+ path->reada = 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) {
+ extent_slot--;
+ btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+ extent_slot);
+ if (found_key.objectid != bytenr)
+ break;
+ if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
+ found_key.offset == num_bytes) {
+ found_extent = 1;
+ break;
+ }
+ if (path->slots[0] - extent_slot > 5)
+ break;
+ }
+ if (!found_extent) {
+ 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);
+ 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(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];
+ ei = btrfs_item_ptr(leaf, extent_slot,
+ struct btrfs_extent_item);
+ refs = btrfs_extent_refs(leaf, ei);
+ BUG_ON(refs == 0);
+ refs -= 1;
+ btrfs_set_extent_refs(leaf, ei, refs);
+
+ btrfs_mark_buffer_dirty(leaf);
+
+ if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
+ struct btrfs_extent_ref *ref;
+ ref = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_extent_ref);
+ BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
+ /* if the back ref and the extent are next to each other
+ * they get deleted below in one shot
+ */
+ path->slots[0] = extent_slot;
+ num_to_del = 2;
+ } else if (found_extent) {
+ /* otherwise delete the extent back ref */
+ ret = remove_extent_backref(trans, extent_root, path);
+ BUG_ON(ret);
+ /* if refs are 0, we need to setup the path for deletion */
+ if (refs == 0) {
+ btrfs_release_path(extent_root, path);
+ ret = btrfs_search_slot(trans, extent_root, &key, path,
+ -1, 1);
+ BUG_ON(ret);
+ }
+ }
+
+ if (refs == 0) {
+ u64 super_used;
+ u64 root_used;
+
+ 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(&info->delalloc_lock);
+ super_used = btrfs_super_bytes_used(&info->super_copy);
+ btrfs_set_super_bytes_used(&info->super_copy,
+ super_used - num_bytes);
+
+ /* 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);
+ }
+ btrfs_free_path(path);
+ finish_current_insert(trans, extent_root, 0);
+ return ret;
+}
+
+/*
+ * 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, int all)
+{
+ int ret;
+ int err = 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;
+
+ INIT_LIST_HEAD(&delete_list);
+ extent_ins = &extent_root->fs_info->extent_ins;
+ pending_del = &extent_root->fs_info->pending_del;
+
+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_WRITEBACK,
+ GFP_NOFS);
+ if (!test_range_bit(extent_ins, start, end,
+ EXTENT_WRITEBACK, 0)) {
+ list_add_tail(&extent_op->list, &delete_list);
+ nr++;
+ } else {
+ kfree(extent_op);
+
+ ret = get_state_private(&info->extent_ins, start,
+ &priv);
+ BUG_ON(ret);
+ extent_op = (struct pending_extent_op *)
+ (unsigned long)priv;
+
+ clear_extent_bits(&info->extent_ins, start, end,
+ EXTENT_WRITEBACK, GFP_NOFS);
+
+ if (extent_op->type == PENDING_BACKREF_UPDATE) {
+ list_add_tail(&extent_op->list, &delete_list);
+ search = end + 1;
+ nr++;
+ continue;
+ }
+
+ 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, 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(&info->extent_ins_mutex);
+ cond_resched();
+ 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;
+ }
+
+ return err;
+}
+
+/*
+ * remove an extent from the root, returns 0 on success
+ */
+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, int pin)
+{
+ struct btrfs_root *extent_root = root->fs_info->extent_root;
+ int pending_ret;
+ int ret;
+
+ WARN_ON(num_bytes < root->sectorsize);
+ if (root == extent_root) {
+ 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);
+
+ extent_op->type = PENDING_EXTENT_DELETE;
+ extent_op->bytenr = bytenr;
+ extent_op->num_bytes = num_bytes;
+ extent_op->parent = parent;
+ extent_op->orig_parent = parent;
+ 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_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) {
+ 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;
+ }
+ pin = 1;
+ }
+
+ /* if data pin when any transaction has committed this */
+ if (ref_generation != trans->transid)
+ pin = 1;
+
+ ret = __free_extent(trans, root, bytenr, num_bytes, parent,
+ root_objectid, ref_generation,
+ owner_objectid, pin, pin == 0);
+
+ 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;
+}
+
+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, int pin)
+{
+ int ret;
+
+ ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
+ root_objectid, ref_generation,
+ owner_objectid, pin);
+ return ret;
+}
+
+static u64 stripe_align(struct btrfs_root *root, u64 val)
+{
+ u64 mask = ((u64)root->stripesize - 1);
+ u64 ret = (val + mask) & ~mask;
+ return ret;
+}
+
+/*
+ * walks the btree of allocated extents and find a hole of a given size.
+ * The key ins is changed to record the hole:
+ * ins->objectid == block start
+ * ins->flags = BTRFS_EXTENT_ITEM_KEY
+ * ins->offset == number of blocks
+ * Any available blocks before search_start are skipped.
+ */
+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,
+ u64 hint_byte, struct btrfs_key *ins,
+ u64 exclude_start, u64 exclude_nr,
+ int data)
+{
+ int ret = 0;
+ struct btrfs_root *root = orig_root->fs_info->extent_root;
+ u64 total_needed = num_bytes;
+ u64 *last_ptr = NULL;
+ 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 = 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) {
+ hint_byte = *last_ptr;
+ last_wanted = *last_ptr;
+ } else
+ empty_size += empty_cluster;
+ } else {
+ empty_cluster = 0;
+ }
+ search_start = max(search_start, first_logical_byte(root, 0));
+ search_start = max(search_start, hint_byte);
+
+ 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(root->fs_info,
+ search_start);
+ space_info = __find_space_info(root->fs_info, data);
+
+ down_read(&space_info->groups_sem);
+ while (1) {
+ struct btrfs_free_space *free_space;
+ /*
+ * 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 (empty_size)
+ extra_loop = 1;
+
+ 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;
+ }
+
+ 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;
+ }
+
+ 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;
+ }
+
+ /* else we go to the next block group */
+ goto new_group;
+ }
+
+ ins->objectid = search_start;
+ ins->offset = num_bytes;
+
+ btrfs_remove_free_space_lock(block_group, search_start,
+ num_bytes);
+ /* we are all good, lets return */
+ mutex_unlock(&block_group->alloc_mutex);
+ break;
+ }
+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;
+
+ /*
+ * 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;
+ }
+
+ block_group = list_entry(cur, struct btrfs_block_group_cache,
+ list);
+ atomic_inc(&block_group->count);
+
+ search_start = block_group->key.objectid;
+ cur = cur->next;
+ }
+
+ /* 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;
+
+ 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 ");
+
+ down_read(&info->groups_sem);
+ list_for_each_entry(cache, &info->block_groups, list) {
+ spin_lock(&cache->lock);
+ 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);
+ }
+ up_read(&info->groups_sem);
+}
+
+static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ u64 num_bytes, u64 min_alloc_size,
+ u64 empty_size, u64 hint_byte,
+ u64 search_end, struct btrfs_key *ins,
+ u64 data)
+{
+ int ret;
+ u64 search_start = 0;
+ u64 alloc_profile;
+ struct btrfs_fs_info *info = root->fs_info;
+
+ if (data) {
+ alloc_profile = info->avail_data_alloc_bits &
+ 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;
+ data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
+ } else {
+ alloc_profile = info->avail_metadata_alloc_bits &
+ info->metadata_alloc_profile;
+ data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
+ }
+again:
+ 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
+ */
+ if (empty_size || root->ref_cows) {
+ if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
+ ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+ 2 * 1024 * 1024,
+ BTRFS_BLOCK_GROUP_METADATA |
+ (info->metadata_alloc_profile &
+ info->avail_metadata_alloc_bits), 0);
+ }
+ ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+ num_bytes + 2 * 1024 * 1024, data, 0);
+ }
+
+ WARN_ON(num_bytes < root->sectorsize);
+ ret = find_free_extent(trans, root, num_bytes, empty_size,
+ search_start, search_end, hint_byte, ins,
+ trans->alloc_exclude_start,
+ trans->alloc_exclude_nr, data);
+
+ if (ret == -ENOSPC && num_bytes > min_alloc_size) {
+ num_bytes = num_bytes >> 1;
+ num_bytes = num_bytes & ~(root->sectorsize - 1);
+ num_bytes = max(num_bytes, min_alloc_size);
+ do_chunk_alloc(trans, root->fs_info->extent_root,
+ num_bytes, data, 1);
+ goto again;
+ }
+ if (ret) {
+ struct btrfs_space_info *sinfo;
+
+ sinfo = __find_space_info(root->fs_info, data);
+ 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();
+ }
+
+ return ret;
+}
+
+int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
+{
+ struct btrfs_block_group_cache *cache;
+ int ret = 0;
+
+ cache = btrfs_lookup_block_group(root->fs_info, start);
+ if (!cache) {
+ 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);
+
+ return ret;
+}
+
+int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ u64 num_bytes, u64 min_alloc_size,
+ u64 empty_size, u64 hint_byte,
+ u64 search_end, struct btrfs_key *ins,
+ u64 data)
+{
+ int ret;
+ 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);
+ 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, struct btrfs_key *ins)
+{
+ int ret;
+ int pending_ret;
+ u64 super_used;
+ u64 root_used;
+ u64 num_bytes = ins->offset;
+ u32 sizes[2];
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_root *extent_root = info->extent_root;
+ struct btrfs_extent_item *extent_item;
+ struct btrfs_extent_ref *ref;
+ struct btrfs_path *path;
+ struct btrfs_key keys[2];
+
+ if (parent == 0)
+ parent = ins->objectid;
+
+ /* block accounting for super block */
+ 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);
+
+ /* 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;
+
+ extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+ BUG_ON(!extent_op);
+
+ extent_op->type = PENDING_EXTENT_INSERT;
+ extent_op->bytenr = ins->objectid;
+ extent_op->num_bytes = ins->offset;
+ extent_op->parent = parent;
+ extent_op->orig_parent = 0;
+ 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_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;
+ }
+
+ memcpy(&keys[0], ins, sizeof(*ins));
+ keys[1].objectid = ins->objectid;
+ keys[1].type = BTRFS_EXTENT_REF_KEY;
+ keys[1].offset = parent;
+ sizes[0] = sizeof(*extent_item);
+ sizes[1] = sizeof(*ref);
+
+ path = btrfs_alloc_path();
+ BUG_ON(!path);
+
+ ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
+ sizes, 2);
+ BUG_ON(ret);
+
+ extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+ struct btrfs_extent_item);
+ btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
+ ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
+ struct btrfs_extent_ref);
+
+ 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_num_refs(path->nodes[0], ref, 1);
+
+ btrfs_mark_buffer_dirty(path->nodes[0]);
+
+ trans->alloc_exclude_start = 0;
+ trans->alloc_exclude_nr = 0;
+ btrfs_free_path(path);
+ finish_current_insert(trans, extent_root, 0);
+ pending_ret = del_pending_extents(trans, extent_root, 0);
+
+ if (ret)
+ goto out;
+ if (pending_ret) {
+ ret = pending_ret;
+ goto out;
+ }
+
+update_block:
+ ret = update_block_group(trans, root, ins->objectid,
+ ins->offset, 1, 0);
+ if (ret) {
+ printk(KERN_ERR "btrfs update block group failed for %llu "
+ "%llu\n", (unsigned long long)ins->objectid,
+ (unsigned long long)ins->offset);
+ BUG();
+ }
+out:
+ return ret;
+}
+
+int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root, u64 parent,
+ u64 root_objectid, u64 ref_generation,
+ u64 owner, struct btrfs_key *ins)
+{
+ int ret;
+
+ if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
+ return 0;
+ ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
+ ref_generation, owner, ins);
+ update_reserved_extents(root, ins->objectid, ins->offset, 0);
+ return ret;
+}
+
+/*
+ * this is used by the tree logging recovery code. It records that
+ * an extent has been allocated and makes sure to clear the free
+ * space cache bits as well
+ */
+int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root, u64 parent,
+ u64 root_objectid, u64 ref_generation,
+ u64 owner, struct btrfs_key *ins)
+{
+ int ret;
+ struct btrfs_block_group_cache *block_group;
+
+ 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);
+ BUG_ON(ret);
+ put_block_group(block_group);
+ ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
+ ref_generation, owner, ins);
+ return ret;
+}
+
+/*
+ * finds a free extent and does all the dirty work required for allocation
+ * returns the key for the extent through ins, and a tree buffer for
+ * the first block of the extent through buf.
+ *
+ * returns 0 if everything worked, non-zero otherwise.
+ */
+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 empty_size, u64 hint_byte,
+ u64 search_end, struct btrfs_key *ins, u64 data)
+{
+ int ret;
+
+ ret = __btrfs_reserve_extent(trans, root, num_bytes,
+ min_alloc_size, empty_size, hint_byte,
+ search_end, ins, data);
+ BUG_ON(ret);
+ if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
+ ret = __btrfs_alloc_reserved_extent(trans, root, parent,
+ root_objectid, ref_generation,
+ owner_objectid, ins);
+ BUG_ON(ret);
+
+ } else {
+ update_reserved_extents(root, ins->objectid, ins->offset, 1);
+ }
+ return ret;
+}
+
+struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ u64 bytenr, u32 blocksize)
+{
+ struct extent_buffer *buf;
+
+ buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
+ if (!buf)
+ return ERR_PTR(-ENOMEM);
+ btrfs_set_header_generation(buf, trans->transid);
+ 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);
+ } else {
+ set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
+ buf->start + buf->len - 1, GFP_NOFS);
+ }
+ trans->blocks_used++;
+ /* this returns a buffer locked for blocking */
+ return buf;
+}
+
+/*
+ * helper function to allocate a block for a given tree
+ * returns the tree buffer or NULL.
+ */
+struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ u32 blocksize, u64 parent,
+ u64 root_objectid,
+ u64 ref_generation,
+ int level,
+ u64 hint,
+ u64 empty_size)
+{
+ struct btrfs_key ins;
+ int ret;
+ struct extent_buffer *buf;
+
+ ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
+ 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);
+ return buf;
+}
+
+int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root, struct extent_buffer *leaf)
+{
+ 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;
+
+ disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+
+ /* holes don't have refs */
+ if (disk_bytenr == 0)
+ continue;
+
+ 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, 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 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;
+ struct refsort *sorted;
+
+ if (ref->nritems == 0)
+ return 0;
+
+ sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
+ for (i = 0; i < ref->nritems; i++) {
+ 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, 0);
+
+ atomic_inc(&root->fs_info->throttle_gen);
+ wake_up(&root->fs_info->transaction_throttle);
+ cond_resched();
+
+ BUG_ON(ret);
+ info++;
+ }
+
+ kfree(sorted);
+ return 0;
+}
+
+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 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.
+ */
+ if (*refs != 1) {
+ struct extent_buffer *eb = NULL;
+ eb = btrfs_find_create_tree_block(root, start, len);
+ if (eb)
+ btrfs_tree_lock(eb);
+
+ mutex_lock(&root->fs_info->alloc_mutex);
+ ret = lookup_extent_ref(NULL, root, start, len, refs);
+ BUG_ON(ret);
+ mutex_unlock(&root->fs_info->alloc_mutex);
+
+ if (eb) {
+ btrfs_tree_unlock(eb);
+ free_extent_buffer(eb);
+ }
+ if (*refs == 1) {
+ printk(KERN_ERR "btrfs block %llu went down to one "
+ "during drop_snap\n", (unsigned long long)start);
+ }
+
+ }
+#endif
+
+ cond_resched();
+ return ret;
+}
+
+/*
+ * 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 noinline int walk_down_tree(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ struct btrfs_path *path, int *level)
+{
+ u64 root_owner;
+ u64 root_gen;
+ u64 bytenr;
+ u64 ptr_gen;
+ struct extent_buffer *next;
+ struct extent_buffer *cur;
+ struct extent_buffer *parent;
+ u32 blocksize;
+ int ret;
+ u32 refs;
+
+ WARN_ON(*level < 0);
+ WARN_ON(*level >= BTRFS_MAX_LEVEL);
+ ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
+ path->nodes[*level]->len, &refs);
+ BUG_ON(ret);
+ if (refs > 1)
+ goto out;
+
+ /*
+ * walk down to the last node level and free all the leaves
+ */
+ while (*level >= 0) {
+ WARN_ON(*level < 0);
+ WARN_ON(*level >= BTRFS_MAX_LEVEL);
+ cur = path->nodes[*level];
+
+ if (btrfs_header_level(cur) != *level)
+ WARN_ON(1);
+
+ 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]++;
+
+ ret = __btrfs_free_extent(trans, root, bytenr,
+ blocksize, parent->start,
+ root_owner, root_gen,
+ *level - 1, 1);
+ BUG_ON(ret);
+
+ atomic_inc(&root->fs_info->throttle_gen);
+ wake_up(&root->fs_info->transaction_throttle);
+ cond_resched();
+
+ continue;
+ }
+
+ /*
+ * we need to keep freeing things in the next level down.
+ * read the block and loop around to process it
+ */
+ 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;
+ }
+
+ 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);
+ path->slots[*level]++;
+ btrfs_tree_unlock(next);
+ free_extent_buffer(next);
+ continue;
+ }
+
+ *level = btrfs_header_level(next);
+ path->nodes[*level] = next;
+ path->slots[*level] = 0;
+ path->locks[*level] = 1;
+ cond_resched();
+ }
+out:
+ parent = path->nodes[*level + 1];
+ bytenr = path->nodes[*level]->start;
+ blocksize = path->nodes[*level]->len;
+
+ ret = btrfs_free_extent(trans, root, bytenr, blocksize,
+ parent->start, btrfs_header_owner(parent),
+ btrfs_header_generation(parent), *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 += 1;
+ cond_resched();
+ return 0;
+}
+
+/*
+ * helper for dropping snapshots. This walks back up the tree in the path
+ * to find the first node higher up where we haven't yet gone through
+ * all the slots
+ */
+static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ struct btrfs_path *path,
+ int *level, int max_level)
+{
+ u64 root_owner;
+ u64 root_gen;
+ struct btrfs_root_item *root_item = &root->root_item;
+ int i;
+ int slot;
+ int ret;
+
+ 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;
+ WARN_ON(*level == 0);
+ btrfs_node_key(node, &disk_key, path->slots[i]);
+ memcpy(&root_item->drop_progress,
+ &disk_key, sizeof(disk_key));
+ root_item->drop_level = i;
+ 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
+ parent = path->nodes[*level + 1];
+
+ 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, *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;
+ }
+ }
+ return 1;
+}
+
+/*
+ * drop the reference count on the tree rooted at 'snap'. This traverses
+ * the tree freeing any blocks that have a ref count of zero after being
+ * decremented.
+ */
+int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root)
+{
+ int ret = 0;
+ int wret;
+ int level;
+ struct btrfs_path *path;
+ int i;
+ int orig_level;
+ struct btrfs_root_item *root_item = &root->root_item;
+
+ WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
+ path = btrfs_alloc_path();
+ BUG_ON(!path);
+
+ level = btrfs_header_level(root->node);
+ orig_level = level;
+ if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
+ path->nodes[level] = root->node;
+ extent_buffer_get(root->node);
+ path->slots[level] = 0;
+ } else {
+ struct btrfs_key key;
+ struct btrfs_disk_key found_key;
+ struct extent_buffer *node;
+
+ btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
+ level = root_item->drop_level;