@ -1894,12 +2479,13 @@ static int __free_extent(struct btrfs_trans_handle *trans, return -ENOMEM; path->reada = 1; - ret = lookup_extent_backref(trans, extent_root, path, bytenr, parent, - root_objectid, ref_generation, 1); + ret = lookup_extent_backref(trans, extent_root, path, + bytenr, parent, root_objectid, + ref_generation, owner_objectid, 1); if (ret == 0) { struct btrfs_key found_key; extent_slot = path->slots[0]; - while(extent_slot > 0) { + while (extent_slot > 0) { extent_slot--; btrfs_item_key_to_cpu(path->nodes[0], &found_key, extent_slot); @@ -1919,16 +2505,24 @@ static int __free_extent(struct btrfs_trans_handle *trans, btrfs_release_path(extent_root, path); ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); + if (ret) { + printk(KERN_ERR "umm, got %d back from search" + ", was looking for %llu\n", ret, + (unsigned long long)bytenr); + btrfs_print_leaf(extent_root, path->nodes[0]); + } BUG_ON(ret); extent_slot = path->slots[0]; } } else { btrfs_print_leaf(extent_root, path->nodes[0]); WARN_ON(1); - printk("Unable to find ref byte nr %Lu root %Lu " - " gen %Lu owner %Lu offset %Lu\n", bytenr, - root_objectid, ref_generation, owner_objectid, - owner_offset); + printk(KERN_ERR "btrfs unable to find ref byte nr %llu " + "root %llu gen %llu owner %llu\n", + (unsigned long long)bytenr, + (unsigned long long)root_objectid, + (unsigned long long)ref_generation, + (unsigned long long)owner_objectid); } leaf = path->nodes[0]; @@ -1967,59 +2561,43 @@ static int __free_extent(struct btrfs_trans_handle *trans, if (refs == 0) { u64 super_used; u64 root_used; -#ifdef BIO_RW_DISCARD - u64 map_length = num_bytes; - struct btrfs_multi_bio *multi = NULL; -#endif if (pin) { + mutex_lock(&root->fs_info->pinned_mutex); ret = pin_down_bytes(trans, root, bytenr, num_bytes, owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); + mutex_unlock(&root->fs_info->pinned_mutex); if (ret > 0) mark_free = 1; BUG_ON(ret < 0); } - /* block accounting for super block */ - spin_lock_irq(&info->delalloc_lock); + spin_lock(&info->delalloc_lock); super_used = btrfs_super_bytes_used(&info->super_copy); btrfs_set_super_bytes_used(&info->super_copy, super_used - num_bytes); - spin_unlock_irq(&info->delalloc_lock); /* block accounting for root item */ root_used = btrfs_root_used(&root->root_item); btrfs_set_root_used(&root->root_item, root_used - num_bytes); + spin_unlock(&info->delalloc_lock); ret = btrfs_del_items(trans, extent_root, path, path->slots[0], num_to_del); BUG_ON(ret); + btrfs_release_path(extent_root, path); + + if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { + ret = btrfs_del_csums(trans, root, bytenr, num_bytes); + BUG_ON(ret); + } + ret = update_block_group(trans, root, bytenr, num_bytes, 0, mark_free); 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, - bytenr, &map_length, &multi, 0); - if (!ret) { - struct btrfs_bio_stripe *stripe = multi->stripes; - int i; - - if (map_length > num_bytes) - map_length = num_bytes; - - for (i = 0; i < multi->num_stripes; i++, stripe++) { - blkdev_issue_discard(stripe->dev->bdev, - stripe->physical >> 9, - map_length >> 9); - } - kfree(multi); - } -#endif } btrfs_free_path(path); - finish_current_insert(trans, extent_root); + finish_current_insert(trans, extent_root, 0); return ret; } @@ -2027,76 +2605,123 @@ static int __free_extent(struct btrfs_trans_handle *trans, * find all the blocks marked as pending in the radix tree and remove * them from the extent map */ -static int del_pending_extents(struct btrfs_trans_handle *trans, struct - btrfs_root *extent_root) +static int del_pending_extents(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, int all) { int ret; int err = 0; - int mark_free = 0; u64 start; u64 end; u64 priv; + u64 search = 0; + int nr = 0, skipped = 0; struct extent_io_tree *pending_del; struct extent_io_tree *extent_ins; struct pending_extent_op *extent_op; + struct btrfs_fs_info *info = extent_root->fs_info; + struct list_head delete_list; - WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); + INIT_LIST_HEAD(&delete_list); extent_ins = &extent_root->fs_info->extent_ins; pending_del = &extent_root->fs_info->pending_del; - while(1) { - ret = find_first_extent_bit(pending_del, 0, &start, &end, - EXTENT_LOCKED); - if (ret) +again: + mutex_lock(&info->extent_ins_mutex); + while (1) { + ret = find_first_extent_bit(pending_del, search, &start, &end, + EXTENT_WRITEBACK); + if (ret) { + if (all && skipped && !nr) { + search = 0; + skipped = 0; + continue; + } + mutex_unlock(&info->extent_ins_mutex); break; + } + + ret = try_lock_extent(extent_ins, start, end, GFP_NOFS); + if (!ret) { + search = end+1; + skipped = 1; + + if (need_resched()) { + mutex_unlock(&info->extent_ins_mutex); + cond_resched(); + mutex_lock(&info->extent_ins_mutex); + } + + continue; + } + BUG_ON(ret < 0); ret = get_state_private(pending_del, start, &priv); BUG_ON(ret); extent_op = (struct pending_extent_op *)(unsigned long)priv; - clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, + clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK, GFP_NOFS); - - ret = pin_down_bytes(trans, extent_root, start, - end + 1 - start, 0); - mark_free = ret > 0; if (!test_range_bit(extent_ins, start, end, - EXTENT_LOCKED, 0)) { -free_extent: - ret = __free_extent(trans, extent_root, - start, end + 1 - start, - extent_op->orig_parent, - extent_root->root_key.objectid, - extent_op->orig_generation, - extent_op->level, 0, 0, mark_free); - kfree(extent_op); + EXTENT_WRITEBACK, 0)) { + list_add_tail(&extent_op->list, &delete_list); + nr++; } else { kfree(extent_op); - ret = get_state_private(extent_ins, start, &priv); + + ret = get_state_private(&info->extent_ins, start, + &priv); BUG_ON(ret); extent_op = (struct pending_extent_op *) - (unsigned long)priv; + (unsigned long)priv; + + clear_extent_bits(&info->extent_ins, start, end, + EXTENT_WRITEBACK, GFP_NOFS); - clear_extent_bits(extent_ins, start, end, - EXTENT_LOCKED, GFP_NOFS); + if (extent_op->type == PENDING_BACKREF_UPDATE) { + list_add_tail(&extent_op->list, &delete_list); + search = end + 1; + nr++; + continue; + } - if (extent_op->type == PENDING_BACKREF_UPDATE) - goto free_extent; + mutex_lock(&extent_root->fs_info->pinned_mutex); + ret = pin_down_bytes(trans, extent_root, start, + end + 1 - start, 0); + mutex_unlock(&extent_root->fs_info->pinned_mutex); ret = update_block_group(trans, extent_root, start, - end + 1 - start, 0, mark_free); + end + 1 - start, 0, ret > 0); + + unlock_extent(extent_ins, start, end, GFP_NOFS); BUG_ON(ret); kfree(extent_op); } if (ret) err = ret; + search = end + 1; + if (need_resched()) { - mutex_unlock(&extent_root->fs_info->alloc_mutex); + mutex_unlock(&info->extent_ins_mutex); cond_resched(); - mutex_lock(&extent_root->fs_info->alloc_mutex); + mutex_lock(&info->extent_ins_mutex); } } + + if (nr) { + ret = free_extents(trans, extent_root, &delete_list); + BUG_ON(ret); + } + + if (all && skipped) { + INIT_LIST_HEAD(&delete_list); + search = 0; + nr = 0; + goto again; + } + + if (!err) + finish_current_insert(trans, extent_root, 0); return err; } @@ -2107,7 +2732,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner_objectid, u64 owner_offset, int pin) + u64 owner_objectid, int pin) { struct btrfs_root *extent_root = root->fs_info->extent_root; int pending_ret; @@ -2115,7 +2740,29 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans, WARN_ON(num_bytes < root->sectorsize); if (root == extent_root) { - struct pending_extent_op *extent_op; + struct pending_extent_op *extent_op = NULL; + + mutex_lock(&root->fs_info->extent_ins_mutex); + if (test_range_bit(&root->fs_info->extent_ins, bytenr, + bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { + u64 priv; + ret = get_state_private(&root->fs_info->extent_ins, + bytenr, &priv); + BUG_ON(ret); + extent_op = (struct pending_extent_op *) + (unsigned long)priv; + + extent_op->del = 1; + if (extent_op->type == PENDING_EXTENT_INSERT) { + mutex_unlock(&root->fs_info->extent_ins_mutex); + return 0; + } + } + + if (extent_op) { + ref_generation = extent_op->orig_generation; + parent = extent_op->orig_parent; + } extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); BUG_ON(!extent_op); @@ -2128,23 +2775,23 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans, extent_op->generation = ref_generation; extent_op->orig_generation = ref_generation; extent_op->level = (int)owner_objectid; + INIT_LIST_HEAD(&extent_op->list); + extent_op->del = 0; set_extent_bits(&root->fs_info->pending_del, bytenr, bytenr + num_bytes - 1, - EXTENT_LOCKED, GFP_NOFS); + EXTENT_WRITEBACK, GFP_NOFS); set_state_private(&root->fs_info->pending_del, bytenr, (unsigned long)extent_op); + mutex_unlock(&root->fs_info->extent_ins_mutex); return 0; } /* if metadata always pin */ if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { - struct btrfs_block_group_cache *cache; - - /* btrfs_free_reserved_extent */ - cache = btrfs_lookup_block_group(root->fs_info, bytenr); - BUG_ON(!cache); - btrfs_add_free_space(cache, bytenr, num_bytes); + mutex_lock(&root->fs_info->pinned_mutex); + btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); + mutex_unlock(&root->fs_info->pinned_mutex); update_reserved_extents(root, bytenr, num_bytes, 0); return 0; } @@ -2156,11 +2803,11 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans, pin = 1; ret = __free_extent(trans, root, bytenr, num_bytes, parent, - root_objectid, ref_generation, owner_objectid, - owner_offset, pin, pin == 0); + root_objectid, ref_generation, + owner_objectid, pin, pin == 0); - finish_current_insert(trans, root->fs_info->extent_root); - pending_ret = del_pending_extents(trans, root->fs_info->extent_root); + finish_current_insert(trans, root->fs_info->extent_root, 0); + pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); return ret ? ret : pending_ret; } @@ -2168,15 +2815,13 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner_objectid, u64 owner_offset, int pin) + u64 owner_objectid, int pin) { int ret; - maybe_lock_mutex(root); ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, root_objectid, ref_generation, - owner_objectid, owner_offset, pin); - maybe_unlock_mutex(root); + owner_objectid, pin); return ret; } @@ -2195,7 +2840,7 @@ static u64 stripe_align(struct btrfs_root *root, u64 val) * ins->offset == number of blocks * Any available blocks before search_start are skipped. */ -static int noinline find_free_extent(struct btrfs_trans_handle *trans, +static noinline int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *orig_root, u64 num_bytes, u64 empty_size, u64 search_start, u64 search_end, @@ -2203,208 +2848,278 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans, u64 exclude_start, u64 exclude_nr, int data) { - int ret; - u64 orig_search_start; - struct btrfs_root * root = orig_root->fs_info->extent_root; - struct btrfs_fs_info *info = root->fs_info; + int ret = 0; + struct btrfs_root *root = orig_root->fs_info->extent_root; u64 total_needed = num_bytes; u64 *last_ptr = NULL; - struct btrfs_block_group_cache *block_group; + u64 last_wanted = 0; + struct btrfs_block_group_cache *block_group = NULL; int chunk_alloc_done = 0; int empty_cluster = 2 * 1024 * 1024; int allowed_chunk_alloc = 0; + struct list_head *head = NULL, *cur = NULL; + int loop = 0; + int extra_loop = 0; + struct btrfs_space_info *space_info; WARN_ON(num_bytes < root->sectorsize); btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); + ins->objectid = 0; + ins->offset = 0; if (orig_root->ref_cows || empty_size) allowed_chunk_alloc = 1; if (data & BTRFS_BLOCK_GROUP_METADATA) { last_ptr = &root->fs_info->last_alloc; - empty_cluster = 256 * 1024; + if (!btrfs_test_opt(root, SSD)) + empty_cluster = 64 * 1024; } if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) last_ptr = &root->fs_info->last_data_alloc; if (last_ptr) { - if (*last_ptr) + if (*last_ptr) { hint_byte = *last_ptr; - else + last_wanted = *last_ptr; + } else empty_size += empty_cluster; + } else { + empty_cluster = 0; } - search_start = max(search_start, first_logical_byte(root, 0)); - orig_search_start = search_start; - search_start = max(search_start, hint_byte); - total_needed += empty_size; -new_group: - block_group = btrfs_lookup_block_group(info, search_start); + if (last_wanted && search_start != last_wanted) { + last_wanted = 0; + empty_size += empty_cluster; + } + + total_needed += empty_size; + block_group = btrfs_lookup_block_group(root->fs_info, search_start); if (!block_group) - block_group = btrfs_lookup_first_block_group(info, + block_group = btrfs_lookup_first_block_group(root->fs_info, search_start); + space_info = __find_space_info(root->fs_info, data); - /* - * Ok this looks a little tricky, buts its really simple. First if we - * didn't find a block group obviously we want to start over. - * Secondly, if the block group we found does not match the type we - * need, and we have a last_ptr and its not 0, chances are the last - * allocation we made was at the end of the block group, so lets go - * ahead and skip the looking through the rest of the block groups and - * start at the beginning. This helps with metadata allocations, - * since you are likely to have a bunch of data block groups to search - * through first before you realize that you need to start over, so go - * ahead and start over and save the time. - */ - if (!block_group || (!block_group_bits(block_group, data) && - last_ptr && *last_ptr)) { - if (search_start != orig_search_start) { - if (last_ptr && *last_ptr) { - total_needed += empty_cluster; - *last_ptr = 0; - } - search_start = orig_search_start; - goto new_group; - } else if (!chunk_alloc_done && allowed_chunk_alloc) { - ret = do_chunk_alloc(trans, root, - num_bytes + 2 * 1024 * 1024, - data, 1); - if (ret < 0) - goto error; - BUG_ON(ret); - chunk_alloc_done = 1; - search_start = orig_search_start; - goto new_group; - } else { - ret = -ENOSPC; - goto error; - } - } - - /* - * this is going to seach through all of the existing block groups it - * can find, so if we don't find something we need to see if we can - * allocate what we need. - */ - ret = find_free_space(root, &block_group, &search_start, - total_needed, data); - if (ret == -ENOSPC) { + down_read(&space_info->groups_sem); + while (1) { + struct btrfs_free_space *free_space; /* - * instead of allocating, start at the original search start - * and see if there is something to be found, if not then we - * allocate + * the only way this happens if our hint points to a block + * group thats not of the proper type, while looping this + * should never happen */ - if (search_start != orig_search_start) { - if (last_ptr && *last_ptr) { - *last_ptr = 0; - total_needed += empty_cluster; - } - search_start = orig_search_start; - goto new_group; - } + if (empty_size) + extra_loop = 1; - /* - * we've already allocated, we're pretty screwed - */ - if (chunk_alloc_done) { - goto error; - } else if (!allowed_chunk_alloc && block_group && - block_group_bits(block_group, data)) { - block_group->space_info->force_alloc = 1; - goto error; - } else if (!allowed_chunk_alloc) { - goto error; + if (!block_group) + goto new_group_no_lock; + + if (unlikely(!block_group->cached)) { + mutex_lock(&block_group->cache_mutex); + ret = cache_block_group(root, block_group); + mutex_unlock(&block_group->cache_mutex); + if (ret) + break; } - ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024, - data, 1); - if (ret < 0) - goto error; + mutex_lock(&block_group->alloc_mutex); + if (unlikely(!block_group_bits(block_group, data))) + goto new_group; + + if (unlikely(block_group->ro)) + goto new_group; + + free_space = btrfs_find_free_space(block_group, search_start, + total_needed); + if (free_space) { + u64 start = block_group->key.objectid; + u64 end = block_group->key.objectid + + block_group->key.offset; + + search_start = stripe_align(root, free_space->offset); + + /* move on to the next group */ + if (search_start + num_bytes >= search_end) + goto new_group; + + /* move on to the next group */ + if (search_start + num_bytes > end) + goto new_group; + + if (last_wanted && search_start != last_wanted) { + total_needed += empty_cluster; + empty_size += empty_cluster; + last_wanted = 0; + /* + * if search_start is still in this block group + * then we just re-search this block group + */ + if (search_start >= start && + search_start < end) { + mutex_unlock(&block_group->alloc_mutex); + continue; + } + + /* else we go to the next block group */ + goto new_group; + } - BUG_ON(ret); - chunk_alloc_done = 1; - if (block_group) - search_start = block_group->key.objectid + - block_group->key.offset; - else - search_start = orig_search_start; - goto new_group; - } + if (exclude_nr > 0 && + (search_start + num_bytes > exclude_start && + search_start < exclude_start + exclude_nr)) { + search_start = exclude_start + exclude_nr; + /* + * if search_start is still in this block group + * then we just re-search this block group + */ + if (search_start >= start && + search_start < end) { + mutex_unlock(&block_group->alloc_mutex); + last_wanted = 0; + continue; + } - if (ret) - goto error; + /* else we go to the next block group */ + goto new_group; + } - search_start = stripe_align(root, search_start); - ins->objectid = search_start; - ins->offset = num_bytes; + ins->objectid = search_start; + ins->offset = num_bytes; - if (ins->objectid + num_bytes >= search_end) { - search_start = orig_search_start; - if (chunk_alloc_done) { - ret = -ENOSPC; - goto error; + btrfs_remove_free_space_lock(block_group, search_start, + num_bytes); + /* we are all good, lets return */ + mutex_unlock(&block_group->alloc_mutex); + break; } - goto new_group; - } +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 + */ + last_wanted = 0; - if (ins->objectid + num_bytes > - block_group->key.objectid + block_group->key.offset) { - if (search_start == orig_search_start && chunk_alloc_done) { - ret = -ENOSPC; - goto error; + /* + * Here's how this works. + * loop == 0: we were searching a block group via a hint + * and didn't find anything, so we start at + * the head of the block groups and keep searching + * loop == 1: we're searching through all of the block groups + * if we hit the head again we have searched + * all of the block groups for this space and we + * need to try and allocate, if we cant error out. + * loop == 2: we allocated more space and are looping through + * all of the block groups again. + */ + if (loop == 0) { + head = &space_info->block_groups; + cur = head->next; + loop++; + } else if (loop == 1 && cur == head) { + int keep_going; + + /* at this point we give up on the empty_size + * allocations and just try to allocate the min + * space. + * + * The extra_loop field was set if an empty_size + * allocation was attempted above, and if this + * is try we need to try the loop again without + * the additional empty_size. + */ + total_needed -= empty_size; + empty_size = 0; + keep_going = extra_loop; + loop++; + + if (allowed_chunk_alloc && !chunk_alloc_done) { + up_read(&space_info->groups_sem); + ret = do_chunk_alloc(trans, root, num_bytes + + 2 * 1024 * 1024, data, 1); + down_read(&space_info->groups_sem); + if (ret < 0) + goto loop_check; + head = &space_info->block_groups; + /* + * we've allocated a new chunk, keep + * trying + */ + keep_going = 1; + chunk_alloc_done = 1; + } else if (!allowed_chunk_alloc) { + space_info->force_alloc = 1; + } +loop_check: + if (keep_going) { + cur = head->next; + extra_loop = 0; + } else { + break; + } + } else if (cur == head) { + break; } - search_start = block_group->key.objectid + - block_group->key.offset; - goto new_group; - } - - if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && - ins->objectid < exclude_start + exclude_nr)) { - search_start = exclude_start + exclude_nr; - goto new_group; - } - if (!(data & BTRFS_BLOCK_GROUP_DATA)) - trans->block_group = block_group; + block_group = list_entry(cur, struct btrfs_block_group_cache, + list); + atomic_inc(&block_group->count); - ins->offset = num_bytes; - if (last_ptr) { - *last_ptr = ins->objectid + ins->offset; - if (*last_ptr == - btrfs_super_total_bytes(&root->fs_info->super_copy)) - *last_ptr = 0; + search_start = block_group->key.objectid; + cur = cur->next; } - ret = 0; -error: + /* we found what we needed */ + if (ins->objectid) { + if (!(data & BTRFS_BLOCK_GROUP_DATA)) + trans->block_group = block_group->key.objectid; + + if (last_ptr) + *last_ptr = ins->objectid + ins->offset; + ret = 0; + } else if (!ret) { + printk(KERN_ERR "btrfs searching for %llu bytes, " + "num_bytes %llu, loop %d, allowed_alloc %d\n", + (unsigned long long)total_needed, + (unsigned long long)num_bytes, + loop, allowed_chunk_alloc); + ret = -ENOSPC; + } + if (block_group) + put_block_group(block_group); + + up_read(&space_info->groups_sem); return ret; } static void dump_space_info(struct btrfs_space_info *info, u64 bytes) { struct btrfs_block_group_cache *cache; - struct list_head *l; - printk(KERN_INFO "space_info has %Lu free, is %sfull\n", - info->total_bytes - info->bytes_used - info->bytes_pinned - - info->bytes_reserved, (info->full) ? "" : "not "); + printk(KERN_INFO "space_info has %llu free, is %sfull\n", + (unsigned long long)(info->total_bytes - info->bytes_used - + info->bytes_pinned - info->bytes_reserved), + (info->full) ? "" : "not "); - spin_lock(&info->lock); - list_for_each(l, &info->block_groups) { - cache = list_entry(l, struct btrfs_block_group_cache, list); + down_read(&info->groups_sem); + list_for_each_entry(cache, &info->block_groups, list) { spin_lock(&cache->lock); - printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used " - "%Lu pinned %Lu reserved\n", - cache->key.objectid, cache->key.offset, - btrfs_block_group_used(&cache->item), - cache->pinned, cache->reserved); + printk(KERN_INFO "block group %llu has %llu bytes, %llu used " + "%llu pinned %llu reserved\n", + (unsigned long long)cache->key.objectid, + (unsigned long long)cache->key.offset, + (unsigned long long)btrfs_block_group_used(&cache->item), + (unsigned long long)cache->pinned, + (unsigned long long)cache->reserved); btrfs_dump_free_space(cache, bytes); spin_unlock(&cache->lock); } - spin_unlock(&info->lock); + up_read(&info->groups_sem); } static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, @@ -2418,23 +3133,22 @@ static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, u64 search_start = 0; u64 alloc_profile; struct btrfs_fs_info *info = root->fs_info; - struct btrfs_block_group_cache *cache; if (data) { alloc_profile = info->avail_data_alloc_bits & - info->data_alloc_profile; + info->data_alloc_profile; data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; } else if (root == root->fs_info->chunk_root) { alloc_profile = info->avail_system_alloc_bits & - info->system_alloc_profile; + info->system_alloc_profile; data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; } else { alloc_profile = info->avail_metadata_alloc_bits & - info->metadata_alloc_profile; + info->metadata_alloc_profile; data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; } again: - data = reduce_alloc_profile(root, data); + data = btrfs_reduce_alloc_profile(root, data); /* * the only place that sets empty_size is btrfs_realloc_node, which * is not called recursively on allocations @@ -2469,18 +3183,12 @@ again: struct btrfs_space_info *sinfo; sinfo = __find_space_info(root->fs_info, data); - printk("allocation failed flags %Lu, wanted %Lu\n", - data, num_bytes); + printk(KERN_ERR "btrfs allocation failed flags %llu, " + "wanted %llu\n", (unsigned long long)data, + (unsigned long long)num_bytes); dump_space_info(sinfo, num_bytes); BUG(); } - cache = btrfs_lookup_block_group(root->fs_info, ins->objectid); - if (!cache) { - printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid); - return -ENOSPC; - } - - ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset); return ret; } @@ -2488,18 +3196,22 @@ again: int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) { struct btrfs_block_group_cache *cache; + int ret = 0; - maybe_lock_mutex(root); cache = btrfs_lookup_block_group(root->fs_info, start); if (!cache) { - printk(KERN_ERR "Unable to find block group for %Lu\n", start); - maybe_unlock_mutex(root); + printk(KERN_ERR "Unable to find block group for %llu\n", + (unsigned long long)start); return -ENOSPC; } + + ret = btrfs_discard_extent(root, start, len); + btrfs_add_free_space(cache, start, len); + put_block_group(cache); update_reserved_extents(root, start, len, 0); - maybe_unlock_mutex(root); - return 0; + + return ret; } int btrfs_reserve_extent(struct btrfs_trans_handle *trans, @@ -2510,20 +3222,17 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans, u64 data) { int ret; - maybe_lock_mutex(root); ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, empty_size, hint_byte, search_end, ins, data); update_reserved_extents(root, ins->objectid, ins->offset, 1); - maybe_unlock_mutex(root); return ret; } static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner, u64 owner_offset, - struct btrfs_key *ins) + u64 owner, struct btrfs_key *ins) { int ret; int pending_ret; @@ -2542,14 +3251,14 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, parent = ins->objectid; /* block accounting for super block */ - spin_lock_irq(&info->delalloc_lock); + spin_lock(&info->delalloc_lock); super_used = btrfs_super_bytes_used(&info->super_copy); btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); - spin_unlock_irq(&info->delalloc_lock); /* block accounting for root item */ root_used = btrfs_root_used(&root->root_item); btrfs_set_root_used(&root->root_item, root_used + num_bytes); + spin_unlock(&info->delalloc_lock); if (root == extent_root) { struct pending_extent_op *extent_op; @@ -2565,12 +3274,16 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, extent_op->generation = ref_generation; extent_op->orig_generation = 0; extent_op->level = (int)owner; + INIT_LIST_HEAD(&extent_op->list); + extent_op->del = 0; + mutex_lock(&root->fs_info->extent_ins_mutex); set_extent_bits(&root->fs_info->extent_ins, ins->objectid, ins->objectid + ins->offset - 1, - EXTENT_LOCKED, GFP_NOFS); + EXTENT_WRITEBACK, GFP_NOFS); set_state_private(&root->fs_info->extent_ins, ins->objectid, (unsigned long)extent_op); + mutex_unlock(&root->fs_info->extent_ins_mutex); goto update_block; } @@ -2597,7 +3310,6 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, btrfs_set_ref_root(path->nodes[0], ref, root_objectid); btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); btrfs_set_ref_objectid(path->nodes[0], ref, owner); - btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); btrfs_set_ref_num_refs(path->nodes[0], ref, 1); btrfs_mark_buffer_dirty(path->nodes[0]); @@ -2605,8 +3317,8 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, trans->alloc_exclude_start = 0; trans->alloc_exclude_nr = 0; btrfs_free_path(path); - finish_current_insert(trans, extent_root); - pending_ret = del_pending_extents(trans, extent_root); + finish_current_insert(trans, extent_root, 0); + pending_ret = del_pending_extents(trans, extent_root, 0); if (ret) goto out; @@ -2616,10 +3328,12 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, } update_block: - ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); + ret = update_block_group(trans, root, ins->objectid, + ins->offset, 1, 0); if (ret) { - printk("update block group failed for %Lu %Lu\n", - ins->objectid, ins->offset); + printk(KERN_ERR "btrfs update block group failed for %llu " + "%llu\n", (unsigned long long)ins->objectid, + (unsigned long long)ins->offset); BUG(); } out: @@ -2629,19 +3343,15 @@ out: int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner, u64 owner_offset, - struct btrfs_key *ins) + u64 owner, struct btrfs_key *ins) { int ret; if (root_objectid == BTRFS_TREE_LOG_OBJECTID) return 0; - maybe_lock_mutex(root); - ret = __btrfs_alloc_reserved_extent(trans, root, parent, - root_objectid, ref_generation, - owner, owner_offset, ins); + ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, + ref_generation, owner, ins); update_reserved_extents(root, ins->objectid, ins->offset, 0); - maybe_unlock_mutex(root); return ret; } @@ -2653,22 +3363,22 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner, u64 owner_offset, - struct btrfs_key *ins) + u64 owner, struct btrfs_key *ins) { int ret; struct btrfs_block_group_cache *block_group; - maybe_lock_mutex(root); block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); + mutex_lock(&block_group->cache_mutex); cache_block_group(root, block_group); + mutex_unlock(&block_group->cache_mutex); - ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset); + ret = btrfs_remove_free_space(block_group, ins->objectid, + ins->offset); BUG_ON(ret); - ret = __btrfs_alloc_reserved_extent(trans, root, parent, - root_objectid, ref_generation, - owner, owner_offset, ins); - maybe_unlock_mutex(root); + put_block_group(block_group); + ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, + ref_generation, owner, ins); return ret; } @@ -2683,14 +3393,11 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 num_bytes, u64 parent, u64 min_alloc_size, u64 root_objectid, u64 ref_generation, - u64 owner_objectid, u64 owner_offset, - u64 empty_size, u64 hint_byte, + u64 owner_objectid, u64 empty_size, u64 hint_byte, u64 search_end, struct btrfs_key *ins, u64 data) { int ret; - maybe_lock_mutex(root); - ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, empty_size, hint_byte, search_end, ins, data); @@ -2698,19 +3405,19 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans, if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, ref_generation, - owner_objectid, owner_offset, ins); + owner_objectid, ins); BUG_ON(ret); } else { update_reserved_extents(root, ins->objectid, ins->offset, 1); } - maybe_unlock_mutex(root); return ret; } struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, - u64 bytenr, u32 blocksize) + u64 bytenr, u32 blocksize, + int level) { struct extent_buffer *buf; @@ -2718,9 +3425,13 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, if (!buf) return ERR_PTR(-ENOMEM); btrfs_set_header_generation(buf, trans->transid); + btrfs_set_buffer_lockdep_class(buf, level); btrfs_tree_lock(buf); clean_tree_block(trans, root, buf); + + btrfs_set_lock_blocking(buf); btrfs_set_buffer_uptodate(buf); + if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { set_extent_dirty(&root->dirty_log_pages, buf->start, buf->start + buf->len - 1, GFP_NOFS); @@ -2729,6 +3440,7 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, buf->start + buf->len - 1, GFP_NOFS); } trans->blocks_used++; + /* this returns a buffer locked for blocking */ return buf; } @@ -2750,14 +3462,15 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, struct extent_buffer *buf; ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize, - root_objectid, ref_generation, level, 0, + root_objectid, ref_generation, level, empty_size, hint, (u64)-1, &ins, 0); if (ret) { BUG_ON(ret > 0); return ERR_PTR(ret); } - buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize); + buf = btrfs_init_new_buffer(trans, root, ins.objectid, + blocksize, level); return buf; } @@ -2766,66 +3479,117 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, { u64 leaf_owner; u64 leaf_generation; + struct refsort *sorted; struct btrfs_key key; struct btrfs_file_extent_item *fi; int i; int nritems; int ret; + int refi = 0; + int slot; BUG_ON(!btrfs_is_leaf(leaf)); nritems = btrfs_header_nritems(leaf); leaf_owner = btrfs_header_owner(leaf); leaf_generation = btrfs_header_generation(leaf); + sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); + /* we do this loop twice. The first time we build a list + * of the extents we have a reference on, then we sort the list + * by bytenr. The second time around we actually do the + * extent freeing. + */ for (i = 0; i < nritems; i++) { u64 disk_bytenr; cond_resched(); btrfs_item_key_to_cpu(leaf, &key, i); + + /* only extents have references, skip everything else */ if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) continue; + fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); + + /* inline extents live in the btree, they don't have refs */ if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) continue; - /* - * FIXME make sure to insert a trans record that - * repeats the snapshot del on crash - */ + disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); + + /* holes don't have refs */ if (disk_bytenr == 0) continue; - mutex_lock(&root->fs_info->alloc_mutex); + sorted[refi].bytenr = disk_bytenr; + sorted[refi].slot = i; + refi++; + } + + if (refi == 0) + goto out; + + sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); + + for (i = 0; i < refi; i++) { + u64 disk_bytenr; + + disk_bytenr = sorted[i].bytenr; + slot = sorted[i].slot; + + cond_resched(); + + btrfs_item_key_to_cpu(leaf, &key, slot); + if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) + continue; + + fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); + ret = __btrfs_free_extent(trans, root, disk_bytenr, btrfs_file_extent_disk_num_bytes(leaf, fi), leaf->start, leaf_owner, leaf_generation, - key.objectid, key.offset, 0); - mutex_unlock(&root->fs_info->alloc_mutex); + key.objectid, 0); BUG_ON(ret); atomic_inc(&root->fs_info->throttle_gen); wake_up(&root->fs_info->transaction_throttle); cond_resched(); } +out: + kfree(sorted); return 0; } -static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans, +static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_leaf_ref *ref) { int i; int ret; - struct btrfs_extent_info *info = ref->extents; + struct btrfs_extent_info *info; + struct refsort *sorted; + + if (ref->nritems == 0) + return 0; + sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS); for (i = 0; i < ref->nritems; i++) { - mutex_lock(&root->fs_info->alloc_mutex); + sorted[i].bytenr = ref->extents[i].bytenr; + sorted[i].slot = i; + } + sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL); + + /* + * the items in the ref were sorted when the ref was inserted + * into the ref cache, so this is already in order + */ + for (i = 0; i < ref->nritems; i++) { + info = ref->extents + sorted[i].slot; ret = __btrfs_free_extent(trans, root, info->bytenr, info->num_bytes, ref->bytenr, ref->owner, ref->generation, - info->objectid, info->offset, 0); - mutex_unlock(&root->fs_info->alloc_mutex); + info->objectid, 0); atomic_inc(&root->fs_info->throttle_gen); wake_up(&root->fs_info->transaction_throttle); @@ -2835,18 +3599,19 @@ static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans, info++; } + kfree(sorted); return 0; } -int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, - u32 *refs) +static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, + u64 len, u32 *refs) { int ret; ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs); BUG_ON(ret); -#if 0 // some debugging code in case we see problems here +#if 0 /* some debugging code in case we see problems here */ /* if the refs count is one, it won't get increased again. But * if the ref count is > 1, someone may be decreasing it at * the same time we are. @@ -2867,8 +3632,8 @@ int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, free_extent_buffer(eb); } if (*refs == 1) { - printk("block %llu went down to one during drop_snap\n", - (unsigned long long)start); + printk(KERN_ERR "btrfs block %llu went down to one " + "during drop_snap\n", (unsigned long long)start); } } @@ -2879,10 +3644,156 @@ int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, } /* + * this is used while deleting old snapshots, and it drops the refs + * on a whole subtree starting from a level 1 node. + * + * The idea is to sort all the leaf pointers, and then drop the + * ref on all the leaves in order. Most of the time the leaves + * will have ref cache entries, so no leaf IOs will be required to + * find the extents they have references on. + * + * For each leaf, any references it has are also dropped in order + * + * This ends up dropping the references in something close to optimal + * order for reading and modifying the extent allocation tree. + */ +static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path) +{ + u64 bytenr; + u64 root_owner; + u64 root_gen; + struct extent_buffer *eb = path->nodes[1]; + struct extent_buffer *leaf; + struct btrfs_leaf_ref *ref; + struct refsort *sorted = NULL; + int nritems = btrfs_header_nritems(eb); + int ret; + int i; + int refi = 0; + int slot = path->slots[1]; + u32 blocksize = btrfs_level_size(root, 0); + u32 refs; + + if (nritems == 0) + goto out; + + root_owner = btrfs_header_owner(eb); + root_gen = btrfs_header_generation(eb); + sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); + + /* + * step one, sort all the leaf pointers so we don't scribble + * randomly into the extent allocation tree + */ + for (i = slot; i < nritems; i++) { + sorted[refi].bytenr = btrfs_node_blockptr(eb, i); + sorted[refi].slot = i; + refi++; + } + + /* + * nritems won't be zero, but if we're picking up drop_snapshot + * after a crash, slot might be > 0, so double check things + * just in case. + */ + if (refi == 0) + goto out; + + sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); + + /* + * the first loop frees everything the leaves point to + */ + for (i = 0; i < refi; i++) { + u64 ptr_gen; + + bytenr = sorted[i].bytenr; + + /* + * check the reference count on this leaf. If it is > 1 + * we just decrement it below and don't update any + * of the refs the leaf points to. + */ + ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); + BUG_ON(ret); + if (refs != 1) + continue; + + ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); + + /* + * the leaf only had one reference, which means the + * only thing pointing to this leaf is the snapshot + * we're deleting. It isn't possible for the reference + * count to increase again later + * + * The reference cache is checked for the leaf, + * and if found we'll be able to drop any refs held by + * the leaf without needing to read it in. + */ + ref = btrfs_lookup_leaf_ref(root, bytenr); + if (ref && ref->generation != ptr_gen) { + btrfs_free_leaf_ref(root, ref); + ref = NULL; + } + if (ref) { + ret = cache_drop_leaf_ref(trans, root, ref); + BUG_ON(ret); + btrfs_remove_leaf_ref(root, ref); + btrfs_free_leaf_ref(root, ref); + } else { + /* + * the leaf wasn't in the reference cache, so + * we have to read it. + */ + leaf = read_tree_block(root, bytenr, blocksize, + ptr_gen); + ret = btrfs_drop_leaf_ref(trans, root, leaf); + BUG_ON(ret); + free_extent_buffer(leaf); + } + atomic_inc(&root->fs_info->throttle_gen); + wake_up(&root->fs_info->transaction_throttle); + cond_resched(); + } + + /* + * run through the loop again to free the refs on the leaves. + * This is faster than doing it in the loop above because + * the leaves are likely to be clustered together. We end up + * working in nice chunks on the extent allocation tree. + */ + for (i = 0; i < refi; i++) { + bytenr = sorted[i].bytenr; + ret = __btrfs_free_extent(trans, root, bytenr, + blocksize, eb->start, + root_owner, root_gen, 0, 1); + BUG_ON(ret); + + atomic_inc(&root->fs_info->throttle_gen); + wake_up(&root->fs_info->transaction_throttle); + cond_resched(); + } +out: + kfree(sorted); + + /* + * update the path to show we've processed the entire level 1 + * node. This will get saved into the root's drop_snapshot_progress + * field so these drops are not repeated again if this transaction + * commits. + */ + path->slots[1] = nritems; + return 0; +} + +/* * helper function for drop_snapshot, this walks down the tree dropping ref * counts as it goes. */ -static int noinline walk_down_tree(struct btrfs_trans_handle *trans, +static noinline int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, int *level) { @@ -2893,7 +3804,6 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, struct extent_buffer *next; struct extent_buffer *cur; struct extent_buffer *parent; - struct btrfs_leaf_ref *ref; u32 blocksize; int ret; u32 refs; @@ -2909,7 +3819,7 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, /* * walk down to the last node level and free all the leaves */ - while(*level >= 0) { + while (*level >= 0) { WARN_ON(*level < 0); WARN_ON(*level >= BTRFS_MAX_LEVEL); cur = path->nodes[*level]; @@ -2920,29 +3830,57 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, if (path->slots[*level] >= btrfs_header_nritems(cur)) break; + + /* the new code goes down to level 1 and does all the + * leaves pointed to that node in bulk. So, this check + * for level 0 will always be false. + * + * But, the disk format allows the drop_snapshot_progress + * field in the root to leave things in a state where + * a leaf will need cleaning up here. If someone crashes + * with the old code and then boots with the new code, + * we might find a leaf here. + */ if (*level == 0) { ret = btrfs_drop_leaf_ref(trans, root, cur); BUG_ON(ret); break; } + + /* + * once we get to level one, process the whole node + * at once, including everything below it. + */ + if (*level == 1) { + ret = drop_level_one_refs(trans, root, path); + BUG_ON(ret); + break; + } + bytenr = btrfs_node_blockptr(cur, path->slots[*level]); ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); blocksize = btrfs_level_size(root, *level - 1); ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); BUG_ON(ret); + + /* + * if there is more than one reference, we don't need + * to read that node to drop any references it has. We + * just drop the ref we hold on that node and move on to the + * next slot in this level. + */ if (refs != 1) { parent = path->nodes[*level]; root_owner = btrfs_header_owner(parent); root_gen = btrfs_header_generation(parent); path->slots[*level]++; - mutex_lock(&root->fs_info->alloc_mutex); ret = __btrfs_free_extent(trans, root, bytenr, blocksize, parent->start, - root_owner, root_gen, 0, 0, 1); + root_owner, root_gen, + *level - 1, 1); BUG_ON(ret); - mutex_unlock(&root->fs_info->alloc_mutex); atomic_inc(&root->fs_info->throttle_gen); wake_up(&root->fs_info->transaction_throttle); @@ -2950,83 +3888,139 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, continue; } + /* - * at this point, we have a single ref, and since the - * only place referencing this extent is a dead root - * the reference count should never go higher. - * So, we don't need to check it again + * we need to keep freeing things in the next level down. + * read the block and loop around to process it */ - if (*level == 1) { - ref = btrfs_lookup_leaf_ref(root, bytenr); - if (ref && ref->generation != ptr_gen) { - btrfs_free_leaf_ref(root, ref); - ref = NULL; - } - if (ref) { - ret = cache_drop_leaf_ref(trans, root, ref); - BUG_ON(ret); - btrfs_remove_leaf_ref(root, ref); - btrfs_free_leaf_ref(root, ref); - *level = 0; - break; - } - if (printk_ratelimit()) - printk("leaf ref miss for bytenr %llu\n", - (unsigned long long)bytenr); + next = read_tree_block(root, bytenr, blocksize, ptr_gen); + WARN_ON(*level <= 0); + if (path->nodes[*level-1]) + free_extent_buffer(path->nodes[*level-1]); + path->nodes[*level-1] = next; + *level = btrfs_header_level(next); + path->slots[*level] = 0; + cond_resched(); + } +out: + WARN_ON(*level < 0); + WARN_ON(*level >= BTRFS_MAX_LEVEL); + + if (path->nodes[*level] == root->node) { + parent = path->nodes[*level]; + bytenr = path->nodes[*level]->start; + } else { + parent = path->nodes[*level + 1]; + bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); + } + + blocksize = btrfs_level_size(root, *level); + root_owner = btrfs_header_owner(parent); + root_gen = btrfs_header_generation(parent); + + /* + * cleanup and free the reference on the last node + * we processed + */ + ret = __btrfs_free_extent(trans, root, bytenr, blocksize, + parent->start, root_owner, root_gen, + *level, 1); + free_extent_buffer(path->nodes[*level]); + path->nodes[*level] = NULL; + + *level += 1; + BUG_ON(ret); + + cond_resched(); + return 0; +} + +/* + * helper function for drop_subtree, this function is similar to + * walk_down_tree. The main difference is that it checks reference + * counts while tree blocks are locked. + */ +static noinline int walk_down_subtree(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, int *level) +{ + struct extent_buffer *next; + struct extent_buffer *cur; + struct extent_buffer *parent; + u64 bytenr; + u64 ptr_gen; + u32 blocksize; + u32 refs; + int ret; + + cur = path->nodes[*level]; + ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len, + &refs); + BUG_ON(ret); + if (refs > 1) + goto out; + + while (*level >= 0) { + cur = path->nodes[*level]; + if (*level == 0) { + ret = btrfs_drop_leaf_ref(trans, root, cur); + BUG_ON(ret); + clean_tree_block(trans, root, cur); + break; + } + if (path->slots[*level] >= btrfs_header_nritems(cur)) { + clean_tree_block(trans, root, cur); + break; } - next = btrfs_find_tree_block(root, bytenr, blocksize); - if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { - free_extent_buffer(next); - next = read_tree_block(root, bytenr, blocksize, - ptr_gen); - cond_resched(); -#if 0 - /* - * this is a debugging check and can go away - * the ref should never go all the way down to 1 - * at this point - */ - ret = lookup_extent_ref(NULL, root, bytenr, blocksize, - &refs); + bytenr = btrfs_node_blockptr(cur, path->slots[*level]); + blocksize = btrfs_level_size(root, *level - 1); + ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); + + next = read_tree_block(root, bytenr, blocksize, ptr_gen); + btrfs_tree_lock(next); + btrfs_set_lock_blocking(next); + + ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize, + &refs); + BUG_ON(ret); + if (refs > 1) { + parent = path->nodes[*level]; + ret = btrfs_free_extent(trans, root, bytenr, + blocksize, parent->start, + btrfs_header_owner(parent), + btrfs_header_generation(parent), + *level - 1, 1); BUG_ON(ret); - WARN_ON(refs != 1); -#endif + path->slots[*level]++; + btrfs_tree_unlock(next); + free_extent_buffer(next); + continue; } - WARN_ON(*level <= 0); - if (path->nodes[*level-1]) - free_extent_buffer(path->nodes[*level-1]); - path->nodes[*level-1] = next; + *level = btrfs_header_level(next); + path->nodes[*level] = next; path->slots[*level] = 0; + path->locks[*level] = 1; cond_resched(); } out: - WARN_ON(*level < 0); - WARN_ON(*level >= BTRFS_MAX_LEVEL); - - if (path->nodes[*level] == root->node) { - parent = path->nodes[*level]; - bytenr = path->nodes[*level]->start; - } else { - parent = path->nodes[*level + 1]; - bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); - } + parent = path->nodes[*level + 1]; + bytenr = path->nodes[*level]->start; + blocksize = path->nodes[*level]->len; - blocksize = btrfs_level_size(root, *level); - root_owner = btrfs_header_owner(parent); - root_gen = btrfs_header_generation(parent); + ret = btrfs_free_extent(trans, root, bytenr, blocksize, + parent->start, btrfs_header_owner(parent), + btrfs_header_generation(parent), *level, 1); + BUG_ON(ret); - mutex_lock(&root->fs_info->alloc_mutex); - ret = __btrfs_free_extent(trans, root, bytenr, blocksize, - parent->start, root_owner, root_gen, - 0, 0, 1); - mutex_unlock(&root->fs_info->alloc_mutex); + if (path->locks[*level]) { + btrfs_tree_unlock(path->nodes[*level]); + path->locks[*level] = 0; + } free_extent_buffer(path->nodes[*level]); path->nodes[*level] = NULL; *level += 1; - BUG_ON(ret); - cond_resched(); return 0; } @@ -3036,9 +4030,10 @@ out: * to find the first node higher up where we haven't yet gone through * all the slots */ -static int noinline walk_up_tree(struct btrfs_trans_handle *trans, +static noinline int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root *root, - struct btrfs_path *path, int *level) + struct btrfs_path *path, + int *level, int max_level) { u64 root_owner; u64 root_gen; @@ -3047,11 +4042,18 @@ static int noinline walk_up_tree(struct btrfs_trans_handle *trans, int slot; int ret; - for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { + for (i = *level; i < max_level && path->nodes[i]; i++) { slot = path->slots[i]; if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { struct extent_buffer *node; struct btrfs_disk_key disk_key; + + /* + * there is more work to do in this level. + * Update the drop_progress marker to reflect + * the work we've done so far, and then bump + * the slot number + */ node = path->nodes[i]; path->slots[i]++; *level = i; @@ -3063,6 +4065,11 @@ static int noinline walk_up_tree(struct btrfs_trans_handle *trans, return 0; } else { struct extent_buffer *parent; + + /* + * this whole node is done, free our reference + * on it and go up one level + */ if (path->nodes[*level] == root->node) parent = path->nodes[*level]; else @@ -3070,12 +4077,18 @@ static int noinline walk_up_tree(struct btrfs_trans_handle *trans, root_owner = btrfs_header_owner(parent); root_gen = btrfs_header_generation(parent); + + clean_tree_block(trans, root, path->nodes[*level]); ret = btrfs_free_extent(trans, root, path->nodes[*level]->start, path->nodes[*level]->len, - parent->start, - root_owner, root_gen, 0, 0, 1); + parent->start, root_owner, + root_gen, *level, 1); BUG_ON(ret); + if (path->locks[*level]) { + btrfs_tree_unlock(path->nodes[*level]); + path->locks[*level] = 0; + } free_extent_buffer(path->nodes[*level]); path->nodes[*level] = NULL; *level = i + 1; @@ -3138,14 +4151,15 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root } } } - while(1) { + while (1) { wret = walk_down_tree(trans, root, path, &level); if (wret > 0) break; if (wret < 0) ret = wret; - wret = walk_up_tree(trans, root, path, &level); + wret = walk_up_tree(trans, root, path, &level, + BTRFS_MAX_LEVEL); if (wret > 0) break; if (wret < 0) @@ -3168,13 +4182,57 @@ out: return ret; } +int btrfs_drop_subtree(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct extent_buffer *node, + struct extent_buffer *parent) +{ + struct btrfs_path *path; + int level; + int parent_level; + int ret = 0; + int wret; + + path = btrfs_alloc_path(); + BUG_ON(!path); + + BUG_ON(!btrfs_tree_locked(parent)); + parent_level = btrfs_header_level(parent); + extent_buffer_get(parent); + path->nodes[parent_level] = parent; + path->slots[parent_level] = btrfs_header_nritems(parent); + + BUG_ON(!btrfs_tree_locked(node)); + level = btrfs_header_level(node); + extent_buffer_get(node); + path->nodes[level] = node; + path->slots[level] = 0; + + while (1) { + wret = walk_down_subtree(trans, root, path, &level); + if (wret < 0) + ret = wret; + if (wret != 0) + break; + + wret = walk_up_tree(trans, root, path, &level, parent_level); + if (wret < 0) + ret = wret; + if (wret != 0) + break; + } + + btrfs_free_path(path); + return ret; +} + static unsigned long calc_ra(unsigned long start, unsigned long last, unsigned long nr) { return min(last, start + nr - 1); } -static int noinline relocate_inode_pages(struct inode *inode, u64 start, +static noinline int relocate_inode_pages(struct inode *inode, u64 start, u64 len) { u64 page_start; @@ -3245,10 +4303,10 @@ again: } set_page_extent_mapped(page); - btrfs_set_extent_delalloc(inode, page_start, page_end); if (i == first_index) set_extent_bits(io_tree, page_start, page_end, EXTENT_BOUNDARY, GFP_NOFS); + btrfs_set_extent_delalloc(inode, page_start, page_end); set_page_dirty(page); total_dirty++; @@ -3265,25 +4323,28 @@ out_unlock: return ret; } -static int noinline relocate_data_extent(struct inode *reloc_inode, +static noinline int relocate_data_extent(struct inode *reloc_inode, struct btrfs_key *extent_key, u64 offset) { struct btrfs_root *root = BTRFS_I(reloc_inode)->root; struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; struct extent_map *em; + u64 start = extent_key->objectid - offset; + u64 end = start + extent_key->offset - 1; em = alloc_extent_map(GFP_NOFS); BUG_ON(!em || IS_ERR(em)); - em->start = extent_key->objectid - offset; + em->start = start; em->len = extent_key->offset; + em->block_len = extent_key->offset; em->block_start = extent_key->objectid; em->bdev = root->fs_info->fs_devices->latest_bdev; set_bit(EXTENT_FLAG_PINNED, &em->flags); /* setup extent map to cheat btrfs_readpage */ - mutex_lock(&BTRFS_I(reloc_inode)->extent_mutex); + lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); while (1) { int ret; spin_lock(&em_tree->lock); @@ -3293,13 +4354,11 @@ static int noinline relocate_data_extent(struct inode *reloc_inode, free_extent_map(em); break; } - btrfs_drop_extent_cache(reloc_inode, em->start, - em->start + em->len - 1, 0); + btrfs_drop_extent_cache(reloc_inode, start, end, 0); } - mutex_unlock(&BTRFS_I(reloc_inode)->extent_mutex); + unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); - return relocate_inode_pages(reloc_inode, extent_key->objectid - offset, - extent_key->offset); + return relocate_inode_pages(reloc_inode, start, extent_key->offset); } struct btrfs_ref_path { @@ -3308,17 +4367,24 @@ struct btrfs_ref_path { u64 root_objectid; u64 root_generation; u64 owner_objectid; - u64 owner_offset; u32 num_refs; int lowest_level; int current_level; + int shared_level; + + struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; + u64 new_nodes[BTRFS_MAX_LEVEL]; }; struct disk_extent { + u64 ram_bytes; u64 disk_bytenr; u64 disk_num_bytes; u64 offset; u64 num_bytes; + u8 compression; + u8 encryption; + u16 other_encoding; }; static int is_cowonly_root(u64 root_objectid) @@ -3327,12 +4393,13 @@ 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; } -static int noinline __next_ref_path(struct btrfs_trans_handle *trans, +static noinline int __next_ref_path(struct btrfs_trans_handle *trans, struct btrfs_root *extent_root, struct btrfs_ref_path *ref_path, int first_time) @@ -3351,11 +4418,10 @@ static int noinline __next_ref_path(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; - mutex_lock(&extent_root->fs_info->alloc_mutex); - if (first_time) { ref_path->lowest_level = -1; ref_path->current_level = -1; + ref_path->shared_level = -1; goto walk_up; } walk_down: @@ -3365,11 +4431,10 @@ walk_down: if (level < ref_path->lowest_level) break; - if (level >= 0) { + if (level >= 0) bytenr = ref_path->nodes[level]; - } else { + else bytenr = ref_path->extent_start; - } BUG_ON(bytenr == 0); parent = ref_path->nodes[level + 1]; @@ -3399,16 +4464,15 @@ walk_down: btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); if (found_key.objectid == bytenr && - found_key.type == BTRFS_EXTENT_REF_KEY) + found_key.type == BTRFS_EXTENT_REF_KEY) { + if (level < ref_path->shared_level) + ref_path->shared_level = level; goto found; + } next: level--; btrfs_release_path(extent_root, path); - if (need_resched()) { - mutex_unlock(&extent_root->fs_info->alloc_mutex); - cond_resched(); - mutex_lock(&extent_root->fs_info->alloc_mutex); - } + cond_resched(); } /* reached lowest level */ ret = 1; @@ -3417,11 +4481,12 @@ walk_up: level = ref_path->current_level; while (level < BTRFS_MAX_LEVEL - 1) { u64 ref_objectid; - if (level >= 0) { + + if (level >= 0) bytenr = ref_path->nodes[level]; - } else { + else bytenr = ref_path->extent_start; - } + BUG_ON(bytenr == 0); key.objectid = bytenr; @@ -3480,7 +4545,6 @@ found: if (ref_path->lowest_level == level) { ref_path->owner_objectid = ref_objectid; - ref_path->owner_offset = btrfs_ref_offset(leaf, ref); ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); } @@ -3520,16 +4584,11 @@ found: } btrfs_release_path(extent_root, path); - if (need_resched()) { - mutex_unlock(&extent_root->fs_info->alloc_mutex); - cond_resched(); - mutex_lock(&extent_root->fs_info->alloc_mutex); - } + cond_resched(); } /* reached max tree level, but no tree root found. */ BUG(); out: - mutex_unlock(&extent_root->fs_info->alloc_mutex); btrfs_free_path(path); return ret; } @@ -3552,7 +4611,7 @@ static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, return __next_ref_path(trans, extent_root, ref_path, 0); } -static int noinline get_new_locations(struct inode *reloc_inode, +static noinline int get_new_locations(struct inode *reloc_inode, struct btrfs_key *extent_key, u64 offset, int no_fragment, struct disk_extent **extents, @@ -3633,8 +4692,14 @@ static int noinline get_new_locations(struct inode *reloc_inode, btrfs_file_extent_disk_num_bytes(leaf, fi); exts[nr].offset = btrfs_file_extent_offset(leaf, fi); exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); - WARN_ON(exts[nr].offset > 0); - WARN_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); + exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); + exts[nr].compression = btrfs_file_extent_compression(leaf, fi); + exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); + exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, + fi); + BUG_ON(exts[nr].offset > 0); + BUG_ON(exts[nr].compression || exts[nr].encryption); + BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); cur_pos += exts[nr].num_bytes; nr++; @@ -3649,7 +4714,7 @@ static int noinline get_new_locations(struct inode *reloc_inode, path->slots[0]++; } - WARN_ON(cur_pos + offset > last_byte); + BUG_ON(cur_pos + offset > last_byte); if (cur_pos + offset < last_byte) { ret = -ENOENT; goto out; @@ -3667,7 +4732,7 @@ out: return ret; } -static int noinline replace_one_extent(struct btrfs_trans_handle *trans, +static noinline int replace_one_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *extent_key, @@ -3684,18 +4749,22 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans, u64 lock_end = 0; u64 num_bytes; u64 ext_offset; - u64 first_pos; + u64 search_end = (u64)-1; u32 nritems; + int nr_scaned = 0; int extent_locked = 0; + int extent_type; int ret; - first_pos = ref_path->owner_offset; + memcpy(&key, leaf_key, sizeof(key)); if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { - key.objectid = ref_path->owner_objectid; - key.offset = ref_path->owner_offset; - key.type = BTRFS_EXTENT_DATA_KEY; - } else { - memcpy(&key, leaf_key, sizeof(key)); + if (key.objectid < ref_path->owner_objectid || + (key.objectid == ref_path->owner_objectid && + key.type < BTRFS_EXTENT_DATA_KEY)) { + key.objectid = ref_path->owner_objectid; + key.type = BTRFS_EXTENT_DATA_KEY; + key.offset = 0; + } } while (1) { @@ -3711,15 +4780,13 @@ next: * the file extent item was modified by someone * before the extent got locked. */ - mutex_unlock(&BTRFS_I(inode)->extent_mutex); unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, lock_end, GFP_NOFS); extent_locked = 0; } if (path->slots[0] >= nritems) { - if (ref_path->owner_objectid == - BTRFS_MULTIPLE_OBJECTIDS) + if (++nr_scaned > 2) break; BUG_ON(extent_locked); @@ -3738,7 +4805,7 @@ next: if ((key.objectid > ref_path->owner_objectid) || (key.objectid == ref_path->owner_objectid && key.type > BTRFS_EXTENT_DATA_KEY) || - (key.offset >= first_pos + extent_key->offset)) + key.offset >= search_end) break; } @@ -3758,8 +4825,9 @@ next: } fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); - if ((btrfs_file_extent_type(leaf, fi) != - BTRFS_FILE_EXTENT_REG) || + extent_type = btrfs_file_extent_type(leaf, fi); + if ((extent_type != BTRFS_FILE_EXTENT_REG && + extent_type != BTRFS_FILE_EXTENT_PREALLOC) || (btrfs_file_extent_disk_bytenr(leaf, fi) != extent_key->objectid)) { path->slots[0]++; @@ -3770,15 +4838,21 @@ next: num_bytes = btrfs_file_extent_num_bytes(leaf, fi); ext_offset = btrfs_file_extent_offset(leaf, fi); - if (first_pos > key.offset - ext_offset) - first_pos = key.offset - ext_offset; + if (search_end == (u64)-1) { + search_end = key.offset - ext_offset + + btrfs_file_extent_ram_bytes(leaf, fi); + } if (!extent_locked) { lock_start = key.offset; lock_end = lock_start + num_bytes - 1; } else { - BUG_ON(lock_start != key.offset); - BUG_ON(lock_end - lock_start + 1 < num_bytes); + if (lock_start > key.offset || + lock_end + 1 < key.offset + num_bytes) { + unlock_extent(&BTRFS_I(inode)->io_tree, + lock_start, lock_end, GFP_NOFS); + extent_locked = 0; + } } if (!inode) { @@ -3832,21 +4906,16 @@ next: if (ordered) btrfs_put_ordered_extent(ordered); - mutex_lock(&BTRFS_I(inode)->extent_mutex); extent_locked = 1; continue; } if (nr_extents == 1) { /* update extent pointer in place */ - btrfs_set_file_extent_generation(leaf, fi, - trans->transid); btrfs_set_file_extent_disk_bytenr(leaf, fi, new_extents[0].disk_bytenr); btrfs_set_file_extent_disk_num_bytes(leaf, fi, new_extents[0].disk_num_bytes); - ext_offset += new_extents[0].offset; - btrfs_set_file_extent_offset(leaf, fi, ext_offset); btrfs_mark_buffer_dirty(leaf); btrfs_drop_extent_cache(inode, key.offset, @@ -3858,7 +4927,7 @@ next: leaf->start, root->root_key.objectid, trans->transid, - key.objectid, key.offset); + key.objectid); BUG_ON(ret); ret = btrfs_free_extent(trans, root, @@ -3867,12 +4936,14 @@ next: leaf->start, btrfs_header_owner(leaf), btrfs_header_generation(leaf), - key.objectid, key.offset, 0); + key.objectid, 0); BUG_ON(ret); btrfs_release_path(root, path); key.offset += num_bytes; } else { + BUG_ON(1); +#if 0 u64 alloc_hint; u64 extent_len; int i; @@ -3910,6 +4981,16 @@ next: new_extents[i].disk_bytenr); btrfs_set_file_extent_disk_num_bytes(leaf, fi, new_extents[i].disk_num_bytes); + btrfs_set_file_extent_ram_bytes(leaf, fi, + new_extents[i].ram_bytes); + + btrfs_set_file_extent_compression(leaf, fi, + new_extents[i].compression); + btrfs_set_file_extent_encryption(leaf, fi, + new_extents[i].encryption); + btrfs_set_file_extent_other_encoding(leaf, fi, + new_extents[i].other_encoding); + btrfs_set_file_extent_num_bytes(leaf, fi, extent_len); ext_offset += new_extents[i].offset; @@ -3925,12 +5006,11 @@ next: new_extents[i].disk_num_bytes, leaf->start, root->root_key.objectid, - trans->transid, - key.objectid, key.offset); + trans->transid, key.objectid); BUG_ON(ret); btrfs_release_path(root, path); - inode->i_blocks += extent_len >> 9; + inode_add_bytes(inode, extent_len); ext_offset = 0; num_bytes -= extent_len; @@ -3940,17 +5020,17 @@ next: break; } BUG_ON(i >= nr_extents); +#endif } if (extent_locked) { - mutex_unlock(&BTRFS_I(inode)->extent_mutex); unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, lock_end, GFP_NOFS); extent_locked = 0; } skip: if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && - key.offset >= first_pos + extent_key->offset) + key.offset >= search_end) break; cond_resched(); @@ -3961,7 +5041,6 @@ out: if (inode) { mutex_unlock(&inode->i_mutex); if (extent_locked) { - mutex_unlock(&BTRFS_I(inode)->extent_mutex); unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, lock_end, GFP_NOFS); } @@ -3970,51 +5049,6 @@ out: return ret; } -int btrfs_add_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr, - u64 num_bytes, u64 new_bytenr) -{ - set_extent_bits(&root->fs_info->reloc_mapping_tree, - orig_bytenr, orig_bytenr + num_bytes - 1, - EXTENT_LOCKED, GFP_NOFS); - set_state_private(&root->fs_info->reloc_mapping_tree, - orig_bytenr, new_bytenr); - return 0; -} - -int btrfs_get_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr, - u64 num_bytes, u64 *new_bytenr) -{ - u64 bytenr; - u64 cur_bytenr = orig_bytenr; - u64 prev_bytenr = orig_bytenr; - int ret; - - while (1) { - ret = get_state_private(&root->fs_info->reloc_mapping_tree, - cur_bytenr, &bytenr); - if (ret) - break; - prev_bytenr = cur_bytenr; - cur_bytenr = bytenr; - } - - if (orig_bytenr == cur_bytenr) - return -ENOENT; - - if (prev_bytenr != orig_bytenr) { - set_state_private(&root->fs_info->reloc_mapping_tree, - orig_bytenr, cur_bytenr); - } - *new_bytenr = cur_bytenr; - return 0; -} - -void btrfs_free_reloc_mappings(struct btrfs_root *root) -{ - clear_extent_bits(&root->fs_info->reloc_mapping_tree, - 0, (u64)-1, -1, GFP_NOFS); -} - int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct extent_buffer *buf, u64 orig_start) @@ -4050,6 +5084,7 @@ int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, ref->bytenr = buf->start; ref->owner = btrfs_header_owner(buf); ref->generation = btrfs_header_generation(buf); + ret = btrfs_add_leaf_ref(root, ref, 0); WARN_ON(ret); btrfs_free_leaf_ref(root, ref); @@ -4057,7 +5092,7 @@ int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, return 0; } -static int noinline invalidate_extent_cache(struct btrfs_root *root, +static noinline int invalidate_extent_cache(struct btrfs_root *root, struct extent_buffer *leaf, struct btrfs_block_group_cache *group, struct btrfs_root *target_root) @@ -4095,10 +5130,8 @@ static int noinline invalidate_extent_cache(struct btrfs_root *root, lock_extent(&BTRFS_I(inode)->io_tree, key.offset, key.offset + num_bytes - 1, GFP_NOFS); - mutex_lock(&BTRFS_I(inode)->extent_mutex); btrfs_drop_extent_cache(inode, key.offset, key.offset + num_bytes - 1, 1); - mutex_unlock(&BTRFS_I(inode)->extent_mutex); unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, key.offset + num_bytes - 1, GFP_NOFS); cond_resched(); @@ -4107,7 +5140,7 @@ static int noinline invalidate_extent_cache(struct btrfs_root *root, return 0; } -static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, +static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct extent_buffer *leaf, struct btrfs_block_group_cache *group, @@ -4168,13 +5201,10 @@ static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, ref->extents[ext_index].bytenr = new_extent->disk_bytenr; ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; - btrfs_set_file_extent_generation(leaf, fi, trans->transid); btrfs_set_file_extent_disk_bytenr(leaf, fi, new_extent->disk_bytenr); btrfs_set_file_extent_disk_num_bytes(leaf, fi, new_extent->disk_num_bytes); - new_extent->offset += btrfs_file_extent_offset(leaf, fi); - btrfs_set_file_extent_offset(leaf, fi, new_extent->offset); btrfs_mark_buffer_dirty(leaf); ret = btrfs_inc_extent_ref(trans, root, @@ -4182,14 +5212,13 @@ static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, new_extent->disk_num_bytes, leaf->start, root->root_key.objectid, - trans->transid, - key.objectid, key.offset); + trans->transid, key.objectid); BUG_ON(ret); ret = btrfs_free_extent(trans, root, bytenr, num_bytes, leaf->start, btrfs_header_owner(leaf), btrfs_header_generation(leaf), - key.objectid, key.offset, 0); + key.objectid, 0); BUG_ON(ret); cond_resched(); } @@ -4199,15 +5228,30 @@ static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, return 0; } -int btrfs_free_reloc_root(struct btrfs_root *root) +int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, + struct btrfs_root *root) { struct btrfs_root *reloc_root; + int ret; if (root->reloc_root) { reloc_root = root->reloc_root; root->reloc_root = NULL; list_add(&reloc_root->dead_list, &root->fs_info->dead_reloc_roots); + + btrfs_set_root_bytenr(&reloc_root->root_item, + reloc_root->node->start); + btrfs_set_root_level(&root->root_item, + btrfs_header_level(reloc_root->node)); + memset(&reloc_root->root_item.drop_progress, 0, + sizeof(struct btrfs_disk_key)); + reloc_root->root_item.drop_level = 0; + + ret = btrfs_update_root(trans, root->fs_info->tree_root, + &reloc_root->root_key, + &reloc_root->root_item); + BUG_ON(ret); } return 0; } @@ -4305,7 +5349,7 @@ int btrfs_cleanup_reloc_trees(struct btrfs_root *root) return 0; } -static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, +static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, struct btrfs_root *root) { struct btrfs_root *reloc_root; @@ -4333,8 +5377,7 @@ static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, btrfs_set_root_refs(root_item, 0); btrfs_set_root_bytenr(root_item, eb->start); btrfs_set_root_level(root_item, btrfs_header_level(eb)); - memset(&root_item->drop_progress, 0, sizeof(root_item->drop_progress)); - root_item->drop_level = 0; + btrfs_set_root_generation(root_item, trans->transid); btrfs_tree_unlock(eb); free_extent_buffer(eb); @@ -4359,17 +5402,21 @@ static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, * Core function of space balance. * * The idea is using reloc trees to relocate tree blocks in reference - * counted roots. There is one reloc tree for each subvol, all reloc - * trees share same key objectid. Reloc trees are snapshots of the - * latest committed roots (subvol root->commit_root). To relocate a tree - * block referenced by a subvol, the code COW the block through the reloc - * tree, then update pointer in the subvol to point to the new block. - * Since all reloc trees share same key objectid, we can easily do special - * handing to share tree blocks between reloc trees. Once a tree block has - * been COWed in one reloc tree, we can use the result when the same block - * is COWed again through other reloc trees. + * counted roots. There is one reloc tree for each subvol, and all + * reloc trees share same root key objectid. Reloc trees are snapshots + * of the latest committed roots of subvols (root->commit_root). + * + * To relocate a tree block referenced by a subvol, there are two steps. + * COW the block through subvol's reloc tree, then update block pointer + * in the subvol to point to the new block. Since all reloc trees share + * same root key objectid, doing special handing for tree blocks owned + * by them is easy. Once a tree block has been COWed in one reloc tree, + * we can use the resulting new block directly when the same block is + * required to COW again through other reloc trees. By this way, relocated + * tree blocks are shared between reloc trees, so they are also shared + * between subvols. */ -static int noinline relocate_one_path(struct btrfs_trans_handle *trans, +static noinline int relocate_one_path(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *first_key, @@ -4382,15 +5429,14 @@ static int noinline relocate_one_path(struct btrfs_trans_handle *trans, struct btrfs_key *keys; u64 *nodes; int level; - int lowest_merge; + int shared_level; int lowest_level = 0; - int update_refs; int ret; if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) lowest_level = ref_path->owner_objectid; - if (is_cowonly_root(ref_path->root_objectid)) { + if (!root->ref_cows) { path->lowest_level = lowest_level; ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); BUG_ON(ret < 0); @@ -4399,91 +5445,50 @@ static int noinline relocate_one_path(struct btrfs_trans_handle *trans, return 0; } - keys = kzalloc(sizeof(*keys) * BTRFS_MAX_LEVEL, GFP_NOFS); - BUG_ON(!keys); - nodes = kzalloc(sizeof(*nodes) * BTRFS_MAX_LEVEL, GFP_NOFS); - BUG_ON(!nodes); - mutex_lock(&root->fs_info->tree_reloc_mutex); ret = init_reloc_tree(trans, root); BUG_ON(ret); reloc_root = root->reloc_root; - path->lowest_level = lowest_level; - ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 0); - BUG_ON(ret); - /* - * get relocation mapping for tree blocks in the path - */ - lowest_merge = BTRFS_MAX_LEVEL; - for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) { - u64 new_bytenr; - eb = path->nodes[level]; - if (!eb || eb == reloc_root->node) - continue; - ret = btrfs_get_reloc_mapping(reloc_root, eb->start, eb->len, - &new_bytenr); - if (ret) - continue; - if (level == 0) - btrfs_item_key_to_cpu(eb, &keys[level], 0); - else - btrfs_node_key_to_cpu(eb, &keys[level], 0); - nodes[level] = new_bytenr; - lowest_merge = level; - } + shared_level = ref_path->shared_level; + ref_path->shared_level = BTRFS_MAX_LEVEL - 1; - update_refs = 0; - if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { - eb = path->nodes[0]; - if (btrfs_header_generation(eb) < trans->transid) - update_refs = 1; - } + keys = ref_path->node_keys; + nodes = ref_path->new_nodes; + memset(&keys[shared_level + 1], 0, + sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); + memset(&nodes[shared_level + 1], 0, + sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); - btrfs_release_path(reloc_root, path); - /* - * merge tree blocks that already relocated in other reloc trees - */ - if (lowest_merge != BTRFS_MAX_LEVEL) { + if (nodes[lowest_level] == 0) { + path->lowest_level = lowest_level; + ret = btrfs_search_slot(trans, reloc_root, first_key, path, + 0, 1); + BUG_ON(ret); + for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { + eb = path->nodes[level]; + if (!eb || eb == reloc_root->node) + break; + nodes[level] = eb->start; + if (level == 0) + btrfs_item_key_to_cpu(eb, &keys[level], 0); + else + btrfs_node_key_to_cpu(eb, &keys[level], 0); + } + if (nodes[0] && + ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { + eb = path->nodes[0]; + ret = replace_extents_in_leaf(trans, reloc_root, eb, + group, reloc_inode); + BUG_ON(ret); + } + btrfs_release_path(reloc_root, path); + } else { ret = btrfs_merge_path(trans, reloc_root, keys, nodes, - lowest_merge); - BUG_ON(ret < 0); - } - /* - * cow any tree blocks that still haven't been relocated - */ - ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 1); - BUG_ON(ret); - /* - * if we are relocating data block group, update extent pointers - * in the newly created tree leaf. - */ - eb = path->nodes[0]; - if (update_refs && nodes[0] != eb->start) { - ret = replace_extents_in_leaf(trans, reloc_root, eb, group, - reloc_inode); + lowest_level); BUG_ON(ret); } - memset(keys, 0, sizeof(*keys) * BTRFS_MAX_LEVEL); - memset(nodes, 0, sizeof(*nodes) * BTRFS_MAX_LEVEL); - for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) { - eb = path->nodes[level]; - if (!eb || eb == reloc_root->node) - continue; - BUG_ON(btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID); - nodes[level] = eb->start; - if (level == 0) - btrfs_item_key_to_cpu(eb, &keys[level], 0); - else - btrfs_node_key_to_cpu(eb, &keys[level], 0); - } - - if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { - eb = path->nodes[0]; - extent_buffer_get(eb); - } - btrfs_release_path(reloc_root, path); /* * replace tree blocks in the fs tree with tree blocks in * the reloc tree. @@ -4492,33 +5497,29 @@ static int noinline relocate_one_path(struct btrfs_trans_handle *trans, BUG_ON(ret < 0); if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { + ret = btrfs_search_slot(trans, reloc_root, first_key, path, + 0, 0); + BUG_ON(ret); + extent_buffer_get(path->nodes[0]); + eb = path->nodes[0]; + btrfs_release_path(reloc_root, path); ret = invalidate_extent_cache(reloc_root, eb, group, root); BUG_ON(ret); free_extent_buffer(eb); } - mutex_unlock(&root->fs_info->tree_reloc_mutex); + mutex_unlock(&root->fs_info->tree_reloc_mutex); path->lowest_level = 0; - kfree(nodes); - kfree(keys); return 0; } -static int noinline relocate_tree_block(struct btrfs_trans_handle *trans, +static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *first_key, struct btrfs_ref_path *ref_path) { int ret; - int needs_lock = 0; - - if (root == root->fs_info->extent_root || - root == root->fs_info->chunk_root || - root == root->fs_info->dev_root) { - needs_lock = 1; - mutex_lock(&root->fs_info->alloc_mutex); - } ret = relocate_one_path(trans, root, path, first_key, ref_path, NULL, NULL); @@ -4526,31 +5527,27 @@ static int noinline relocate_tree_block(struct btrfs_trans_handle *trans, if (root == root->fs_info->extent_root) btrfs_extent_post_op(trans, root); - if (needs_lock) - mutex_unlock(&root->fs_info->alloc_mutex); return 0; } -static int noinline del_extent_zero(struct btrfs_trans_handle *trans, +static noinline int del_extent_zero(struct btrfs_trans_handle *trans, struct btrfs_root *extent_root, struct btrfs_path *path, struct btrfs_key *extent_key) { int ret; - mutex_lock(&extent_root->fs_info->alloc_mutex); ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); if (ret) goto out; ret = btrfs_del_item(trans, extent_root, path); out: btrfs_release_path(extent_root, path); - mutex_unlock(&extent_root->fs_info->alloc_mutex); return ret; } -static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info, +static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, struct btrfs_ref_path *ref_path) { struct btrfs_key root_key; @@ -4565,7 +5562,7 @@ static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info, return btrfs_read_fs_root_no_name(fs_info, &root_key); } -static int noinline relocate_one_extent(struct btrfs_root *extent_root, +static noinline int relocate_one_extent(struct btrfs_root *extent_root, struct btrfs_path *path, struct btrfs_key *extent_key, struct btrfs_block_group_cache *group, @@ -4582,7 +5579,6 @@ static int noinline relocate_one_extent(struct btrfs_root *extent_root, struct btrfs_key first_key; u64 prev_block = 0; - mutex_unlock(&extent_root->fs_info->alloc_mutex); trans = btrfs_start_transaction(extent_root, 1); BUG_ON(!trans); @@ -4594,8 +5590,8 @@ static int noinline relocate_one_extent(struct btrfs_root *extent_root, ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); if (!ref_path) { - ret = -ENOMEM; - goto out; + ret = -ENOMEM; + goto out; } for (loops = 0; ; loops++) { @@ -4662,44 +5658,44 @@ static int noinline relocate_one_extent(struct btrfs_root *extent_root, prev_block = block_start; } - if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID && - pass >= 2) { + btrfs_record_root_in_trans(found_root); + if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { + /* + * try to update data extent references while + * keeping metadata shared between snapshots. + */ + if (pass == 1) { + ret = relocate_one_path(trans, found_root, + path, &first_key, ref_path, + group, reloc_inode); + if (ret < 0) + goto out; + continue; + } /* * use fallback method to process the remaining * references. */ if (!new_extents) { u64 group_start = group->key.objectid; + new_extents = kmalloc(sizeof(*new_extents), + GFP_NOFS); + nr_extents = 1; ret = get_new_locations(reloc_inode, extent_key, - group_start, 0, + group_start, 1, &new_extents, &nr_extents); - if (ret < 0) + if (ret) goto out; } - btrfs_record_root_in_trans(found_root); ret = replace_one_extent(trans, found_root, path, extent_key, &first_key, ref_path, new_extents, nr_extents); - if (ret < 0) - goto out; - continue; - } - - btrfs_record_root_in_trans(found_root); - if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { + } else { ret = relocate_tree_block(trans, found_root, path, &first_key, ref_path); - } else { - /* - * try to update data extent references while - * keeping metadata shared between snapshots. - */ - ret = relocate_one_path(trans, found_root, path, - &first_key, ref_path, - group, reloc_inode); } if (ret < 0) goto out; @@ -4709,7 +5705,6 @@ out: btrfs_end_transaction(trans, extent_root); kfree(new_extents); kfree(ref_path); - mutex_lock(&extent_root->fs_info->alloc_mutex); return ret; } @@ -4719,7 +5714,7 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; - num_devices = root->fs_info->fs_devices->num_devices; + num_devices = root->fs_info->fs_devices->rw_devices; if (num_devices == 1) { stripped |= BTRFS_BLOCK_GROUP_DUP; stripped = flags & ~stripped; @@ -4751,7 +5746,7 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) return flags; } -int __alloc_chunk_for_shrink(struct btrfs_root *root, +static int __alloc_chunk_for_shrink(struct btrfs_root *root, struct btrfs_block_group_cache *shrink_block_group, int force) { @@ -4762,10 +5757,8 @@ int __alloc_chunk_for_shrink(struct btrfs_root *root, spin_lock(&shrink_block_group->lock); if (btrfs_block_group_used(&shrink_block_group->item) > 0) { spin_unlock(&shrink_block_group->lock); - mutex_unlock(&root->fs_info->alloc_mutex); trans = btrfs_start_transaction(root, 1); - mutex_lock(&root->fs_info->alloc_mutex); spin_lock(&shrink_block_group->lock); new_alloc_flags = update_block_group_flags(root, @@ -4781,9 +5774,7 @@ int __alloc_chunk_for_shrink(struct btrfs_root *root, do_chunk_alloc(trans, root->fs_info->extent_root, calc + 2 * 1024 * 1024, new_alloc_flags, force); - mutex_unlock(&root->fs_info->alloc_mutex); btrfs_end_transaction(trans, root); - mutex_lock(&root->fs_info->alloc_mutex); } else spin_unlock(&shrink_block_group->lock); return 0; @@ -4812,7 +5803,7 @@ 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_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS); btrfs_mark_buffer_dirty(leaf); btrfs_release_path(root, path); out: @@ -4820,7 +5811,7 @@ out: return ret; } -static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, +static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, struct btrfs_block_group_cache *group) { struct inode *inode = NULL; @@ -4848,7 +5839,8 @@ static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, BUG_ON(err); err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0, - group->key.offset, 0); + group->key.offset, 0, group->key.offset, + 0, 0, 0); BUG_ON(err); inode = btrfs_iget_locked(root->fs_info->sb, objectid, root); @@ -4863,6 +5855,7 @@ static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, } else { BUG_ON(1); } + BTRFS_I(inode)->index_cnt = group->key.objectid; err = btrfs_orphan_add(trans, inode); out: @@ -4875,6 +5868,47 @@ out: return inode; } +int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) +{ + + struct btrfs_ordered_sum *sums; + struct btrfs_sector_sum *sector_sum; + struct btrfs_ordered_extent *ordered; + struct btrfs_root *root = BTRFS_I(inode)->root; + struct list_head list; + size_t offset; + int ret; + u64 disk_bytenr; + + INIT_LIST_HEAD(&list); + + ordered = btrfs_lookup_ordered_extent(inode, file_pos); + BUG_ON(ordered->file_offset != file_pos || ordered->len != len); + + disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; + ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, + disk_bytenr + len - 1, &list); + + while (!list_empty(&list)) { + sums = list_entry(list.next, struct btrfs_ordered_sum, list); + list_del_init(&sums->list); + + sector_sum = sums->sums; + sums->bytenr = ordered->start; + + offset = 0; + while (offset < sums->len) { + sector_sum->bytenr += ordered->start - disk_bytenr; + sector_sum++; + offset += root->sectorsize; + } + + btrfs_add_ordered_sum(inode, ordered, sums); + } + btrfs_put_ordered_extent(ordered); + return 0; +} + int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) { struct btrfs_trans_handle *trans; @@ -4884,6 +5918,7 @@ int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) struct inode *reloc_inode; struct btrfs_block_group_cache *block_group; struct btrfs_key key; + u64 skipped; u64 cur_byte; u64 total_found; u32 nritems; @@ -4896,7 +5931,7 @@ int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) block_group = btrfs_lookup_block_group(info, group_start); BUG_ON(!block_group); - printk("btrfs relocating block group %llu flags %llu\n", + printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n", (unsigned long long)block_group->key.objectid, (unsigned long long)block_group->flags); @@ -4906,17 +5941,13 @@ int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) reloc_inode = create_reloc_inode(info, block_group); BUG_ON(IS_ERR(reloc_inode)); - mutex_lock(&root->fs_info->alloc_mutex); - __alloc_chunk_for_shrink(root, block_group, 1); - block_group->ro = 1; - block_group->space_info->total_bytes -= block_group->key.offset; - - mutex_unlock(&root->fs_info->alloc_mutex); + set_block_group_readonly(block_group); btrfs_start_delalloc_inodes(info->tree_root); btrfs_wait_ordered_extents(info->tree_root, 0); again: + skipped = 0; total_found = 0; progress = 0; key.objectid = block_group->key.objectid; @@ -4932,9 +5963,7 @@ again: btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); mutex_unlock(&root->fs_info->cleaner_mutex); - mutex_lock(&root->fs_info->alloc_mutex); - - while(1) { + while (1) { ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) goto out; @@ -4961,9 +5990,7 @@ next: if (progress && need_resched()) { btrfs_release_path(root, path); - mutex_unlock(&root->fs_info->alloc_mutex); cond_resched(); - mutex_lock(&root->fs_info->alloc_mutex); progress = 0; continue; } @@ -4983,6 +6010,8 @@ next: ret = relocate_one_extent(root, path, &key, block_group, reloc_inode, pass); BUG_ON(ret < 0); + if (ret > 0) + skipped++; key.objectid = cur_byte; key.type = 0; @@ -4990,18 +6019,21 @@ next: } btrfs_release_path(root, path); - mutex_unlock(&root->fs_info->alloc_mutex); if (pass == 0) { btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); - WARN_ON(reloc_inode->i_mapping->nrpages); } if (total_found > 0) { - printk("btrfs found %llu extents in pass %d\n", + printk(KERN_INFO "btrfs found %llu extents in pass %d\n", (unsigned long long)total_found, pass); pass++; + if (total_found == skipped && pass > 2) { + iput(reloc_inode); + reloc_inode = create_reloc_inode(info, block_group); + pass = 0; + } goto again; } @@ -5012,22 +6044,20 @@ next: trans = btrfs_start_transaction(info->tree_root, 1); btrfs_commit_transaction(trans, info->tree_root); - mutex_lock(&root->fs_info->alloc_mutex); - spin_lock(&block_group->lock); WARN_ON(block_group->pinned > 0); 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: - mutex_unlock(&root->fs_info->alloc_mutex); btrfs_free_path(path); return ret; } -int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, - struct btrfs_key *key) +static int find_first_block_group(struct btrfs_root *root, + struct btrfs_path *path, struct btrfs_key *key) { int ret = 0; struct btrfs_key found_key; @@ -5038,7 +6068,7 @@ int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, if (ret < 0) goto out; - while(1) { + while (1) { slot = path->slots[0]; leaf = path->nodes[0]; if (slot >= btrfs_header_nritems(leaf)) { @@ -5068,25 +6098,25 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info) struct btrfs_block_group_cache *block_group; struct rb_node *n; - mutex_lock(&info->alloc_mutex); spin_lock(&info->block_group_cache_lock); while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { block_group = rb_entry(n, struct btrfs_block_group_cache, cache_node); - - spin_unlock(&info->block_group_cache_lock); - btrfs_remove_free_space_cache(block_group); - spin_lock(&info->block_group_cache_lock); - rb_erase(&block_group->cache_node, &info->block_group_cache_tree); - spin_lock(&block_group->space_info->lock); + spin_unlock(&info->block_group_cache_lock); + + btrfs_remove_free_space_cache(block_group); + down_write(&block_group->space_info->groups_sem); list_del(&block_group->list); - spin_unlock(&block_group->space_info->lock); + 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); } spin_unlock(&info->block_group_cache_lock); - mutex_unlock(&info->alloc_mutex); return 0; } @@ -5109,8 +6139,7 @@ int btrfs_read_block_groups(struct btrfs_root *root) if (!path) return -ENOMEM; - mutex_lock(&root->fs_info->alloc_mutex); - while(1) { + while (1) { ret = find_first_block_group(root, path, &key); if (ret > 0) { ret = 0; @@ -5127,7 +6156,10 @@ 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); INIT_LIST_HEAD(&cache->list); read_extent_buffer(leaf, &cache->item, btrfs_item_ptr_offset(leaf, path->slots[0]), @@ -5143,19 +6175,20 @@ int btrfs_read_block_groups(struct btrfs_root *root) &space_info); BUG_ON(ret); cache->space_info = space_info; - spin_lock(&space_info->lock); - list_add(&cache->list, &space_info->block_groups); - spin_unlock(&space_info->lock); + down_write(&space_info->groups_sem); + list_add_tail(&cache->list, &space_info->block_groups); + up_write(&space_info->groups_sem); ret = btrfs_add_block_group_cache(root->fs_info, cache); BUG_ON(ret); set_avail_alloc_bits(root->fs_info, cache->flags); + if (btrfs_chunk_readonly(root, cache->key.objectid)) + set_block_group_readonly(cache); } ret = 0; error: btrfs_free_path(path); - mutex_unlock(&root->fs_info->alloc_mutex); return ret; } @@ -5168,7 +6201,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, struct btrfs_root *extent_root; struct btrfs_block_group_cache *cache; - WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); extent_root = root->fs_info->extent_root; root->fs_info->last_trans_new_blockgroup = trans->transid; @@ -5179,9 +6211,12 @@ 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); @@ -5191,9 +6226,9 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, &cache->space_info); BUG_ON(ret); - spin_lock(&cache->space_info->lock); - list_add(&cache->list, &cache->space_info->block_groups); - spin_unlock(&cache->space_info->lock); + down_write(&cache->space_info->groups_sem); + list_add_tail(&cache->list, &cache->space_info->block_groups); + up_write(&cache->space_info->groups_sem); ret = btrfs_add_block_group_cache(root->fs_info, cache); BUG_ON(ret); @@ -5202,8 +6237,8 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, sizeof(cache->item)); BUG_ON(ret); - finish_current_insert(trans, extent_root); - ret = del_pending_extents(trans, extent_root); + finish_current_insert(trans, extent_root, 0); + ret = del_pending_extents(trans, extent_root, 0); BUG_ON(ret); set_avail_alloc_bits(extent_root->fs_info, type); @@ -5218,28 +6253,34 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans, struct btrfs_key key; int ret; - BUG_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); root = root->fs_info->extent_root; block_group = btrfs_lookup_block_group(root->fs_info, group_start); BUG_ON(!block_group); + BUG_ON(!block_group->ro); memcpy(&key, &block_group->key, sizeof(key)); path = btrfs_alloc_path(); BUG_ON(!path); - btrfs_remove_free_space_cache(block_group); + spin_lock(&root->fs_info->block_group_cache_lock); rb_erase(&block_group->cache_node, &root->fs_info->block_group_cache_tree); - spin_lock(&block_group->space_info->lock); + spin_unlock(&root->fs_info->block_group_cache_lock); + btrfs_remove_free_space_cache(block_group); + down_write(&block_group->space_info->groups_sem); list_del(&block_group->list); + up_write(&block_group->space_info->groups_sem); + + spin_lock(&block_group->space_info->lock); + block_group->space_info->total_bytes -= block_group->key.offset; + block_group->space_info->bytes_readonly -= block_group->key.offset; 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)