2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "free-space-cache.h"
35 static int update_block_group(struct btrfs_trans_handle *trans,
36 struct btrfs_root *root,
37 u64 bytenr, u64 num_bytes, int alloc,
39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61 struct btrfs_root *extent_root, u64 alloc_bytes,
62 u64 flags, int force);
63 static int pin_down_bytes(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 struct btrfs_path *path,
66 u64 bytenr, u64 num_bytes,
67 int is_data, int reserved,
68 struct extent_buffer **must_clean);
69 static int find_next_key(struct btrfs_path *path, int level,
70 struct btrfs_key *key);
71 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
72 int dump_block_groups);
75 block_group_cache_done(struct btrfs_block_group_cache *cache)
78 return cache->cached == BTRFS_CACHE_FINISHED;
81 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
83 return (cache->flags & bits) == bits;
87 * this adds the block group to the fs_info rb tree for the block group
90 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
91 struct btrfs_block_group_cache *block_group)
94 struct rb_node *parent = NULL;
95 struct btrfs_block_group_cache *cache;
97 spin_lock(&info->block_group_cache_lock);
98 p = &info->block_group_cache_tree.rb_node;
102 cache = rb_entry(parent, struct btrfs_block_group_cache,
104 if (block_group->key.objectid < cache->key.objectid) {
106 } else if (block_group->key.objectid > cache->key.objectid) {
109 spin_unlock(&info->block_group_cache_lock);
114 rb_link_node(&block_group->cache_node, parent, p);
115 rb_insert_color(&block_group->cache_node,
116 &info->block_group_cache_tree);
117 spin_unlock(&info->block_group_cache_lock);
123 * This will return the block group at or after bytenr if contains is 0, else
124 * it will return the block group that contains the bytenr
126 static struct btrfs_block_group_cache *
127 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
130 struct btrfs_block_group_cache *cache, *ret = NULL;
134 spin_lock(&info->block_group_cache_lock);
135 n = info->block_group_cache_tree.rb_node;
138 cache = rb_entry(n, struct btrfs_block_group_cache,
140 end = cache->key.objectid + cache->key.offset - 1;
141 start = cache->key.objectid;
143 if (bytenr < start) {
144 if (!contains && (!ret || start < ret->key.objectid))
147 } else if (bytenr > start) {
148 if (contains && bytenr <= end) {
159 atomic_inc(&ret->count);
160 spin_unlock(&info->block_group_cache_lock);
165 static int add_excluded_extent(struct btrfs_root *root,
166 u64 start, u64 num_bytes)
168 u64 end = start + num_bytes - 1;
169 set_extent_bits(&root->fs_info->freed_extents[0],
170 start, end, EXTENT_UPTODATE, GFP_NOFS);
171 set_extent_bits(&root->fs_info->freed_extents[1],
172 start, end, EXTENT_UPTODATE, GFP_NOFS);
176 static void free_excluded_extents(struct btrfs_root *root,
177 struct btrfs_block_group_cache *cache)
181 start = cache->key.objectid;
182 end = start + cache->key.offset - 1;
184 clear_extent_bits(&root->fs_info->freed_extents[0],
185 start, end, EXTENT_UPTODATE, GFP_NOFS);
186 clear_extent_bits(&root->fs_info->freed_extents[1],
187 start, end, EXTENT_UPTODATE, GFP_NOFS);
190 static int exclude_super_stripes(struct btrfs_root *root,
191 struct btrfs_block_group_cache *cache)
198 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
199 bytenr = btrfs_sb_offset(i);
200 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
201 cache->key.objectid, bytenr,
202 0, &logical, &nr, &stripe_len);
206 cache->bytes_super += stripe_len;
207 ret = add_excluded_extent(root, logical[nr],
217 static struct btrfs_caching_control *
218 get_caching_control(struct btrfs_block_group_cache *cache)
220 struct btrfs_caching_control *ctl;
222 spin_lock(&cache->lock);
223 if (cache->cached != BTRFS_CACHE_STARTED) {
224 spin_unlock(&cache->lock);
228 ctl = cache->caching_ctl;
229 atomic_inc(&ctl->count);
230 spin_unlock(&cache->lock);
234 static void put_caching_control(struct btrfs_caching_control *ctl)
236 if (atomic_dec_and_test(&ctl->count))
241 * this is only called by cache_block_group, since we could have freed extents
242 * we need to check the pinned_extents for any extents that can't be used yet
243 * since their free space will be released as soon as the transaction commits.
245 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
246 struct btrfs_fs_info *info, u64 start, u64 end)
248 u64 extent_start, extent_end, size, total_added = 0;
251 while (start < end) {
252 ret = find_first_extent_bit(info->pinned_extents, start,
253 &extent_start, &extent_end,
254 EXTENT_DIRTY | EXTENT_UPTODATE);
258 if (extent_start == start) {
259 start = extent_end + 1;
260 } else if (extent_start > start && extent_start < end) {
261 size = extent_start - start;
263 ret = btrfs_add_free_space(block_group, start,
266 start = extent_end + 1;
275 ret = btrfs_add_free_space(block_group, start, size);
282 static int caching_kthread(void *data)
284 struct btrfs_block_group_cache *block_group = data;
285 struct btrfs_fs_info *fs_info = block_group->fs_info;
286 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
287 struct btrfs_root *extent_root = fs_info->extent_root;
288 struct btrfs_path *path;
289 struct extent_buffer *leaf;
290 struct btrfs_key key;
296 path = btrfs_alloc_path();
300 exclude_super_stripes(extent_root, block_group);
301 spin_lock(&block_group->space_info->lock);
302 block_group->space_info->bytes_super += block_group->bytes_super;
303 spin_unlock(&block_group->space_info->lock);
305 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
308 * We don't want to deadlock with somebody trying to allocate a new
309 * extent for the extent root while also trying to search the extent
310 * root to add free space. So we skip locking and search the commit
311 * root, since its read-only
313 path->skip_locking = 1;
314 path->search_commit_root = 1;
319 key.type = BTRFS_EXTENT_ITEM_KEY;
321 mutex_lock(&caching_ctl->mutex);
322 /* need to make sure the commit_root doesn't disappear */
323 down_read(&fs_info->extent_commit_sem);
325 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
329 leaf = path->nodes[0];
330 nritems = btrfs_header_nritems(leaf);
334 if (fs_info->closing > 1) {
339 if (path->slots[0] < nritems) {
340 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
342 ret = find_next_key(path, 0, &key);
346 caching_ctl->progress = last;
347 btrfs_release_path(extent_root, path);
348 up_read(&fs_info->extent_commit_sem);
349 mutex_unlock(&caching_ctl->mutex);
350 if (btrfs_transaction_in_commit(fs_info))
357 if (key.objectid < block_group->key.objectid) {
362 if (key.objectid >= block_group->key.objectid +
363 block_group->key.offset)
366 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
367 total_found += add_new_free_space(block_group,
370 last = key.objectid + key.offset;
372 if (total_found > (1024 * 1024 * 2)) {
374 wake_up(&caching_ctl->wait);
381 total_found += add_new_free_space(block_group, fs_info, last,
382 block_group->key.objectid +
383 block_group->key.offset);
384 caching_ctl->progress = (u64)-1;
386 spin_lock(&block_group->lock);
387 block_group->caching_ctl = NULL;
388 block_group->cached = BTRFS_CACHE_FINISHED;
389 spin_unlock(&block_group->lock);
392 btrfs_free_path(path);
393 up_read(&fs_info->extent_commit_sem);
395 free_excluded_extents(extent_root, block_group);
397 mutex_unlock(&caching_ctl->mutex);
398 wake_up(&caching_ctl->wait);
400 put_caching_control(caching_ctl);
401 atomic_dec(&block_group->space_info->caching_threads);
405 static int cache_block_group(struct btrfs_block_group_cache *cache)
407 struct btrfs_fs_info *fs_info = cache->fs_info;
408 struct btrfs_caching_control *caching_ctl;
409 struct task_struct *tsk;
413 if (cache->cached != BTRFS_CACHE_NO)
416 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
417 BUG_ON(!caching_ctl);
419 INIT_LIST_HEAD(&caching_ctl->list);
420 mutex_init(&caching_ctl->mutex);
421 init_waitqueue_head(&caching_ctl->wait);
422 caching_ctl->block_group = cache;
423 caching_ctl->progress = cache->key.objectid;
424 /* one for caching kthread, one for caching block group list */
425 atomic_set(&caching_ctl->count, 2);
427 spin_lock(&cache->lock);
428 if (cache->cached != BTRFS_CACHE_NO) {
429 spin_unlock(&cache->lock);
433 cache->caching_ctl = caching_ctl;
434 cache->cached = BTRFS_CACHE_STARTED;
435 spin_unlock(&cache->lock);
437 down_write(&fs_info->extent_commit_sem);
438 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
439 up_write(&fs_info->extent_commit_sem);
441 atomic_inc(&cache->space_info->caching_threads);
443 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
444 cache->key.objectid);
447 printk(KERN_ERR "error running thread %d\n", ret);
455 * return the block group that starts at or after bytenr
457 static struct btrfs_block_group_cache *
458 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
460 struct btrfs_block_group_cache *cache;
462 cache = block_group_cache_tree_search(info, bytenr, 0);
468 * return the block group that contains the given bytenr
470 struct btrfs_block_group_cache *btrfs_lookup_block_group(
471 struct btrfs_fs_info *info,
474 struct btrfs_block_group_cache *cache;
476 cache = block_group_cache_tree_search(info, bytenr, 1);
481 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
483 if (atomic_dec_and_test(&cache->count))
487 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
490 struct list_head *head = &info->space_info;
491 struct btrfs_space_info *found;
494 list_for_each_entry_rcu(found, head, list) {
495 if (found->flags == flags) {
505 * after adding space to the filesystem, we need to clear the full flags
506 * on all the space infos.
508 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
510 struct list_head *head = &info->space_info;
511 struct btrfs_space_info *found;
514 list_for_each_entry_rcu(found, head, list)
519 static u64 div_factor(u64 num, int factor)
528 u64 btrfs_find_block_group(struct btrfs_root *root,
529 u64 search_start, u64 search_hint, int owner)
531 struct btrfs_block_group_cache *cache;
533 u64 last = max(search_hint, search_start);
540 cache = btrfs_lookup_first_block_group(root->fs_info, last);
544 spin_lock(&cache->lock);
545 last = cache->key.objectid + cache->key.offset;
546 used = btrfs_block_group_used(&cache->item);
548 if ((full_search || !cache->ro) &&
549 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
550 if (used + cache->pinned + cache->reserved <
551 div_factor(cache->key.offset, factor)) {
552 group_start = cache->key.objectid;
553 spin_unlock(&cache->lock);
554 btrfs_put_block_group(cache);
558 spin_unlock(&cache->lock);
559 btrfs_put_block_group(cache);
567 if (!full_search && factor < 10) {
577 /* simple helper to search for an existing extent at a given offset */
578 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
581 struct btrfs_key key;
582 struct btrfs_path *path;
584 path = btrfs_alloc_path();
586 key.objectid = start;
588 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
589 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
591 btrfs_free_path(path);
596 * Back reference rules. Back refs have three main goals:
598 * 1) differentiate between all holders of references to an extent so that
599 * when a reference is dropped we can make sure it was a valid reference
600 * before freeing the extent.
602 * 2) Provide enough information to quickly find the holders of an extent
603 * if we notice a given block is corrupted or bad.
605 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
606 * maintenance. This is actually the same as #2, but with a slightly
607 * different use case.
609 * There are two kinds of back refs. The implicit back refs is optimized
610 * for pointers in non-shared tree blocks. For a given pointer in a block,
611 * back refs of this kind provide information about the block's owner tree
612 * and the pointer's key. These information allow us to find the block by
613 * b-tree searching. The full back refs is for pointers in tree blocks not
614 * referenced by their owner trees. The location of tree block is recorded
615 * in the back refs. Actually the full back refs is generic, and can be
616 * used in all cases the implicit back refs is used. The major shortcoming
617 * of the full back refs is its overhead. Every time a tree block gets
618 * COWed, we have to update back refs entry for all pointers in it.
620 * For a newly allocated tree block, we use implicit back refs for
621 * pointers in it. This means most tree related operations only involve
622 * implicit back refs. For a tree block created in old transaction, the
623 * only way to drop a reference to it is COW it. So we can detect the
624 * event that tree block loses its owner tree's reference and do the
625 * back refs conversion.
627 * When a tree block is COW'd through a tree, there are four cases:
629 * The reference count of the block is one and the tree is the block's
630 * owner tree. Nothing to do in this case.
632 * The reference count of the block is one and the tree is not the
633 * block's owner tree. In this case, full back refs is used for pointers
634 * in the block. Remove these full back refs, add implicit back refs for
635 * every pointers in the new block.
637 * The reference count of the block is greater than one and the tree is
638 * the block's owner tree. In this case, implicit back refs is used for
639 * pointers in the block. Add full back refs for every pointers in the
640 * block, increase lower level extents' reference counts. The original
641 * implicit back refs are entailed to the new block.
643 * The reference count of the block is greater than one and the tree is
644 * not the block's owner tree. Add implicit back refs for every pointer in
645 * the new block, increase lower level extents' reference count.
647 * Back Reference Key composing:
649 * The key objectid corresponds to the first byte in the extent,
650 * The key type is used to differentiate between types of back refs.
651 * There are different meanings of the key offset for different types
654 * File extents can be referenced by:
656 * - multiple snapshots, subvolumes, or different generations in one subvol
657 * - different files inside a single subvolume
658 * - different offsets inside a file (bookend extents in file.c)
660 * The extent ref structure for the implicit back refs has fields for:
662 * - Objectid of the subvolume root
663 * - objectid of the file holding the reference
664 * - original offset in the file
665 * - how many bookend extents
667 * The key offset for the implicit back refs is hash of the first
670 * The extent ref structure for the full back refs has field for:
672 * - number of pointers in the tree leaf
674 * The key offset for the implicit back refs is the first byte of
677 * When a file extent is allocated, The implicit back refs is used.
678 * the fields are filled in:
680 * (root_key.objectid, inode objectid, offset in file, 1)
682 * When a file extent is removed file truncation, we find the
683 * corresponding implicit back refs and check the following fields:
685 * (btrfs_header_owner(leaf), inode objectid, offset in file)
687 * Btree extents can be referenced by:
689 * - Different subvolumes
691 * Both the implicit back refs and the full back refs for tree blocks
692 * only consist of key. The key offset for the implicit back refs is
693 * objectid of block's owner tree. The key offset for the full back refs
694 * is the first byte of parent block.
696 * When implicit back refs is used, information about the lowest key and
697 * level of the tree block are required. These information are stored in
698 * tree block info structure.
701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
702 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
703 struct btrfs_root *root,
704 struct btrfs_path *path,
705 u64 owner, u32 extra_size)
707 struct btrfs_extent_item *item;
708 struct btrfs_extent_item_v0 *ei0;
709 struct btrfs_extent_ref_v0 *ref0;
710 struct btrfs_tree_block_info *bi;
711 struct extent_buffer *leaf;
712 struct btrfs_key key;
713 struct btrfs_key found_key;
714 u32 new_size = sizeof(*item);
718 leaf = path->nodes[0];
719 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
721 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
722 ei0 = btrfs_item_ptr(leaf, path->slots[0],
723 struct btrfs_extent_item_v0);
724 refs = btrfs_extent_refs_v0(leaf, ei0);
726 if (owner == (u64)-1) {
728 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
729 ret = btrfs_next_leaf(root, path);
733 leaf = path->nodes[0];
735 btrfs_item_key_to_cpu(leaf, &found_key,
737 BUG_ON(key.objectid != found_key.objectid);
738 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
742 ref0 = btrfs_item_ptr(leaf, path->slots[0],
743 struct btrfs_extent_ref_v0);
744 owner = btrfs_ref_objectid_v0(leaf, ref0);
748 btrfs_release_path(root, path);
750 if (owner < BTRFS_FIRST_FREE_OBJECTID)
751 new_size += sizeof(*bi);
753 new_size -= sizeof(*ei0);
754 ret = btrfs_search_slot(trans, root, &key, path,
755 new_size + extra_size, 1);
760 ret = btrfs_extend_item(trans, root, path, new_size);
763 leaf = path->nodes[0];
764 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
765 btrfs_set_extent_refs(leaf, item, refs);
766 /* FIXME: get real generation */
767 btrfs_set_extent_generation(leaf, item, 0);
768 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
769 btrfs_set_extent_flags(leaf, item,
770 BTRFS_EXTENT_FLAG_TREE_BLOCK |
771 BTRFS_BLOCK_FLAG_FULL_BACKREF);
772 bi = (struct btrfs_tree_block_info *)(item + 1);
773 /* FIXME: get first key of the block */
774 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
775 btrfs_set_tree_block_level(leaf, bi, (int)owner);
777 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
779 btrfs_mark_buffer_dirty(leaf);
784 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
786 u32 high_crc = ~(u32)0;
787 u32 low_crc = ~(u32)0;
790 lenum = cpu_to_le64(root_objectid);
791 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
792 lenum = cpu_to_le64(owner);
793 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
794 lenum = cpu_to_le64(offset);
795 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
797 return ((u64)high_crc << 31) ^ (u64)low_crc;
800 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
801 struct btrfs_extent_data_ref *ref)
803 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
804 btrfs_extent_data_ref_objectid(leaf, ref),
805 btrfs_extent_data_ref_offset(leaf, ref));
808 static int match_extent_data_ref(struct extent_buffer *leaf,
809 struct btrfs_extent_data_ref *ref,
810 u64 root_objectid, u64 owner, u64 offset)
812 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
813 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
814 btrfs_extent_data_ref_offset(leaf, ref) != offset)
819 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
820 struct btrfs_root *root,
821 struct btrfs_path *path,
822 u64 bytenr, u64 parent,
824 u64 owner, u64 offset)
826 struct btrfs_key key;
827 struct btrfs_extent_data_ref *ref;
828 struct extent_buffer *leaf;
834 key.objectid = bytenr;
836 key.type = BTRFS_SHARED_DATA_REF_KEY;
839 key.type = BTRFS_EXTENT_DATA_REF_KEY;
840 key.offset = hash_extent_data_ref(root_objectid,
845 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
854 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
855 key.type = BTRFS_EXTENT_REF_V0_KEY;
856 btrfs_release_path(root, path);
857 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
868 leaf = path->nodes[0];
869 nritems = btrfs_header_nritems(leaf);
871 if (path->slots[0] >= nritems) {
872 ret = btrfs_next_leaf(root, path);
878 leaf = path->nodes[0];
879 nritems = btrfs_header_nritems(leaf);
883 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
884 if (key.objectid != bytenr ||
885 key.type != BTRFS_EXTENT_DATA_REF_KEY)
888 ref = btrfs_item_ptr(leaf, path->slots[0],
889 struct btrfs_extent_data_ref);
891 if (match_extent_data_ref(leaf, ref, root_objectid,
894 btrfs_release_path(root, path);
906 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
907 struct btrfs_root *root,
908 struct btrfs_path *path,
909 u64 bytenr, u64 parent,
910 u64 root_objectid, u64 owner,
911 u64 offset, int refs_to_add)
913 struct btrfs_key key;
914 struct extent_buffer *leaf;
919 key.objectid = bytenr;
921 key.type = BTRFS_SHARED_DATA_REF_KEY;
923 size = sizeof(struct btrfs_shared_data_ref);
925 key.type = BTRFS_EXTENT_DATA_REF_KEY;
926 key.offset = hash_extent_data_ref(root_objectid,
928 size = sizeof(struct btrfs_extent_data_ref);
931 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
932 if (ret && ret != -EEXIST)
935 leaf = path->nodes[0];
937 struct btrfs_shared_data_ref *ref;
938 ref = btrfs_item_ptr(leaf, path->slots[0],
939 struct btrfs_shared_data_ref);
941 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
943 num_refs = btrfs_shared_data_ref_count(leaf, ref);
944 num_refs += refs_to_add;
945 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
948 struct btrfs_extent_data_ref *ref;
949 while (ret == -EEXIST) {
950 ref = btrfs_item_ptr(leaf, path->slots[0],
951 struct btrfs_extent_data_ref);
952 if (match_extent_data_ref(leaf, ref, root_objectid,
955 btrfs_release_path(root, path);
957 ret = btrfs_insert_empty_item(trans, root, path, &key,
959 if (ret && ret != -EEXIST)
962 leaf = path->nodes[0];
964 ref = btrfs_item_ptr(leaf, path->slots[0],
965 struct btrfs_extent_data_ref);
967 btrfs_set_extent_data_ref_root(leaf, ref,
969 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
970 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
971 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
973 num_refs = btrfs_extent_data_ref_count(leaf, ref);
974 num_refs += refs_to_add;
975 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
978 btrfs_mark_buffer_dirty(leaf);
981 btrfs_release_path(root, path);
985 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root,
987 struct btrfs_path *path,
990 struct btrfs_key key;
991 struct btrfs_extent_data_ref *ref1 = NULL;
992 struct btrfs_shared_data_ref *ref2 = NULL;
993 struct extent_buffer *leaf;
997 leaf = path->nodes[0];
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1001 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1002 struct btrfs_extent_data_ref);
1003 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1004 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1005 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_shared_data_ref);
1007 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1009 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1010 struct btrfs_extent_ref_v0 *ref0;
1011 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1012 struct btrfs_extent_ref_v0);
1013 num_refs = btrfs_ref_count_v0(leaf, ref0);
1019 BUG_ON(num_refs < refs_to_drop);
1020 num_refs -= refs_to_drop;
1022 if (num_refs == 0) {
1023 ret = btrfs_del_item(trans, root, path);
1025 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1026 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1027 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1028 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1031 struct btrfs_extent_ref_v0 *ref0;
1032 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1033 struct btrfs_extent_ref_v0);
1034 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1037 btrfs_mark_buffer_dirty(leaf);
1042 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1043 struct btrfs_path *path,
1044 struct btrfs_extent_inline_ref *iref)
1046 struct btrfs_key key;
1047 struct extent_buffer *leaf;
1048 struct btrfs_extent_data_ref *ref1;
1049 struct btrfs_shared_data_ref *ref2;
1052 leaf = path->nodes[0];
1053 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1055 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1056 BTRFS_EXTENT_DATA_REF_KEY) {
1057 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1058 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1060 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1061 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1063 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1064 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1065 struct btrfs_extent_data_ref);
1066 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1067 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1068 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1069 struct btrfs_shared_data_ref);
1070 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1071 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1072 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1073 struct btrfs_extent_ref_v0 *ref0;
1074 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1075 struct btrfs_extent_ref_v0);
1076 num_refs = btrfs_ref_count_v0(leaf, ref0);
1084 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1085 struct btrfs_root *root,
1086 struct btrfs_path *path,
1087 u64 bytenr, u64 parent,
1090 struct btrfs_key key;
1093 key.objectid = bytenr;
1095 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1096 key.offset = parent;
1098 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1099 key.offset = root_objectid;
1102 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 if (ret == -ENOENT && parent) {
1107 btrfs_release_path(root, path);
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1117 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1118 struct btrfs_root *root,
1119 struct btrfs_path *path,
1120 u64 bytenr, u64 parent,
1123 struct btrfs_key key;
1126 key.objectid = bytenr;
1128 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1129 key.offset = parent;
1131 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1132 key.offset = root_objectid;
1135 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1136 btrfs_release_path(root, path);
1140 static inline int extent_ref_type(u64 parent, u64 owner)
1143 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1145 type = BTRFS_SHARED_BLOCK_REF_KEY;
1147 type = BTRFS_TREE_BLOCK_REF_KEY;
1150 type = BTRFS_SHARED_DATA_REF_KEY;
1152 type = BTRFS_EXTENT_DATA_REF_KEY;
1157 static int find_next_key(struct btrfs_path *path, int level,
1158 struct btrfs_key *key)
1161 for (; level < BTRFS_MAX_LEVEL; level++) {
1162 if (!path->nodes[level])
1164 if (path->slots[level] + 1 >=
1165 btrfs_header_nritems(path->nodes[level]))
1168 btrfs_item_key_to_cpu(path->nodes[level], key,
1169 path->slots[level] + 1);
1171 btrfs_node_key_to_cpu(path->nodes[level], key,
1172 path->slots[level] + 1);
1179 * look for inline back ref. if back ref is found, *ref_ret is set
1180 * to the address of inline back ref, and 0 is returned.
1182 * if back ref isn't found, *ref_ret is set to the address where it
1183 * should be inserted, and -ENOENT is returned.
1185 * if insert is true and there are too many inline back refs, the path
1186 * points to the extent item, and -EAGAIN is returned.
1188 * NOTE: inline back refs are ordered in the same way that back ref
1189 * items in the tree are ordered.
1191 static noinline_for_stack
1192 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_path *path,
1195 struct btrfs_extent_inline_ref **ref_ret,
1196 u64 bytenr, u64 num_bytes,
1197 u64 parent, u64 root_objectid,
1198 u64 owner, u64 offset, int insert)
1200 struct btrfs_key key;
1201 struct extent_buffer *leaf;
1202 struct btrfs_extent_item *ei;
1203 struct btrfs_extent_inline_ref *iref;
1214 key.objectid = bytenr;
1215 key.type = BTRFS_EXTENT_ITEM_KEY;
1216 key.offset = num_bytes;
1218 want = extent_ref_type(parent, owner);
1220 extra_size = btrfs_extent_inline_ref_size(want);
1221 path->keep_locks = 1;
1224 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1231 leaf = path->nodes[0];
1232 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234 if (item_size < sizeof(*ei)) {
1239 ret = convert_extent_item_v0(trans, root, path, owner,
1245 leaf = path->nodes[0];
1246 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1249 BUG_ON(item_size < sizeof(*ei));
1251 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1252 flags = btrfs_extent_flags(leaf, ei);
1254 ptr = (unsigned long)(ei + 1);
1255 end = (unsigned long)ei + item_size;
1257 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1258 ptr += sizeof(struct btrfs_tree_block_info);
1261 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1270 iref = (struct btrfs_extent_inline_ref *)ptr;
1271 type = btrfs_extent_inline_ref_type(leaf, iref);
1275 ptr += btrfs_extent_inline_ref_size(type);
1279 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1280 struct btrfs_extent_data_ref *dref;
1281 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1282 if (match_extent_data_ref(leaf, dref, root_objectid,
1287 if (hash_extent_data_ref_item(leaf, dref) <
1288 hash_extent_data_ref(root_objectid, owner, offset))
1292 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1294 if (parent == ref_offset) {
1298 if (ref_offset < parent)
1301 if (root_objectid == ref_offset) {
1305 if (ref_offset < root_objectid)
1309 ptr += btrfs_extent_inline_ref_size(type);
1311 if (err == -ENOENT && insert) {
1312 if (item_size + extra_size >=
1313 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1318 * To add new inline back ref, we have to make sure
1319 * there is no corresponding back ref item.
1320 * For simplicity, we just do not add new inline back
1321 * ref if there is any kind of item for this block
1323 if (find_next_key(path, 0, &key) == 0 &&
1324 key.objectid == bytenr &&
1325 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1330 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1333 path->keep_locks = 0;
1334 btrfs_unlock_up_safe(path, 1);
1340 * helper to add new inline back ref
1342 static noinline_for_stack
1343 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1344 struct btrfs_root *root,
1345 struct btrfs_path *path,
1346 struct btrfs_extent_inline_ref *iref,
1347 u64 parent, u64 root_objectid,
1348 u64 owner, u64 offset, int refs_to_add,
1349 struct btrfs_delayed_extent_op *extent_op)
1351 struct extent_buffer *leaf;
1352 struct btrfs_extent_item *ei;
1355 unsigned long item_offset;
1361 leaf = path->nodes[0];
1362 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1363 item_offset = (unsigned long)iref - (unsigned long)ei;
1365 type = extent_ref_type(parent, owner);
1366 size = btrfs_extent_inline_ref_size(type);
1368 ret = btrfs_extend_item(trans, root, path, size);
1371 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1372 refs = btrfs_extent_refs(leaf, ei);
1373 refs += refs_to_add;
1374 btrfs_set_extent_refs(leaf, ei, refs);
1376 __run_delayed_extent_op(extent_op, leaf, ei);
1378 ptr = (unsigned long)ei + item_offset;
1379 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1380 if (ptr < end - size)
1381 memmove_extent_buffer(leaf, ptr + size, ptr,
1384 iref = (struct btrfs_extent_inline_ref *)ptr;
1385 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1386 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1387 struct btrfs_extent_data_ref *dref;
1388 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1389 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1390 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1391 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1392 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1393 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1394 struct btrfs_shared_data_ref *sref;
1395 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1396 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1397 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1398 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1399 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1401 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1403 btrfs_mark_buffer_dirty(leaf);
1407 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root,
1409 struct btrfs_path *path,
1410 struct btrfs_extent_inline_ref **ref_ret,
1411 u64 bytenr, u64 num_bytes, u64 parent,
1412 u64 root_objectid, u64 owner, u64 offset)
1416 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1417 bytenr, num_bytes, parent,
1418 root_objectid, owner, offset, 0);
1422 btrfs_release_path(root, path);
1425 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1426 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1429 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1430 root_objectid, owner, offset);
1436 * helper to update/remove inline back ref
1438 static noinline_for_stack
1439 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root,
1441 struct btrfs_path *path,
1442 struct btrfs_extent_inline_ref *iref,
1444 struct btrfs_delayed_extent_op *extent_op)
1446 struct extent_buffer *leaf;
1447 struct btrfs_extent_item *ei;
1448 struct btrfs_extent_data_ref *dref = NULL;
1449 struct btrfs_shared_data_ref *sref = NULL;
1458 leaf = path->nodes[0];
1459 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1460 refs = btrfs_extent_refs(leaf, ei);
1461 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1462 refs += refs_to_mod;
1463 btrfs_set_extent_refs(leaf, ei, refs);
1465 __run_delayed_extent_op(extent_op, leaf, ei);
1467 type = btrfs_extent_inline_ref_type(leaf, iref);
1469 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1470 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471 refs = btrfs_extent_data_ref_count(leaf, dref);
1472 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1473 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1474 refs = btrfs_shared_data_ref_count(leaf, sref);
1477 BUG_ON(refs_to_mod != -1);
1480 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1481 refs += refs_to_mod;
1484 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1485 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1487 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1489 size = btrfs_extent_inline_ref_size(type);
1490 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1491 ptr = (unsigned long)iref;
1492 end = (unsigned long)ei + item_size;
1493 if (ptr + size < end)
1494 memmove_extent_buffer(leaf, ptr, ptr + size,
1497 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1500 btrfs_mark_buffer_dirty(leaf);
1504 static noinline_for_stack
1505 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root,
1507 struct btrfs_path *path,
1508 u64 bytenr, u64 num_bytes, u64 parent,
1509 u64 root_objectid, u64 owner,
1510 u64 offset, int refs_to_add,
1511 struct btrfs_delayed_extent_op *extent_op)
1513 struct btrfs_extent_inline_ref *iref;
1516 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1517 bytenr, num_bytes, parent,
1518 root_objectid, owner, offset, 1);
1520 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1521 ret = update_inline_extent_backref(trans, root, path, iref,
1522 refs_to_add, extent_op);
1523 } else if (ret == -ENOENT) {
1524 ret = setup_inline_extent_backref(trans, root, path, iref,
1525 parent, root_objectid,
1526 owner, offset, refs_to_add,
1532 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root,
1534 struct btrfs_path *path,
1535 u64 bytenr, u64 parent, u64 root_objectid,
1536 u64 owner, u64 offset, int refs_to_add)
1539 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1540 BUG_ON(refs_to_add != 1);
1541 ret = insert_tree_block_ref(trans, root, path, bytenr,
1542 parent, root_objectid);
1544 ret = insert_extent_data_ref(trans, root, path, bytenr,
1545 parent, root_objectid,
1546 owner, offset, refs_to_add);
1551 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1552 struct btrfs_root *root,
1553 struct btrfs_path *path,
1554 struct btrfs_extent_inline_ref *iref,
1555 int refs_to_drop, int is_data)
1559 BUG_ON(!is_data && refs_to_drop != 1);
1561 ret = update_inline_extent_backref(trans, root, path, iref,
1562 -refs_to_drop, NULL);
1563 } else if (is_data) {
1564 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1566 ret = btrfs_del_item(trans, root, path);
1571 #ifdef BIO_RW_DISCARD
1572 static void btrfs_issue_discard(struct block_device *bdev,
1575 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1576 DISCARD_FL_BARRIER);
1580 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1583 #ifdef BIO_RW_DISCARD
1585 u64 map_length = num_bytes;
1586 struct btrfs_multi_bio *multi = NULL;
1588 /* Tell the block device(s) that the sectors can be discarded */
1589 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1590 bytenr, &map_length, &multi, 0);
1592 struct btrfs_bio_stripe *stripe = multi->stripes;
1595 if (map_length > num_bytes)
1596 map_length = num_bytes;
1598 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1599 btrfs_issue_discard(stripe->dev->bdev,
1612 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1613 struct btrfs_root *root,
1614 u64 bytenr, u64 num_bytes, u64 parent,
1615 u64 root_objectid, u64 owner, u64 offset)
1618 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1619 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1621 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1622 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1623 parent, root_objectid, (int)owner,
1624 BTRFS_ADD_DELAYED_REF, NULL);
1626 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1627 parent, root_objectid, owner, offset,
1628 BTRFS_ADD_DELAYED_REF, NULL);
1633 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1634 struct btrfs_root *root,
1635 u64 bytenr, u64 num_bytes,
1636 u64 parent, u64 root_objectid,
1637 u64 owner, u64 offset, int refs_to_add,
1638 struct btrfs_delayed_extent_op *extent_op)
1640 struct btrfs_path *path;
1641 struct extent_buffer *leaf;
1642 struct btrfs_extent_item *item;
1647 path = btrfs_alloc_path();
1652 path->leave_spinning = 1;
1653 /* this will setup the path even if it fails to insert the back ref */
1654 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1655 path, bytenr, num_bytes, parent,
1656 root_objectid, owner, offset,
1657 refs_to_add, extent_op);
1661 if (ret != -EAGAIN) {
1666 leaf = path->nodes[0];
1667 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1668 refs = btrfs_extent_refs(leaf, item);
1669 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1671 __run_delayed_extent_op(extent_op, leaf, item);
1673 btrfs_mark_buffer_dirty(leaf);
1674 btrfs_release_path(root->fs_info->extent_root, path);
1677 path->leave_spinning = 1;
1679 /* now insert the actual backref */
1680 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1681 path, bytenr, parent, root_objectid,
1682 owner, offset, refs_to_add);
1685 btrfs_free_path(path);
1689 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1690 struct btrfs_root *root,
1691 struct btrfs_delayed_ref_node *node,
1692 struct btrfs_delayed_extent_op *extent_op,
1693 int insert_reserved)
1696 struct btrfs_delayed_data_ref *ref;
1697 struct btrfs_key ins;
1702 ins.objectid = node->bytenr;
1703 ins.offset = node->num_bytes;
1704 ins.type = BTRFS_EXTENT_ITEM_KEY;
1706 ref = btrfs_delayed_node_to_data_ref(node);
1707 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1708 parent = ref->parent;
1710 ref_root = ref->root;
1712 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1714 BUG_ON(extent_op->update_key);
1715 flags |= extent_op->flags_to_set;
1717 ret = alloc_reserved_file_extent(trans, root,
1718 parent, ref_root, flags,
1719 ref->objectid, ref->offset,
1720 &ins, node->ref_mod);
1721 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1722 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1723 node->num_bytes, parent,
1724 ref_root, ref->objectid,
1725 ref->offset, node->ref_mod,
1727 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1728 ret = __btrfs_free_extent(trans, root, node->bytenr,
1729 node->num_bytes, parent,
1730 ref_root, ref->objectid,
1731 ref->offset, node->ref_mod,
1739 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1740 struct extent_buffer *leaf,
1741 struct btrfs_extent_item *ei)
1743 u64 flags = btrfs_extent_flags(leaf, ei);
1744 if (extent_op->update_flags) {
1745 flags |= extent_op->flags_to_set;
1746 btrfs_set_extent_flags(leaf, ei, flags);
1749 if (extent_op->update_key) {
1750 struct btrfs_tree_block_info *bi;
1751 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1752 bi = (struct btrfs_tree_block_info *)(ei + 1);
1753 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1757 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1758 struct btrfs_root *root,
1759 struct btrfs_delayed_ref_node *node,
1760 struct btrfs_delayed_extent_op *extent_op)
1762 struct btrfs_key key;
1763 struct btrfs_path *path;
1764 struct btrfs_extent_item *ei;
1765 struct extent_buffer *leaf;
1770 path = btrfs_alloc_path();
1774 key.objectid = node->bytenr;
1775 key.type = BTRFS_EXTENT_ITEM_KEY;
1776 key.offset = node->num_bytes;
1779 path->leave_spinning = 1;
1780 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1791 leaf = path->nodes[0];
1792 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1793 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1794 if (item_size < sizeof(*ei)) {
1795 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1801 leaf = path->nodes[0];
1802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1805 BUG_ON(item_size < sizeof(*ei));
1806 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1807 __run_delayed_extent_op(extent_op, leaf, ei);
1809 btrfs_mark_buffer_dirty(leaf);
1811 btrfs_free_path(path);
1815 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1816 struct btrfs_root *root,
1817 struct btrfs_delayed_ref_node *node,
1818 struct btrfs_delayed_extent_op *extent_op,
1819 int insert_reserved)
1822 struct btrfs_delayed_tree_ref *ref;
1823 struct btrfs_key ins;
1827 ins.objectid = node->bytenr;
1828 ins.offset = node->num_bytes;
1829 ins.type = BTRFS_EXTENT_ITEM_KEY;
1831 ref = btrfs_delayed_node_to_tree_ref(node);
1832 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1833 parent = ref->parent;
1835 ref_root = ref->root;
1837 BUG_ON(node->ref_mod != 1);
1838 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1839 BUG_ON(!extent_op || !extent_op->update_flags ||
1840 !extent_op->update_key);
1841 ret = alloc_reserved_tree_block(trans, root,
1843 extent_op->flags_to_set,
1846 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1847 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1848 node->num_bytes, parent, ref_root,
1849 ref->level, 0, 1, extent_op);
1850 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1851 ret = __btrfs_free_extent(trans, root, node->bytenr,
1852 node->num_bytes, parent, ref_root,
1853 ref->level, 0, 1, extent_op);
1861 /* helper function to actually process a single delayed ref entry */
1862 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1863 struct btrfs_root *root,
1864 struct btrfs_delayed_ref_node *node,
1865 struct btrfs_delayed_extent_op *extent_op,
1866 int insert_reserved)
1869 if (btrfs_delayed_ref_is_head(node)) {
1870 struct btrfs_delayed_ref_head *head;
1872 * we've hit the end of the chain and we were supposed
1873 * to insert this extent into the tree. But, it got
1874 * deleted before we ever needed to insert it, so all
1875 * we have to do is clean up the accounting
1878 head = btrfs_delayed_node_to_head(node);
1879 if (insert_reserved) {
1881 struct extent_buffer *must_clean = NULL;
1883 ret = pin_down_bytes(trans, root, NULL,
1884 node->bytenr, node->num_bytes,
1885 head->is_data, 1, &must_clean);
1890 clean_tree_block(NULL, root, must_clean);
1891 btrfs_tree_unlock(must_clean);
1892 free_extent_buffer(must_clean);
1894 if (head->is_data) {
1895 ret = btrfs_del_csums(trans, root,
1901 ret = btrfs_free_reserved_extent(root,
1907 mutex_unlock(&head->mutex);
1911 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1912 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1913 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1915 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1916 node->type == BTRFS_SHARED_DATA_REF_KEY)
1917 ret = run_delayed_data_ref(trans, root, node, extent_op,
1924 static noinline struct btrfs_delayed_ref_node *
1925 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1927 struct rb_node *node;
1928 struct btrfs_delayed_ref_node *ref;
1929 int action = BTRFS_ADD_DELAYED_REF;
1932 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1933 * this prevents ref count from going down to zero when
1934 * there still are pending delayed ref.
1936 node = rb_prev(&head->node.rb_node);
1940 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1942 if (ref->bytenr != head->node.bytenr)
1944 if (ref->action == action)
1946 node = rb_prev(node);
1948 if (action == BTRFS_ADD_DELAYED_REF) {
1949 action = BTRFS_DROP_DELAYED_REF;
1955 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1956 struct btrfs_root *root,
1957 struct list_head *cluster)
1959 struct btrfs_delayed_ref_root *delayed_refs;
1960 struct btrfs_delayed_ref_node *ref;
1961 struct btrfs_delayed_ref_head *locked_ref = NULL;
1962 struct btrfs_delayed_extent_op *extent_op;
1965 int must_insert_reserved = 0;
1967 delayed_refs = &trans->transaction->delayed_refs;
1970 /* pick a new head ref from the cluster list */
1971 if (list_empty(cluster))
1974 locked_ref = list_entry(cluster->next,
1975 struct btrfs_delayed_ref_head, cluster);
1977 /* grab the lock that says we are going to process
1978 * all the refs for this head */
1979 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1982 * we may have dropped the spin lock to get the head
1983 * mutex lock, and that might have given someone else
1984 * time to free the head. If that's true, it has been
1985 * removed from our list and we can move on.
1987 if (ret == -EAGAIN) {
1995 * record the must insert reserved flag before we
1996 * drop the spin lock.
1998 must_insert_reserved = locked_ref->must_insert_reserved;
1999 locked_ref->must_insert_reserved = 0;
2001 extent_op = locked_ref->extent_op;
2002 locked_ref->extent_op = NULL;
2005 * locked_ref is the head node, so we have to go one
2006 * node back for any delayed ref updates
2008 ref = select_delayed_ref(locked_ref);
2010 /* All delayed refs have been processed, Go ahead
2011 * and send the head node to run_one_delayed_ref,
2012 * so that any accounting fixes can happen
2014 ref = &locked_ref->node;
2016 if (extent_op && must_insert_reserved) {
2022 spin_unlock(&delayed_refs->lock);
2024 ret = run_delayed_extent_op(trans, root,
2030 spin_lock(&delayed_refs->lock);
2034 list_del_init(&locked_ref->cluster);
2039 rb_erase(&ref->rb_node, &delayed_refs->root);
2040 delayed_refs->num_entries--;
2042 spin_unlock(&delayed_refs->lock);
2044 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2045 must_insert_reserved);
2048 btrfs_put_delayed_ref(ref);
2053 spin_lock(&delayed_refs->lock);
2059 * this starts processing the delayed reference count updates and
2060 * extent insertions we have queued up so far. count can be
2061 * 0, which means to process everything in the tree at the start
2062 * of the run (but not newly added entries), or it can be some target
2063 * number you'd like to process.
2065 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2066 struct btrfs_root *root, unsigned long count)
2068 struct rb_node *node;
2069 struct btrfs_delayed_ref_root *delayed_refs;
2070 struct btrfs_delayed_ref_node *ref;
2071 struct list_head cluster;
2073 int run_all = count == (unsigned long)-1;
2076 if (root == root->fs_info->extent_root)
2077 root = root->fs_info->tree_root;
2079 delayed_refs = &trans->transaction->delayed_refs;
2080 INIT_LIST_HEAD(&cluster);
2082 spin_lock(&delayed_refs->lock);
2084 count = delayed_refs->num_entries * 2;
2088 if (!(run_all || run_most) &&
2089 delayed_refs->num_heads_ready < 64)
2093 * go find something we can process in the rbtree. We start at
2094 * the beginning of the tree, and then build a cluster
2095 * of refs to process starting at the first one we are able to
2098 ret = btrfs_find_ref_cluster(trans, &cluster,
2099 delayed_refs->run_delayed_start);
2103 ret = run_clustered_refs(trans, root, &cluster);
2106 count -= min_t(unsigned long, ret, count);
2113 node = rb_first(&delayed_refs->root);
2116 count = (unsigned long)-1;
2119 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2121 if (btrfs_delayed_ref_is_head(ref)) {
2122 struct btrfs_delayed_ref_head *head;
2124 head = btrfs_delayed_node_to_head(ref);
2125 atomic_inc(&ref->refs);
2127 spin_unlock(&delayed_refs->lock);
2128 mutex_lock(&head->mutex);
2129 mutex_unlock(&head->mutex);
2131 btrfs_put_delayed_ref(ref);
2135 node = rb_next(node);
2137 spin_unlock(&delayed_refs->lock);
2138 schedule_timeout(1);
2142 spin_unlock(&delayed_refs->lock);
2146 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2147 struct btrfs_root *root,
2148 u64 bytenr, u64 num_bytes, u64 flags,
2151 struct btrfs_delayed_extent_op *extent_op;
2154 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2158 extent_op->flags_to_set = flags;
2159 extent_op->update_flags = 1;
2160 extent_op->update_key = 0;
2161 extent_op->is_data = is_data ? 1 : 0;
2163 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2169 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2170 struct btrfs_root *root,
2171 struct btrfs_path *path,
2172 u64 objectid, u64 offset, u64 bytenr)
2174 struct btrfs_delayed_ref_head *head;
2175 struct btrfs_delayed_ref_node *ref;
2176 struct btrfs_delayed_data_ref *data_ref;
2177 struct btrfs_delayed_ref_root *delayed_refs;
2178 struct rb_node *node;
2182 delayed_refs = &trans->transaction->delayed_refs;
2183 spin_lock(&delayed_refs->lock);
2184 head = btrfs_find_delayed_ref_head(trans, bytenr);
2188 if (!mutex_trylock(&head->mutex)) {
2189 atomic_inc(&head->node.refs);
2190 spin_unlock(&delayed_refs->lock);
2192 btrfs_release_path(root->fs_info->extent_root, path);
2194 mutex_lock(&head->mutex);
2195 mutex_unlock(&head->mutex);
2196 btrfs_put_delayed_ref(&head->node);
2200 node = rb_prev(&head->node.rb_node);
2204 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2206 if (ref->bytenr != bytenr)
2210 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2213 data_ref = btrfs_delayed_node_to_data_ref(ref);
2215 node = rb_prev(node);
2217 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2218 if (ref->bytenr == bytenr)
2222 if (data_ref->root != root->root_key.objectid ||
2223 data_ref->objectid != objectid || data_ref->offset != offset)
2228 mutex_unlock(&head->mutex);
2230 spin_unlock(&delayed_refs->lock);
2234 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root,
2236 struct btrfs_path *path,
2237 u64 objectid, u64 offset, u64 bytenr)
2239 struct btrfs_root *extent_root = root->fs_info->extent_root;
2240 struct extent_buffer *leaf;
2241 struct btrfs_extent_data_ref *ref;
2242 struct btrfs_extent_inline_ref *iref;
2243 struct btrfs_extent_item *ei;
2244 struct btrfs_key key;
2248 key.objectid = bytenr;
2249 key.offset = (u64)-1;
2250 key.type = BTRFS_EXTENT_ITEM_KEY;
2252 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2258 if (path->slots[0] == 0)
2262 leaf = path->nodes[0];
2263 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2265 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2269 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2270 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2271 if (item_size < sizeof(*ei)) {
2272 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2276 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2278 if (item_size != sizeof(*ei) +
2279 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2282 if (btrfs_extent_generation(leaf, ei) <=
2283 btrfs_root_last_snapshot(&root->root_item))
2286 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2287 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2288 BTRFS_EXTENT_DATA_REF_KEY)
2291 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2292 if (btrfs_extent_refs(leaf, ei) !=
2293 btrfs_extent_data_ref_count(leaf, ref) ||
2294 btrfs_extent_data_ref_root(leaf, ref) !=
2295 root->root_key.objectid ||
2296 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2297 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2305 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2306 struct btrfs_root *root,
2307 u64 objectid, u64 offset, u64 bytenr)
2309 struct btrfs_path *path;
2313 path = btrfs_alloc_path();
2318 ret = check_committed_ref(trans, root, path, objectid,
2320 if (ret && ret != -ENOENT)
2323 ret2 = check_delayed_ref(trans, root, path, objectid,
2325 } while (ret2 == -EAGAIN);
2327 if (ret2 && ret2 != -ENOENT) {
2332 if (ret != -ENOENT || ret2 != -ENOENT)
2335 btrfs_free_path(path);
2340 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2341 struct extent_buffer *buf, u32 nr_extents)
2343 struct btrfs_key key;
2344 struct btrfs_file_extent_item *fi;
2352 if (!root->ref_cows)
2355 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2357 root_gen = root->root_key.offset;
2360 root_gen = trans->transid - 1;
2363 level = btrfs_header_level(buf);
2364 nritems = btrfs_header_nritems(buf);
2367 struct btrfs_leaf_ref *ref;
2368 struct btrfs_extent_info *info;
2370 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2376 ref->root_gen = root_gen;
2377 ref->bytenr = buf->start;
2378 ref->owner = btrfs_header_owner(buf);
2379 ref->generation = btrfs_header_generation(buf);
2380 ref->nritems = nr_extents;
2381 info = ref->extents;
2383 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2385 btrfs_item_key_to_cpu(buf, &key, i);
2386 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2388 fi = btrfs_item_ptr(buf, i,
2389 struct btrfs_file_extent_item);
2390 if (btrfs_file_extent_type(buf, fi) ==
2391 BTRFS_FILE_EXTENT_INLINE)
2393 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2394 if (disk_bytenr == 0)
2397 info->bytenr = disk_bytenr;
2399 btrfs_file_extent_disk_num_bytes(buf, fi);
2400 info->objectid = key.objectid;
2401 info->offset = key.offset;
2405 ret = btrfs_add_leaf_ref(root, ref, shared);
2406 if (ret == -EEXIST && shared) {
2407 struct btrfs_leaf_ref *old;
2408 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2410 btrfs_remove_leaf_ref(root, old);
2411 btrfs_free_leaf_ref(root, old);
2412 ret = btrfs_add_leaf_ref(root, ref, shared);
2415 btrfs_free_leaf_ref(root, ref);
2421 /* when a block goes through cow, we update the reference counts of
2422 * everything that block points to. The internal pointers of the block
2423 * can be in just about any order, and it is likely to have clusters of
2424 * things that are close together and clusters of things that are not.
2426 * To help reduce the seeks that come with updating all of these reference
2427 * counts, sort them by byte number before actual updates are done.
2429 * struct refsort is used to match byte number to slot in the btree block.
2430 * we sort based on the byte number and then use the slot to actually
2433 * struct refsort is smaller than strcut btrfs_item and smaller than
2434 * struct btrfs_key_ptr. Since we're currently limited to the page size
2435 * for a btree block, there's no way for a kmalloc of refsorts for a
2436 * single node to be bigger than a page.
2444 * for passing into sort()
2446 static int refsort_cmp(const void *a_void, const void *b_void)
2448 const struct refsort *a = a_void;
2449 const struct refsort *b = b_void;
2451 if (a->bytenr < b->bytenr)
2453 if (a->bytenr > b->bytenr)
2459 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2460 struct btrfs_root *root,
2461 struct extent_buffer *buf,
2462 int full_backref, int inc)
2469 struct btrfs_key key;
2470 struct btrfs_file_extent_item *fi;
2474 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2475 u64, u64, u64, u64, u64, u64);
2477 ref_root = btrfs_header_owner(buf);
2478 nritems = btrfs_header_nritems(buf);
2479 level = btrfs_header_level(buf);
2481 if (!root->ref_cows && level == 0)
2485 process_func = btrfs_inc_extent_ref;
2487 process_func = btrfs_free_extent;
2490 parent = buf->start;
2494 for (i = 0; i < nritems; i++) {
2496 btrfs_item_key_to_cpu(buf, &key, i);
2497 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2499 fi = btrfs_item_ptr(buf, i,
2500 struct btrfs_file_extent_item);
2501 if (btrfs_file_extent_type(buf, fi) ==
2502 BTRFS_FILE_EXTENT_INLINE)
2504 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2508 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2509 key.offset -= btrfs_file_extent_offset(buf, fi);
2510 ret = process_func(trans, root, bytenr, num_bytes,
2511 parent, ref_root, key.objectid,
2516 bytenr = btrfs_node_blockptr(buf, i);
2517 num_bytes = btrfs_level_size(root, level - 1);
2518 ret = process_func(trans, root, bytenr, num_bytes,
2519 parent, ref_root, level - 1, 0);
2530 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2531 struct extent_buffer *buf, int full_backref)
2533 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2536 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2537 struct extent_buffer *buf, int full_backref)
2539 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2542 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2543 struct btrfs_root *root,
2544 struct btrfs_path *path,
2545 struct btrfs_block_group_cache *cache)
2548 struct btrfs_root *extent_root = root->fs_info->extent_root;
2550 struct extent_buffer *leaf;
2552 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2557 leaf = path->nodes[0];
2558 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2559 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2560 btrfs_mark_buffer_dirty(leaf);
2561 btrfs_release_path(extent_root, path);
2569 static struct btrfs_block_group_cache *
2570 next_block_group(struct btrfs_root *root,
2571 struct btrfs_block_group_cache *cache)
2573 struct rb_node *node;
2574 spin_lock(&root->fs_info->block_group_cache_lock);
2575 node = rb_next(&cache->cache_node);
2576 btrfs_put_block_group(cache);
2578 cache = rb_entry(node, struct btrfs_block_group_cache,
2580 atomic_inc(&cache->count);
2583 spin_unlock(&root->fs_info->block_group_cache_lock);
2587 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2588 struct btrfs_root *root)
2590 struct btrfs_block_group_cache *cache;
2592 struct btrfs_path *path;
2595 path = btrfs_alloc_path();
2601 err = btrfs_run_delayed_refs(trans, root,
2606 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2610 cache = next_block_group(root, cache);
2620 last = cache->key.objectid + cache->key.offset;
2622 err = write_one_cache_group(trans, root, path, cache);
2624 btrfs_put_block_group(cache);
2627 btrfs_free_path(path);
2631 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2633 struct btrfs_block_group_cache *block_group;
2636 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2637 if (!block_group || block_group->ro)
2640 btrfs_put_block_group(block_group);
2644 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2645 u64 total_bytes, u64 bytes_used,
2646 struct btrfs_space_info **space_info)
2648 struct btrfs_space_info *found;
2650 found = __find_space_info(info, flags);
2652 spin_lock(&found->lock);
2653 found->total_bytes += total_bytes;
2654 found->bytes_used += bytes_used;
2656 spin_unlock(&found->lock);
2657 *space_info = found;
2660 found = kzalloc(sizeof(*found), GFP_NOFS);
2664 INIT_LIST_HEAD(&found->block_groups);
2665 init_rwsem(&found->groups_sem);
2666 spin_lock_init(&found->lock);
2667 found->flags = flags;
2668 found->total_bytes = total_bytes;
2669 found->bytes_used = bytes_used;
2670 found->bytes_pinned = 0;
2671 found->bytes_reserved = 0;
2672 found->bytes_readonly = 0;
2673 found->bytes_delalloc = 0;
2675 found->force_alloc = 0;
2676 *space_info = found;
2677 list_add_rcu(&found->list, &info->space_info);
2678 atomic_set(&found->caching_threads, 0);
2682 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2684 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2685 BTRFS_BLOCK_GROUP_RAID1 |
2686 BTRFS_BLOCK_GROUP_RAID10 |
2687 BTRFS_BLOCK_GROUP_DUP);
2689 if (flags & BTRFS_BLOCK_GROUP_DATA)
2690 fs_info->avail_data_alloc_bits |= extra_flags;
2691 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2692 fs_info->avail_metadata_alloc_bits |= extra_flags;
2693 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2694 fs_info->avail_system_alloc_bits |= extra_flags;
2698 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2700 spin_lock(&cache->space_info->lock);
2701 spin_lock(&cache->lock);
2703 cache->space_info->bytes_readonly += cache->key.offset -
2704 btrfs_block_group_used(&cache->item);
2707 spin_unlock(&cache->lock);
2708 spin_unlock(&cache->space_info->lock);
2711 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2713 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2715 if (num_devices == 1)
2716 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2717 if (num_devices < 4)
2718 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2720 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2721 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2722 BTRFS_BLOCK_GROUP_RAID10))) {
2723 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2726 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2727 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2728 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2731 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2732 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2733 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2734 (flags & BTRFS_BLOCK_GROUP_DUP)))
2735 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2739 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2741 struct btrfs_fs_info *info = root->fs_info;
2745 alloc_profile = info->avail_data_alloc_bits &
2746 info->data_alloc_profile;
2747 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2748 } else if (root == root->fs_info->chunk_root) {
2749 alloc_profile = info->avail_system_alloc_bits &
2750 info->system_alloc_profile;
2751 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2753 alloc_profile = info->avail_metadata_alloc_bits &
2754 info->metadata_alloc_profile;
2755 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2758 return btrfs_reduce_alloc_profile(root, data);
2761 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2765 alloc_target = btrfs_get_alloc_profile(root, 1);
2766 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2770 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2775 level = BTRFS_MAX_LEVEL - 2;
2777 * NOTE: these calculations are absolutely the worst possible case.
2778 * This assumes that _every_ item we insert will require a new leaf, and
2779 * that the tree has grown to its maximum level size.
2783 * for every item we insert we could insert both an extent item and a
2784 * extent ref item. Then for ever item we insert, we will need to cow
2785 * both the original leaf, plus the leaf to the left and right of it.
2787 * Unless we are talking about the extent root, then we just want the
2788 * number of items * 2, since we just need the extent item plus its ref.
2790 if (root == root->fs_info->extent_root)
2791 num_bytes = num_items * 2;
2793 num_bytes = (num_items + (2 * num_items)) * 3;
2796 * num_bytes is total number of leaves we could need times the leaf
2797 * size, and then for every leaf we could end up cow'ing 2 nodes per
2798 * level, down to the leaf level.
2800 num_bytes = (num_bytes * root->leafsize) +
2801 (num_bytes * (level * 2)) * root->nodesize;
2807 * Unreserve metadata space for delalloc. If we have less reserved credits than
2808 * we have extents, this function does nothing.
2810 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2811 struct inode *inode, int num_items)
2813 struct btrfs_fs_info *info = root->fs_info;
2814 struct btrfs_space_info *meta_sinfo;
2819 /* get the space info for where the metadata will live */
2820 alloc_target = btrfs_get_alloc_profile(root, 0);
2821 meta_sinfo = __find_space_info(info, alloc_target);
2823 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2826 spin_lock(&meta_sinfo->lock);
2827 if (BTRFS_I(inode)->delalloc_reserved_extents <=
2828 BTRFS_I(inode)->delalloc_extents) {
2829 spin_unlock(&meta_sinfo->lock);
2833 BTRFS_I(inode)->delalloc_reserved_extents--;
2834 BUG_ON(BTRFS_I(inode)->delalloc_reserved_extents < 0);
2836 if (meta_sinfo->bytes_delalloc < num_bytes) {
2838 meta_sinfo->bytes_delalloc = 0;
2840 meta_sinfo->bytes_delalloc -= num_bytes;
2842 spin_unlock(&meta_sinfo->lock);
2849 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2853 thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2854 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2855 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2856 meta_sinfo->bytes_may_use;
2858 thresh = meta_sinfo->total_bytes - thresh;
2860 do_div(thresh, 100);
2861 if (thresh <= meta_sinfo->bytes_delalloc)
2862 meta_sinfo->force_delalloc = 1;
2864 meta_sinfo->force_delalloc = 0;
2867 static int maybe_allocate_chunk(struct btrfs_root *root,
2868 struct btrfs_space_info *info)
2870 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2871 struct btrfs_trans_handle *trans;
2877 free_space = btrfs_super_total_bytes(disk_super);
2879 * we allow the metadata to grow to a max of either 5gb or 5% of the
2880 * space in the volume.
2882 min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2883 div64_u64(free_space * 5, 100));
2884 if (info->total_bytes >= min_metadata) {
2885 spin_unlock(&info->lock);
2890 spin_unlock(&info->lock);
2894 if (!info->allocating_chunk) {
2895 info->force_alloc = 1;
2896 info->allocating_chunk = 1;
2897 init_waitqueue_head(&info->wait);
2902 spin_unlock(&info->lock);
2905 wait_event(info->wait,
2906 !info->allocating_chunk);
2910 trans = btrfs_start_transaction(root, 1);
2916 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2917 4096 + 2 * 1024 * 1024,
2919 btrfs_end_transaction(trans, root);
2923 spin_lock(&info->lock);
2924 info->allocating_chunk = 0;
2925 spin_unlock(&info->lock);
2926 wake_up(&info->wait);
2934 * Reserve metadata space for delalloc.
2936 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
2937 struct inode *inode, int num_items)
2939 struct btrfs_fs_info *info = root->fs_info;
2940 struct btrfs_space_info *meta_sinfo;
2947 /* get the space info for where the metadata will live */
2948 alloc_target = btrfs_get_alloc_profile(root, 0);
2949 meta_sinfo = __find_space_info(info, alloc_target);
2951 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2954 spin_lock(&meta_sinfo->lock);
2956 force_delalloc = meta_sinfo->force_delalloc;
2958 if (unlikely(!meta_sinfo->bytes_root))
2959 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
2962 meta_sinfo->bytes_delalloc += num_bytes;
2964 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2965 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2966 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2967 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
2969 if (used > meta_sinfo->total_bytes) {
2973 if (maybe_allocate_chunk(root, meta_sinfo))
2977 spin_unlock(&meta_sinfo->lock);
2981 filemap_flush(inode->i_mapping);
2983 } else if (flushed == 3) {
2984 btrfs_start_delalloc_inodes(root);
2985 btrfs_wait_ordered_extents(root, 0);
2988 spin_lock(&meta_sinfo->lock);
2989 meta_sinfo->bytes_delalloc -= num_bytes;
2990 spin_unlock(&meta_sinfo->lock);
2991 printk(KERN_ERR "enospc, has %d, reserved %d\n",
2992 BTRFS_I(inode)->delalloc_extents,
2993 BTRFS_I(inode)->delalloc_reserved_extents);
2994 dump_space_info(meta_sinfo, 0, 0);
2998 BTRFS_I(inode)->delalloc_reserved_extents++;
2999 check_force_delalloc(meta_sinfo);
3000 spin_unlock(&meta_sinfo->lock);
3002 if (!flushed && force_delalloc)
3003 filemap_flush(inode->i_mapping);
3009 * unreserve num_items number of items worth of metadata space. This needs to
3010 * be paired with btrfs_reserve_metadata_space.
3012 * NOTE: if you have the option, run this _AFTER_ you do a
3013 * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3014 * oprations which will result in more used metadata, so we want to make sure we
3015 * can do that without issue.
3017 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3019 struct btrfs_fs_info *info = root->fs_info;
3020 struct btrfs_space_info *meta_sinfo;
3025 /* get the space info for where the metadata will live */
3026 alloc_target = btrfs_get_alloc_profile(root, 0);
3027 meta_sinfo = __find_space_info(info, alloc_target);
3029 num_bytes = calculate_bytes_needed(root, num_items);
3031 spin_lock(&meta_sinfo->lock);
3032 if (meta_sinfo->bytes_may_use < num_bytes) {
3034 meta_sinfo->bytes_may_use = 0;
3036 meta_sinfo->bytes_may_use -= num_bytes;
3038 spin_unlock(&meta_sinfo->lock);
3046 * Reserve some metadata space for use. We'll calculate the worste case number
3047 * of bytes that would be needed to modify num_items number of items. If we
3048 * have space, fantastic, if not, you get -ENOSPC. Please call
3049 * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3050 * items you reserved, since whatever metadata you needed should have already
3053 * This will commit the transaction to make more space if we don't have enough
3054 * metadata space. THe only time we don't do this is if we're reserving space
3055 * inside of a transaction, then we will just return -ENOSPC and it is the
3056 * callers responsibility to handle it properly.
3058 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3060 struct btrfs_fs_info *info = root->fs_info;
3061 struct btrfs_space_info *meta_sinfo;
3067 /* get the space info for where the metadata will live */
3068 alloc_target = btrfs_get_alloc_profile(root, 0);
3069 meta_sinfo = __find_space_info(info, alloc_target);
3071 num_bytes = calculate_bytes_needed(root, num_items);
3073 spin_lock(&meta_sinfo->lock);
3075 if (unlikely(!meta_sinfo->bytes_root))
3076 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3079 meta_sinfo->bytes_may_use += num_bytes;
3081 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3082 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3083 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3084 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3086 if (used > meta_sinfo->total_bytes) {
3089 if (maybe_allocate_chunk(root, meta_sinfo))
3093 spin_unlock(&meta_sinfo->lock);
3097 btrfs_start_delalloc_inodes(root);
3098 btrfs_wait_ordered_extents(root, 0);
3101 spin_lock(&meta_sinfo->lock);
3102 meta_sinfo->bytes_may_use -= num_bytes;
3103 spin_unlock(&meta_sinfo->lock);
3105 dump_space_info(meta_sinfo, 0, 0);
3109 check_force_delalloc(meta_sinfo);
3110 spin_unlock(&meta_sinfo->lock);
3116 * This will check the space that the inode allocates from to make sure we have
3117 * enough space for bytes.
3119 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3122 struct btrfs_space_info *data_sinfo;
3123 int ret = 0, committed = 0;
3125 /* make sure bytes are sectorsize aligned */
3126 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3128 data_sinfo = BTRFS_I(inode)->space_info;
3133 /* make sure we have enough space to handle the data first */
3134 spin_lock(&data_sinfo->lock);
3135 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3136 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3137 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3138 data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3139 struct btrfs_trans_handle *trans;
3142 * if we don't have enough free bytes in this space then we need
3143 * to alloc a new chunk.
3145 if (!data_sinfo->full) {
3148 data_sinfo->force_alloc = 1;
3149 spin_unlock(&data_sinfo->lock);
3151 alloc_target = btrfs_get_alloc_profile(root, 1);
3152 trans = btrfs_start_transaction(root, 1);
3156 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3157 bytes + 2 * 1024 * 1024,
3159 btrfs_end_transaction(trans, root);
3164 btrfs_set_inode_space_info(root, inode);
3165 data_sinfo = BTRFS_I(inode)->space_info;
3169 spin_unlock(&data_sinfo->lock);
3171 /* commit the current transaction and try again */
3172 if (!committed && !root->fs_info->open_ioctl_trans) {
3174 trans = btrfs_join_transaction(root, 1);
3177 ret = btrfs_commit_transaction(trans, root);
3183 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3184 ", %llu bytes_used, %llu bytes_reserved, "
3185 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3186 "%llu total\n", (unsigned long long)bytes,
3187 (unsigned long long)data_sinfo->bytes_delalloc,
3188 (unsigned long long)data_sinfo->bytes_used,
3189 (unsigned long long)data_sinfo->bytes_reserved,
3190 (unsigned long long)data_sinfo->bytes_pinned,
3191 (unsigned long long)data_sinfo->bytes_readonly,
3192 (unsigned long long)data_sinfo->bytes_may_use,
3193 (unsigned long long)data_sinfo->total_bytes);
3196 data_sinfo->bytes_may_use += bytes;
3197 BTRFS_I(inode)->reserved_bytes += bytes;
3198 spin_unlock(&data_sinfo->lock);
3204 * if there was an error for whatever reason after calling
3205 * btrfs_check_data_free_space, call this so we can cleanup the counters.
3207 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3208 struct inode *inode, u64 bytes)
3210 struct btrfs_space_info *data_sinfo;
3212 /* make sure bytes are sectorsize aligned */
3213 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3215 data_sinfo = BTRFS_I(inode)->space_info;
3216 spin_lock(&data_sinfo->lock);
3217 data_sinfo->bytes_may_use -= bytes;
3218 BTRFS_I(inode)->reserved_bytes -= bytes;
3219 spin_unlock(&data_sinfo->lock);
3222 /* called when we are adding a delalloc extent to the inode's io_tree */
3223 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3226 struct btrfs_space_info *data_sinfo;
3228 /* get the space info for where this inode will be storing its data */
3229 data_sinfo = BTRFS_I(inode)->space_info;
3231 /* make sure we have enough space to handle the data first */
3232 spin_lock(&data_sinfo->lock);
3233 data_sinfo->bytes_delalloc += bytes;
3236 * we are adding a delalloc extent without calling
3237 * btrfs_check_data_free_space first. This happens on a weird
3238 * writepage condition, but shouldn't hurt our accounting
3240 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3241 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3242 BTRFS_I(inode)->reserved_bytes = 0;
3244 data_sinfo->bytes_may_use -= bytes;
3245 BTRFS_I(inode)->reserved_bytes -= bytes;
3248 spin_unlock(&data_sinfo->lock);
3251 /* called when we are clearing an delalloc extent from the inode's io_tree */
3252 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3255 struct btrfs_space_info *info;
3257 info = BTRFS_I(inode)->space_info;
3259 spin_lock(&info->lock);
3260 info->bytes_delalloc -= bytes;
3261 spin_unlock(&info->lock);
3264 static void force_metadata_allocation(struct btrfs_fs_info *info)
3266 struct list_head *head = &info->space_info;
3267 struct btrfs_space_info *found;
3270 list_for_each_entry_rcu(found, head, list) {
3271 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3272 found->force_alloc = 1;
3277 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *extent_root, u64 alloc_bytes,
3279 u64 flags, int force)
3281 struct btrfs_space_info *space_info;
3282 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3286 mutex_lock(&fs_info->chunk_mutex);
3288 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3290 space_info = __find_space_info(extent_root->fs_info, flags);
3292 ret = update_space_info(extent_root->fs_info, flags,
3296 BUG_ON(!space_info);
3298 spin_lock(&space_info->lock);
3299 if (space_info->force_alloc)
3301 if (space_info->full) {
3302 spin_unlock(&space_info->lock);
3306 thresh = space_info->total_bytes - space_info->bytes_readonly;
3307 thresh = div_factor(thresh, 8);
3309 (space_info->bytes_used + space_info->bytes_pinned +
3310 space_info->bytes_reserved + alloc_bytes) < thresh) {
3311 spin_unlock(&space_info->lock);
3314 spin_unlock(&space_info->lock);
3317 * if we're doing a data chunk, go ahead and make sure that
3318 * we keep a reasonable number of metadata chunks allocated in the
3321 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3322 fs_info->data_chunk_allocations++;
3323 if (!(fs_info->data_chunk_allocations %
3324 fs_info->metadata_ratio))
3325 force_metadata_allocation(fs_info);
3328 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3329 spin_lock(&space_info->lock);
3331 space_info->full = 1;
3332 space_info->force_alloc = 0;
3333 spin_unlock(&space_info->lock);
3335 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3339 static int update_block_group(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *root,
3341 u64 bytenr, u64 num_bytes, int alloc,
3344 struct btrfs_block_group_cache *cache;
3345 struct btrfs_fs_info *info = root->fs_info;
3346 u64 total = num_bytes;
3350 /* block accounting for super block */
3351 spin_lock(&info->delalloc_lock);
3352 old_val = btrfs_super_bytes_used(&info->super_copy);
3354 old_val += num_bytes;
3356 old_val -= num_bytes;
3357 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3359 /* block accounting for root item */
3360 old_val = btrfs_root_used(&root->root_item);
3362 old_val += num_bytes;
3364 old_val -= num_bytes;
3365 btrfs_set_root_used(&root->root_item, old_val);
3366 spin_unlock(&info->delalloc_lock);
3369 cache = btrfs_lookup_block_group(info, bytenr);
3372 byte_in_group = bytenr - cache->key.objectid;
3373 WARN_ON(byte_in_group > cache->key.offset);
3375 spin_lock(&cache->space_info->lock);
3376 spin_lock(&cache->lock);
3378 old_val = btrfs_block_group_used(&cache->item);
3379 num_bytes = min(total, cache->key.offset - byte_in_group);
3381 old_val += num_bytes;
3382 btrfs_set_block_group_used(&cache->item, old_val);
3383 cache->reserved -= num_bytes;
3384 cache->space_info->bytes_used += num_bytes;
3385 cache->space_info->bytes_reserved -= num_bytes;
3387 cache->space_info->bytes_readonly -= num_bytes;
3388 spin_unlock(&cache->lock);
3389 spin_unlock(&cache->space_info->lock);
3391 old_val -= num_bytes;
3392 cache->space_info->bytes_used -= num_bytes;
3394 cache->space_info->bytes_readonly += num_bytes;
3395 btrfs_set_block_group_used(&cache->item, old_val);
3396 spin_unlock(&cache->lock);
3397 spin_unlock(&cache->space_info->lock);
3401 ret = btrfs_discard_extent(root, bytenr,
3405 ret = btrfs_add_free_space(cache, bytenr,
3410 btrfs_put_block_group(cache);
3412 bytenr += num_bytes;
3417 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3419 struct btrfs_block_group_cache *cache;
3422 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3426 bytenr = cache->key.objectid;
3427 btrfs_put_block_group(cache);
3433 * this function must be called within transaction
3435 int btrfs_pin_extent(struct btrfs_root *root,
3436 u64 bytenr, u64 num_bytes, int reserved)
3438 struct btrfs_fs_info *fs_info = root->fs_info;
3439 struct btrfs_block_group_cache *cache;
3441 cache = btrfs_lookup_block_group(fs_info, bytenr);
3444 spin_lock(&cache->space_info->lock);
3445 spin_lock(&cache->lock);
3446 cache->pinned += num_bytes;
3447 cache->space_info->bytes_pinned += num_bytes;
3449 cache->reserved -= num_bytes;
3450 cache->space_info->bytes_reserved -= num_bytes;
3452 spin_unlock(&cache->lock);
3453 spin_unlock(&cache->space_info->lock);
3455 btrfs_put_block_group(cache);
3457 set_extent_dirty(fs_info->pinned_extents,
3458 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3462 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3463 u64 num_bytes, int reserve)
3465 spin_lock(&cache->space_info->lock);
3466 spin_lock(&cache->lock);
3468 cache->reserved += num_bytes;
3469 cache->space_info->bytes_reserved += num_bytes;
3471 cache->reserved -= num_bytes;
3472 cache->space_info->bytes_reserved -= num_bytes;
3474 spin_unlock(&cache->lock);
3475 spin_unlock(&cache->space_info->lock);
3479 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3480 struct btrfs_root *root)
3482 struct btrfs_fs_info *fs_info = root->fs_info;
3483 struct btrfs_caching_control *next;
3484 struct btrfs_caching_control *caching_ctl;
3485 struct btrfs_block_group_cache *cache;
3487 down_write(&fs_info->extent_commit_sem);
3489 list_for_each_entry_safe(caching_ctl, next,
3490 &fs_info->caching_block_groups, list) {
3491 cache = caching_ctl->block_group;
3492 if (block_group_cache_done(cache)) {
3493 cache->last_byte_to_unpin = (u64)-1;
3494 list_del_init(&caching_ctl->list);
3495 put_caching_control(caching_ctl);
3497 cache->last_byte_to_unpin = caching_ctl->progress;
3501 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3502 fs_info->pinned_extents = &fs_info->freed_extents[1];
3504 fs_info->pinned_extents = &fs_info->freed_extents[0];
3506 up_write(&fs_info->extent_commit_sem);
3510 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3512 struct btrfs_fs_info *fs_info = root->fs_info;
3513 struct btrfs_block_group_cache *cache = NULL;
3516 while (start <= end) {
3518 start >= cache->key.objectid + cache->key.offset) {
3520 btrfs_put_block_group(cache);
3521 cache = btrfs_lookup_block_group(fs_info, start);
3525 len = cache->key.objectid + cache->key.offset - start;
3526 len = min(len, end + 1 - start);
3528 if (start < cache->last_byte_to_unpin) {
3529 len = min(len, cache->last_byte_to_unpin - start);
3530 btrfs_add_free_space(cache, start, len);
3533 spin_lock(&cache->space_info->lock);
3534 spin_lock(&cache->lock);
3535 cache->pinned -= len;
3536 cache->space_info->bytes_pinned -= len;
3537 spin_unlock(&cache->lock);
3538 spin_unlock(&cache->space_info->lock);
3544 btrfs_put_block_group(cache);
3548 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3549 struct btrfs_root *root)
3551 struct btrfs_fs_info *fs_info = root->fs_info;
3552 struct extent_io_tree *unpin;
3557 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3558 unpin = &fs_info->freed_extents[1];
3560 unpin = &fs_info->freed_extents[0];
3563 ret = find_first_extent_bit(unpin, 0, &start, &end,
3568 ret = btrfs_discard_extent(root, start, end + 1 - start);
3570 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3571 unpin_extent_range(root, start, end);
3578 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3579 struct btrfs_root *root,
3580 struct btrfs_path *path,
3581 u64 bytenr, u64 num_bytes,
3582 int is_data, int reserved,
3583 struct extent_buffer **must_clean)
3586 struct extent_buffer *buf;
3591 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3595 /* we can reuse a block if it hasn't been written
3596 * and it is from this transaction. We can't
3597 * reuse anything from the tree log root because
3598 * it has tiny sub-transactions.
3600 if (btrfs_buffer_uptodate(buf, 0) &&
3601 btrfs_try_tree_lock(buf)) {
3602 u64 header_owner = btrfs_header_owner(buf);
3603 u64 header_transid = btrfs_header_generation(buf);
3604 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3605 header_transid == trans->transid &&
3606 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3610 btrfs_tree_unlock(buf);
3612 free_extent_buffer(buf);
3615 btrfs_set_path_blocking(path);
3616 /* unlocks the pinned mutex */
3617 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3623 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3624 struct btrfs_root *root,
3625 u64 bytenr, u64 num_bytes, u64 parent,
3626 u64 root_objectid, u64 owner_objectid,
3627 u64 owner_offset, int refs_to_drop,
3628 struct btrfs_delayed_extent_op *extent_op)
3630 struct btrfs_key key;
3631 struct btrfs_path *path;
3632 struct btrfs_fs_info *info = root->fs_info;
3633 struct btrfs_root *extent_root = info->extent_root;
3634 struct extent_buffer *leaf;
3635 struct btrfs_extent_item *ei;
3636 struct btrfs_extent_inline_ref *iref;
3639 int extent_slot = 0;
3640 int found_extent = 0;
3645 path = btrfs_alloc_path();
3650 path->leave_spinning = 1;
3652 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3653 BUG_ON(!is_data && refs_to_drop != 1);
3655 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3656 bytenr, num_bytes, parent,
3657 root_objectid, owner_objectid,
3660 extent_slot = path->slots[0];
3661 while (extent_slot >= 0) {
3662 btrfs_item_key_to_cpu(path->nodes[0], &key,
3664 if (key.objectid != bytenr)
3666 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3667 key.offset == num_bytes) {
3671 if (path->slots[0] - extent_slot > 5)
3675 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3676 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3677 if (found_extent && item_size < sizeof(*ei))
3680 if (!found_extent) {
3682 ret = remove_extent_backref(trans, extent_root, path,
3686 btrfs_release_path(extent_root, path);
3687 path->leave_spinning = 1;
3689 key.objectid = bytenr;
3690 key.type = BTRFS_EXTENT_ITEM_KEY;
3691 key.offset = num_bytes;
3693 ret = btrfs_search_slot(trans, extent_root,
3696 printk(KERN_ERR "umm, got %d back from search"
3697 ", was looking for %llu\n", ret,
3698 (unsigned long long)bytenr);
3699 btrfs_print_leaf(extent_root, path->nodes[0]);
3702 extent_slot = path->slots[0];
3705 btrfs_print_leaf(extent_root, path->nodes[0]);
3707 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3708 "parent %llu root %llu owner %llu offset %llu\n",
3709 (unsigned long long)bytenr,
3710 (unsigned long long)parent,
3711 (unsigned long long)root_objectid,
3712 (unsigned long long)owner_objectid,
3713 (unsigned long long)owner_offset);
3716 leaf = path->nodes[0];
3717 item_size = btrfs_item_size_nr(leaf, extent_slot);
3718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3719 if (item_size < sizeof(*ei)) {
3720 BUG_ON(found_extent || extent_slot != path->slots[0]);
3721 ret = convert_extent_item_v0(trans, extent_root, path,
3725 btrfs_release_path(extent_root, path);
3726 path->leave_spinning = 1;
3728 key.objectid = bytenr;
3729 key.type = BTRFS_EXTENT_ITEM_KEY;
3730 key.offset = num_bytes;
3732 ret = btrfs_search_slot(trans, extent_root, &key, path,
3735 printk(KERN_ERR "umm, got %d back from search"
3736 ", was looking for %llu\n", ret,
3737 (unsigned long long)bytenr);
3738 btrfs_print_leaf(extent_root, path->nodes[0]);
3741 extent_slot = path->slots[0];
3742 leaf = path->nodes[0];
3743 item_size = btrfs_item_size_nr(leaf, extent_slot);
3746 BUG_ON(item_size < sizeof(*ei));
3747 ei = btrfs_item_ptr(leaf, extent_slot,
3748 struct btrfs_extent_item);
3749 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3750 struct btrfs_tree_block_info *bi;
3751 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3752 bi = (struct btrfs_tree_block_info *)(ei + 1);
3753 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3756 refs = btrfs_extent_refs(leaf, ei);
3757 BUG_ON(refs < refs_to_drop);
3758 refs -= refs_to_drop;
3762 __run_delayed_extent_op(extent_op, leaf, ei);
3764 * In the case of inline back ref, reference count will
3765 * be updated by remove_extent_backref
3768 BUG_ON(!found_extent);
3770 btrfs_set_extent_refs(leaf, ei, refs);
3771 btrfs_mark_buffer_dirty(leaf);
3774 ret = remove_extent_backref(trans, extent_root, path,
3781 struct extent_buffer *must_clean = NULL;
3784 BUG_ON(is_data && refs_to_drop !=
3785 extent_data_ref_count(root, path, iref));
3787 BUG_ON(path->slots[0] != extent_slot);
3789 BUG_ON(path->slots[0] != extent_slot + 1);
3790 path->slots[0] = extent_slot;
3795 ret = pin_down_bytes(trans, root, path, bytenr,
3796 num_bytes, is_data, 0, &must_clean);
3801 * it is going to be very rare for someone to be waiting
3802 * on the block we're freeing. del_items might need to
3803 * schedule, so rather than get fancy, just force it
3807 btrfs_set_lock_blocking(must_clean);
3809 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3812 btrfs_release_path(extent_root, path);
3815 clean_tree_block(NULL, root, must_clean);
3816 btrfs_tree_unlock(must_clean);
3817 free_extent_buffer(must_clean);
3821 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3824 invalidate_mapping_pages(info->btree_inode->i_mapping,
3825 bytenr >> PAGE_CACHE_SHIFT,
3826 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3829 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3833 btrfs_free_path(path);
3838 * when we free an extent, it is possible (and likely) that we free the last
3839 * delayed ref for that extent as well. This searches the delayed ref tree for
3840 * a given extent, and if there are no other delayed refs to be processed, it
3841 * removes it from the tree.
3843 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3844 struct btrfs_root *root, u64 bytenr)
3846 struct btrfs_delayed_ref_head *head;
3847 struct btrfs_delayed_ref_root *delayed_refs;
3848 struct btrfs_delayed_ref_node *ref;
3849 struct rb_node *node;
3852 delayed_refs = &trans->transaction->delayed_refs;
3853 spin_lock(&delayed_refs->lock);
3854 head = btrfs_find_delayed_ref_head(trans, bytenr);
3858 node = rb_prev(&head->node.rb_node);
3862 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3864 /* there are still entries for this ref, we can't drop it */
3865 if (ref->bytenr == bytenr)
3868 if (head->extent_op) {
3869 if (!head->must_insert_reserved)
3871 kfree(head->extent_op);
3872 head->extent_op = NULL;
3876 * waiting for the lock here would deadlock. If someone else has it
3877 * locked they are already in the process of dropping it anyway
3879 if (!mutex_trylock(&head->mutex))
3883 * at this point we have a head with no other entries. Go
3884 * ahead and process it.
3886 head->node.in_tree = 0;
3887 rb_erase(&head->node.rb_node, &delayed_refs->root);
3889 delayed_refs->num_entries--;
3892 * we don't take a ref on the node because we're removing it from the
3893 * tree, so we just steal the ref the tree was holding.
3895 delayed_refs->num_heads--;
3896 if (list_empty(&head->cluster))
3897 delayed_refs->num_heads_ready--;
3899 list_del_init(&head->cluster);
3900 spin_unlock(&delayed_refs->lock);
3902 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3903 &head->node, head->extent_op,
3904 head->must_insert_reserved);
3906 btrfs_put_delayed_ref(&head->node);
3909 spin_unlock(&delayed_refs->lock);
3913 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3914 struct btrfs_root *root,
3915 u64 bytenr, u64 num_bytes, u64 parent,
3916 u64 root_objectid, u64 owner, u64 offset)
3921 * tree log blocks never actually go into the extent allocation
3922 * tree, just update pinning info and exit early.
3924 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3925 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3926 /* unlocks the pinned mutex */
3927 btrfs_pin_extent(root, bytenr, num_bytes, 1);
3929 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3930 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3931 parent, root_objectid, (int)owner,
3932 BTRFS_DROP_DELAYED_REF, NULL);
3934 ret = check_ref_cleanup(trans, root, bytenr);
3937 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3938 parent, root_objectid, owner,
3939 offset, BTRFS_DROP_DELAYED_REF, NULL);
3945 static u64 stripe_align(struct btrfs_root *root, u64 val)
3947 u64 mask = ((u64)root->stripesize - 1);
3948 u64 ret = (val + mask) & ~mask;
3953 * when we wait for progress in the block group caching, its because
3954 * our allocation attempt failed at least once. So, we must sleep
3955 * and let some progress happen before we try again.
3957 * This function will sleep at least once waiting for new free space to
3958 * show up, and then it will check the block group free space numbers
3959 * for our min num_bytes. Another option is to have it go ahead
3960 * and look in the rbtree for a free extent of a given size, but this
3964 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3967 struct btrfs_caching_control *caching_ctl;
3970 caching_ctl = get_caching_control(cache);
3974 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
3975 (cache->free_space >= num_bytes));
3977 put_caching_control(caching_ctl);
3982 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
3984 struct btrfs_caching_control *caching_ctl;
3987 caching_ctl = get_caching_control(cache);
3991 wait_event(caching_ctl->wait, block_group_cache_done(cache));
3993 put_caching_control(caching_ctl);
3997 enum btrfs_loop_type {
3998 LOOP_CACHED_ONLY = 0,
3999 LOOP_CACHING_NOWAIT = 1,
4000 LOOP_CACHING_WAIT = 2,
4001 LOOP_ALLOC_CHUNK = 3,
4002 LOOP_NO_EMPTY_SIZE = 4,
4006 * walks the btree of allocated extents and find a hole of a given size.
4007 * The key ins is changed to record the hole:
4008 * ins->objectid == block start
4009 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4010 * ins->offset == number of blocks
4011 * Any available blocks before search_start are skipped.
4013 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4014 struct btrfs_root *orig_root,
4015 u64 num_bytes, u64 empty_size,
4016 u64 search_start, u64 search_end,
4017 u64 hint_byte, struct btrfs_key *ins,
4018 u64 exclude_start, u64 exclude_nr,
4022 struct btrfs_root *root = orig_root->fs_info->extent_root;
4023 struct btrfs_free_cluster *last_ptr = NULL;
4024 struct btrfs_block_group_cache *block_group = NULL;
4025 int empty_cluster = 2 * 1024 * 1024;
4026 int allowed_chunk_alloc = 0;
4027 struct btrfs_space_info *space_info;
4028 int last_ptr_loop = 0;
4030 bool found_uncached_bg = false;
4031 bool failed_cluster_refill = false;
4033 WARN_ON(num_bytes < root->sectorsize);
4034 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4038 space_info = __find_space_info(root->fs_info, data);
4040 if (orig_root->ref_cows || empty_size)
4041 allowed_chunk_alloc = 1;
4043 if (data & BTRFS_BLOCK_GROUP_METADATA) {
4044 last_ptr = &root->fs_info->meta_alloc_cluster;
4045 if (!btrfs_test_opt(root, SSD))
4046 empty_cluster = 64 * 1024;
4049 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4050 last_ptr = &root->fs_info->data_alloc_cluster;
4054 spin_lock(&last_ptr->lock);
4055 if (last_ptr->block_group)
4056 hint_byte = last_ptr->window_start;
4057 spin_unlock(&last_ptr->lock);
4060 search_start = max(search_start, first_logical_byte(root, 0));
4061 search_start = max(search_start, hint_byte);
4066 if (search_start == hint_byte) {
4067 block_group = btrfs_lookup_block_group(root->fs_info,
4070 * we don't want to use the block group if it doesn't match our
4071 * allocation bits, or if its not cached.
4073 if (block_group && block_group_bits(block_group, data) &&
4074 block_group_cache_done(block_group)) {
4075 down_read(&space_info->groups_sem);
4076 if (list_empty(&block_group->list) ||
4079 * someone is removing this block group,
4080 * we can't jump into the have_block_group
4081 * target because our list pointers are not
4084 btrfs_put_block_group(block_group);
4085 up_read(&space_info->groups_sem);
4087 goto have_block_group;
4088 } else if (block_group) {
4089 btrfs_put_block_group(block_group);
4094 down_read(&space_info->groups_sem);
4095 list_for_each_entry(block_group, &space_info->block_groups, list) {
4099 atomic_inc(&block_group->count);
4100 search_start = block_group->key.objectid;
4103 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4105 * we want to start caching kthreads, but not too many
4106 * right off the bat so we don't overwhelm the system,
4107 * so only start them if there are less than 2 and we're
4108 * in the initial allocation phase.
4110 if (loop > LOOP_CACHING_NOWAIT ||
4111 atomic_read(&space_info->caching_threads) < 2) {
4112 ret = cache_block_group(block_group);
4117 cached = block_group_cache_done(block_group);
4118 if (unlikely(!cached)) {
4119 found_uncached_bg = true;
4121 /* if we only want cached bgs, loop */
4122 if (loop == LOOP_CACHED_ONLY)
4126 if (unlikely(block_group->ro))
4130 * Ok we want to try and use the cluster allocator, so lets look
4131 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4132 * have tried the cluster allocator plenty of times at this
4133 * point and not have found anything, so we are likely way too
4134 * fragmented for the clustering stuff to find anything, so lets
4135 * just skip it and let the allocator find whatever block it can
4138 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4140 * the refill lock keeps out other
4141 * people trying to start a new cluster
4143 spin_lock(&last_ptr->refill_lock);
4144 if (last_ptr->block_group &&
4145 (last_ptr->block_group->ro ||
4146 !block_group_bits(last_ptr->block_group, data))) {
4148 goto refill_cluster;
4151 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4152 num_bytes, search_start);
4154 /* we have a block, we're done */
4155 spin_unlock(&last_ptr->refill_lock);
4159 spin_lock(&last_ptr->lock);
4161 * whoops, this cluster doesn't actually point to
4162 * this block group. Get a ref on the block
4163 * group is does point to and try again
4165 if (!last_ptr_loop && last_ptr->block_group &&
4166 last_ptr->block_group != block_group) {
4168 btrfs_put_block_group(block_group);
4169 block_group = last_ptr->block_group;
4170 atomic_inc(&block_group->count);
4171 spin_unlock(&last_ptr->lock);
4172 spin_unlock(&last_ptr->refill_lock);
4175 search_start = block_group->key.objectid;
4177 * we know this block group is properly
4178 * in the list because
4179 * btrfs_remove_block_group, drops the
4180 * cluster before it removes the block
4181 * group from the list
4183 goto have_block_group;
4185 spin_unlock(&last_ptr->lock);
4188 * this cluster didn't work out, free it and
4191 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4195 /* allocate a cluster in this block group */
4196 ret = btrfs_find_space_cluster(trans, root,
4197 block_group, last_ptr,
4199 empty_cluster + empty_size);
4202 * now pull our allocation out of this
4205 offset = btrfs_alloc_from_cluster(block_group,
4206 last_ptr, num_bytes,
4209 /* we found one, proceed */
4210 spin_unlock(&last_ptr->refill_lock);
4213 } else if (!cached && loop > LOOP_CACHING_NOWAIT
4214 && !failed_cluster_refill) {
4215 spin_unlock(&last_ptr->refill_lock);
4217 failed_cluster_refill = true;
4218 wait_block_group_cache_progress(block_group,
4219 num_bytes + empty_cluster + empty_size);
4220 goto have_block_group;
4224 * at this point we either didn't find a cluster
4225 * or we weren't able to allocate a block from our
4226 * cluster. Free the cluster we've been trying
4227 * to use, and go to the next block group
4229 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4230 spin_unlock(&last_ptr->refill_lock);
4234 offset = btrfs_find_space_for_alloc(block_group, search_start,
4235 num_bytes, empty_size);
4236 if (!offset && (cached || (!cached &&
4237 loop == LOOP_CACHING_NOWAIT))) {
4239 } else if (!offset && (!cached &&
4240 loop > LOOP_CACHING_NOWAIT)) {
4241 wait_block_group_cache_progress(block_group,
4242 num_bytes + empty_size);
4243 goto have_block_group;
4246 search_start = stripe_align(root, offset);
4247 /* move on to the next group */
4248 if (search_start + num_bytes >= search_end) {
4249 btrfs_add_free_space(block_group, offset, num_bytes);
4253 /* move on to the next group */
4254 if (search_start + num_bytes >
4255 block_group->key.objectid + block_group->key.offset) {
4256 btrfs_add_free_space(block_group, offset, num_bytes);
4260 if (exclude_nr > 0 &&
4261 (search_start + num_bytes > exclude_start &&
4262 search_start < exclude_start + exclude_nr)) {
4263 search_start = exclude_start + exclude_nr;
4265 btrfs_add_free_space(block_group, offset, num_bytes);
4267 * if search_start is still in this block group
4268 * then we just re-search this block group
4270 if (search_start >= block_group->key.objectid &&
4271 search_start < (block_group->key.objectid +
4272 block_group->key.offset))
4273 goto have_block_group;
4277 ins->objectid = search_start;
4278 ins->offset = num_bytes;
4280 if (offset < search_start)
4281 btrfs_add_free_space(block_group, offset,
4282 search_start - offset);
4283 BUG_ON(offset > search_start);
4285 update_reserved_extents(block_group, num_bytes, 1);
4287 /* we are all good, lets return */
4290 failed_cluster_refill = false;
4291 btrfs_put_block_group(block_group);
4293 up_read(&space_info->groups_sem);
4295 /* LOOP_CACHED_ONLY, only search fully cached block groups
4296 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4297 * dont wait foR them to finish caching
4298 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4299 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4300 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4303 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4304 (found_uncached_bg || empty_size || empty_cluster ||
4305 allowed_chunk_alloc)) {
4306 if (found_uncached_bg) {
4307 found_uncached_bg = false;
4308 if (loop < LOOP_CACHING_WAIT) {
4314 if (loop == LOOP_ALLOC_CHUNK) {
4319 if (allowed_chunk_alloc) {
4320 ret = do_chunk_alloc(trans, root, num_bytes +
4321 2 * 1024 * 1024, data, 1);
4322 allowed_chunk_alloc = 0;
4324 space_info->force_alloc = 1;
4327 if (loop < LOOP_NO_EMPTY_SIZE) {
4332 } else if (!ins->objectid) {
4336 /* we found what we needed */
4337 if (ins->objectid) {
4338 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4339 trans->block_group = block_group->key.objectid;
4341 btrfs_put_block_group(block_group);
4348 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4349 int dump_block_groups)
4351 struct btrfs_block_group_cache *cache;
4353 spin_lock(&info->lock);
4354 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4355 (unsigned long long)(info->total_bytes - info->bytes_used -
4356 info->bytes_pinned - info->bytes_reserved -
4358 (info->full) ? "" : "not ");
4359 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4360 " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4362 (unsigned long long)info->total_bytes,
4363 (unsigned long long)info->bytes_pinned,
4364 (unsigned long long)info->bytes_delalloc,
4365 (unsigned long long)info->bytes_may_use,
4366 (unsigned long long)info->bytes_used,
4367 (unsigned long long)info->bytes_root,
4368 (unsigned long long)info->bytes_super,
4369 (unsigned long long)info->bytes_reserved);
4370 spin_unlock(&info->lock);
4372 if (!dump_block_groups)
4375 down_read(&info->groups_sem);
4376 list_for_each_entry(cache, &info->block_groups, list) {
4377 spin_lock(&cache->lock);
4378 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4379 "%llu pinned %llu reserved\n",
4380 (unsigned long long)cache->key.objectid,
4381 (unsigned long long)cache->key.offset,
4382 (unsigned long long)btrfs_block_group_used(&cache->item),
4383 (unsigned long long)cache->pinned,
4384 (unsigned long long)cache->reserved);
4385 btrfs_dump_free_space(cache, bytes);
4386 spin_unlock(&cache->lock);
4388 up_read(&info->groups_sem);
4391 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4392 struct btrfs_root *root,
4393 u64 num_bytes, u64 min_alloc_size,
4394 u64 empty_size, u64 hint_byte,
4395 u64 search_end, struct btrfs_key *ins,
4399 u64 search_start = 0;
4400 struct btrfs_fs_info *info = root->fs_info;
4402 data = btrfs_get_alloc_profile(root, data);
4405 * the only place that sets empty_size is btrfs_realloc_node, which
4406 * is not called recursively on allocations
4408 if (empty_size || root->ref_cows) {
4409 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4410 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4412 BTRFS_BLOCK_GROUP_METADATA |
4413 (info->metadata_alloc_profile &
4414 info->avail_metadata_alloc_bits), 0);
4416 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4417 num_bytes + 2 * 1024 * 1024, data, 0);
4420 WARN_ON(num_bytes < root->sectorsize);
4421 ret = find_free_extent(trans, root, num_bytes, empty_size,
4422 search_start, search_end, hint_byte, ins,
4423 trans->alloc_exclude_start,
4424 trans->alloc_exclude_nr, data);
4426 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4427 num_bytes = num_bytes >> 1;
4428 num_bytes = num_bytes & ~(root->sectorsize - 1);
4429 num_bytes = max(num_bytes, min_alloc_size);
4430 do_chunk_alloc(trans, root->fs_info->extent_root,
4431 num_bytes, data, 1);
4434 if (ret == -ENOSPC) {
4435 struct btrfs_space_info *sinfo;
4437 sinfo = __find_space_info(root->fs_info, data);
4438 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4439 "wanted %llu\n", (unsigned long long)data,
4440 (unsigned long long)num_bytes);
4441 dump_space_info(sinfo, num_bytes, 1);
4447 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4449 struct btrfs_block_group_cache *cache;
4452 cache = btrfs_lookup_block_group(root->fs_info, start);
4454 printk(KERN_ERR "Unable to find block group for %llu\n",
4455 (unsigned long long)start);
4459 ret = btrfs_discard_extent(root, start, len);
4461 btrfs_add_free_space(cache, start, len);
4462 update_reserved_extents(cache, len, 0);
4463 btrfs_put_block_group(cache);
4468 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4469 struct btrfs_root *root,
4470 u64 parent, u64 root_objectid,
4471 u64 flags, u64 owner, u64 offset,
4472 struct btrfs_key *ins, int ref_mod)
4475 struct btrfs_fs_info *fs_info = root->fs_info;
4476 struct btrfs_extent_item *extent_item;
4477 struct btrfs_extent_inline_ref *iref;
4478 struct btrfs_path *path;
4479 struct extent_buffer *leaf;
4484 type = BTRFS_SHARED_DATA_REF_KEY;
4486 type = BTRFS_EXTENT_DATA_REF_KEY;
4488 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4490 path = btrfs_alloc_path();
4493 path->leave_spinning = 1;
4494 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4498 leaf = path->nodes[0];
4499 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4500 struct btrfs_extent_item);
4501 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4502 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4503 btrfs_set_extent_flags(leaf, extent_item,
4504 flags | BTRFS_EXTENT_FLAG_DATA);
4506 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4507 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4509 struct btrfs_shared_data_ref *ref;
4510 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4511 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4512 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4514 struct btrfs_extent_data_ref *ref;
4515 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4516 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4517 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4518 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4519 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4522 btrfs_mark_buffer_dirty(path->nodes[0]);
4523 btrfs_free_path(path);
4525 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4528 printk(KERN_ERR "btrfs update block group failed for %llu "
4529 "%llu\n", (unsigned long long)ins->objectid,
4530 (unsigned long long)ins->offset);
4536 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4537 struct btrfs_root *root,
4538 u64 parent, u64 root_objectid,
4539 u64 flags, struct btrfs_disk_key *key,
4540 int level, struct btrfs_key *ins)
4543 struct btrfs_fs_info *fs_info = root->fs_info;
4544 struct btrfs_extent_item *extent_item;
4545 struct btrfs_tree_block_info *block_info;
4546 struct btrfs_extent_inline_ref *iref;
4547 struct btrfs_path *path;
4548 struct extent_buffer *leaf;
4549 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4551 path = btrfs_alloc_path();
4554 path->leave_spinning = 1;
4555 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4559 leaf = path->nodes[0];
4560 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4561 struct btrfs_extent_item);
4562 btrfs_set_extent_refs(leaf, extent_item, 1);
4563 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4564 btrfs_set_extent_flags(leaf, extent_item,
4565 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4566 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4568 btrfs_set_tree_block_key(leaf, block_info, key);
4569 btrfs_set_tree_block_level(leaf, block_info, level);
4571 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4573 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4574 btrfs_set_extent_inline_ref_type(leaf, iref,
4575 BTRFS_SHARED_BLOCK_REF_KEY);
4576 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4578 btrfs_set_extent_inline_ref_type(leaf, iref,
4579 BTRFS_TREE_BLOCK_REF_KEY);
4580 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4583 btrfs_mark_buffer_dirty(leaf);
4584 btrfs_free_path(path);
4586 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4589 printk(KERN_ERR "btrfs update block group failed for %llu "
4590 "%llu\n", (unsigned long long)ins->objectid,
4591 (unsigned long long)ins->offset);
4597 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4598 struct btrfs_root *root,
4599 u64 root_objectid, u64 owner,
4600 u64 offset, struct btrfs_key *ins)
4604 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4606 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4607 0, root_objectid, owner, offset,
4608 BTRFS_ADD_DELAYED_EXTENT, NULL);
4613 * this is used by the tree logging recovery code. It records that
4614 * an extent has been allocated and makes sure to clear the free
4615 * space cache bits as well
4617 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4618 struct btrfs_root *root,
4619 u64 root_objectid, u64 owner, u64 offset,
4620 struct btrfs_key *ins)
4623 struct btrfs_block_group_cache *block_group;
4624 struct btrfs_caching_control *caching_ctl;
4625 u64 start = ins->objectid;
4626 u64 num_bytes = ins->offset;
4628 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4629 cache_block_group(block_group);
4630 caching_ctl = get_caching_control(block_group);
4633 BUG_ON(!block_group_cache_done(block_group));
4634 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4637 mutex_lock(&caching_ctl->mutex);
4639 if (start >= caching_ctl->progress) {
4640 ret = add_excluded_extent(root, start, num_bytes);
4642 } else if (start + num_bytes <= caching_ctl->progress) {
4643 ret = btrfs_remove_free_space(block_group,
4647 num_bytes = caching_ctl->progress - start;
4648 ret = btrfs_remove_free_space(block_group,
4652 start = caching_ctl->progress;
4653 num_bytes = ins->objectid + ins->offset -
4654 caching_ctl->progress;
4655 ret = add_excluded_extent(root, start, num_bytes);
4659 mutex_unlock(&caching_ctl->mutex);
4660 put_caching_control(caching_ctl);
4663 update_reserved_extents(block_group, ins->offset, 1);
4664 btrfs_put_block_group(block_group);
4665 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4666 0, owner, offset, ins, 1);
4671 * finds a free extent and does all the dirty work required for allocation
4672 * returns the key for the extent through ins, and a tree buffer for
4673 * the first block of the extent through buf.
4675 * returns 0 if everything worked, non-zero otherwise.
4677 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4678 struct btrfs_root *root,
4679 u64 num_bytes, u64 parent, u64 root_objectid,
4680 struct btrfs_disk_key *key, int level,
4681 u64 empty_size, u64 hint_byte, u64 search_end,
4682 struct btrfs_key *ins)
4687 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4688 empty_size, hint_byte, search_end,
4693 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4695 parent = ins->objectid;
4696 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4700 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4701 struct btrfs_delayed_extent_op *extent_op;
4702 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4705 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4707 memset(&extent_op->key, 0, sizeof(extent_op->key));
4708 extent_op->flags_to_set = flags;
4709 extent_op->update_key = 1;
4710 extent_op->update_flags = 1;
4711 extent_op->is_data = 0;
4713 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4714 ins->offset, parent, root_objectid,
4715 level, BTRFS_ADD_DELAYED_EXTENT,
4722 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4723 struct btrfs_root *root,
4724 u64 bytenr, u32 blocksize,
4727 struct extent_buffer *buf;
4729 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4731 return ERR_PTR(-ENOMEM);
4732 btrfs_set_header_generation(buf, trans->transid);
4733 btrfs_set_buffer_lockdep_class(buf, level);
4734 btrfs_tree_lock(buf);
4735 clean_tree_block(trans, root, buf);
4737 btrfs_set_lock_blocking(buf);
4738 btrfs_set_buffer_uptodate(buf);
4740 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4741 set_extent_dirty(&root->dirty_log_pages, buf->start,
4742 buf->start + buf->len - 1, GFP_NOFS);
4744 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4745 buf->start + buf->len - 1, GFP_NOFS);
4747 trans->blocks_used++;
4748 /* this returns a buffer locked for blocking */
4753 * helper function to allocate a block for a given tree
4754 * returns the tree buffer or NULL.
4756 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4757 struct btrfs_root *root, u32 blocksize,
4758 u64 parent, u64 root_objectid,
4759 struct btrfs_disk_key *key, int level,
4760 u64 hint, u64 empty_size)
4762 struct btrfs_key ins;
4764 struct extent_buffer *buf;
4766 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4767 key, level, empty_size, hint, (u64)-1, &ins);
4770 return ERR_PTR(ret);
4773 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4778 struct walk_control {
4779 u64 refs[BTRFS_MAX_LEVEL];
4780 u64 flags[BTRFS_MAX_LEVEL];
4781 struct btrfs_key update_progress;
4791 #define DROP_REFERENCE 1
4792 #define UPDATE_BACKREF 2
4794 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4795 struct btrfs_root *root,
4796 struct walk_control *wc,
4797 struct btrfs_path *path)
4805 struct btrfs_key key;
4806 struct extent_buffer *eb;
4811 if (path->slots[wc->level] < wc->reada_slot) {
4812 wc->reada_count = wc->reada_count * 2 / 3;
4813 wc->reada_count = max(wc->reada_count, 2);
4815 wc->reada_count = wc->reada_count * 3 / 2;
4816 wc->reada_count = min_t(int, wc->reada_count,
4817 BTRFS_NODEPTRS_PER_BLOCK(root));
4820 eb = path->nodes[wc->level];
4821 nritems = btrfs_header_nritems(eb);
4822 blocksize = btrfs_level_size(root, wc->level - 1);
4824 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4825 if (nread >= wc->reada_count)
4829 bytenr = btrfs_node_blockptr(eb, slot);
4830 generation = btrfs_node_ptr_generation(eb, slot);
4832 if (slot == path->slots[wc->level])
4835 if (wc->stage == UPDATE_BACKREF &&
4836 generation <= root->root_key.offset)
4839 if (wc->stage == DROP_REFERENCE) {
4840 ret = btrfs_lookup_extent_info(trans, root,
4848 if (!wc->update_ref ||
4849 generation <= root->root_key.offset)
4851 btrfs_node_key_to_cpu(eb, &key, slot);
4852 ret = btrfs_comp_cpu_keys(&key,
4853 &wc->update_progress);
4858 ret = readahead_tree_block(root, bytenr, blocksize,
4862 last = bytenr + blocksize;
4865 wc->reada_slot = slot;
4869 * hepler to process tree block while walking down the tree.
4871 * when wc->stage == UPDATE_BACKREF, this function updates
4872 * back refs for pointers in the block.
4874 * NOTE: return value 1 means we should stop walking down.
4876 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4877 struct btrfs_root *root,
4878 struct btrfs_path *path,
4879 struct walk_control *wc)
4881 int level = wc->level;
4882 struct extent_buffer *eb = path->nodes[level];
4883 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4886 if (wc->stage == UPDATE_BACKREF &&
4887 btrfs_header_owner(eb) != root->root_key.objectid)
4891 * when reference count of tree block is 1, it won't increase
4892 * again. once full backref flag is set, we never clear it.
4894 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4895 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4896 BUG_ON(!path->locks[level]);
4897 ret = btrfs_lookup_extent_info(trans, root,
4902 BUG_ON(wc->refs[level] == 0);
4905 if (wc->stage == DROP_REFERENCE) {
4906 if (wc->refs[level] > 1)
4909 if (path->locks[level] && !wc->keep_locks) {
4910 btrfs_tree_unlock(eb);
4911 path->locks[level] = 0;
4916 /* wc->stage == UPDATE_BACKREF */
4917 if (!(wc->flags[level] & flag)) {
4918 BUG_ON(!path->locks[level]);
4919 ret = btrfs_inc_ref(trans, root, eb, 1);
4921 ret = btrfs_dec_ref(trans, root, eb, 0);
4923 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4926 wc->flags[level] |= flag;
4930 * the block is shared by multiple trees, so it's not good to
4931 * keep the tree lock
4933 if (path->locks[level] && level > 0) {
4934 btrfs_tree_unlock(eb);
4935 path->locks[level] = 0;
4941 * hepler to process tree block pointer.
4943 * when wc->stage == DROP_REFERENCE, this function checks
4944 * reference count of the block pointed to. if the block
4945 * is shared and we need update back refs for the subtree
4946 * rooted at the block, this function changes wc->stage to
4947 * UPDATE_BACKREF. if the block is shared and there is no
4948 * need to update back, this function drops the reference
4951 * NOTE: return value 1 means we should stop walking down.
4953 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4954 struct btrfs_root *root,
4955 struct btrfs_path *path,
4956 struct walk_control *wc)
4962 struct btrfs_key key;
4963 struct extent_buffer *next;
4964 int level = wc->level;
4968 generation = btrfs_node_ptr_generation(path->nodes[level],
4969 path->slots[level]);
4971 * if the lower level block was created before the snapshot
4972 * was created, we know there is no need to update back refs
4975 if (wc->stage == UPDATE_BACKREF &&
4976 generation <= root->root_key.offset)
4979 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4980 blocksize = btrfs_level_size(root, level - 1);
4982 next = btrfs_find_tree_block(root, bytenr, blocksize);
4984 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
4987 btrfs_tree_lock(next);
4988 btrfs_set_lock_blocking(next);
4990 if (wc->stage == DROP_REFERENCE) {
4991 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4992 &wc->refs[level - 1],
4993 &wc->flags[level - 1]);
4995 BUG_ON(wc->refs[level - 1] == 0);
4997 if (wc->refs[level - 1] > 1) {
4998 if (!wc->update_ref ||
4999 generation <= root->root_key.offset)
5002 btrfs_node_key_to_cpu(path->nodes[level], &key,
5003 path->slots[level]);
5004 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5008 wc->stage = UPDATE_BACKREF;
5009 wc->shared_level = level - 1;
5013 if (!btrfs_buffer_uptodate(next, generation)) {
5014 btrfs_tree_unlock(next);
5015 free_extent_buffer(next);
5020 if (reada && level == 1)
5021 reada_walk_down(trans, root, wc, path);
5022 next = read_tree_block(root, bytenr, blocksize, generation);
5023 btrfs_tree_lock(next);
5024 btrfs_set_lock_blocking(next);
5028 BUG_ON(level != btrfs_header_level(next));
5029 path->nodes[level] = next;
5030 path->slots[level] = 0;
5031 path->locks[level] = 1;
5037 wc->refs[level - 1] = 0;
5038 wc->flags[level - 1] = 0;
5040 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5041 parent = path->nodes[level]->start;
5043 BUG_ON(root->root_key.objectid !=
5044 btrfs_header_owner(path->nodes[level]));
5048 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5049 root->root_key.objectid, level - 1, 0);
5052 btrfs_tree_unlock(next);
5053 free_extent_buffer(next);
5058 * hepler to process tree block while walking up the tree.
5060 * when wc->stage == DROP_REFERENCE, this function drops
5061 * reference count on the block.
5063 * when wc->stage == UPDATE_BACKREF, this function changes
5064 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5065 * to UPDATE_BACKREF previously while processing the block.
5067 * NOTE: return value 1 means we should stop walking up.
5069 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5070 struct btrfs_root *root,
5071 struct btrfs_path *path,
5072 struct walk_control *wc)
5075 int level = wc->level;
5076 struct extent_buffer *eb = path->nodes[level];
5079 if (wc->stage == UPDATE_BACKREF) {
5080 BUG_ON(wc->shared_level < level);
5081 if (level < wc->shared_level)
5084 ret = find_next_key(path, level + 1, &wc->update_progress);
5088 wc->stage = DROP_REFERENCE;
5089 wc->shared_level = -1;
5090 path->slots[level] = 0;
5093 * check reference count again if the block isn't locked.
5094 * we should start walking down the tree again if reference
5097 if (!path->locks[level]) {
5099 btrfs_tree_lock(eb);
5100 btrfs_set_lock_blocking(eb);
5101 path->locks[level] = 1;
5103 ret = btrfs_lookup_extent_info(trans, root,
5108 BUG_ON(wc->refs[level] == 0);
5109 if (wc->refs[level] == 1) {
5110 btrfs_tree_unlock(eb);
5111 path->locks[level] = 0;
5117 /* wc->stage == DROP_REFERENCE */
5118 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5120 if (wc->refs[level] == 1) {
5122 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5123 ret = btrfs_dec_ref(trans, root, eb, 1);
5125 ret = btrfs_dec_ref(trans, root, eb, 0);
5128 /* make block locked assertion in clean_tree_block happy */
5129 if (!path->locks[level] &&
5130 btrfs_header_generation(eb) == trans->transid) {
5131 btrfs_tree_lock(eb);
5132 btrfs_set_lock_blocking(eb);
5133 path->locks[level] = 1;
5135 clean_tree_block(trans, root, eb);
5138 if (eb == root->node) {
5139 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5142 BUG_ON(root->root_key.objectid !=
5143 btrfs_header_owner(eb));
5145 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5146 parent = path->nodes[level + 1]->start;
5148 BUG_ON(root->root_key.objectid !=
5149 btrfs_header_owner(path->nodes[level + 1]));
5152 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5153 root->root_key.objectid, level, 0);
5156 wc->refs[level] = 0;
5157 wc->flags[level] = 0;
5161 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5162 struct btrfs_root *root,
5163 struct btrfs_path *path,
5164 struct walk_control *wc)
5166 int level = wc->level;
5169 while (level >= 0) {
5170 if (path->slots[level] >=
5171 btrfs_header_nritems(path->nodes[level]))
5174 ret = walk_down_proc(trans, root, path, wc);
5181 ret = do_walk_down(trans, root, path, wc);
5183 path->slots[level]++;
5191 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5192 struct btrfs_root *root,
5193 struct btrfs_path *path,
5194 struct walk_control *wc, int max_level)
5196 int level = wc->level;
5199 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5200 while (level < max_level && path->nodes[level]) {
5202 if (path->slots[level] + 1 <
5203 btrfs_header_nritems(path->nodes[level])) {
5204 path->slots[level]++;
5207 ret = walk_up_proc(trans, root, path, wc);
5211 if (path->locks[level]) {
5212 btrfs_tree_unlock(path->nodes[level]);
5213 path->locks[level] = 0;
5215 free_extent_buffer(path->nodes[level]);
5216 path->nodes[level] = NULL;
5224 * drop a subvolume tree.
5226 * this function traverses the tree freeing any blocks that only
5227 * referenced by the tree.
5229 * when a shared tree block is found. this function decreases its
5230 * reference count by one. if update_ref is true, this function
5231 * also make sure backrefs for the shared block and all lower level
5232 * blocks are properly updated.
5234 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5236 struct btrfs_path *path;
5237 struct btrfs_trans_handle *trans;
5238 struct btrfs_root *tree_root = root->fs_info->tree_root;
5239 struct btrfs_root_item *root_item = &root->root_item;
5240 struct walk_control *wc;
5241 struct btrfs_key key;
5246 path = btrfs_alloc_path();
5249 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5252 trans = btrfs_start_transaction(tree_root, 1);
5254 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5255 level = btrfs_header_level(root->node);
5256 path->nodes[level] = btrfs_lock_root_node(root);
5257 btrfs_set_lock_blocking(path->nodes[level]);
5258 path->slots[level] = 0;
5259 path->locks[level] = 1;
5260 memset(&wc->update_progress, 0,
5261 sizeof(wc->update_progress));
5263 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5264 memcpy(&wc->update_progress, &key,
5265 sizeof(wc->update_progress));
5267 level = root_item->drop_level;
5269 path->lowest_level = level;
5270 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5271 path->lowest_level = 0;
5279 * unlock our path, this is safe because only this
5280 * function is allowed to delete this snapshot
5282 btrfs_unlock_up_safe(path, 0);
5284 level = btrfs_header_level(root->node);
5286 btrfs_tree_lock(path->nodes[level]);
5287 btrfs_set_lock_blocking(path->nodes[level]);
5289 ret = btrfs_lookup_extent_info(trans, root,
5290 path->nodes[level]->start,
5291 path->nodes[level]->len,
5295 BUG_ON(wc->refs[level] == 0);
5297 if (level == root_item->drop_level)
5300 btrfs_tree_unlock(path->nodes[level]);
5301 WARN_ON(wc->refs[level] != 1);
5307 wc->shared_level = -1;
5308 wc->stage = DROP_REFERENCE;
5309 wc->update_ref = update_ref;
5311 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5314 ret = walk_down_tree(trans, root, path, wc);
5320 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5327 BUG_ON(wc->stage != DROP_REFERENCE);
5331 if (wc->stage == DROP_REFERENCE) {
5333 btrfs_node_key(path->nodes[level],
5334 &root_item->drop_progress,
5335 path->slots[level]);
5336 root_item->drop_level = level;
5339 BUG_ON(wc->level == 0);
5340 if (trans->transaction->in_commit ||
5341 trans->transaction->delayed_refs.flushing) {
5342 ret = btrfs_update_root(trans, tree_root,
5347 btrfs_end_transaction(trans, tree_root);
5348 trans = btrfs_start_transaction(tree_root, 1);
5350 unsigned long update;
5351 update = trans->delayed_ref_updates;
5352 trans->delayed_ref_updates = 0;
5354 btrfs_run_delayed_refs(trans, tree_root,
5358 btrfs_release_path(root, path);
5361 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5364 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5365 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5369 ret = btrfs_del_orphan_item(trans, tree_root,
5370 root->root_key.objectid);
5375 if (root->in_radix) {
5376 btrfs_free_fs_root(tree_root->fs_info, root);
5378 free_extent_buffer(root->node);
5379 free_extent_buffer(root->commit_root);
5383 btrfs_end_transaction(trans, tree_root);
5385 btrfs_free_path(path);
5390 * drop subtree rooted at tree block 'node'.
5392 * NOTE: this function will unlock and release tree block 'node'
5394 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5395 struct btrfs_root *root,
5396 struct extent_buffer *node,
5397 struct extent_buffer *parent)
5399 struct btrfs_path *path;
5400 struct walk_control *wc;
5406 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5408 path = btrfs_alloc_path();
5411 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5414 btrfs_assert_tree_locked(parent);
5415 parent_level = btrfs_header_level(parent);
5416 extent_buffer_get(parent);
5417 path->nodes[parent_level] = parent;
5418 path->slots[parent_level] = btrfs_header_nritems(parent);
5420 btrfs_assert_tree_locked(node);
5421 level = btrfs_header_level(node);
5422 path->nodes[level] = node;
5423 path->slots[level] = 0;
5424 path->locks[level] = 1;
5426 wc->refs[parent_level] = 1;
5427 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5429 wc->shared_level = -1;
5430 wc->stage = DROP_REFERENCE;
5433 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5436 wret = walk_down_tree(trans, root, path, wc);
5442 wret = walk_up_tree(trans, root, path, wc, parent_level);
5450 btrfs_free_path(path);
5455 static unsigned long calc_ra(unsigned long start, unsigned long last,
5458 return min(last, start + nr - 1);
5461 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5466 unsigned long first_index;
5467 unsigned long last_index;
5470 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5471 struct file_ra_state *ra;
5472 struct btrfs_ordered_extent *ordered;
5473 unsigned int total_read = 0;
5474 unsigned int total_dirty = 0;
5477 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5479 mutex_lock(&inode->i_mutex);
5480 first_index = start >> PAGE_CACHE_SHIFT;
5481 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5483 /* make sure the dirty trick played by the caller work */
5484 ret = invalidate_inode_pages2_range(inode->i_mapping,
5485 first_index, last_index);
5489 file_ra_state_init(ra, inode->i_mapping);
5491 for (i = first_index ; i <= last_index; i++) {
5492 if (total_read % ra->ra_pages == 0) {
5493 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5494 calc_ra(i, last_index, ra->ra_pages));
5498 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5500 page = grab_cache_page(inode->i_mapping, i);
5505 if (!PageUptodate(page)) {
5506 btrfs_readpage(NULL, page);
5508 if (!PageUptodate(page)) {
5510 page_cache_release(page);
5515 wait_on_page_writeback(page);
5517 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5518 page_end = page_start + PAGE_CACHE_SIZE - 1;
5519 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5521 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5523 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5525 page_cache_release(page);
5526 btrfs_start_ordered_extent(inode, ordered, 1);
5527 btrfs_put_ordered_extent(ordered);
5530 set_page_extent_mapped(page);
5532 if (i == first_index)
5533 set_extent_bits(io_tree, page_start, page_end,
5534 EXTENT_BOUNDARY, GFP_NOFS);
5535 btrfs_set_extent_delalloc(inode, page_start, page_end);
5537 set_page_dirty(page);
5540 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5542 page_cache_release(page);
5547 mutex_unlock(&inode->i_mutex);
5548 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5552 static noinline int relocate_data_extent(struct inode *reloc_inode,
5553 struct btrfs_key *extent_key,
5556 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5557 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5558 struct extent_map *em;
5559 u64 start = extent_key->objectid - offset;
5560 u64 end = start + extent_key->offset - 1;
5562 em = alloc_extent_map(GFP_NOFS);
5563 BUG_ON(!em || IS_ERR(em));
5566 em->len = extent_key->offset;
5567 em->block_len = extent_key->offset;
5568 em->block_start = extent_key->objectid;
5569 em->bdev = root->fs_info->fs_devices->latest_bdev;
5570 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5572 /* setup extent map to cheat btrfs_readpage */
5573 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5576 write_lock(&em_tree->lock);
5577 ret = add_extent_mapping(em_tree, em);
5578 write_unlock(&em_tree->lock);
5579 if (ret != -EEXIST) {
5580 free_extent_map(em);
5583 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5585 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5587 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5590 struct btrfs_ref_path {
5592 u64 nodes[BTRFS_MAX_LEVEL];
5594 u64 root_generation;
5601 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5602 u64 new_nodes[BTRFS_MAX_LEVEL];
5605 struct disk_extent {
5616 static int is_cowonly_root(u64 root_objectid)
5618 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5619 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5620 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5621 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5622 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5623 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5628 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5629 struct btrfs_root *extent_root,
5630 struct btrfs_ref_path *ref_path,
5633 struct extent_buffer *leaf;
5634 struct btrfs_path *path;
5635 struct btrfs_extent_ref *ref;
5636 struct btrfs_key key;
5637 struct btrfs_key found_key;
5643 path = btrfs_alloc_path();
5648 ref_path->lowest_level = -1;
5649 ref_path->current_level = -1;
5650 ref_path->shared_level = -1;
5654 level = ref_path->current_level - 1;
5655 while (level >= -1) {
5657 if (level < ref_path->lowest_level)
5661 bytenr = ref_path->nodes[level];
5663 bytenr = ref_path->extent_start;
5664 BUG_ON(bytenr == 0);
5666 parent = ref_path->nodes[level + 1];
5667 ref_path->nodes[level + 1] = 0;
5668 ref_path->current_level = level;
5669 BUG_ON(parent == 0);
5671 key.objectid = bytenr;
5672 key.offset = parent + 1;
5673 key.type = BTRFS_EXTENT_REF_KEY;
5675 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5680 leaf = path->nodes[0];
5681 nritems = btrfs_header_nritems(leaf);
5682 if (path->slots[0] >= nritems) {
5683 ret = btrfs_next_leaf(extent_root, path);
5688 leaf = path->nodes[0];
5691 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5692 if (found_key.objectid == bytenr &&
5693 found_key.type == BTRFS_EXTENT_REF_KEY) {
5694 if (level < ref_path->shared_level)
5695 ref_path->shared_level = level;
5700 btrfs_release_path(extent_root, path);
5703 /* reached lowest level */
5707 level = ref_path->current_level;
5708 while (level < BTRFS_MAX_LEVEL - 1) {
5712 bytenr = ref_path->nodes[level];
5714 bytenr = ref_path->extent_start;
5716 BUG_ON(bytenr == 0);
5718 key.objectid = bytenr;
5720 key.type = BTRFS_EXTENT_REF_KEY;
5722 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5726 leaf = path->nodes[0];
5727 nritems = btrfs_header_nritems(leaf);
5728 if (path->slots[0] >= nritems) {
5729 ret = btrfs_next_leaf(extent_root, path);
5733 /* the extent was freed by someone */
5734 if (ref_path->lowest_level == level)
5736 btrfs_release_path(extent_root, path);
5739 leaf = path->nodes[0];
5742 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5743 if (found_key.objectid != bytenr ||
5744 found_key.type != BTRFS_EXTENT_REF_KEY) {
5745 /* the extent was freed by someone */
5746 if (ref_path->lowest_level == level) {
5750 btrfs_release_path(extent_root, path);
5754 ref = btrfs_item_ptr(leaf, path->slots[0],
5755 struct btrfs_extent_ref);
5756 ref_objectid = btrfs_ref_objectid(leaf, ref);
5757 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5759 level = (int)ref_objectid;
5760 BUG_ON(level >= BTRFS_MAX_LEVEL);
5761 ref_path->lowest_level = level;
5762 ref_path->current_level = level;
5763 ref_path->nodes[level] = bytenr;
5765 WARN_ON(ref_objectid != level);
5768 WARN_ON(level != -1);
5772 if (ref_path->lowest_level == level) {
5773 ref_path->owner_objectid = ref_objectid;
5774 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5778 * the block is tree root or the block isn't in reference
5781 if (found_key.objectid == found_key.offset ||
5782 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5783 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5784 ref_path->root_generation =
5785 btrfs_ref_generation(leaf, ref);
5787 /* special reference from the tree log */
5788 ref_path->nodes[0] = found_key.offset;
5789 ref_path->current_level = 0;
5796 BUG_ON(ref_path->nodes[level] != 0);
5797 ref_path->nodes[level] = found_key.offset;
5798 ref_path->current_level = level;
5801 * the reference was created in the running transaction,
5802 * no need to continue walking up.
5804 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5805 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5806 ref_path->root_generation =
5807 btrfs_ref_generation(leaf, ref);
5812 btrfs_release_path(extent_root, path);
5815 /* reached max tree level, but no tree root found. */
5818 btrfs_free_path(path);
5822 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5823 struct btrfs_root *extent_root,
5824 struct btrfs_ref_path *ref_path,
5827 memset(ref_path, 0, sizeof(*ref_path));
5828 ref_path->extent_start = extent_start;
5830 return __next_ref_path(trans, extent_root, ref_path, 1);
5833 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5834 struct btrfs_root *extent_root,
5835 struct btrfs_ref_path *ref_path)
5837 return __next_ref_path(trans, extent_root, ref_path, 0);
5840 static noinline int get_new_locations(struct inode *reloc_inode,
5841 struct btrfs_key *extent_key,
5842 u64 offset, int no_fragment,
5843 struct disk_extent **extents,
5846 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5847 struct btrfs_path *path;
5848 struct btrfs_file_extent_item *fi;
5849 struct extent_buffer *leaf;
5850 struct disk_extent *exts = *extents;
5851 struct btrfs_key found_key;
5856 int max = *nr_extents;
5859 WARN_ON(!no_fragment && *extents);
5862 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5867 path = btrfs_alloc_path();
5870 cur_pos = extent_key->objectid - offset;
5871 last_byte = extent_key->objectid + extent_key->offset;
5872 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5882 leaf = path->nodes[0];
5883 nritems = btrfs_header_nritems(leaf);
5884 if (path->slots[0] >= nritems) {
5885 ret = btrfs_next_leaf(root, path);
5890 leaf = path->nodes[0];
5893 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5894 if (found_key.offset != cur_pos ||
5895 found_key.type != BTRFS_EXTENT_DATA_KEY ||
5896 found_key.objectid != reloc_inode->i_ino)
5899 fi = btrfs_item_ptr(leaf, path->slots[0],
5900 struct btrfs_file_extent_item);
5901 if (btrfs_file_extent_type(leaf, fi) !=
5902 BTRFS_FILE_EXTENT_REG ||
5903 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5907 struct disk_extent *old = exts;
5909 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5910 memcpy(exts, old, sizeof(*exts) * nr);
5911 if (old != *extents)
5915 exts[nr].disk_bytenr =
5916 btrfs_file_extent_disk_bytenr(leaf, fi);
5917 exts[nr].disk_num_bytes =
5918 btrfs_file_extent_disk_num_bytes(leaf, fi);
5919 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5920 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5921 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5922 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5923 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5924 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5926 BUG_ON(exts[nr].offset > 0);
5927 BUG_ON(exts[nr].compression || exts[nr].encryption);
5928 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5930 cur_pos += exts[nr].num_bytes;
5933 if (cur_pos + offset >= last_byte)
5943 BUG_ON(cur_pos + offset > last_byte);
5944 if (cur_pos + offset < last_byte) {
5950 btrfs_free_path(path);
5952 if (exts != *extents)
5961 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5962 struct btrfs_root *root,
5963 struct btrfs_path *path,
5964 struct btrfs_key *extent_key,
5965 struct btrfs_key *leaf_key,
5966 struct btrfs_ref_path *ref_path,
5967 struct disk_extent *new_extents,
5970 struct extent_buffer *leaf;
5971 struct btrfs_file_extent_item *fi;
5972 struct inode *inode = NULL;
5973 struct btrfs_key key;
5978 u64 search_end = (u64)-1;
5981 int extent_locked = 0;
5985 memcpy(&key, leaf_key, sizeof(key));
5986 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5987 if (key.objectid < ref_path->owner_objectid ||
5988 (key.objectid == ref_path->owner_objectid &&
5989 key.type < BTRFS_EXTENT_DATA_KEY)) {
5990 key.objectid = ref_path->owner_objectid;
5991 key.type = BTRFS_EXTENT_DATA_KEY;
5997 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6001 leaf = path->nodes[0];
6002 nritems = btrfs_header_nritems(leaf);
6004 if (extent_locked && ret > 0) {
6006 * the file extent item was modified by someone
6007 * before the extent got locked.
6009 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6010 lock_end, GFP_NOFS);
6014 if (path->slots[0] >= nritems) {
6015 if (++nr_scaned > 2)
6018 BUG_ON(extent_locked);
6019 ret = btrfs_next_leaf(root, path);
6024 leaf = path->nodes[0];
6025 nritems = btrfs_header_nritems(leaf);
6028 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6030 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6031 if ((key.objectid > ref_path->owner_objectid) ||
6032 (key.objectid == ref_path->owner_objectid &&
6033 key.type > BTRFS_EXTENT_DATA_KEY) ||
6034 key.offset >= search_end)
6038 if (inode && key.objectid != inode->i_ino) {
6039 BUG_ON(extent_locked);
6040 btrfs_release_path(root, path);
6041 mutex_unlock(&inode->i_mutex);
6047 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6052 fi = btrfs_item_ptr(leaf, path->slots[0],
6053 struct btrfs_file_extent_item);
6054 extent_type = btrfs_file_extent_type(leaf, fi);
6055 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6056 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6057 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6058 extent_key->objectid)) {
6064 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6065 ext_offset = btrfs_file_extent_offset(leaf, fi);
6067 if (search_end == (u64)-1) {
6068 search_end = key.offset - ext_offset +
6069 btrfs_file_extent_ram_bytes(leaf, fi);
6072 if (!extent_locked) {
6073 lock_start = key.offset;
6074 lock_end = lock_start + num_bytes - 1;
6076 if (lock_start > key.offset ||
6077 lock_end + 1 < key.offset + num_bytes) {
6078 unlock_extent(&BTRFS_I(inode)->io_tree,
6079 lock_start, lock_end, GFP_NOFS);
6085 btrfs_release_path(root, path);
6087 inode = btrfs_iget_locked(root->fs_info->sb,
6088 key.objectid, root);
6089 if (inode->i_state & I_NEW) {
6090 BTRFS_I(inode)->root = root;
6091 BTRFS_I(inode)->location.objectid =
6093 BTRFS_I(inode)->location.type =
6094 BTRFS_INODE_ITEM_KEY;
6095 BTRFS_I(inode)->location.offset = 0;
6096 btrfs_read_locked_inode(inode);
6097 unlock_new_inode(inode);
6100 * some code call btrfs_commit_transaction while
6101 * holding the i_mutex, so we can't use mutex_lock
6104 if (is_bad_inode(inode) ||
6105 !mutex_trylock(&inode->i_mutex)) {
6108 key.offset = (u64)-1;
6113 if (!extent_locked) {
6114 struct btrfs_ordered_extent *ordered;
6116 btrfs_release_path(root, path);
6118 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6119 lock_end, GFP_NOFS);
6120 ordered = btrfs_lookup_first_ordered_extent(inode,
6123 ordered->file_offset <= lock_end &&
6124 ordered->file_offset + ordered->len > lock_start) {
6125 unlock_extent(&BTRFS_I(inode)->io_tree,
6126 lock_start, lock_end, GFP_NOFS);
6127 btrfs_start_ordered_extent(inode, ordered, 1);
6128 btrfs_put_ordered_extent(ordered);
6129 key.offset += num_bytes;
6133 btrfs_put_ordered_extent(ordered);
6139 if (nr_extents == 1) {
6140 /* update extent pointer in place */
6141 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6142 new_extents[0].disk_bytenr);
6143 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6144 new_extents[0].disk_num_bytes);
6145 btrfs_mark_buffer_dirty(leaf);
6147 btrfs_drop_extent_cache(inode, key.offset,
6148 key.offset + num_bytes - 1, 0);
6150 ret = btrfs_inc_extent_ref(trans, root,
6151 new_extents[0].disk_bytenr,
6152 new_extents[0].disk_num_bytes,
6154 root->root_key.objectid,
6159 ret = btrfs_free_extent(trans, root,
6160 extent_key->objectid,
6163 btrfs_header_owner(leaf),
6164 btrfs_header_generation(leaf),
6168 btrfs_release_path(root, path);
6169 key.offset += num_bytes;
6177 * drop old extent pointer at first, then insert the
6178 * new pointers one bye one
6180 btrfs_release_path(root, path);
6181 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6182 key.offset + num_bytes,
6183 key.offset, &alloc_hint);
6186 for (i = 0; i < nr_extents; i++) {
6187 if (ext_offset >= new_extents[i].num_bytes) {
6188 ext_offset -= new_extents[i].num_bytes;
6191 extent_len = min(new_extents[i].num_bytes -
6192 ext_offset, num_bytes);
6194 ret = btrfs_insert_empty_item(trans, root,
6199 leaf = path->nodes[0];
6200 fi = btrfs_item_ptr(leaf, path->slots[0],
6201 struct btrfs_file_extent_item);
6202 btrfs_set_file_extent_generation(leaf, fi,
6204 btrfs_set_file_extent_type(leaf, fi,
6205 BTRFS_FILE_EXTENT_REG);
6206 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6207 new_extents[i].disk_bytenr);
6208 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6209 new_extents[i].disk_num_bytes);
6210 btrfs_set_file_extent_ram_bytes(leaf, fi,
6211 new_extents[i].ram_bytes);
6213 btrfs_set_file_extent_compression(leaf, fi,
6214 new_extents[i].compression);
6215 btrfs_set_file_extent_encryption(leaf, fi,
6216 new_extents[i].encryption);
6217 btrfs_set_file_extent_other_encoding(leaf, fi,
6218 new_extents[i].other_encoding);
6220 btrfs_set_file_extent_num_bytes(leaf, fi,
6222 ext_offset += new_extents[i].offset;
6223 btrfs_set_file_extent_offset(leaf, fi,
6225 btrfs_mark_buffer_dirty(leaf);
6227 btrfs_drop_extent_cache(inode, key.offset,
6228 key.offset + extent_len - 1, 0);
6230 ret = btrfs_inc_extent_ref(trans, root,
6231 new_extents[i].disk_bytenr,
6232 new_extents[i].disk_num_bytes,
6234 root->root_key.objectid,
6235 trans->transid, key.objectid);
6237 btrfs_release_path(root, path);
6239 inode_add_bytes(inode, extent_len);
6242 num_bytes -= extent_len;
6243 key.offset += extent_len;
6248 BUG_ON(i >= nr_extents);
6252 if (extent_locked) {
6253 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6254 lock_end, GFP_NOFS);
6258 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6259 key.offset >= search_end)
6266 btrfs_release_path(root, path);
6268 mutex_unlock(&inode->i_mutex);
6269 if (extent_locked) {
6270 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6271 lock_end, GFP_NOFS);
6278 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6279 struct btrfs_root *root,
6280 struct extent_buffer *buf, u64 orig_start)
6285 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6286 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6288 level = btrfs_header_level(buf);
6290 struct btrfs_leaf_ref *ref;
6291 struct btrfs_leaf_ref *orig_ref;
6293 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6297 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6299 btrfs_free_leaf_ref(root, orig_ref);
6303 ref->nritems = orig_ref->nritems;
6304 memcpy(ref->extents, orig_ref->extents,
6305 sizeof(ref->extents[0]) * ref->nritems);
6307 btrfs_free_leaf_ref(root, orig_ref);
6309 ref->root_gen = trans->transid;
6310 ref->bytenr = buf->start;
6311 ref->owner = btrfs_header_owner(buf);
6312 ref->generation = btrfs_header_generation(buf);
6314 ret = btrfs_add_leaf_ref(root, ref, 0);
6316 btrfs_free_leaf_ref(root, ref);
6321 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6322 struct extent_buffer *leaf,
6323 struct btrfs_block_group_cache *group,
6324 struct btrfs_root *target_root)
6326 struct btrfs_key key;
6327 struct inode *inode = NULL;
6328 struct btrfs_file_extent_item *fi;
6330 u64 skip_objectid = 0;
6334 nritems = btrfs_header_nritems(leaf);
6335 for (i = 0; i < nritems; i++) {
6336 btrfs_item_key_to_cpu(leaf, &key, i);
6337 if (key.objectid == skip_objectid ||
6338 key.type != BTRFS_EXTENT_DATA_KEY)
6340 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6341 if (btrfs_file_extent_type(leaf, fi) ==
6342 BTRFS_FILE_EXTENT_INLINE)
6344 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6346 if (!inode || inode->i_ino != key.objectid) {
6348 inode = btrfs_ilookup(target_root->fs_info->sb,
6349 key.objectid, target_root, 1);
6352 skip_objectid = key.objectid;
6355 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6357 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6358 key.offset + num_bytes - 1, GFP_NOFS);
6359 btrfs_drop_extent_cache(inode, key.offset,
6360 key.offset + num_bytes - 1, 1);
6361 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6362 key.offset + num_bytes - 1, GFP_NOFS);
6369 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6370 struct btrfs_root *root,
6371 struct extent_buffer *leaf,
6372 struct btrfs_block_group_cache *group,
6373 struct inode *reloc_inode)
6375 struct btrfs_key key;
6376 struct btrfs_key extent_key;
6377 struct btrfs_file_extent_item *fi;
6378 struct btrfs_leaf_ref *ref;
6379 struct disk_extent *new_extent;
6388 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6389 BUG_ON(!new_extent);
6391 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6395 nritems = btrfs_header_nritems(leaf);
6396 for (i = 0; i < nritems; i++) {
6397 btrfs_item_key_to_cpu(leaf, &key, i);
6398 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6400 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6401 if (btrfs_file_extent_type(leaf, fi) ==
6402 BTRFS_FILE_EXTENT_INLINE)
6404 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6405 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6410 if (bytenr >= group->key.objectid + group->key.offset ||
6411 bytenr + num_bytes <= group->key.objectid)
6414 extent_key.objectid = bytenr;
6415 extent_key.offset = num_bytes;
6416 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6418 ret = get_new_locations(reloc_inode, &extent_key,
6419 group->key.objectid, 1,
6420 &new_extent, &nr_extent);
6425 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6426 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6427 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6428 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6430 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6431 new_extent->disk_bytenr);
6432 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6433 new_extent->disk_num_bytes);
6434 btrfs_mark_buffer_dirty(leaf);
6436 ret = btrfs_inc_extent_ref(trans, root,
6437 new_extent->disk_bytenr,
6438 new_extent->disk_num_bytes,
6440 root->root_key.objectid,
6441 trans->transid, key.objectid);
6444 ret = btrfs_free_extent(trans, root,
6445 bytenr, num_bytes, leaf->start,
6446 btrfs_header_owner(leaf),
6447 btrfs_header_generation(leaf),
6453 BUG_ON(ext_index + 1 != ref->nritems);
6454 btrfs_free_leaf_ref(root, ref);
6458 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6459 struct btrfs_root *root)
6461 struct btrfs_root *reloc_root;
6464 if (root->reloc_root) {
6465 reloc_root = root->reloc_root;
6466 root->reloc_root = NULL;
6467 list_add(&reloc_root->dead_list,
6468 &root->fs_info->dead_reloc_roots);
6470 btrfs_set_root_bytenr(&reloc_root->root_item,
6471 reloc_root->node->start);
6472 btrfs_set_root_level(&root->root_item,
6473 btrfs_header_level(reloc_root->node));
6474 memset(&reloc_root->root_item.drop_progress, 0,
6475 sizeof(struct btrfs_disk_key));
6476 reloc_root->root_item.drop_level = 0;
6478 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6479 &reloc_root->root_key,
6480 &reloc_root->root_item);
6486 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6488 struct btrfs_trans_handle *trans;
6489 struct btrfs_root *reloc_root;
6490 struct btrfs_root *prev_root = NULL;
6491 struct list_head dead_roots;
6495 INIT_LIST_HEAD(&dead_roots);
6496 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6498 while (!list_empty(&dead_roots)) {
6499 reloc_root = list_entry(dead_roots.prev,
6500 struct btrfs_root, dead_list);
6501 list_del_init(&reloc_root->dead_list);
6503 BUG_ON(reloc_root->commit_root != NULL);
6505 trans = btrfs_join_transaction(root, 1);
6508 mutex_lock(&root->fs_info->drop_mutex);
6509 ret = btrfs_drop_snapshot(trans, reloc_root);
6512 mutex_unlock(&root->fs_info->drop_mutex);
6514 nr = trans->blocks_used;
6515 ret = btrfs_end_transaction(trans, root);
6517 btrfs_btree_balance_dirty(root, nr);
6520 free_extent_buffer(reloc_root->node);
6522 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6523 &reloc_root->root_key);
6525 mutex_unlock(&root->fs_info->drop_mutex);
6527 nr = trans->blocks_used;
6528 ret = btrfs_end_transaction(trans, root);
6530 btrfs_btree_balance_dirty(root, nr);
6533 prev_root = reloc_root;
6536 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6542 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6544 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6548 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6550 struct btrfs_root *reloc_root;
6551 struct btrfs_trans_handle *trans;
6552 struct btrfs_key location;
6556 mutex_lock(&root->fs_info->tree_reloc_mutex);
6557 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6559 found = !list_empty(&root->fs_info->dead_reloc_roots);
6560 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6563 trans = btrfs_start_transaction(root, 1);
6565 ret = btrfs_commit_transaction(trans, root);
6569 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6570 location.offset = (u64)-1;
6571 location.type = BTRFS_ROOT_ITEM_KEY;
6573 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6574 BUG_ON(!reloc_root);
6575 btrfs_orphan_cleanup(reloc_root);
6579 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6580 struct btrfs_root *root)
6582 struct btrfs_root *reloc_root;
6583 struct extent_buffer *eb;
6584 struct btrfs_root_item *root_item;
6585 struct btrfs_key root_key;
6588 BUG_ON(!root->ref_cows);
6589 if (root->reloc_root)
6592 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6595 ret = btrfs_copy_root(trans, root, root->commit_root,
6596 &eb, BTRFS_TREE_RELOC_OBJECTID);
6599 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6600 root_key.offset = root->root_key.objectid;
6601 root_key.type = BTRFS_ROOT_ITEM_KEY;
6603 memcpy(root_item, &root->root_item, sizeof(root_item));
6604 btrfs_set_root_refs(root_item, 0);
6605 btrfs_set_root_bytenr(root_item, eb->start);
6606 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6607 btrfs_set_root_generation(root_item, trans->transid);
6609 btrfs_tree_unlock(eb);
6610 free_extent_buffer(eb);
6612 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6613 &root_key, root_item);
6617 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6619 BUG_ON(!reloc_root);
6620 reloc_root->last_trans = trans->transid;
6621 reloc_root->commit_root = NULL;
6622 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6624 root->reloc_root = reloc_root;
6629 * Core function of space balance.
6631 * The idea is using reloc trees to relocate tree blocks in reference
6632 * counted roots. There is one reloc tree for each subvol, and all
6633 * reloc trees share same root key objectid. Reloc trees are snapshots
6634 * of the latest committed roots of subvols (root->commit_root).
6636 * To relocate a tree block referenced by a subvol, there are two steps.
6637 * COW the block through subvol's reloc tree, then update block pointer
6638 * in the subvol to point to the new block. Since all reloc trees share
6639 * same root key objectid, doing special handing for tree blocks owned
6640 * by them is easy. Once a tree block has been COWed in one reloc tree,
6641 * we can use the resulting new block directly when the same block is
6642 * required to COW again through other reloc trees. By this way, relocated
6643 * tree blocks are shared between reloc trees, so they are also shared
6646 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6647 struct btrfs_root *root,
6648 struct btrfs_path *path,
6649 struct btrfs_key *first_key,
6650 struct btrfs_ref_path *ref_path,
6651 struct btrfs_block_group_cache *group,
6652 struct inode *reloc_inode)
6654 struct btrfs_root *reloc_root;
6655 struct extent_buffer *eb = NULL;
6656 struct btrfs_key *keys;
6660 int lowest_level = 0;
6663 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6664 lowest_level = ref_path->owner_objectid;
6666 if (!root->ref_cows) {
6667 path->lowest_level = lowest_level;
6668 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6670 path->lowest_level = 0;
6671 btrfs_release_path(root, path);
6675 mutex_lock(&root->fs_info->tree_reloc_mutex);
6676 ret = init_reloc_tree(trans, root);
6678 reloc_root = root->reloc_root;
6680 shared_level = ref_path->shared_level;
6681 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6683 keys = ref_path->node_keys;
6684 nodes = ref_path->new_nodes;
6685 memset(&keys[shared_level + 1], 0,
6686 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6687 memset(&nodes[shared_level + 1], 0,
6688 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6690 if (nodes[lowest_level] == 0) {
6691 path->lowest_level = lowest_level;
6692 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6695 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6696 eb = path->nodes[level];
6697 if (!eb || eb == reloc_root->node)
6699 nodes[level] = eb->start;
6701 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6703 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6706 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6707 eb = path->nodes[0];
6708 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6709 group, reloc_inode);
6712 btrfs_release_path(reloc_root, path);
6714 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6720 * replace tree blocks in the fs tree with tree blocks in
6723 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6726 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6727 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6730 extent_buffer_get(path->nodes[0]);
6731 eb = path->nodes[0];
6732 btrfs_release_path(reloc_root, path);
6733 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6735 free_extent_buffer(eb);
6738 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6739 path->lowest_level = 0;
6743 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6744 struct btrfs_root *root,
6745 struct btrfs_path *path,
6746 struct btrfs_key *first_key,
6747 struct btrfs_ref_path *ref_path)
6751 ret = relocate_one_path(trans, root, path, first_key,
6752 ref_path, NULL, NULL);
6758 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6759 struct btrfs_root *extent_root,
6760 struct btrfs_path *path,
6761 struct btrfs_key *extent_key)
6765 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6768 ret = btrfs_del_item(trans, extent_root, path);
6770 btrfs_release_path(extent_root, path);
6774 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6775 struct btrfs_ref_path *ref_path)
6777 struct btrfs_key root_key;
6779 root_key.objectid = ref_path->root_objectid;
6780 root_key.type = BTRFS_ROOT_ITEM_KEY;
6781 if (is_cowonly_root(ref_path->root_objectid))
6782 root_key.offset = 0;
6784 root_key.offset = (u64)-1;
6786 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6789 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6790 struct btrfs_path *path,
6791 struct btrfs_key *extent_key,
6792 struct btrfs_block_group_cache *group,
6793 struct inode *reloc_inode, int pass)
6795 struct btrfs_trans_handle *trans;
6796 struct btrfs_root *found_root;
6797 struct btrfs_ref_path *ref_path = NULL;
6798 struct disk_extent *new_extents = NULL;
6803 struct btrfs_key first_key;
6807 trans = btrfs_start_transaction(extent_root, 1);
6810 if (extent_key->objectid == 0) {
6811 ret = del_extent_zero(trans, extent_root, path, extent_key);
6815 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6821 for (loops = 0; ; loops++) {
6823 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6824 extent_key->objectid);
6826 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6833 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6834 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6837 found_root = read_ref_root(extent_root->fs_info, ref_path);
6838 BUG_ON(!found_root);
6840 * for reference counted tree, only process reference paths
6841 * rooted at the latest committed root.
6843 if (found_root->ref_cows &&
6844 ref_path->root_generation != found_root->root_key.offset)
6847 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6850 * copy data extents to new locations
6852 u64 group_start = group->key.objectid;
6853 ret = relocate_data_extent(reloc_inode,
6862 level = ref_path->owner_objectid;
6865 if (prev_block != ref_path->nodes[level]) {
6866 struct extent_buffer *eb;
6867 u64 block_start = ref_path->nodes[level];
6868 u64 block_size = btrfs_level_size(found_root, level);
6870 eb = read_tree_block(found_root, block_start,
6872 btrfs_tree_lock(eb);
6873 BUG_ON(level != btrfs_header_level(eb));
6876 btrfs_item_key_to_cpu(eb, &first_key, 0);
6878 btrfs_node_key_to_cpu(eb, &first_key, 0);
6880 btrfs_tree_unlock(eb);
6881 free_extent_buffer(eb);
6882 prev_block = block_start;
6885 mutex_lock(&extent_root->fs_info->trans_mutex);
6886 btrfs_record_root_in_trans(found_root);
6887 mutex_unlock(&extent_root->fs_info->trans_mutex);
6888 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6890 * try to update data extent references while
6891 * keeping metadata shared between snapshots.
6894 ret = relocate_one_path(trans, found_root,
6895 path, &first_key, ref_path,
6896 group, reloc_inode);
6902 * use fallback method to process the remaining
6906 u64 group_start = group->key.objectid;
6907 new_extents = kmalloc(sizeof(*new_extents),
6910 ret = get_new_locations(reloc_inode,
6918 ret = replace_one_extent(trans, found_root,
6920 &first_key, ref_path,
6921 new_extents, nr_extents);
6923 ret = relocate_tree_block(trans, found_root, path,
6924 &first_key, ref_path);
6931 btrfs_end_transaction(trans, extent_root);
6938 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6941 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6942 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6944 num_devices = root->fs_info->fs_devices->rw_devices;
6945 if (num_devices == 1) {
6946 stripped |= BTRFS_BLOCK_GROUP_DUP;
6947 stripped = flags & ~stripped;
6949 /* turn raid0 into single device chunks */
6950 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6953 /* turn mirroring into duplication */
6954 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6955 BTRFS_BLOCK_GROUP_RAID10))
6956 return stripped | BTRFS_BLOCK_GROUP_DUP;
6959 /* they already had raid on here, just return */
6960 if (flags & stripped)
6963 stripped |= BTRFS_BLOCK_GROUP_DUP;
6964 stripped = flags & ~stripped;
6966 /* switch duplicated blocks with raid1 */
6967 if (flags & BTRFS_BLOCK_GROUP_DUP)
6968 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6970 /* turn single device chunks into raid0 */
6971 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6976 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6977 struct btrfs_block_group_cache *shrink_block_group,
6980 struct btrfs_trans_handle *trans;
6981 u64 new_alloc_flags;
6984 spin_lock(&shrink_block_group->lock);
6985 if (btrfs_block_group_used(&shrink_block_group->item) +
6986 shrink_block_group->reserved > 0) {
6987 spin_unlock(&shrink_block_group->lock);
6989 trans = btrfs_start_transaction(root, 1);
6990 spin_lock(&shrink_block_group->lock);
6992 new_alloc_flags = update_block_group_flags(root,
6993 shrink_block_group->flags);
6994 if (new_alloc_flags != shrink_block_group->flags) {
6996 btrfs_block_group_used(&shrink_block_group->item);
6998 calc = shrink_block_group->key.offset;
7000 spin_unlock(&shrink_block_group->lock);
7002 do_chunk_alloc(trans, root->fs_info->extent_root,
7003 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7005 btrfs_end_transaction(trans, root);
7007 spin_unlock(&shrink_block_group->lock);
7012 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7013 struct btrfs_block_group_cache *group)
7016 __alloc_chunk_for_shrink(root, group, 1);
7017 set_block_group_readonly(group);
7022 * checks to see if its even possible to relocate this block group.
7024 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7025 * ok to go ahead and try.
7027 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7029 struct btrfs_block_group_cache *block_group;
7030 struct btrfs_space_info *space_info;
7031 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7032 struct btrfs_device *device;
7036 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7038 /* odd, couldn't find the block group, leave it alone */
7042 /* no bytes used, we're good */
7043 if (!btrfs_block_group_used(&block_group->item))
7046 space_info = block_group->space_info;
7047 spin_lock(&space_info->lock);
7049 full = space_info->full;
7052 * if this is the last block group we have in this space, we can't
7053 * relocate it unless we're able to allocate a new chunk below.
7055 * Otherwise, we need to make sure we have room in the space to handle
7056 * all of the extents from this block group. If we can, we're good
7058 if ((space_info->total_bytes != block_group->key.offset) &&
7059 (space_info->bytes_used + space_info->bytes_reserved +
7060 space_info->bytes_pinned + space_info->bytes_readonly +
7061 btrfs_block_group_used(&block_group->item) <
7062 space_info->total_bytes)) {
7063 spin_unlock(&space_info->lock);
7066 spin_unlock(&space_info->lock);
7069 * ok we don't have enough space, but maybe we have free space on our
7070 * devices to allocate new chunks for relocation, so loop through our
7071 * alloc devices and guess if we have enough space. However, if we
7072 * were marked as full, then we know there aren't enough chunks, and we
7079 mutex_lock(&root->fs_info->chunk_mutex);
7080 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7081 u64 min_free = btrfs_block_group_used(&block_group->item);
7082 u64 dev_offset, max_avail;
7085 * check to make sure we can actually find a chunk with enough
7086 * space to fit our block group in.
7088 if (device->total_bytes > device->bytes_used + min_free) {
7089 ret = find_free_dev_extent(NULL, device, min_free,
7090 &dev_offset, &max_avail);
7096 mutex_unlock(&root->fs_info->chunk_mutex);
7098 btrfs_put_block_group(block_group);
7102 static int find_first_block_group(struct btrfs_root *root,
7103 struct btrfs_path *path, struct btrfs_key *key)
7106 struct btrfs_key found_key;
7107 struct extent_buffer *leaf;
7110 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7115 slot = path->slots[0];
7116 leaf = path->nodes[0];
7117 if (slot >= btrfs_header_nritems(leaf)) {
7118 ret = btrfs_next_leaf(root, path);
7125 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7127 if (found_key.objectid >= key->objectid &&
7128 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7139 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7141 struct btrfs_block_group_cache *block_group;
7142 struct btrfs_space_info *space_info;
7143 struct btrfs_caching_control *caching_ctl;
7146 down_write(&info->extent_commit_sem);
7147 while (!list_empty(&info->caching_block_groups)) {
7148 caching_ctl = list_entry(info->caching_block_groups.next,
7149 struct btrfs_caching_control, list);
7150 list_del(&caching_ctl->list);
7151 put_caching_control(caching_ctl);
7153 up_write(&info->extent_commit_sem);
7155 spin_lock(&info->block_group_cache_lock);
7156 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7157 block_group = rb_entry(n, struct btrfs_block_group_cache,
7159 rb_erase(&block_group->cache_node,
7160 &info->block_group_cache_tree);
7161 spin_unlock(&info->block_group_cache_lock);
7163 down_write(&block_group->space_info->groups_sem);
7164 list_del(&block_group->list);
7165 up_write(&block_group->space_info->groups_sem);
7167 if (block_group->cached == BTRFS_CACHE_STARTED)
7168 wait_block_group_cache_done(block_group);
7170 btrfs_remove_free_space_cache(block_group);
7172 WARN_ON(atomic_read(&block_group->count) != 1);
7175 spin_lock(&info->block_group_cache_lock);
7177 spin_unlock(&info->block_group_cache_lock);
7179 /* now that all the block groups are freed, go through and
7180 * free all the space_info structs. This is only called during
7181 * the final stages of unmount, and so we know nobody is
7182 * using them. We call synchronize_rcu() once before we start,
7183 * just to be on the safe side.
7187 while(!list_empty(&info->space_info)) {
7188 space_info = list_entry(info->space_info.next,
7189 struct btrfs_space_info,
7192 list_del(&space_info->list);
7198 int btrfs_read_block_groups(struct btrfs_root *root)
7200 struct btrfs_path *path;
7202 struct btrfs_block_group_cache *cache;
7203 struct btrfs_fs_info *info = root->fs_info;
7204 struct btrfs_space_info *space_info;
7205 struct btrfs_key key;
7206 struct btrfs_key found_key;
7207 struct extent_buffer *leaf;
7209 root = info->extent_root;
7212 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7213 path = btrfs_alloc_path();
7218 ret = find_first_block_group(root, path, &key);
7226 leaf = path->nodes[0];
7227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7228 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7234 atomic_set(&cache->count, 1);
7235 spin_lock_init(&cache->lock);
7236 spin_lock_init(&cache->tree_lock);
7237 cache->fs_info = info;
7238 INIT_LIST_HEAD(&cache->list);
7239 INIT_LIST_HEAD(&cache->cluster_list);
7242 * we only want to have 32k of ram per block group for keeping
7243 * track of free space, and if we pass 1/2 of that we want to
7244 * start converting things over to using bitmaps
7246 cache->extents_thresh = ((1024 * 32) / 2) /
7247 sizeof(struct btrfs_free_space);
7249 read_extent_buffer(leaf, &cache->item,
7250 btrfs_item_ptr_offset(leaf, path->slots[0]),
7251 sizeof(cache->item));
7252 memcpy(&cache->key, &found_key, sizeof(found_key));
7254 key.objectid = found_key.objectid + found_key.offset;
7255 btrfs_release_path(root, path);
7256 cache->flags = btrfs_block_group_flags(&cache->item);
7257 cache->sectorsize = root->sectorsize;
7260 * check for two cases, either we are full, and therefore
7261 * don't need to bother with the caching work since we won't
7262 * find any space, or we are empty, and we can just add all
7263 * the space in and be done with it. This saves us _alot_ of
7264 * time, particularly in the full case.
7266 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7267 exclude_super_stripes(root, cache);
7268 cache->last_byte_to_unpin = (u64)-1;
7269 cache->cached = BTRFS_CACHE_FINISHED;
7270 free_excluded_extents(root, cache);
7271 } else if (btrfs_block_group_used(&cache->item) == 0) {
7272 exclude_super_stripes(root, cache);
7273 cache->last_byte_to_unpin = (u64)-1;
7274 cache->cached = BTRFS_CACHE_FINISHED;
7275 add_new_free_space(cache, root->fs_info,
7277 found_key.objectid +
7279 free_excluded_extents(root, cache);
7282 ret = update_space_info(info, cache->flags, found_key.offset,
7283 btrfs_block_group_used(&cache->item),
7286 cache->space_info = space_info;
7287 spin_lock(&cache->space_info->lock);
7288 cache->space_info->bytes_super += cache->bytes_super;
7289 spin_unlock(&cache->space_info->lock);
7291 down_write(&space_info->groups_sem);
7292 list_add_tail(&cache->list, &space_info->block_groups);
7293 up_write(&space_info->groups_sem);
7295 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7298 set_avail_alloc_bits(root->fs_info, cache->flags);
7299 if (btrfs_chunk_readonly(root, cache->key.objectid))
7300 set_block_group_readonly(cache);
7304 btrfs_free_path(path);
7308 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7309 struct btrfs_root *root, u64 bytes_used,
7310 u64 type, u64 chunk_objectid, u64 chunk_offset,
7314 struct btrfs_root *extent_root;
7315 struct btrfs_block_group_cache *cache;
7317 extent_root = root->fs_info->extent_root;
7319 root->fs_info->last_trans_log_full_commit = trans->transid;
7321 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7325 cache->key.objectid = chunk_offset;
7326 cache->key.offset = size;
7327 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7328 cache->sectorsize = root->sectorsize;
7331 * we only want to have 32k of ram per block group for keeping track
7332 * of free space, and if we pass 1/2 of that we want to start
7333 * converting things over to using bitmaps
7335 cache->extents_thresh = ((1024 * 32) / 2) /
7336 sizeof(struct btrfs_free_space);
7337 atomic_set(&cache->count, 1);
7338 spin_lock_init(&cache->lock);
7339 spin_lock_init(&cache->tree_lock);
7340 INIT_LIST_HEAD(&cache->list);
7341 INIT_LIST_HEAD(&cache->cluster_list);
7343 btrfs_set_block_group_used(&cache->item, bytes_used);
7344 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7345 cache->flags = type;
7346 btrfs_set_block_group_flags(&cache->item, type);
7348 cache->last_byte_to_unpin = (u64)-1;
7349 cache->cached = BTRFS_CACHE_FINISHED;
7350 exclude_super_stripes(root, cache);
7352 add_new_free_space(cache, root->fs_info, chunk_offset,
7353 chunk_offset + size);
7355 free_excluded_extents(root, cache);
7357 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7358 &cache->space_info);
7361 spin_lock(&cache->space_info->lock);
7362 cache->space_info->bytes_super += cache->bytes_super;
7363 spin_unlock(&cache->space_info->lock);
7365 down_write(&cache->space_info->groups_sem);
7366 list_add_tail(&cache->list, &cache->space_info->block_groups);
7367 up_write(&cache->space_info->groups_sem);
7369 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7372 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7373 sizeof(cache->item));
7376 set_avail_alloc_bits(extent_root->fs_info, type);
7381 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7382 struct btrfs_root *root, u64 group_start)
7384 struct btrfs_path *path;
7385 struct btrfs_block_group_cache *block_group;
7386 struct btrfs_free_cluster *cluster;
7387 struct btrfs_key key;
7390 root = root->fs_info->extent_root;
7392 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7393 BUG_ON(!block_group);
7394 BUG_ON(!block_group->ro);
7396 memcpy(&key, &block_group->key, sizeof(key));
7398 /* make sure this block group isn't part of an allocation cluster */
7399 cluster = &root->fs_info->data_alloc_cluster;
7400 spin_lock(&cluster->refill_lock);
7401 btrfs_return_cluster_to_free_space(block_group, cluster);
7402 spin_unlock(&cluster->refill_lock);
7405 * make sure this block group isn't part of a metadata
7406 * allocation cluster
7408 cluster = &root->fs_info->meta_alloc_cluster;
7409 spin_lock(&cluster->refill_lock);
7410 btrfs_return_cluster_to_free_space(block_group, cluster);
7411 spin_unlock(&cluster->refill_lock);
7413 path = btrfs_alloc_path();
7416 spin_lock(&root->fs_info->block_group_cache_lock);
7417 rb_erase(&block_group->cache_node,
7418 &root->fs_info->block_group_cache_tree);
7419 spin_unlock(&root->fs_info->block_group_cache_lock);
7421 down_write(&block_group->space_info->groups_sem);
7423 * we must use list_del_init so people can check to see if they
7424 * are still on the list after taking the semaphore
7426 list_del_init(&block_group->list);
7427 up_write(&block_group->space_info->groups_sem);
7429 if (block_group->cached == BTRFS_CACHE_STARTED)
7430 wait_block_group_cache_done(block_group);
7432 btrfs_remove_free_space_cache(block_group);
7434 spin_lock(&block_group->space_info->lock);
7435 block_group->space_info->total_bytes -= block_group->key.offset;
7436 block_group->space_info->bytes_readonly -= block_group->key.offset;
7437 spin_unlock(&block_group->space_info->lock);
7439 btrfs_clear_space_info_full(root->fs_info);
7441 btrfs_put_block_group(block_group);
7442 btrfs_put_block_group(block_group);
7444 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7450 ret = btrfs_del_item(trans, root, path);
7452 btrfs_free_path(path);