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);
1579 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1582 #ifdef BIO_RW_DISCARD
1584 u64 map_length = num_bytes;
1585 struct btrfs_multi_bio *multi = NULL;
1587 if (!btrfs_test_opt(root, DISCARD))
1590 /* Tell the block device(s) that the sectors can be discarded */
1591 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1592 bytenr, &map_length, &multi, 0);
1594 struct btrfs_bio_stripe *stripe = multi->stripes;
1597 if (map_length > num_bytes)
1598 map_length = num_bytes;
1600 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1601 btrfs_issue_discard(stripe->dev->bdev,
1614 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1615 struct btrfs_root *root,
1616 u64 bytenr, u64 num_bytes, u64 parent,
1617 u64 root_objectid, u64 owner, u64 offset)
1620 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1621 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1623 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1624 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1625 parent, root_objectid, (int)owner,
1626 BTRFS_ADD_DELAYED_REF, NULL);
1628 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1629 parent, root_objectid, owner, offset,
1630 BTRFS_ADD_DELAYED_REF, NULL);
1635 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1636 struct btrfs_root *root,
1637 u64 bytenr, u64 num_bytes,
1638 u64 parent, u64 root_objectid,
1639 u64 owner, u64 offset, int refs_to_add,
1640 struct btrfs_delayed_extent_op *extent_op)
1642 struct btrfs_path *path;
1643 struct extent_buffer *leaf;
1644 struct btrfs_extent_item *item;
1649 path = btrfs_alloc_path();
1654 path->leave_spinning = 1;
1655 /* this will setup the path even if it fails to insert the back ref */
1656 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1657 path, bytenr, num_bytes, parent,
1658 root_objectid, owner, offset,
1659 refs_to_add, extent_op);
1663 if (ret != -EAGAIN) {
1668 leaf = path->nodes[0];
1669 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1670 refs = btrfs_extent_refs(leaf, item);
1671 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1673 __run_delayed_extent_op(extent_op, leaf, item);
1675 btrfs_mark_buffer_dirty(leaf);
1676 btrfs_release_path(root->fs_info->extent_root, path);
1679 path->leave_spinning = 1;
1681 /* now insert the actual backref */
1682 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1683 path, bytenr, parent, root_objectid,
1684 owner, offset, refs_to_add);
1687 btrfs_free_path(path);
1691 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1692 struct btrfs_root *root,
1693 struct btrfs_delayed_ref_node *node,
1694 struct btrfs_delayed_extent_op *extent_op,
1695 int insert_reserved)
1698 struct btrfs_delayed_data_ref *ref;
1699 struct btrfs_key ins;
1704 ins.objectid = node->bytenr;
1705 ins.offset = node->num_bytes;
1706 ins.type = BTRFS_EXTENT_ITEM_KEY;
1708 ref = btrfs_delayed_node_to_data_ref(node);
1709 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1710 parent = ref->parent;
1712 ref_root = ref->root;
1714 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1716 BUG_ON(extent_op->update_key);
1717 flags |= extent_op->flags_to_set;
1719 ret = alloc_reserved_file_extent(trans, root,
1720 parent, ref_root, flags,
1721 ref->objectid, ref->offset,
1722 &ins, node->ref_mod);
1723 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1724 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1725 node->num_bytes, parent,
1726 ref_root, ref->objectid,
1727 ref->offset, node->ref_mod,
1729 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1730 ret = __btrfs_free_extent(trans, root, node->bytenr,
1731 node->num_bytes, parent,
1732 ref_root, ref->objectid,
1733 ref->offset, node->ref_mod,
1741 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1742 struct extent_buffer *leaf,
1743 struct btrfs_extent_item *ei)
1745 u64 flags = btrfs_extent_flags(leaf, ei);
1746 if (extent_op->update_flags) {
1747 flags |= extent_op->flags_to_set;
1748 btrfs_set_extent_flags(leaf, ei, flags);
1751 if (extent_op->update_key) {
1752 struct btrfs_tree_block_info *bi;
1753 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1754 bi = (struct btrfs_tree_block_info *)(ei + 1);
1755 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1759 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1760 struct btrfs_root *root,
1761 struct btrfs_delayed_ref_node *node,
1762 struct btrfs_delayed_extent_op *extent_op)
1764 struct btrfs_key key;
1765 struct btrfs_path *path;
1766 struct btrfs_extent_item *ei;
1767 struct extent_buffer *leaf;
1772 path = btrfs_alloc_path();
1776 key.objectid = node->bytenr;
1777 key.type = BTRFS_EXTENT_ITEM_KEY;
1778 key.offset = node->num_bytes;
1781 path->leave_spinning = 1;
1782 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1793 leaf = path->nodes[0];
1794 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1795 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1796 if (item_size < sizeof(*ei)) {
1797 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1803 leaf = path->nodes[0];
1804 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807 BUG_ON(item_size < sizeof(*ei));
1808 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1809 __run_delayed_extent_op(extent_op, leaf, ei);
1811 btrfs_mark_buffer_dirty(leaf);
1813 btrfs_free_path(path);
1817 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1818 struct btrfs_root *root,
1819 struct btrfs_delayed_ref_node *node,
1820 struct btrfs_delayed_extent_op *extent_op,
1821 int insert_reserved)
1824 struct btrfs_delayed_tree_ref *ref;
1825 struct btrfs_key ins;
1829 ins.objectid = node->bytenr;
1830 ins.offset = node->num_bytes;
1831 ins.type = BTRFS_EXTENT_ITEM_KEY;
1833 ref = btrfs_delayed_node_to_tree_ref(node);
1834 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1835 parent = ref->parent;
1837 ref_root = ref->root;
1839 BUG_ON(node->ref_mod != 1);
1840 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1841 BUG_ON(!extent_op || !extent_op->update_flags ||
1842 !extent_op->update_key);
1843 ret = alloc_reserved_tree_block(trans, root,
1845 extent_op->flags_to_set,
1848 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1849 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1850 node->num_bytes, parent, ref_root,
1851 ref->level, 0, 1, extent_op);
1852 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1853 ret = __btrfs_free_extent(trans, root, node->bytenr,
1854 node->num_bytes, parent, ref_root,
1855 ref->level, 0, 1, extent_op);
1863 /* helper function to actually process a single delayed ref entry */
1864 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_delayed_ref_node *node,
1867 struct btrfs_delayed_extent_op *extent_op,
1868 int insert_reserved)
1871 if (btrfs_delayed_ref_is_head(node)) {
1872 struct btrfs_delayed_ref_head *head;
1874 * we've hit the end of the chain and we were supposed
1875 * to insert this extent into the tree. But, it got
1876 * deleted before we ever needed to insert it, so all
1877 * we have to do is clean up the accounting
1880 head = btrfs_delayed_node_to_head(node);
1881 if (insert_reserved) {
1883 struct extent_buffer *must_clean = NULL;
1885 ret = pin_down_bytes(trans, root, NULL,
1886 node->bytenr, node->num_bytes,
1887 head->is_data, 1, &must_clean);
1892 clean_tree_block(NULL, root, must_clean);
1893 btrfs_tree_unlock(must_clean);
1894 free_extent_buffer(must_clean);
1896 if (head->is_data) {
1897 ret = btrfs_del_csums(trans, root,
1903 ret = btrfs_free_reserved_extent(root,
1909 mutex_unlock(&head->mutex);
1913 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1914 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1915 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1917 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1918 node->type == BTRFS_SHARED_DATA_REF_KEY)
1919 ret = run_delayed_data_ref(trans, root, node, extent_op,
1926 static noinline struct btrfs_delayed_ref_node *
1927 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1929 struct rb_node *node;
1930 struct btrfs_delayed_ref_node *ref;
1931 int action = BTRFS_ADD_DELAYED_REF;
1934 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1935 * this prevents ref count from going down to zero when
1936 * there still are pending delayed ref.
1938 node = rb_prev(&head->node.rb_node);
1942 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1944 if (ref->bytenr != head->node.bytenr)
1946 if (ref->action == action)
1948 node = rb_prev(node);
1950 if (action == BTRFS_ADD_DELAYED_REF) {
1951 action = BTRFS_DROP_DELAYED_REF;
1957 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct list_head *cluster)
1961 struct btrfs_delayed_ref_root *delayed_refs;
1962 struct btrfs_delayed_ref_node *ref;
1963 struct btrfs_delayed_ref_head *locked_ref = NULL;
1964 struct btrfs_delayed_extent_op *extent_op;
1967 int must_insert_reserved = 0;
1969 delayed_refs = &trans->transaction->delayed_refs;
1972 /* pick a new head ref from the cluster list */
1973 if (list_empty(cluster))
1976 locked_ref = list_entry(cluster->next,
1977 struct btrfs_delayed_ref_head, cluster);
1979 /* grab the lock that says we are going to process
1980 * all the refs for this head */
1981 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1984 * we may have dropped the spin lock to get the head
1985 * mutex lock, and that might have given someone else
1986 * time to free the head. If that's true, it has been
1987 * removed from our list and we can move on.
1989 if (ret == -EAGAIN) {
1997 * record the must insert reserved flag before we
1998 * drop the spin lock.
2000 must_insert_reserved = locked_ref->must_insert_reserved;
2001 locked_ref->must_insert_reserved = 0;
2003 extent_op = locked_ref->extent_op;
2004 locked_ref->extent_op = NULL;
2007 * locked_ref is the head node, so we have to go one
2008 * node back for any delayed ref updates
2010 ref = select_delayed_ref(locked_ref);
2012 /* All delayed refs have been processed, Go ahead
2013 * and send the head node to run_one_delayed_ref,
2014 * so that any accounting fixes can happen
2016 ref = &locked_ref->node;
2018 if (extent_op && must_insert_reserved) {
2024 spin_unlock(&delayed_refs->lock);
2026 ret = run_delayed_extent_op(trans, root,
2032 spin_lock(&delayed_refs->lock);
2036 list_del_init(&locked_ref->cluster);
2041 rb_erase(&ref->rb_node, &delayed_refs->root);
2042 delayed_refs->num_entries--;
2044 spin_unlock(&delayed_refs->lock);
2046 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2047 must_insert_reserved);
2050 btrfs_put_delayed_ref(ref);
2055 spin_lock(&delayed_refs->lock);
2061 * this starts processing the delayed reference count updates and
2062 * extent insertions we have queued up so far. count can be
2063 * 0, which means to process everything in the tree at the start
2064 * of the run (but not newly added entries), or it can be some target
2065 * number you'd like to process.
2067 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2068 struct btrfs_root *root, unsigned long count)
2070 struct rb_node *node;
2071 struct btrfs_delayed_ref_root *delayed_refs;
2072 struct btrfs_delayed_ref_node *ref;
2073 struct list_head cluster;
2075 int run_all = count == (unsigned long)-1;
2078 if (root == root->fs_info->extent_root)
2079 root = root->fs_info->tree_root;
2081 delayed_refs = &trans->transaction->delayed_refs;
2082 INIT_LIST_HEAD(&cluster);
2084 spin_lock(&delayed_refs->lock);
2086 count = delayed_refs->num_entries * 2;
2090 if (!(run_all || run_most) &&
2091 delayed_refs->num_heads_ready < 64)
2095 * go find something we can process in the rbtree. We start at
2096 * the beginning of the tree, and then build a cluster
2097 * of refs to process starting at the first one we are able to
2100 ret = btrfs_find_ref_cluster(trans, &cluster,
2101 delayed_refs->run_delayed_start);
2105 ret = run_clustered_refs(trans, root, &cluster);
2108 count -= min_t(unsigned long, ret, count);
2115 node = rb_first(&delayed_refs->root);
2118 count = (unsigned long)-1;
2121 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2123 if (btrfs_delayed_ref_is_head(ref)) {
2124 struct btrfs_delayed_ref_head *head;
2126 head = btrfs_delayed_node_to_head(ref);
2127 atomic_inc(&ref->refs);
2129 spin_unlock(&delayed_refs->lock);
2130 mutex_lock(&head->mutex);
2131 mutex_unlock(&head->mutex);
2133 btrfs_put_delayed_ref(ref);
2137 node = rb_next(node);
2139 spin_unlock(&delayed_refs->lock);
2140 schedule_timeout(1);
2144 spin_unlock(&delayed_refs->lock);
2148 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *root,
2150 u64 bytenr, u64 num_bytes, u64 flags,
2153 struct btrfs_delayed_extent_op *extent_op;
2156 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2160 extent_op->flags_to_set = flags;
2161 extent_op->update_flags = 1;
2162 extent_op->update_key = 0;
2163 extent_op->is_data = is_data ? 1 : 0;
2165 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2171 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2172 struct btrfs_root *root,
2173 struct btrfs_path *path,
2174 u64 objectid, u64 offset, u64 bytenr)
2176 struct btrfs_delayed_ref_head *head;
2177 struct btrfs_delayed_ref_node *ref;
2178 struct btrfs_delayed_data_ref *data_ref;
2179 struct btrfs_delayed_ref_root *delayed_refs;
2180 struct rb_node *node;
2184 delayed_refs = &trans->transaction->delayed_refs;
2185 spin_lock(&delayed_refs->lock);
2186 head = btrfs_find_delayed_ref_head(trans, bytenr);
2190 if (!mutex_trylock(&head->mutex)) {
2191 atomic_inc(&head->node.refs);
2192 spin_unlock(&delayed_refs->lock);
2194 btrfs_release_path(root->fs_info->extent_root, path);
2196 mutex_lock(&head->mutex);
2197 mutex_unlock(&head->mutex);
2198 btrfs_put_delayed_ref(&head->node);
2202 node = rb_prev(&head->node.rb_node);
2206 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2208 if (ref->bytenr != bytenr)
2212 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2215 data_ref = btrfs_delayed_node_to_data_ref(ref);
2217 node = rb_prev(node);
2219 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2220 if (ref->bytenr == bytenr)
2224 if (data_ref->root != root->root_key.objectid ||
2225 data_ref->objectid != objectid || data_ref->offset != offset)
2230 mutex_unlock(&head->mutex);
2232 spin_unlock(&delayed_refs->lock);
2236 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2237 struct btrfs_root *root,
2238 struct btrfs_path *path,
2239 u64 objectid, u64 offset, u64 bytenr)
2241 struct btrfs_root *extent_root = root->fs_info->extent_root;
2242 struct extent_buffer *leaf;
2243 struct btrfs_extent_data_ref *ref;
2244 struct btrfs_extent_inline_ref *iref;
2245 struct btrfs_extent_item *ei;
2246 struct btrfs_key key;
2250 key.objectid = bytenr;
2251 key.offset = (u64)-1;
2252 key.type = BTRFS_EXTENT_ITEM_KEY;
2254 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2260 if (path->slots[0] == 0)
2264 leaf = path->nodes[0];
2265 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2267 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2271 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2272 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2273 if (item_size < sizeof(*ei)) {
2274 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2278 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2280 if (item_size != sizeof(*ei) +
2281 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2284 if (btrfs_extent_generation(leaf, ei) <=
2285 btrfs_root_last_snapshot(&root->root_item))
2288 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2289 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2290 BTRFS_EXTENT_DATA_REF_KEY)
2293 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2294 if (btrfs_extent_refs(leaf, ei) !=
2295 btrfs_extent_data_ref_count(leaf, ref) ||
2296 btrfs_extent_data_ref_root(leaf, ref) !=
2297 root->root_key.objectid ||
2298 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2299 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2307 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2308 struct btrfs_root *root,
2309 u64 objectid, u64 offset, u64 bytenr)
2311 struct btrfs_path *path;
2315 path = btrfs_alloc_path();
2320 ret = check_committed_ref(trans, root, path, objectid,
2322 if (ret && ret != -ENOENT)
2325 ret2 = check_delayed_ref(trans, root, path, objectid,
2327 } while (ret2 == -EAGAIN);
2329 if (ret2 && ret2 != -ENOENT) {
2334 if (ret != -ENOENT || ret2 != -ENOENT)
2337 btrfs_free_path(path);
2342 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2343 struct extent_buffer *buf, u32 nr_extents)
2345 struct btrfs_key key;
2346 struct btrfs_file_extent_item *fi;
2354 if (!root->ref_cows)
2357 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2359 root_gen = root->root_key.offset;
2362 root_gen = trans->transid - 1;
2365 level = btrfs_header_level(buf);
2366 nritems = btrfs_header_nritems(buf);
2369 struct btrfs_leaf_ref *ref;
2370 struct btrfs_extent_info *info;
2372 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2378 ref->root_gen = root_gen;
2379 ref->bytenr = buf->start;
2380 ref->owner = btrfs_header_owner(buf);
2381 ref->generation = btrfs_header_generation(buf);
2382 ref->nritems = nr_extents;
2383 info = ref->extents;
2385 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2387 btrfs_item_key_to_cpu(buf, &key, i);
2388 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2390 fi = btrfs_item_ptr(buf, i,
2391 struct btrfs_file_extent_item);
2392 if (btrfs_file_extent_type(buf, fi) ==
2393 BTRFS_FILE_EXTENT_INLINE)
2395 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2396 if (disk_bytenr == 0)
2399 info->bytenr = disk_bytenr;
2401 btrfs_file_extent_disk_num_bytes(buf, fi);
2402 info->objectid = key.objectid;
2403 info->offset = key.offset;
2407 ret = btrfs_add_leaf_ref(root, ref, shared);
2408 if (ret == -EEXIST && shared) {
2409 struct btrfs_leaf_ref *old;
2410 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2412 btrfs_remove_leaf_ref(root, old);
2413 btrfs_free_leaf_ref(root, old);
2414 ret = btrfs_add_leaf_ref(root, ref, shared);
2417 btrfs_free_leaf_ref(root, ref);
2423 /* when a block goes through cow, we update the reference counts of
2424 * everything that block points to. The internal pointers of the block
2425 * can be in just about any order, and it is likely to have clusters of
2426 * things that are close together and clusters of things that are not.
2428 * To help reduce the seeks that come with updating all of these reference
2429 * counts, sort them by byte number before actual updates are done.
2431 * struct refsort is used to match byte number to slot in the btree block.
2432 * we sort based on the byte number and then use the slot to actually
2435 * struct refsort is smaller than strcut btrfs_item and smaller than
2436 * struct btrfs_key_ptr. Since we're currently limited to the page size
2437 * for a btree block, there's no way for a kmalloc of refsorts for a
2438 * single node to be bigger than a page.
2446 * for passing into sort()
2448 static int refsort_cmp(const void *a_void, const void *b_void)
2450 const struct refsort *a = a_void;
2451 const struct refsort *b = b_void;
2453 if (a->bytenr < b->bytenr)
2455 if (a->bytenr > b->bytenr)
2461 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2462 struct btrfs_root *root,
2463 struct extent_buffer *buf,
2464 int full_backref, int inc)
2471 struct btrfs_key key;
2472 struct btrfs_file_extent_item *fi;
2476 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2477 u64, u64, u64, u64, u64, u64);
2479 ref_root = btrfs_header_owner(buf);
2480 nritems = btrfs_header_nritems(buf);
2481 level = btrfs_header_level(buf);
2483 if (!root->ref_cows && level == 0)
2487 process_func = btrfs_inc_extent_ref;
2489 process_func = btrfs_free_extent;
2492 parent = buf->start;
2496 for (i = 0; i < nritems; i++) {
2498 btrfs_item_key_to_cpu(buf, &key, i);
2499 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2501 fi = btrfs_item_ptr(buf, i,
2502 struct btrfs_file_extent_item);
2503 if (btrfs_file_extent_type(buf, fi) ==
2504 BTRFS_FILE_EXTENT_INLINE)
2506 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2510 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2511 key.offset -= btrfs_file_extent_offset(buf, fi);
2512 ret = process_func(trans, root, bytenr, num_bytes,
2513 parent, ref_root, key.objectid,
2518 bytenr = btrfs_node_blockptr(buf, i);
2519 num_bytes = btrfs_level_size(root, level - 1);
2520 ret = process_func(trans, root, bytenr, num_bytes,
2521 parent, ref_root, level - 1, 0);
2532 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2533 struct extent_buffer *buf, int full_backref)
2535 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2538 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2539 struct extent_buffer *buf, int full_backref)
2541 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2544 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2545 struct btrfs_root *root,
2546 struct btrfs_path *path,
2547 struct btrfs_block_group_cache *cache)
2550 struct btrfs_root *extent_root = root->fs_info->extent_root;
2552 struct extent_buffer *leaf;
2554 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2559 leaf = path->nodes[0];
2560 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2561 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2562 btrfs_mark_buffer_dirty(leaf);
2563 btrfs_release_path(extent_root, path);
2571 static struct btrfs_block_group_cache *
2572 next_block_group(struct btrfs_root *root,
2573 struct btrfs_block_group_cache *cache)
2575 struct rb_node *node;
2576 spin_lock(&root->fs_info->block_group_cache_lock);
2577 node = rb_next(&cache->cache_node);
2578 btrfs_put_block_group(cache);
2580 cache = rb_entry(node, struct btrfs_block_group_cache,
2582 atomic_inc(&cache->count);
2585 spin_unlock(&root->fs_info->block_group_cache_lock);
2589 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2590 struct btrfs_root *root)
2592 struct btrfs_block_group_cache *cache;
2594 struct btrfs_path *path;
2597 path = btrfs_alloc_path();
2603 err = btrfs_run_delayed_refs(trans, root,
2608 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2612 cache = next_block_group(root, cache);
2622 last = cache->key.objectid + cache->key.offset;
2624 err = write_one_cache_group(trans, root, path, cache);
2626 btrfs_put_block_group(cache);
2629 btrfs_free_path(path);
2633 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2635 struct btrfs_block_group_cache *block_group;
2638 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2639 if (!block_group || block_group->ro)
2642 btrfs_put_block_group(block_group);
2646 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2647 u64 total_bytes, u64 bytes_used,
2648 struct btrfs_space_info **space_info)
2650 struct btrfs_space_info *found;
2652 found = __find_space_info(info, flags);
2654 spin_lock(&found->lock);
2655 found->total_bytes += total_bytes;
2656 found->bytes_used += bytes_used;
2658 spin_unlock(&found->lock);
2659 *space_info = found;
2662 found = kzalloc(sizeof(*found), GFP_NOFS);
2666 INIT_LIST_HEAD(&found->block_groups);
2667 init_rwsem(&found->groups_sem);
2668 spin_lock_init(&found->lock);
2669 found->flags = flags;
2670 found->total_bytes = total_bytes;
2671 found->bytes_used = bytes_used;
2672 found->bytes_pinned = 0;
2673 found->bytes_reserved = 0;
2674 found->bytes_readonly = 0;
2675 found->bytes_delalloc = 0;
2677 found->force_alloc = 0;
2678 *space_info = found;
2679 list_add_rcu(&found->list, &info->space_info);
2680 atomic_set(&found->caching_threads, 0);
2684 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2686 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2687 BTRFS_BLOCK_GROUP_RAID1 |
2688 BTRFS_BLOCK_GROUP_RAID10 |
2689 BTRFS_BLOCK_GROUP_DUP);
2691 if (flags & BTRFS_BLOCK_GROUP_DATA)
2692 fs_info->avail_data_alloc_bits |= extra_flags;
2693 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2694 fs_info->avail_metadata_alloc_bits |= extra_flags;
2695 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2696 fs_info->avail_system_alloc_bits |= extra_flags;
2700 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2702 spin_lock(&cache->space_info->lock);
2703 spin_lock(&cache->lock);
2705 cache->space_info->bytes_readonly += cache->key.offset -
2706 btrfs_block_group_used(&cache->item);
2709 spin_unlock(&cache->lock);
2710 spin_unlock(&cache->space_info->lock);
2713 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2715 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2717 if (num_devices == 1)
2718 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2719 if (num_devices < 4)
2720 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2722 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2723 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2724 BTRFS_BLOCK_GROUP_RAID10))) {
2725 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2728 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2729 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2730 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2733 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2734 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2735 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2736 (flags & BTRFS_BLOCK_GROUP_DUP)))
2737 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2741 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2743 struct btrfs_fs_info *info = root->fs_info;
2747 alloc_profile = info->avail_data_alloc_bits &
2748 info->data_alloc_profile;
2749 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2750 } else if (root == root->fs_info->chunk_root) {
2751 alloc_profile = info->avail_system_alloc_bits &
2752 info->system_alloc_profile;
2753 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2755 alloc_profile = info->avail_metadata_alloc_bits &
2756 info->metadata_alloc_profile;
2757 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2760 return btrfs_reduce_alloc_profile(root, data);
2763 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2767 alloc_target = btrfs_get_alloc_profile(root, 1);
2768 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2772 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2777 level = BTRFS_MAX_LEVEL - 2;
2779 * NOTE: these calculations are absolutely the worst possible case.
2780 * This assumes that _every_ item we insert will require a new leaf, and
2781 * that the tree has grown to its maximum level size.
2785 * for every item we insert we could insert both an extent item and a
2786 * extent ref item. Then for ever item we insert, we will need to cow
2787 * both the original leaf, plus the leaf to the left and right of it.
2789 * Unless we are talking about the extent root, then we just want the
2790 * number of items * 2, since we just need the extent item plus its ref.
2792 if (root == root->fs_info->extent_root)
2793 num_bytes = num_items * 2;
2795 num_bytes = (num_items + (2 * num_items)) * 3;
2798 * num_bytes is total number of leaves we could need times the leaf
2799 * size, and then for every leaf we could end up cow'ing 2 nodes per
2800 * level, down to the leaf level.
2802 num_bytes = (num_bytes * root->leafsize) +
2803 (num_bytes * (level * 2)) * root->nodesize;
2809 * Unreserve metadata space for delalloc. If we have less reserved credits than
2810 * we have extents, this function does nothing.
2812 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2813 struct inode *inode, int num_items)
2815 struct btrfs_fs_info *info = root->fs_info;
2816 struct btrfs_space_info *meta_sinfo;
2821 /* get the space info for where the metadata will live */
2822 alloc_target = btrfs_get_alloc_profile(root, 0);
2823 meta_sinfo = __find_space_info(info, alloc_target);
2825 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2828 spin_lock(&meta_sinfo->lock);
2829 spin_lock(&BTRFS_I(inode)->accounting_lock);
2830 if (BTRFS_I(inode)->reserved_extents <=
2831 BTRFS_I(inode)->outstanding_extents) {
2832 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2833 spin_unlock(&meta_sinfo->lock);
2836 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2838 BTRFS_I(inode)->reserved_extents--;
2839 BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2841 if (meta_sinfo->bytes_delalloc < num_bytes) {
2843 meta_sinfo->bytes_delalloc = 0;
2845 meta_sinfo->bytes_delalloc -= num_bytes;
2847 spin_unlock(&meta_sinfo->lock);
2854 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2858 thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2859 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2860 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2861 meta_sinfo->bytes_may_use;
2863 thresh = meta_sinfo->total_bytes - thresh;
2865 do_div(thresh, 100);
2866 if (thresh <= meta_sinfo->bytes_delalloc)
2867 meta_sinfo->force_delalloc = 1;
2869 meta_sinfo->force_delalloc = 0;
2872 struct async_flush {
2873 struct btrfs_root *root;
2874 struct btrfs_space_info *info;
2875 struct btrfs_work work;
2878 static noinline void flush_delalloc_async(struct btrfs_work *work)
2880 struct async_flush *async;
2881 struct btrfs_root *root;
2882 struct btrfs_space_info *info;
2884 async = container_of(work, struct async_flush, work);
2888 btrfs_start_delalloc_inodes(root);
2889 wake_up(&info->flush_wait);
2890 btrfs_wait_ordered_extents(root, 0);
2892 spin_lock(&info->lock);
2894 spin_unlock(&info->lock);
2895 wake_up(&info->flush_wait);
2900 static void wait_on_flush(struct btrfs_space_info *info)
2906 prepare_to_wait(&info->flush_wait, &wait,
2907 TASK_UNINTERRUPTIBLE);
2908 spin_lock(&info->lock);
2909 if (!info->flushing) {
2910 spin_unlock(&info->lock);
2914 used = info->bytes_used + info->bytes_reserved +
2915 info->bytes_pinned + info->bytes_readonly +
2916 info->bytes_super + info->bytes_root +
2917 info->bytes_may_use + info->bytes_delalloc;
2918 if (used < info->total_bytes) {
2919 spin_unlock(&info->lock);
2922 spin_unlock(&info->lock);
2925 finish_wait(&info->flush_wait, &wait);
2928 static void flush_delalloc(struct btrfs_root *root,
2929 struct btrfs_space_info *info)
2931 struct async_flush *async;
2934 spin_lock(&info->lock);
2936 if (!info->flushing) {
2938 init_waitqueue_head(&info->flush_wait);
2943 spin_unlock(&info->lock);
2946 wait_on_flush(info);
2950 async = kzalloc(sizeof(*async), GFP_NOFS);
2956 async->work.func = flush_delalloc_async;
2958 btrfs_queue_worker(&root->fs_info->enospc_workers,
2960 wait_on_flush(info);
2964 btrfs_start_delalloc_inodes(root);
2965 btrfs_wait_ordered_extents(root, 0);
2967 spin_lock(&info->lock);
2969 spin_unlock(&info->lock);
2970 wake_up(&info->flush_wait);
2973 static int maybe_allocate_chunk(struct btrfs_root *root,
2974 struct btrfs_space_info *info)
2976 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2977 struct btrfs_trans_handle *trans;
2983 free_space = btrfs_super_total_bytes(disk_super);
2985 * we allow the metadata to grow to a max of either 5gb or 5% of the
2986 * space in the volume.
2988 min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2989 div64_u64(free_space * 5, 100));
2990 if (info->total_bytes >= min_metadata) {
2991 spin_unlock(&info->lock);
2996 spin_unlock(&info->lock);
3000 if (!info->allocating_chunk) {
3001 info->force_alloc = 1;
3002 info->allocating_chunk = 1;
3003 init_waitqueue_head(&info->allocate_wait);
3008 spin_unlock(&info->lock);
3011 wait_event(info->allocate_wait,
3012 !info->allocating_chunk);
3016 trans = btrfs_start_transaction(root, 1);
3022 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3023 4096 + 2 * 1024 * 1024,
3025 btrfs_end_transaction(trans, root);
3029 spin_lock(&info->lock);
3030 info->allocating_chunk = 0;
3031 spin_unlock(&info->lock);
3032 wake_up(&info->allocate_wait);
3040 * Reserve metadata space for delalloc.
3042 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3043 struct inode *inode, int num_items)
3045 struct btrfs_fs_info *info = root->fs_info;
3046 struct btrfs_space_info *meta_sinfo;
3053 /* get the space info for where the metadata will live */
3054 alloc_target = btrfs_get_alloc_profile(root, 0);
3055 meta_sinfo = __find_space_info(info, alloc_target);
3057 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3060 spin_lock(&meta_sinfo->lock);
3062 force_delalloc = meta_sinfo->force_delalloc;
3064 if (unlikely(!meta_sinfo->bytes_root))
3065 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3068 meta_sinfo->bytes_delalloc += num_bytes;
3070 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3071 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3072 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3073 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3075 if (used > meta_sinfo->total_bytes) {
3079 if (maybe_allocate_chunk(root, meta_sinfo))
3083 spin_unlock(&meta_sinfo->lock);
3087 filemap_flush(inode->i_mapping);
3089 } else if (flushed == 3) {
3090 flush_delalloc(root, meta_sinfo);
3093 spin_lock(&meta_sinfo->lock);
3094 meta_sinfo->bytes_delalloc -= num_bytes;
3095 spin_unlock(&meta_sinfo->lock);
3096 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3097 BTRFS_I(inode)->outstanding_extents,
3098 BTRFS_I(inode)->reserved_extents);
3099 dump_space_info(meta_sinfo, 0, 0);
3103 BTRFS_I(inode)->reserved_extents++;
3104 check_force_delalloc(meta_sinfo);
3105 spin_unlock(&meta_sinfo->lock);
3107 if (!flushed && force_delalloc)
3108 filemap_flush(inode->i_mapping);
3114 * unreserve num_items number of items worth of metadata space. This needs to
3115 * be paired with btrfs_reserve_metadata_space.
3117 * NOTE: if you have the option, run this _AFTER_ you do a
3118 * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3119 * oprations which will result in more used metadata, so we want to make sure we
3120 * can do that without issue.
3122 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3124 struct btrfs_fs_info *info = root->fs_info;
3125 struct btrfs_space_info *meta_sinfo;
3130 /* get the space info for where the metadata will live */
3131 alloc_target = btrfs_get_alloc_profile(root, 0);
3132 meta_sinfo = __find_space_info(info, alloc_target);
3134 num_bytes = calculate_bytes_needed(root, num_items);
3136 spin_lock(&meta_sinfo->lock);
3137 if (meta_sinfo->bytes_may_use < num_bytes) {
3139 meta_sinfo->bytes_may_use = 0;
3141 meta_sinfo->bytes_may_use -= num_bytes;
3143 spin_unlock(&meta_sinfo->lock);
3151 * Reserve some metadata space for use. We'll calculate the worste case number
3152 * of bytes that would be needed to modify num_items number of items. If we
3153 * have space, fantastic, if not, you get -ENOSPC. Please call
3154 * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3155 * items you reserved, since whatever metadata you needed should have already
3158 * This will commit the transaction to make more space if we don't have enough
3159 * metadata space. THe only time we don't do this is if we're reserving space
3160 * inside of a transaction, then we will just return -ENOSPC and it is the
3161 * callers responsibility to handle it properly.
3163 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3165 struct btrfs_fs_info *info = root->fs_info;
3166 struct btrfs_space_info *meta_sinfo;
3172 /* get the space info for where the metadata will live */
3173 alloc_target = btrfs_get_alloc_profile(root, 0);
3174 meta_sinfo = __find_space_info(info, alloc_target);
3176 num_bytes = calculate_bytes_needed(root, num_items);
3178 spin_lock(&meta_sinfo->lock);
3180 if (unlikely(!meta_sinfo->bytes_root))
3181 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3184 meta_sinfo->bytes_may_use += num_bytes;
3186 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3187 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3188 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3189 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3191 if (used > meta_sinfo->total_bytes) {
3194 if (maybe_allocate_chunk(root, meta_sinfo))
3198 spin_unlock(&meta_sinfo->lock);
3202 flush_delalloc(root, meta_sinfo);
3205 spin_lock(&meta_sinfo->lock);
3206 meta_sinfo->bytes_may_use -= num_bytes;
3207 spin_unlock(&meta_sinfo->lock);
3209 dump_space_info(meta_sinfo, 0, 0);
3213 check_force_delalloc(meta_sinfo);
3214 spin_unlock(&meta_sinfo->lock);
3220 * This will check the space that the inode allocates from to make sure we have
3221 * enough space for bytes.
3223 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3226 struct btrfs_space_info *data_sinfo;
3227 int ret = 0, committed = 0;
3229 /* make sure bytes are sectorsize aligned */
3230 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3232 data_sinfo = BTRFS_I(inode)->space_info;
3237 /* make sure we have enough space to handle the data first */
3238 spin_lock(&data_sinfo->lock);
3239 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3240 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3241 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3242 data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3243 struct btrfs_trans_handle *trans;
3246 * if we don't have enough free bytes in this space then we need
3247 * to alloc a new chunk.
3249 if (!data_sinfo->full) {
3252 data_sinfo->force_alloc = 1;
3253 spin_unlock(&data_sinfo->lock);
3255 alloc_target = btrfs_get_alloc_profile(root, 1);
3256 trans = btrfs_start_transaction(root, 1);
3260 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3261 bytes + 2 * 1024 * 1024,
3263 btrfs_end_transaction(trans, root);
3268 btrfs_set_inode_space_info(root, inode);
3269 data_sinfo = BTRFS_I(inode)->space_info;
3273 spin_unlock(&data_sinfo->lock);
3275 /* commit the current transaction and try again */
3276 if (!committed && !root->fs_info->open_ioctl_trans) {
3278 trans = btrfs_join_transaction(root, 1);
3281 ret = btrfs_commit_transaction(trans, root);
3287 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3288 ", %llu bytes_used, %llu bytes_reserved, "
3289 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3290 "%llu total\n", (unsigned long long)bytes,
3291 (unsigned long long)data_sinfo->bytes_delalloc,
3292 (unsigned long long)data_sinfo->bytes_used,
3293 (unsigned long long)data_sinfo->bytes_reserved,
3294 (unsigned long long)data_sinfo->bytes_pinned,
3295 (unsigned long long)data_sinfo->bytes_readonly,
3296 (unsigned long long)data_sinfo->bytes_may_use,
3297 (unsigned long long)data_sinfo->total_bytes);
3300 data_sinfo->bytes_may_use += bytes;
3301 BTRFS_I(inode)->reserved_bytes += bytes;
3302 spin_unlock(&data_sinfo->lock);
3308 * if there was an error for whatever reason after calling
3309 * btrfs_check_data_free_space, call this so we can cleanup the counters.
3311 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3312 struct inode *inode, u64 bytes)
3314 struct btrfs_space_info *data_sinfo;
3316 /* make sure bytes are sectorsize aligned */
3317 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3319 data_sinfo = BTRFS_I(inode)->space_info;
3320 spin_lock(&data_sinfo->lock);
3321 data_sinfo->bytes_may_use -= bytes;
3322 BTRFS_I(inode)->reserved_bytes -= bytes;
3323 spin_unlock(&data_sinfo->lock);
3326 /* called when we are adding a delalloc extent to the inode's io_tree */
3327 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3330 struct btrfs_space_info *data_sinfo;
3332 /* get the space info for where this inode will be storing its data */
3333 data_sinfo = BTRFS_I(inode)->space_info;
3335 /* make sure we have enough space to handle the data first */
3336 spin_lock(&data_sinfo->lock);
3337 data_sinfo->bytes_delalloc += bytes;
3340 * we are adding a delalloc extent without calling
3341 * btrfs_check_data_free_space first. This happens on a weird
3342 * writepage condition, but shouldn't hurt our accounting
3344 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3345 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3346 BTRFS_I(inode)->reserved_bytes = 0;
3348 data_sinfo->bytes_may_use -= bytes;
3349 BTRFS_I(inode)->reserved_bytes -= bytes;
3352 spin_unlock(&data_sinfo->lock);
3355 /* called when we are clearing an delalloc extent from the inode's io_tree */
3356 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3359 struct btrfs_space_info *info;
3361 info = BTRFS_I(inode)->space_info;
3363 spin_lock(&info->lock);
3364 info->bytes_delalloc -= bytes;
3365 spin_unlock(&info->lock);
3368 static void force_metadata_allocation(struct btrfs_fs_info *info)
3370 struct list_head *head = &info->space_info;
3371 struct btrfs_space_info *found;
3374 list_for_each_entry_rcu(found, head, list) {
3375 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3376 found->force_alloc = 1;
3381 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3382 struct btrfs_root *extent_root, u64 alloc_bytes,
3383 u64 flags, int force)
3385 struct btrfs_space_info *space_info;
3386 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3390 mutex_lock(&fs_info->chunk_mutex);
3392 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3394 space_info = __find_space_info(extent_root->fs_info, flags);
3396 ret = update_space_info(extent_root->fs_info, flags,
3400 BUG_ON(!space_info);
3402 spin_lock(&space_info->lock);
3403 if (space_info->force_alloc)
3405 if (space_info->full) {
3406 spin_unlock(&space_info->lock);
3410 thresh = space_info->total_bytes - space_info->bytes_readonly;
3411 thresh = div_factor(thresh, 8);
3413 (space_info->bytes_used + space_info->bytes_pinned +
3414 space_info->bytes_reserved + alloc_bytes) < thresh) {
3415 spin_unlock(&space_info->lock);
3418 spin_unlock(&space_info->lock);
3421 * if we're doing a data chunk, go ahead and make sure that
3422 * we keep a reasonable number of metadata chunks allocated in the
3425 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3426 fs_info->data_chunk_allocations++;
3427 if (!(fs_info->data_chunk_allocations %
3428 fs_info->metadata_ratio))
3429 force_metadata_allocation(fs_info);
3432 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3433 spin_lock(&space_info->lock);
3435 space_info->full = 1;
3436 space_info->force_alloc = 0;
3437 spin_unlock(&space_info->lock);
3439 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3443 static int update_block_group(struct btrfs_trans_handle *trans,
3444 struct btrfs_root *root,
3445 u64 bytenr, u64 num_bytes, int alloc,
3448 struct btrfs_block_group_cache *cache;
3449 struct btrfs_fs_info *info = root->fs_info;
3450 u64 total = num_bytes;
3454 /* block accounting for super block */
3455 spin_lock(&info->delalloc_lock);
3456 old_val = btrfs_super_bytes_used(&info->super_copy);
3458 old_val += num_bytes;
3460 old_val -= num_bytes;
3461 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3463 /* block accounting for root item */
3464 old_val = btrfs_root_used(&root->root_item);
3466 old_val += num_bytes;
3468 old_val -= num_bytes;
3469 btrfs_set_root_used(&root->root_item, old_val);
3470 spin_unlock(&info->delalloc_lock);
3473 cache = btrfs_lookup_block_group(info, bytenr);
3476 byte_in_group = bytenr - cache->key.objectid;
3477 WARN_ON(byte_in_group > cache->key.offset);
3479 spin_lock(&cache->space_info->lock);
3480 spin_lock(&cache->lock);
3482 old_val = btrfs_block_group_used(&cache->item);
3483 num_bytes = min(total, cache->key.offset - byte_in_group);
3485 old_val += num_bytes;
3486 btrfs_set_block_group_used(&cache->item, old_val);
3487 cache->reserved -= num_bytes;
3488 cache->space_info->bytes_used += num_bytes;
3489 cache->space_info->bytes_reserved -= num_bytes;
3491 cache->space_info->bytes_readonly -= num_bytes;
3492 spin_unlock(&cache->lock);
3493 spin_unlock(&cache->space_info->lock);
3495 old_val -= num_bytes;
3496 cache->space_info->bytes_used -= num_bytes;
3498 cache->space_info->bytes_readonly += num_bytes;
3499 btrfs_set_block_group_used(&cache->item, old_val);
3500 spin_unlock(&cache->lock);
3501 spin_unlock(&cache->space_info->lock);
3505 ret = btrfs_discard_extent(root, bytenr,
3509 ret = btrfs_add_free_space(cache, bytenr,
3514 btrfs_put_block_group(cache);
3516 bytenr += num_bytes;
3521 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3523 struct btrfs_block_group_cache *cache;
3526 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3530 bytenr = cache->key.objectid;
3531 btrfs_put_block_group(cache);
3537 * this function must be called within transaction
3539 int btrfs_pin_extent(struct btrfs_root *root,
3540 u64 bytenr, u64 num_bytes, int reserved)
3542 struct btrfs_fs_info *fs_info = root->fs_info;
3543 struct btrfs_block_group_cache *cache;
3545 cache = btrfs_lookup_block_group(fs_info, bytenr);
3548 spin_lock(&cache->space_info->lock);
3549 spin_lock(&cache->lock);
3550 cache->pinned += num_bytes;
3551 cache->space_info->bytes_pinned += num_bytes;
3553 cache->reserved -= num_bytes;
3554 cache->space_info->bytes_reserved -= num_bytes;
3556 spin_unlock(&cache->lock);
3557 spin_unlock(&cache->space_info->lock);
3559 btrfs_put_block_group(cache);
3561 set_extent_dirty(fs_info->pinned_extents,
3562 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3566 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3567 u64 num_bytes, int reserve)
3569 spin_lock(&cache->space_info->lock);
3570 spin_lock(&cache->lock);
3572 cache->reserved += num_bytes;
3573 cache->space_info->bytes_reserved += num_bytes;
3575 cache->reserved -= num_bytes;
3576 cache->space_info->bytes_reserved -= num_bytes;
3578 spin_unlock(&cache->lock);
3579 spin_unlock(&cache->space_info->lock);
3583 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3584 struct btrfs_root *root)
3586 struct btrfs_fs_info *fs_info = root->fs_info;
3587 struct btrfs_caching_control *next;
3588 struct btrfs_caching_control *caching_ctl;
3589 struct btrfs_block_group_cache *cache;
3591 down_write(&fs_info->extent_commit_sem);
3593 list_for_each_entry_safe(caching_ctl, next,
3594 &fs_info->caching_block_groups, list) {
3595 cache = caching_ctl->block_group;
3596 if (block_group_cache_done(cache)) {
3597 cache->last_byte_to_unpin = (u64)-1;
3598 list_del_init(&caching_ctl->list);
3599 put_caching_control(caching_ctl);
3601 cache->last_byte_to_unpin = caching_ctl->progress;
3605 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3606 fs_info->pinned_extents = &fs_info->freed_extents[1];
3608 fs_info->pinned_extents = &fs_info->freed_extents[0];
3610 up_write(&fs_info->extent_commit_sem);
3614 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3616 struct btrfs_fs_info *fs_info = root->fs_info;
3617 struct btrfs_block_group_cache *cache = NULL;
3620 while (start <= end) {
3622 start >= cache->key.objectid + cache->key.offset) {
3624 btrfs_put_block_group(cache);
3625 cache = btrfs_lookup_block_group(fs_info, start);
3629 len = cache->key.objectid + cache->key.offset - start;
3630 len = min(len, end + 1 - start);
3632 if (start < cache->last_byte_to_unpin) {
3633 len = min(len, cache->last_byte_to_unpin - start);
3634 btrfs_add_free_space(cache, start, len);
3637 spin_lock(&cache->space_info->lock);
3638 spin_lock(&cache->lock);
3639 cache->pinned -= len;
3640 cache->space_info->bytes_pinned -= len;
3641 spin_unlock(&cache->lock);
3642 spin_unlock(&cache->space_info->lock);
3648 btrfs_put_block_group(cache);
3652 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3653 struct btrfs_root *root)
3655 struct btrfs_fs_info *fs_info = root->fs_info;
3656 struct extent_io_tree *unpin;
3661 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3662 unpin = &fs_info->freed_extents[1];
3664 unpin = &fs_info->freed_extents[0];
3667 ret = find_first_extent_bit(unpin, 0, &start, &end,
3672 ret = btrfs_discard_extent(root, start, end + 1 - start);
3674 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3675 unpin_extent_range(root, start, end);
3682 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3683 struct btrfs_root *root,
3684 struct btrfs_path *path,
3685 u64 bytenr, u64 num_bytes,
3686 int is_data, int reserved,
3687 struct extent_buffer **must_clean)
3690 struct extent_buffer *buf;
3695 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3699 /* we can reuse a block if it hasn't been written
3700 * and it is from this transaction. We can't
3701 * reuse anything from the tree log root because
3702 * it has tiny sub-transactions.
3704 if (btrfs_buffer_uptodate(buf, 0) &&
3705 btrfs_try_tree_lock(buf)) {
3706 u64 header_owner = btrfs_header_owner(buf);
3707 u64 header_transid = btrfs_header_generation(buf);
3708 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3709 header_transid == trans->transid &&
3710 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3714 btrfs_tree_unlock(buf);
3716 free_extent_buffer(buf);
3719 btrfs_set_path_blocking(path);
3720 /* unlocks the pinned mutex */
3721 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3727 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3728 struct btrfs_root *root,
3729 u64 bytenr, u64 num_bytes, u64 parent,
3730 u64 root_objectid, u64 owner_objectid,
3731 u64 owner_offset, int refs_to_drop,
3732 struct btrfs_delayed_extent_op *extent_op)
3734 struct btrfs_key key;
3735 struct btrfs_path *path;
3736 struct btrfs_fs_info *info = root->fs_info;
3737 struct btrfs_root *extent_root = info->extent_root;
3738 struct extent_buffer *leaf;
3739 struct btrfs_extent_item *ei;
3740 struct btrfs_extent_inline_ref *iref;
3743 int extent_slot = 0;
3744 int found_extent = 0;
3749 path = btrfs_alloc_path();
3754 path->leave_spinning = 1;
3756 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3757 BUG_ON(!is_data && refs_to_drop != 1);
3759 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3760 bytenr, num_bytes, parent,
3761 root_objectid, owner_objectid,
3764 extent_slot = path->slots[0];
3765 while (extent_slot >= 0) {
3766 btrfs_item_key_to_cpu(path->nodes[0], &key,
3768 if (key.objectid != bytenr)
3770 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3771 key.offset == num_bytes) {
3775 if (path->slots[0] - extent_slot > 5)
3779 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3780 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3781 if (found_extent && item_size < sizeof(*ei))
3784 if (!found_extent) {
3786 ret = remove_extent_backref(trans, extent_root, path,
3790 btrfs_release_path(extent_root, path);
3791 path->leave_spinning = 1;
3793 key.objectid = bytenr;
3794 key.type = BTRFS_EXTENT_ITEM_KEY;
3795 key.offset = num_bytes;
3797 ret = btrfs_search_slot(trans, extent_root,
3800 printk(KERN_ERR "umm, got %d back from search"
3801 ", was looking for %llu\n", ret,
3802 (unsigned long long)bytenr);
3803 btrfs_print_leaf(extent_root, path->nodes[0]);
3806 extent_slot = path->slots[0];
3809 btrfs_print_leaf(extent_root, path->nodes[0]);
3811 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3812 "parent %llu root %llu owner %llu offset %llu\n",
3813 (unsigned long long)bytenr,
3814 (unsigned long long)parent,
3815 (unsigned long long)root_objectid,
3816 (unsigned long long)owner_objectid,
3817 (unsigned long long)owner_offset);
3820 leaf = path->nodes[0];
3821 item_size = btrfs_item_size_nr(leaf, extent_slot);
3822 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3823 if (item_size < sizeof(*ei)) {
3824 BUG_ON(found_extent || extent_slot != path->slots[0]);
3825 ret = convert_extent_item_v0(trans, extent_root, path,
3829 btrfs_release_path(extent_root, path);
3830 path->leave_spinning = 1;
3832 key.objectid = bytenr;
3833 key.type = BTRFS_EXTENT_ITEM_KEY;
3834 key.offset = num_bytes;
3836 ret = btrfs_search_slot(trans, extent_root, &key, path,
3839 printk(KERN_ERR "umm, got %d back from search"
3840 ", was looking for %llu\n", ret,
3841 (unsigned long long)bytenr);
3842 btrfs_print_leaf(extent_root, path->nodes[0]);
3845 extent_slot = path->slots[0];
3846 leaf = path->nodes[0];
3847 item_size = btrfs_item_size_nr(leaf, extent_slot);
3850 BUG_ON(item_size < sizeof(*ei));
3851 ei = btrfs_item_ptr(leaf, extent_slot,
3852 struct btrfs_extent_item);
3853 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3854 struct btrfs_tree_block_info *bi;
3855 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3856 bi = (struct btrfs_tree_block_info *)(ei + 1);
3857 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3860 refs = btrfs_extent_refs(leaf, ei);
3861 BUG_ON(refs < refs_to_drop);
3862 refs -= refs_to_drop;
3866 __run_delayed_extent_op(extent_op, leaf, ei);
3868 * In the case of inline back ref, reference count will
3869 * be updated by remove_extent_backref
3872 BUG_ON(!found_extent);
3874 btrfs_set_extent_refs(leaf, ei, refs);
3875 btrfs_mark_buffer_dirty(leaf);
3878 ret = remove_extent_backref(trans, extent_root, path,
3885 struct extent_buffer *must_clean = NULL;
3888 BUG_ON(is_data && refs_to_drop !=
3889 extent_data_ref_count(root, path, iref));
3891 BUG_ON(path->slots[0] != extent_slot);
3893 BUG_ON(path->slots[0] != extent_slot + 1);
3894 path->slots[0] = extent_slot;
3899 ret = pin_down_bytes(trans, root, path, bytenr,
3900 num_bytes, is_data, 0, &must_clean);
3905 * it is going to be very rare for someone to be waiting
3906 * on the block we're freeing. del_items might need to
3907 * schedule, so rather than get fancy, just force it
3911 btrfs_set_lock_blocking(must_clean);
3913 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3916 btrfs_release_path(extent_root, path);
3919 clean_tree_block(NULL, root, must_clean);
3920 btrfs_tree_unlock(must_clean);
3921 free_extent_buffer(must_clean);
3925 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3928 invalidate_mapping_pages(info->btree_inode->i_mapping,
3929 bytenr >> PAGE_CACHE_SHIFT,
3930 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3933 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3937 btrfs_free_path(path);
3942 * when we free an extent, it is possible (and likely) that we free the last
3943 * delayed ref for that extent as well. This searches the delayed ref tree for
3944 * a given extent, and if there are no other delayed refs to be processed, it
3945 * removes it from the tree.
3947 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3948 struct btrfs_root *root, u64 bytenr)
3950 struct btrfs_delayed_ref_head *head;
3951 struct btrfs_delayed_ref_root *delayed_refs;
3952 struct btrfs_delayed_ref_node *ref;
3953 struct rb_node *node;
3956 delayed_refs = &trans->transaction->delayed_refs;
3957 spin_lock(&delayed_refs->lock);
3958 head = btrfs_find_delayed_ref_head(trans, bytenr);
3962 node = rb_prev(&head->node.rb_node);
3966 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3968 /* there are still entries for this ref, we can't drop it */
3969 if (ref->bytenr == bytenr)
3972 if (head->extent_op) {
3973 if (!head->must_insert_reserved)
3975 kfree(head->extent_op);
3976 head->extent_op = NULL;
3980 * waiting for the lock here would deadlock. If someone else has it
3981 * locked they are already in the process of dropping it anyway
3983 if (!mutex_trylock(&head->mutex))
3987 * at this point we have a head with no other entries. Go
3988 * ahead and process it.
3990 head->node.in_tree = 0;
3991 rb_erase(&head->node.rb_node, &delayed_refs->root);
3993 delayed_refs->num_entries--;
3996 * we don't take a ref on the node because we're removing it from the
3997 * tree, so we just steal the ref the tree was holding.
3999 delayed_refs->num_heads--;
4000 if (list_empty(&head->cluster))
4001 delayed_refs->num_heads_ready--;
4003 list_del_init(&head->cluster);
4004 spin_unlock(&delayed_refs->lock);
4006 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4007 &head->node, head->extent_op,
4008 head->must_insert_reserved);
4010 btrfs_put_delayed_ref(&head->node);
4013 spin_unlock(&delayed_refs->lock);
4017 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4018 struct btrfs_root *root,
4019 u64 bytenr, u64 num_bytes, u64 parent,
4020 u64 root_objectid, u64 owner, u64 offset)
4025 * tree log blocks never actually go into the extent allocation
4026 * tree, just update pinning info and exit early.
4028 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4029 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4030 /* unlocks the pinned mutex */
4031 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4033 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4034 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4035 parent, root_objectid, (int)owner,
4036 BTRFS_DROP_DELAYED_REF, NULL);
4038 ret = check_ref_cleanup(trans, root, bytenr);
4041 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4042 parent, root_objectid, owner,
4043 offset, BTRFS_DROP_DELAYED_REF, NULL);
4049 static u64 stripe_align(struct btrfs_root *root, u64 val)
4051 u64 mask = ((u64)root->stripesize - 1);
4052 u64 ret = (val + mask) & ~mask;
4057 * when we wait for progress in the block group caching, its because
4058 * our allocation attempt failed at least once. So, we must sleep
4059 * and let some progress happen before we try again.
4061 * This function will sleep at least once waiting for new free space to
4062 * show up, and then it will check the block group free space numbers
4063 * for our min num_bytes. Another option is to have it go ahead
4064 * and look in the rbtree for a free extent of a given size, but this
4068 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4071 struct btrfs_caching_control *caching_ctl;
4074 caching_ctl = get_caching_control(cache);
4078 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4079 (cache->free_space >= num_bytes));
4081 put_caching_control(caching_ctl);
4086 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4088 struct btrfs_caching_control *caching_ctl;
4091 caching_ctl = get_caching_control(cache);
4095 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4097 put_caching_control(caching_ctl);
4101 enum btrfs_loop_type {
4102 LOOP_CACHED_ONLY = 0,
4103 LOOP_CACHING_NOWAIT = 1,
4104 LOOP_CACHING_WAIT = 2,
4105 LOOP_ALLOC_CHUNK = 3,
4106 LOOP_NO_EMPTY_SIZE = 4,
4110 * walks the btree of allocated extents and find a hole of a given size.
4111 * The key ins is changed to record the hole:
4112 * ins->objectid == block start
4113 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4114 * ins->offset == number of blocks
4115 * Any available blocks before search_start are skipped.
4117 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4118 struct btrfs_root *orig_root,
4119 u64 num_bytes, u64 empty_size,
4120 u64 search_start, u64 search_end,
4121 u64 hint_byte, struct btrfs_key *ins,
4122 u64 exclude_start, u64 exclude_nr,
4126 struct btrfs_root *root = orig_root->fs_info->extent_root;
4127 struct btrfs_free_cluster *last_ptr = NULL;
4128 struct btrfs_block_group_cache *block_group = NULL;
4129 int empty_cluster = 2 * 1024 * 1024;
4130 int allowed_chunk_alloc = 0;
4131 struct btrfs_space_info *space_info;
4132 int last_ptr_loop = 0;
4134 bool found_uncached_bg = false;
4135 bool failed_cluster_refill = false;
4136 bool failed_alloc = false;
4138 WARN_ON(num_bytes < root->sectorsize);
4139 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4143 space_info = __find_space_info(root->fs_info, data);
4145 if (orig_root->ref_cows || empty_size)
4146 allowed_chunk_alloc = 1;
4148 if (data & BTRFS_BLOCK_GROUP_METADATA) {
4149 last_ptr = &root->fs_info->meta_alloc_cluster;
4150 if (!btrfs_test_opt(root, SSD))
4151 empty_cluster = 64 * 1024;
4154 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4155 last_ptr = &root->fs_info->data_alloc_cluster;
4159 spin_lock(&last_ptr->lock);
4160 if (last_ptr->block_group)
4161 hint_byte = last_ptr->window_start;
4162 spin_unlock(&last_ptr->lock);
4165 search_start = max(search_start, first_logical_byte(root, 0));
4166 search_start = max(search_start, hint_byte);
4171 if (search_start == hint_byte) {
4172 block_group = btrfs_lookup_block_group(root->fs_info,
4175 * we don't want to use the block group if it doesn't match our
4176 * allocation bits, or if its not cached.
4178 if (block_group && block_group_bits(block_group, data) &&
4179 block_group_cache_done(block_group)) {
4180 down_read(&space_info->groups_sem);
4181 if (list_empty(&block_group->list) ||
4184 * someone is removing this block group,
4185 * we can't jump into the have_block_group
4186 * target because our list pointers are not
4189 btrfs_put_block_group(block_group);
4190 up_read(&space_info->groups_sem);
4192 goto have_block_group;
4193 } else if (block_group) {
4194 btrfs_put_block_group(block_group);
4199 down_read(&space_info->groups_sem);
4200 list_for_each_entry(block_group, &space_info->block_groups, list) {
4204 atomic_inc(&block_group->count);
4205 search_start = block_group->key.objectid;
4208 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4210 * we want to start caching kthreads, but not too many
4211 * right off the bat so we don't overwhelm the system,
4212 * so only start them if there are less than 2 and we're
4213 * in the initial allocation phase.
4215 if (loop > LOOP_CACHING_NOWAIT ||
4216 atomic_read(&space_info->caching_threads) < 2) {
4217 ret = cache_block_group(block_group);
4222 cached = block_group_cache_done(block_group);
4223 if (unlikely(!cached)) {
4224 found_uncached_bg = true;
4226 /* if we only want cached bgs, loop */
4227 if (loop == LOOP_CACHED_ONLY)
4231 if (unlikely(block_group->ro))
4235 * Ok we want to try and use the cluster allocator, so lets look
4236 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4237 * have tried the cluster allocator plenty of times at this
4238 * point and not have found anything, so we are likely way too
4239 * fragmented for the clustering stuff to find anything, so lets
4240 * just skip it and let the allocator find whatever block it can
4243 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4245 * the refill lock keeps out other
4246 * people trying to start a new cluster
4248 spin_lock(&last_ptr->refill_lock);
4249 if (last_ptr->block_group &&
4250 (last_ptr->block_group->ro ||
4251 !block_group_bits(last_ptr->block_group, data))) {
4253 goto refill_cluster;
4256 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4257 num_bytes, search_start);
4259 /* we have a block, we're done */
4260 spin_unlock(&last_ptr->refill_lock);
4264 spin_lock(&last_ptr->lock);
4266 * whoops, this cluster doesn't actually point to
4267 * this block group. Get a ref on the block
4268 * group is does point to and try again
4270 if (!last_ptr_loop && last_ptr->block_group &&
4271 last_ptr->block_group != block_group) {
4273 btrfs_put_block_group(block_group);
4274 block_group = last_ptr->block_group;
4275 atomic_inc(&block_group->count);
4276 spin_unlock(&last_ptr->lock);
4277 spin_unlock(&last_ptr->refill_lock);
4280 search_start = block_group->key.objectid;
4282 * we know this block group is properly
4283 * in the list because
4284 * btrfs_remove_block_group, drops the
4285 * cluster before it removes the block
4286 * group from the list
4288 goto have_block_group;
4290 spin_unlock(&last_ptr->lock);
4293 * this cluster didn't work out, free it and
4296 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4300 /* allocate a cluster in this block group */
4301 ret = btrfs_find_space_cluster(trans, root,
4302 block_group, last_ptr,
4304 empty_cluster + empty_size);
4307 * now pull our allocation out of this
4310 offset = btrfs_alloc_from_cluster(block_group,
4311 last_ptr, num_bytes,
4314 /* we found one, proceed */
4315 spin_unlock(&last_ptr->refill_lock);
4318 } else if (!cached && loop > LOOP_CACHING_NOWAIT
4319 && !failed_cluster_refill) {
4320 spin_unlock(&last_ptr->refill_lock);
4322 failed_cluster_refill = true;
4323 wait_block_group_cache_progress(block_group,
4324 num_bytes + empty_cluster + empty_size);
4325 goto have_block_group;
4329 * at this point we either didn't find a cluster
4330 * or we weren't able to allocate a block from our
4331 * cluster. Free the cluster we've been trying
4332 * to use, and go to the next block group
4334 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4335 spin_unlock(&last_ptr->refill_lock);
4339 offset = btrfs_find_space_for_alloc(block_group, search_start,
4340 num_bytes, empty_size);
4342 * If we didn't find a chunk, and we haven't failed on this
4343 * block group before, and this block group is in the middle of
4344 * caching and we are ok with waiting, then go ahead and wait
4345 * for progress to be made, and set failed_alloc to true.
4347 * If failed_alloc is true then we've already waited on this
4348 * block group once and should move on to the next block group.
4350 if (!offset && !failed_alloc && !cached &&
4351 loop > LOOP_CACHING_NOWAIT) {
4352 wait_block_group_cache_progress(block_group,
4353 num_bytes + empty_size);
4354 failed_alloc = true;
4355 goto have_block_group;
4356 } else if (!offset) {
4360 search_start = stripe_align(root, offset);
4361 /* move on to the next group */
4362 if (search_start + num_bytes >= search_end) {
4363 btrfs_add_free_space(block_group, offset, num_bytes);
4367 /* move on to the next group */
4368 if (search_start + num_bytes >
4369 block_group->key.objectid + block_group->key.offset) {
4370 btrfs_add_free_space(block_group, offset, num_bytes);
4374 if (exclude_nr > 0 &&
4375 (search_start + num_bytes > exclude_start &&
4376 search_start < exclude_start + exclude_nr)) {
4377 search_start = exclude_start + exclude_nr;
4379 btrfs_add_free_space(block_group, offset, num_bytes);
4381 * if search_start is still in this block group
4382 * then we just re-search this block group
4384 if (search_start >= block_group->key.objectid &&
4385 search_start < (block_group->key.objectid +
4386 block_group->key.offset))
4387 goto have_block_group;
4391 ins->objectid = search_start;
4392 ins->offset = num_bytes;
4394 if (offset < search_start)
4395 btrfs_add_free_space(block_group, offset,
4396 search_start - offset);
4397 BUG_ON(offset > search_start);
4399 update_reserved_extents(block_group, num_bytes, 1);
4401 /* we are all good, lets return */
4404 failed_cluster_refill = false;
4405 failed_alloc = false;
4406 btrfs_put_block_group(block_group);
4408 up_read(&space_info->groups_sem);
4410 /* LOOP_CACHED_ONLY, only search fully cached block groups
4411 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4412 * dont wait foR them to finish caching
4413 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4414 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4415 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4418 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4419 (found_uncached_bg || empty_size || empty_cluster ||
4420 allowed_chunk_alloc)) {
4421 if (found_uncached_bg) {
4422 found_uncached_bg = false;
4423 if (loop < LOOP_CACHING_WAIT) {
4429 if (loop == LOOP_ALLOC_CHUNK) {
4434 if (allowed_chunk_alloc) {
4435 ret = do_chunk_alloc(trans, root, num_bytes +
4436 2 * 1024 * 1024, data, 1);
4437 allowed_chunk_alloc = 0;
4439 space_info->force_alloc = 1;
4442 if (loop < LOOP_NO_EMPTY_SIZE) {
4447 } else if (!ins->objectid) {
4451 /* we found what we needed */
4452 if (ins->objectid) {
4453 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4454 trans->block_group = block_group->key.objectid;
4456 btrfs_put_block_group(block_group);
4463 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4464 int dump_block_groups)
4466 struct btrfs_block_group_cache *cache;
4468 spin_lock(&info->lock);
4469 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4470 (unsigned long long)(info->total_bytes - info->bytes_used -
4471 info->bytes_pinned - info->bytes_reserved -
4473 (info->full) ? "" : "not ");
4474 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4475 " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4477 (unsigned long long)info->total_bytes,
4478 (unsigned long long)info->bytes_pinned,
4479 (unsigned long long)info->bytes_delalloc,
4480 (unsigned long long)info->bytes_may_use,
4481 (unsigned long long)info->bytes_used,
4482 (unsigned long long)info->bytes_root,
4483 (unsigned long long)info->bytes_super,
4484 (unsigned long long)info->bytes_reserved);
4485 spin_unlock(&info->lock);
4487 if (!dump_block_groups)
4490 down_read(&info->groups_sem);
4491 list_for_each_entry(cache, &info->block_groups, list) {
4492 spin_lock(&cache->lock);
4493 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4494 "%llu pinned %llu reserved\n",
4495 (unsigned long long)cache->key.objectid,
4496 (unsigned long long)cache->key.offset,
4497 (unsigned long long)btrfs_block_group_used(&cache->item),
4498 (unsigned long long)cache->pinned,
4499 (unsigned long long)cache->reserved);
4500 btrfs_dump_free_space(cache, bytes);
4501 spin_unlock(&cache->lock);
4503 up_read(&info->groups_sem);
4506 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4507 struct btrfs_root *root,
4508 u64 num_bytes, u64 min_alloc_size,
4509 u64 empty_size, u64 hint_byte,
4510 u64 search_end, struct btrfs_key *ins,
4514 u64 search_start = 0;
4515 struct btrfs_fs_info *info = root->fs_info;
4517 data = btrfs_get_alloc_profile(root, data);
4520 * the only place that sets empty_size is btrfs_realloc_node, which
4521 * is not called recursively on allocations
4523 if (empty_size || root->ref_cows) {
4524 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4525 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4527 BTRFS_BLOCK_GROUP_METADATA |
4528 (info->metadata_alloc_profile &
4529 info->avail_metadata_alloc_bits), 0);
4531 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4532 num_bytes + 2 * 1024 * 1024, data, 0);
4535 WARN_ON(num_bytes < root->sectorsize);
4536 ret = find_free_extent(trans, root, num_bytes, empty_size,
4537 search_start, search_end, hint_byte, ins,
4538 trans->alloc_exclude_start,
4539 trans->alloc_exclude_nr, data);
4541 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4542 num_bytes = num_bytes >> 1;
4543 num_bytes = num_bytes & ~(root->sectorsize - 1);
4544 num_bytes = max(num_bytes, min_alloc_size);
4545 do_chunk_alloc(trans, root->fs_info->extent_root,
4546 num_bytes, data, 1);
4549 if (ret == -ENOSPC) {
4550 struct btrfs_space_info *sinfo;
4552 sinfo = __find_space_info(root->fs_info, data);
4553 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4554 "wanted %llu\n", (unsigned long long)data,
4555 (unsigned long long)num_bytes);
4556 dump_space_info(sinfo, num_bytes, 1);
4562 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4564 struct btrfs_block_group_cache *cache;
4567 cache = btrfs_lookup_block_group(root->fs_info, start);
4569 printk(KERN_ERR "Unable to find block group for %llu\n",
4570 (unsigned long long)start);
4574 ret = btrfs_discard_extent(root, start, len);
4576 btrfs_add_free_space(cache, start, len);
4577 update_reserved_extents(cache, len, 0);
4578 btrfs_put_block_group(cache);
4583 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4584 struct btrfs_root *root,
4585 u64 parent, u64 root_objectid,
4586 u64 flags, u64 owner, u64 offset,
4587 struct btrfs_key *ins, int ref_mod)
4590 struct btrfs_fs_info *fs_info = root->fs_info;
4591 struct btrfs_extent_item *extent_item;
4592 struct btrfs_extent_inline_ref *iref;
4593 struct btrfs_path *path;
4594 struct extent_buffer *leaf;
4599 type = BTRFS_SHARED_DATA_REF_KEY;
4601 type = BTRFS_EXTENT_DATA_REF_KEY;
4603 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4605 path = btrfs_alloc_path();
4608 path->leave_spinning = 1;
4609 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4613 leaf = path->nodes[0];
4614 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4615 struct btrfs_extent_item);
4616 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4617 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4618 btrfs_set_extent_flags(leaf, extent_item,
4619 flags | BTRFS_EXTENT_FLAG_DATA);
4621 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4622 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4624 struct btrfs_shared_data_ref *ref;
4625 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4626 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4627 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4629 struct btrfs_extent_data_ref *ref;
4630 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4631 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4632 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4633 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4634 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4637 btrfs_mark_buffer_dirty(path->nodes[0]);
4638 btrfs_free_path(path);
4640 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4643 printk(KERN_ERR "btrfs update block group failed for %llu "
4644 "%llu\n", (unsigned long long)ins->objectid,
4645 (unsigned long long)ins->offset);
4651 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4652 struct btrfs_root *root,
4653 u64 parent, u64 root_objectid,
4654 u64 flags, struct btrfs_disk_key *key,
4655 int level, struct btrfs_key *ins)
4658 struct btrfs_fs_info *fs_info = root->fs_info;
4659 struct btrfs_extent_item *extent_item;
4660 struct btrfs_tree_block_info *block_info;
4661 struct btrfs_extent_inline_ref *iref;
4662 struct btrfs_path *path;
4663 struct extent_buffer *leaf;
4664 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4666 path = btrfs_alloc_path();
4669 path->leave_spinning = 1;
4670 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4674 leaf = path->nodes[0];
4675 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4676 struct btrfs_extent_item);
4677 btrfs_set_extent_refs(leaf, extent_item, 1);
4678 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4679 btrfs_set_extent_flags(leaf, extent_item,
4680 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4681 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4683 btrfs_set_tree_block_key(leaf, block_info, key);
4684 btrfs_set_tree_block_level(leaf, block_info, level);
4686 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4688 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4689 btrfs_set_extent_inline_ref_type(leaf, iref,
4690 BTRFS_SHARED_BLOCK_REF_KEY);
4691 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4693 btrfs_set_extent_inline_ref_type(leaf, iref,
4694 BTRFS_TREE_BLOCK_REF_KEY);
4695 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4698 btrfs_mark_buffer_dirty(leaf);
4699 btrfs_free_path(path);
4701 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4704 printk(KERN_ERR "btrfs update block group failed for %llu "
4705 "%llu\n", (unsigned long long)ins->objectid,
4706 (unsigned long long)ins->offset);
4712 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4713 struct btrfs_root *root,
4714 u64 root_objectid, u64 owner,
4715 u64 offset, struct btrfs_key *ins)
4719 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4721 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4722 0, root_objectid, owner, offset,
4723 BTRFS_ADD_DELAYED_EXTENT, NULL);
4728 * this is used by the tree logging recovery code. It records that
4729 * an extent has been allocated and makes sure to clear the free
4730 * space cache bits as well
4732 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4733 struct btrfs_root *root,
4734 u64 root_objectid, u64 owner, u64 offset,
4735 struct btrfs_key *ins)
4738 struct btrfs_block_group_cache *block_group;
4739 struct btrfs_caching_control *caching_ctl;
4740 u64 start = ins->objectid;
4741 u64 num_bytes = ins->offset;
4743 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4744 cache_block_group(block_group);
4745 caching_ctl = get_caching_control(block_group);
4748 BUG_ON(!block_group_cache_done(block_group));
4749 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4752 mutex_lock(&caching_ctl->mutex);
4754 if (start >= caching_ctl->progress) {
4755 ret = add_excluded_extent(root, start, num_bytes);
4757 } else if (start + num_bytes <= caching_ctl->progress) {
4758 ret = btrfs_remove_free_space(block_group,
4762 num_bytes = caching_ctl->progress - start;
4763 ret = btrfs_remove_free_space(block_group,
4767 start = caching_ctl->progress;
4768 num_bytes = ins->objectid + ins->offset -
4769 caching_ctl->progress;
4770 ret = add_excluded_extent(root, start, num_bytes);
4774 mutex_unlock(&caching_ctl->mutex);
4775 put_caching_control(caching_ctl);
4778 update_reserved_extents(block_group, ins->offset, 1);
4779 btrfs_put_block_group(block_group);
4780 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4781 0, owner, offset, ins, 1);
4786 * finds a free extent and does all the dirty work required for allocation
4787 * returns the key for the extent through ins, and a tree buffer for
4788 * the first block of the extent through buf.
4790 * returns 0 if everything worked, non-zero otherwise.
4792 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4793 struct btrfs_root *root,
4794 u64 num_bytes, u64 parent, u64 root_objectid,
4795 struct btrfs_disk_key *key, int level,
4796 u64 empty_size, u64 hint_byte, u64 search_end,
4797 struct btrfs_key *ins)
4802 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4803 empty_size, hint_byte, search_end,
4808 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4810 parent = ins->objectid;
4811 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4815 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4816 struct btrfs_delayed_extent_op *extent_op;
4817 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4820 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4822 memset(&extent_op->key, 0, sizeof(extent_op->key));
4823 extent_op->flags_to_set = flags;
4824 extent_op->update_key = 1;
4825 extent_op->update_flags = 1;
4826 extent_op->is_data = 0;
4828 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4829 ins->offset, parent, root_objectid,
4830 level, BTRFS_ADD_DELAYED_EXTENT,
4837 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4838 struct btrfs_root *root,
4839 u64 bytenr, u32 blocksize,
4842 struct extent_buffer *buf;
4844 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4846 return ERR_PTR(-ENOMEM);
4847 btrfs_set_header_generation(buf, trans->transid);
4848 btrfs_set_buffer_lockdep_class(buf, level);
4849 btrfs_tree_lock(buf);
4850 clean_tree_block(trans, root, buf);
4852 btrfs_set_lock_blocking(buf);
4853 btrfs_set_buffer_uptodate(buf);
4855 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4856 set_extent_dirty(&root->dirty_log_pages, buf->start,
4857 buf->start + buf->len - 1, GFP_NOFS);
4859 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4860 buf->start + buf->len - 1, GFP_NOFS);
4862 trans->blocks_used++;
4863 /* this returns a buffer locked for blocking */
4868 * helper function to allocate a block for a given tree
4869 * returns the tree buffer or NULL.
4871 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4872 struct btrfs_root *root, u32 blocksize,
4873 u64 parent, u64 root_objectid,
4874 struct btrfs_disk_key *key, int level,
4875 u64 hint, u64 empty_size)
4877 struct btrfs_key ins;
4879 struct extent_buffer *buf;
4881 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4882 key, level, empty_size, hint, (u64)-1, &ins);
4885 return ERR_PTR(ret);
4888 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4893 struct walk_control {
4894 u64 refs[BTRFS_MAX_LEVEL];
4895 u64 flags[BTRFS_MAX_LEVEL];
4896 struct btrfs_key update_progress;
4906 #define DROP_REFERENCE 1
4907 #define UPDATE_BACKREF 2
4909 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4910 struct btrfs_root *root,
4911 struct walk_control *wc,
4912 struct btrfs_path *path)
4921 struct btrfs_key key;
4922 struct extent_buffer *eb;
4927 if (path->slots[wc->level] < wc->reada_slot) {
4928 wc->reada_count = wc->reada_count * 2 / 3;
4929 wc->reada_count = max(wc->reada_count, 2);
4931 wc->reada_count = wc->reada_count * 3 / 2;
4932 wc->reada_count = min_t(int, wc->reada_count,
4933 BTRFS_NODEPTRS_PER_BLOCK(root));
4936 eb = path->nodes[wc->level];
4937 nritems = btrfs_header_nritems(eb);
4938 blocksize = btrfs_level_size(root, wc->level - 1);
4940 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4941 if (nread >= wc->reada_count)
4945 bytenr = btrfs_node_blockptr(eb, slot);
4946 generation = btrfs_node_ptr_generation(eb, slot);
4948 if (slot == path->slots[wc->level])
4951 if (wc->stage == UPDATE_BACKREF &&
4952 generation <= root->root_key.offset)
4955 /* We don't lock the tree block, it's OK to be racy here */
4956 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4961 if (wc->stage == DROP_REFERENCE) {
4965 if (wc->level == 1 &&
4966 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4968 if (!wc->update_ref ||
4969 generation <= root->root_key.offset)
4971 btrfs_node_key_to_cpu(eb, &key, slot);
4972 ret = btrfs_comp_cpu_keys(&key,
4973 &wc->update_progress);
4977 if (wc->level == 1 &&
4978 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4982 ret = readahead_tree_block(root, bytenr, blocksize,
4986 last = bytenr + blocksize;
4989 wc->reada_slot = slot;
4993 * hepler to process tree block while walking down the tree.
4995 * when wc->stage == UPDATE_BACKREF, this function updates
4996 * back refs for pointers in the block.
4998 * NOTE: return value 1 means we should stop walking down.
5000 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5001 struct btrfs_root *root,
5002 struct btrfs_path *path,
5003 struct walk_control *wc, int lookup_info)
5005 int level = wc->level;
5006 struct extent_buffer *eb = path->nodes[level];
5007 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5010 if (wc->stage == UPDATE_BACKREF &&
5011 btrfs_header_owner(eb) != root->root_key.objectid)
5015 * when reference count of tree block is 1, it won't increase
5016 * again. once full backref flag is set, we never clear it.
5019 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5020 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5021 BUG_ON(!path->locks[level]);
5022 ret = btrfs_lookup_extent_info(trans, root,
5027 BUG_ON(wc->refs[level] == 0);
5030 if (wc->stage == DROP_REFERENCE) {
5031 if (wc->refs[level] > 1)
5034 if (path->locks[level] && !wc->keep_locks) {
5035 btrfs_tree_unlock(eb);
5036 path->locks[level] = 0;
5041 /* wc->stage == UPDATE_BACKREF */
5042 if (!(wc->flags[level] & flag)) {
5043 BUG_ON(!path->locks[level]);
5044 ret = btrfs_inc_ref(trans, root, eb, 1);
5046 ret = btrfs_dec_ref(trans, root, eb, 0);
5048 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5051 wc->flags[level] |= flag;
5055 * the block is shared by multiple trees, so it's not good to
5056 * keep the tree lock
5058 if (path->locks[level] && level > 0) {
5059 btrfs_tree_unlock(eb);
5060 path->locks[level] = 0;
5066 * hepler to process tree block pointer.
5068 * when wc->stage == DROP_REFERENCE, this function checks
5069 * reference count of the block pointed to. if the block
5070 * is shared and we need update back refs for the subtree
5071 * rooted at the block, this function changes wc->stage to
5072 * UPDATE_BACKREF. if the block is shared and there is no
5073 * need to update back, this function drops the reference
5076 * NOTE: return value 1 means we should stop walking down.
5078 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5079 struct btrfs_root *root,
5080 struct btrfs_path *path,
5081 struct walk_control *wc, int *lookup_info)
5087 struct btrfs_key key;
5088 struct extent_buffer *next;
5089 int level = wc->level;
5093 generation = btrfs_node_ptr_generation(path->nodes[level],
5094 path->slots[level]);
5096 * if the lower level block was created before the snapshot
5097 * was created, we know there is no need to update back refs
5100 if (wc->stage == UPDATE_BACKREF &&
5101 generation <= root->root_key.offset) {
5106 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5107 blocksize = btrfs_level_size(root, level - 1);
5109 next = btrfs_find_tree_block(root, bytenr, blocksize);
5111 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5114 btrfs_tree_lock(next);
5115 btrfs_set_lock_blocking(next);
5117 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5118 &wc->refs[level - 1],
5119 &wc->flags[level - 1]);
5121 BUG_ON(wc->refs[level - 1] == 0);
5124 if (wc->stage == DROP_REFERENCE) {
5125 if (wc->refs[level - 1] > 1) {
5127 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5130 if (!wc->update_ref ||
5131 generation <= root->root_key.offset)
5134 btrfs_node_key_to_cpu(path->nodes[level], &key,
5135 path->slots[level]);
5136 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5140 wc->stage = UPDATE_BACKREF;
5141 wc->shared_level = level - 1;
5145 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5149 if (!btrfs_buffer_uptodate(next, generation)) {
5150 btrfs_tree_unlock(next);
5151 free_extent_buffer(next);
5157 if (reada && level == 1)
5158 reada_walk_down(trans, root, wc, path);
5159 next = read_tree_block(root, bytenr, blocksize, generation);
5160 btrfs_tree_lock(next);
5161 btrfs_set_lock_blocking(next);
5165 BUG_ON(level != btrfs_header_level(next));
5166 path->nodes[level] = next;
5167 path->slots[level] = 0;
5168 path->locks[level] = 1;
5174 wc->refs[level - 1] = 0;
5175 wc->flags[level - 1] = 0;
5176 if (wc->stage == DROP_REFERENCE) {
5177 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5178 parent = path->nodes[level]->start;
5180 BUG_ON(root->root_key.objectid !=
5181 btrfs_header_owner(path->nodes[level]));
5185 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5186 root->root_key.objectid, level - 1, 0);
5189 btrfs_tree_unlock(next);
5190 free_extent_buffer(next);
5196 * hepler to process tree block while walking up the tree.
5198 * when wc->stage == DROP_REFERENCE, this function drops
5199 * reference count on the block.
5201 * when wc->stage == UPDATE_BACKREF, this function changes
5202 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5203 * to UPDATE_BACKREF previously while processing the block.
5205 * NOTE: return value 1 means we should stop walking up.
5207 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5208 struct btrfs_root *root,
5209 struct btrfs_path *path,
5210 struct walk_control *wc)
5213 int level = wc->level;
5214 struct extent_buffer *eb = path->nodes[level];
5217 if (wc->stage == UPDATE_BACKREF) {
5218 BUG_ON(wc->shared_level < level);
5219 if (level < wc->shared_level)
5222 ret = find_next_key(path, level + 1, &wc->update_progress);
5226 wc->stage = DROP_REFERENCE;
5227 wc->shared_level = -1;
5228 path->slots[level] = 0;
5231 * check reference count again if the block isn't locked.
5232 * we should start walking down the tree again if reference
5235 if (!path->locks[level]) {
5237 btrfs_tree_lock(eb);
5238 btrfs_set_lock_blocking(eb);
5239 path->locks[level] = 1;
5241 ret = btrfs_lookup_extent_info(trans, root,
5246 BUG_ON(wc->refs[level] == 0);
5247 if (wc->refs[level] == 1) {
5248 btrfs_tree_unlock(eb);
5249 path->locks[level] = 0;
5255 /* wc->stage == DROP_REFERENCE */
5256 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5258 if (wc->refs[level] == 1) {
5260 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5261 ret = btrfs_dec_ref(trans, root, eb, 1);
5263 ret = btrfs_dec_ref(trans, root, eb, 0);
5266 /* make block locked assertion in clean_tree_block happy */
5267 if (!path->locks[level] &&
5268 btrfs_header_generation(eb) == trans->transid) {
5269 btrfs_tree_lock(eb);
5270 btrfs_set_lock_blocking(eb);
5271 path->locks[level] = 1;
5273 clean_tree_block(trans, root, eb);
5276 if (eb == root->node) {
5277 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5280 BUG_ON(root->root_key.objectid !=
5281 btrfs_header_owner(eb));
5283 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5284 parent = path->nodes[level + 1]->start;
5286 BUG_ON(root->root_key.objectid !=
5287 btrfs_header_owner(path->nodes[level + 1]));
5290 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5291 root->root_key.objectid, level, 0);
5294 wc->refs[level] = 0;
5295 wc->flags[level] = 0;
5299 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5300 struct btrfs_root *root,
5301 struct btrfs_path *path,
5302 struct walk_control *wc)
5304 int level = wc->level;
5305 int lookup_info = 1;
5308 while (level >= 0) {
5309 if (path->slots[level] >=
5310 btrfs_header_nritems(path->nodes[level]))
5313 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5320 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5322 path->slots[level]++;
5330 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5331 struct btrfs_root *root,
5332 struct btrfs_path *path,
5333 struct walk_control *wc, int max_level)
5335 int level = wc->level;
5338 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5339 while (level < max_level && path->nodes[level]) {
5341 if (path->slots[level] + 1 <
5342 btrfs_header_nritems(path->nodes[level])) {
5343 path->slots[level]++;
5346 ret = walk_up_proc(trans, root, path, wc);
5350 if (path->locks[level]) {
5351 btrfs_tree_unlock(path->nodes[level]);
5352 path->locks[level] = 0;
5354 free_extent_buffer(path->nodes[level]);
5355 path->nodes[level] = NULL;
5363 * drop a subvolume tree.
5365 * this function traverses the tree freeing any blocks that only
5366 * referenced by the tree.
5368 * when a shared tree block is found. this function decreases its
5369 * reference count by one. if update_ref is true, this function
5370 * also make sure backrefs for the shared block and all lower level
5371 * blocks are properly updated.
5373 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5375 struct btrfs_path *path;
5376 struct btrfs_trans_handle *trans;
5377 struct btrfs_root *tree_root = root->fs_info->tree_root;
5378 struct btrfs_root_item *root_item = &root->root_item;
5379 struct walk_control *wc;
5380 struct btrfs_key key;
5385 path = btrfs_alloc_path();
5388 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5391 trans = btrfs_start_transaction(tree_root, 1);
5393 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5394 level = btrfs_header_level(root->node);
5395 path->nodes[level] = btrfs_lock_root_node(root);
5396 btrfs_set_lock_blocking(path->nodes[level]);
5397 path->slots[level] = 0;
5398 path->locks[level] = 1;
5399 memset(&wc->update_progress, 0,
5400 sizeof(wc->update_progress));
5402 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5403 memcpy(&wc->update_progress, &key,
5404 sizeof(wc->update_progress));
5406 level = root_item->drop_level;
5408 path->lowest_level = level;
5409 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5410 path->lowest_level = 0;
5418 * unlock our path, this is safe because only this
5419 * function is allowed to delete this snapshot
5421 btrfs_unlock_up_safe(path, 0);
5423 level = btrfs_header_level(root->node);
5425 btrfs_tree_lock(path->nodes[level]);
5426 btrfs_set_lock_blocking(path->nodes[level]);
5428 ret = btrfs_lookup_extent_info(trans, root,
5429 path->nodes[level]->start,
5430 path->nodes[level]->len,
5434 BUG_ON(wc->refs[level] == 0);
5436 if (level == root_item->drop_level)
5439 btrfs_tree_unlock(path->nodes[level]);
5440 WARN_ON(wc->refs[level] != 1);
5446 wc->shared_level = -1;
5447 wc->stage = DROP_REFERENCE;
5448 wc->update_ref = update_ref;
5450 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5453 ret = walk_down_tree(trans, root, path, wc);
5459 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5466 BUG_ON(wc->stage != DROP_REFERENCE);
5470 if (wc->stage == DROP_REFERENCE) {
5472 btrfs_node_key(path->nodes[level],
5473 &root_item->drop_progress,
5474 path->slots[level]);
5475 root_item->drop_level = level;
5478 BUG_ON(wc->level == 0);
5479 if (trans->transaction->in_commit ||
5480 trans->transaction->delayed_refs.flushing) {
5481 ret = btrfs_update_root(trans, tree_root,
5486 btrfs_end_transaction(trans, tree_root);
5487 trans = btrfs_start_transaction(tree_root, 1);
5489 unsigned long update;
5490 update = trans->delayed_ref_updates;
5491 trans->delayed_ref_updates = 0;
5493 btrfs_run_delayed_refs(trans, tree_root,
5497 btrfs_release_path(root, path);
5500 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5503 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5504 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5508 ret = btrfs_del_orphan_item(trans, tree_root,
5509 root->root_key.objectid);
5514 if (root->in_radix) {
5515 btrfs_free_fs_root(tree_root->fs_info, root);
5517 free_extent_buffer(root->node);
5518 free_extent_buffer(root->commit_root);
5522 btrfs_end_transaction(trans, tree_root);
5524 btrfs_free_path(path);
5529 * drop subtree rooted at tree block 'node'.
5531 * NOTE: this function will unlock and release tree block 'node'
5533 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5534 struct btrfs_root *root,
5535 struct extent_buffer *node,
5536 struct extent_buffer *parent)
5538 struct btrfs_path *path;
5539 struct walk_control *wc;
5545 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5547 path = btrfs_alloc_path();
5550 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5553 btrfs_assert_tree_locked(parent);
5554 parent_level = btrfs_header_level(parent);
5555 extent_buffer_get(parent);
5556 path->nodes[parent_level] = parent;
5557 path->slots[parent_level] = btrfs_header_nritems(parent);
5559 btrfs_assert_tree_locked(node);
5560 level = btrfs_header_level(node);
5561 path->nodes[level] = node;
5562 path->slots[level] = 0;
5563 path->locks[level] = 1;
5565 wc->refs[parent_level] = 1;
5566 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5568 wc->shared_level = -1;
5569 wc->stage = DROP_REFERENCE;
5572 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5575 wret = walk_down_tree(trans, root, path, wc);
5581 wret = walk_up_tree(trans, root, path, wc, parent_level);
5589 btrfs_free_path(path);
5594 static unsigned long calc_ra(unsigned long start, unsigned long last,
5597 return min(last, start + nr - 1);
5600 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5605 unsigned long first_index;
5606 unsigned long last_index;
5609 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5610 struct file_ra_state *ra;
5611 struct btrfs_ordered_extent *ordered;
5612 unsigned int total_read = 0;
5613 unsigned int total_dirty = 0;
5616 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5618 mutex_lock(&inode->i_mutex);
5619 first_index = start >> PAGE_CACHE_SHIFT;
5620 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5622 /* make sure the dirty trick played by the caller work */
5623 ret = invalidate_inode_pages2_range(inode->i_mapping,
5624 first_index, last_index);
5628 file_ra_state_init(ra, inode->i_mapping);
5630 for (i = first_index ; i <= last_index; i++) {
5631 if (total_read % ra->ra_pages == 0) {
5632 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5633 calc_ra(i, last_index, ra->ra_pages));
5637 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5639 page = grab_cache_page(inode->i_mapping, i);
5644 if (!PageUptodate(page)) {
5645 btrfs_readpage(NULL, page);
5647 if (!PageUptodate(page)) {
5649 page_cache_release(page);
5654 wait_on_page_writeback(page);
5656 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5657 page_end = page_start + PAGE_CACHE_SIZE - 1;
5658 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5660 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5662 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5664 page_cache_release(page);
5665 btrfs_start_ordered_extent(inode, ordered, 1);
5666 btrfs_put_ordered_extent(ordered);
5669 set_page_extent_mapped(page);
5671 if (i == first_index)
5672 set_extent_bits(io_tree, page_start, page_end,
5673 EXTENT_BOUNDARY, GFP_NOFS);
5674 btrfs_set_extent_delalloc(inode, page_start, page_end);
5676 set_page_dirty(page);
5679 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5681 page_cache_release(page);
5686 mutex_unlock(&inode->i_mutex);
5687 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5691 static noinline int relocate_data_extent(struct inode *reloc_inode,
5692 struct btrfs_key *extent_key,
5695 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5696 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5697 struct extent_map *em;
5698 u64 start = extent_key->objectid - offset;
5699 u64 end = start + extent_key->offset - 1;
5701 em = alloc_extent_map(GFP_NOFS);
5702 BUG_ON(!em || IS_ERR(em));
5705 em->len = extent_key->offset;
5706 em->block_len = extent_key->offset;
5707 em->block_start = extent_key->objectid;
5708 em->bdev = root->fs_info->fs_devices->latest_bdev;
5709 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5711 /* setup extent map to cheat btrfs_readpage */
5712 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5715 write_lock(&em_tree->lock);
5716 ret = add_extent_mapping(em_tree, em);
5717 write_unlock(&em_tree->lock);
5718 if (ret != -EEXIST) {
5719 free_extent_map(em);
5722 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5724 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5726 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5729 struct btrfs_ref_path {
5731 u64 nodes[BTRFS_MAX_LEVEL];
5733 u64 root_generation;
5740 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5741 u64 new_nodes[BTRFS_MAX_LEVEL];
5744 struct disk_extent {
5755 static int is_cowonly_root(u64 root_objectid)
5757 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5758 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5759 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5760 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5761 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5762 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5767 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5768 struct btrfs_root *extent_root,
5769 struct btrfs_ref_path *ref_path,
5772 struct extent_buffer *leaf;
5773 struct btrfs_path *path;
5774 struct btrfs_extent_ref *ref;
5775 struct btrfs_key key;
5776 struct btrfs_key found_key;
5782 path = btrfs_alloc_path();
5787 ref_path->lowest_level = -1;
5788 ref_path->current_level = -1;
5789 ref_path->shared_level = -1;
5793 level = ref_path->current_level - 1;
5794 while (level >= -1) {
5796 if (level < ref_path->lowest_level)
5800 bytenr = ref_path->nodes[level];
5802 bytenr = ref_path->extent_start;
5803 BUG_ON(bytenr == 0);
5805 parent = ref_path->nodes[level + 1];
5806 ref_path->nodes[level + 1] = 0;
5807 ref_path->current_level = level;
5808 BUG_ON(parent == 0);
5810 key.objectid = bytenr;
5811 key.offset = parent + 1;
5812 key.type = BTRFS_EXTENT_REF_KEY;
5814 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5819 leaf = path->nodes[0];
5820 nritems = btrfs_header_nritems(leaf);
5821 if (path->slots[0] >= nritems) {
5822 ret = btrfs_next_leaf(extent_root, path);
5827 leaf = path->nodes[0];
5830 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5831 if (found_key.objectid == bytenr &&
5832 found_key.type == BTRFS_EXTENT_REF_KEY) {
5833 if (level < ref_path->shared_level)
5834 ref_path->shared_level = level;
5839 btrfs_release_path(extent_root, path);
5842 /* reached lowest level */
5846 level = ref_path->current_level;
5847 while (level < BTRFS_MAX_LEVEL - 1) {
5851 bytenr = ref_path->nodes[level];
5853 bytenr = ref_path->extent_start;
5855 BUG_ON(bytenr == 0);
5857 key.objectid = bytenr;
5859 key.type = BTRFS_EXTENT_REF_KEY;
5861 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5865 leaf = path->nodes[0];
5866 nritems = btrfs_header_nritems(leaf);
5867 if (path->slots[0] >= nritems) {
5868 ret = btrfs_next_leaf(extent_root, path);
5872 /* the extent was freed by someone */
5873 if (ref_path->lowest_level == level)
5875 btrfs_release_path(extent_root, path);
5878 leaf = path->nodes[0];
5881 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5882 if (found_key.objectid != bytenr ||
5883 found_key.type != BTRFS_EXTENT_REF_KEY) {
5884 /* the extent was freed by someone */
5885 if (ref_path->lowest_level == level) {
5889 btrfs_release_path(extent_root, path);
5893 ref = btrfs_item_ptr(leaf, path->slots[0],
5894 struct btrfs_extent_ref);
5895 ref_objectid = btrfs_ref_objectid(leaf, ref);
5896 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5898 level = (int)ref_objectid;
5899 BUG_ON(level >= BTRFS_MAX_LEVEL);
5900 ref_path->lowest_level = level;
5901 ref_path->current_level = level;
5902 ref_path->nodes[level] = bytenr;
5904 WARN_ON(ref_objectid != level);
5907 WARN_ON(level != -1);
5911 if (ref_path->lowest_level == level) {
5912 ref_path->owner_objectid = ref_objectid;
5913 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5917 * the block is tree root or the block isn't in reference
5920 if (found_key.objectid == found_key.offset ||
5921 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5922 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5923 ref_path->root_generation =
5924 btrfs_ref_generation(leaf, ref);
5926 /* special reference from the tree log */
5927 ref_path->nodes[0] = found_key.offset;
5928 ref_path->current_level = 0;
5935 BUG_ON(ref_path->nodes[level] != 0);
5936 ref_path->nodes[level] = found_key.offset;
5937 ref_path->current_level = level;
5940 * the reference was created in the running transaction,
5941 * no need to continue walking up.
5943 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5944 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5945 ref_path->root_generation =
5946 btrfs_ref_generation(leaf, ref);
5951 btrfs_release_path(extent_root, path);
5954 /* reached max tree level, but no tree root found. */
5957 btrfs_free_path(path);
5961 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5962 struct btrfs_root *extent_root,
5963 struct btrfs_ref_path *ref_path,
5966 memset(ref_path, 0, sizeof(*ref_path));
5967 ref_path->extent_start = extent_start;
5969 return __next_ref_path(trans, extent_root, ref_path, 1);
5972 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5973 struct btrfs_root *extent_root,
5974 struct btrfs_ref_path *ref_path)
5976 return __next_ref_path(trans, extent_root, ref_path, 0);
5979 static noinline int get_new_locations(struct inode *reloc_inode,
5980 struct btrfs_key *extent_key,
5981 u64 offset, int no_fragment,
5982 struct disk_extent **extents,
5985 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5986 struct btrfs_path *path;
5987 struct btrfs_file_extent_item *fi;
5988 struct extent_buffer *leaf;
5989 struct disk_extent *exts = *extents;
5990 struct btrfs_key found_key;
5995 int max = *nr_extents;
5998 WARN_ON(!no_fragment && *extents);
6001 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6006 path = btrfs_alloc_path();
6009 cur_pos = extent_key->objectid - offset;
6010 last_byte = extent_key->objectid + extent_key->offset;
6011 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6021 leaf = path->nodes[0];
6022 nritems = btrfs_header_nritems(leaf);
6023 if (path->slots[0] >= nritems) {
6024 ret = btrfs_next_leaf(root, path);
6029 leaf = path->nodes[0];
6032 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6033 if (found_key.offset != cur_pos ||
6034 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6035 found_key.objectid != reloc_inode->i_ino)
6038 fi = btrfs_item_ptr(leaf, path->slots[0],
6039 struct btrfs_file_extent_item);
6040 if (btrfs_file_extent_type(leaf, fi) !=
6041 BTRFS_FILE_EXTENT_REG ||
6042 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6046 struct disk_extent *old = exts;
6048 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6049 memcpy(exts, old, sizeof(*exts) * nr);
6050 if (old != *extents)
6054 exts[nr].disk_bytenr =
6055 btrfs_file_extent_disk_bytenr(leaf, fi);
6056 exts[nr].disk_num_bytes =
6057 btrfs_file_extent_disk_num_bytes(leaf, fi);
6058 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6059 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6060 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6061 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6062 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6063 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6065 BUG_ON(exts[nr].offset > 0);
6066 BUG_ON(exts[nr].compression || exts[nr].encryption);
6067 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6069 cur_pos += exts[nr].num_bytes;
6072 if (cur_pos + offset >= last_byte)
6082 BUG_ON(cur_pos + offset > last_byte);
6083 if (cur_pos + offset < last_byte) {
6089 btrfs_free_path(path);
6091 if (exts != *extents)
6100 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6101 struct btrfs_root *root,
6102 struct btrfs_path *path,
6103 struct btrfs_key *extent_key,
6104 struct btrfs_key *leaf_key,
6105 struct btrfs_ref_path *ref_path,
6106 struct disk_extent *new_extents,
6109 struct extent_buffer *leaf;
6110 struct btrfs_file_extent_item *fi;
6111 struct inode *inode = NULL;
6112 struct btrfs_key key;
6117 u64 search_end = (u64)-1;
6120 int extent_locked = 0;
6124 memcpy(&key, leaf_key, sizeof(key));
6125 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6126 if (key.objectid < ref_path->owner_objectid ||
6127 (key.objectid == ref_path->owner_objectid &&
6128 key.type < BTRFS_EXTENT_DATA_KEY)) {
6129 key.objectid = ref_path->owner_objectid;
6130 key.type = BTRFS_EXTENT_DATA_KEY;
6136 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6140 leaf = path->nodes[0];
6141 nritems = btrfs_header_nritems(leaf);
6143 if (extent_locked && ret > 0) {
6145 * the file extent item was modified by someone
6146 * before the extent got locked.
6148 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6149 lock_end, GFP_NOFS);
6153 if (path->slots[0] >= nritems) {
6154 if (++nr_scaned > 2)
6157 BUG_ON(extent_locked);
6158 ret = btrfs_next_leaf(root, path);
6163 leaf = path->nodes[0];
6164 nritems = btrfs_header_nritems(leaf);
6167 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6169 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6170 if ((key.objectid > ref_path->owner_objectid) ||
6171 (key.objectid == ref_path->owner_objectid &&
6172 key.type > BTRFS_EXTENT_DATA_KEY) ||
6173 key.offset >= search_end)
6177 if (inode && key.objectid != inode->i_ino) {
6178 BUG_ON(extent_locked);
6179 btrfs_release_path(root, path);
6180 mutex_unlock(&inode->i_mutex);
6186 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6191 fi = btrfs_item_ptr(leaf, path->slots[0],
6192 struct btrfs_file_extent_item);
6193 extent_type = btrfs_file_extent_type(leaf, fi);
6194 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6195 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6196 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6197 extent_key->objectid)) {
6203 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6204 ext_offset = btrfs_file_extent_offset(leaf, fi);
6206 if (search_end == (u64)-1) {
6207 search_end = key.offset - ext_offset +
6208 btrfs_file_extent_ram_bytes(leaf, fi);
6211 if (!extent_locked) {
6212 lock_start = key.offset;
6213 lock_end = lock_start + num_bytes - 1;
6215 if (lock_start > key.offset ||
6216 lock_end + 1 < key.offset + num_bytes) {
6217 unlock_extent(&BTRFS_I(inode)->io_tree,
6218 lock_start, lock_end, GFP_NOFS);
6224 btrfs_release_path(root, path);
6226 inode = btrfs_iget_locked(root->fs_info->sb,
6227 key.objectid, root);
6228 if (inode->i_state & I_NEW) {
6229 BTRFS_I(inode)->root = root;
6230 BTRFS_I(inode)->location.objectid =
6232 BTRFS_I(inode)->location.type =
6233 BTRFS_INODE_ITEM_KEY;
6234 BTRFS_I(inode)->location.offset = 0;
6235 btrfs_read_locked_inode(inode);
6236 unlock_new_inode(inode);
6239 * some code call btrfs_commit_transaction while
6240 * holding the i_mutex, so we can't use mutex_lock
6243 if (is_bad_inode(inode) ||
6244 !mutex_trylock(&inode->i_mutex)) {
6247 key.offset = (u64)-1;
6252 if (!extent_locked) {
6253 struct btrfs_ordered_extent *ordered;
6255 btrfs_release_path(root, path);
6257 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6258 lock_end, GFP_NOFS);
6259 ordered = btrfs_lookup_first_ordered_extent(inode,
6262 ordered->file_offset <= lock_end &&
6263 ordered->file_offset + ordered->len > lock_start) {
6264 unlock_extent(&BTRFS_I(inode)->io_tree,
6265 lock_start, lock_end, GFP_NOFS);
6266 btrfs_start_ordered_extent(inode, ordered, 1);
6267 btrfs_put_ordered_extent(ordered);
6268 key.offset += num_bytes;
6272 btrfs_put_ordered_extent(ordered);
6278 if (nr_extents == 1) {
6279 /* update extent pointer in place */
6280 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6281 new_extents[0].disk_bytenr);
6282 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6283 new_extents[0].disk_num_bytes);
6284 btrfs_mark_buffer_dirty(leaf);
6286 btrfs_drop_extent_cache(inode, key.offset,
6287 key.offset + num_bytes - 1, 0);
6289 ret = btrfs_inc_extent_ref(trans, root,
6290 new_extents[0].disk_bytenr,
6291 new_extents[0].disk_num_bytes,
6293 root->root_key.objectid,
6298 ret = btrfs_free_extent(trans, root,
6299 extent_key->objectid,
6302 btrfs_header_owner(leaf),
6303 btrfs_header_generation(leaf),
6307 btrfs_release_path(root, path);
6308 key.offset += num_bytes;
6316 * drop old extent pointer at first, then insert the
6317 * new pointers one bye one
6319 btrfs_release_path(root, path);
6320 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6321 key.offset + num_bytes,
6322 key.offset, &alloc_hint);
6325 for (i = 0; i < nr_extents; i++) {
6326 if (ext_offset >= new_extents[i].num_bytes) {
6327 ext_offset -= new_extents[i].num_bytes;
6330 extent_len = min(new_extents[i].num_bytes -
6331 ext_offset, num_bytes);
6333 ret = btrfs_insert_empty_item(trans, root,
6338 leaf = path->nodes[0];
6339 fi = btrfs_item_ptr(leaf, path->slots[0],
6340 struct btrfs_file_extent_item);
6341 btrfs_set_file_extent_generation(leaf, fi,
6343 btrfs_set_file_extent_type(leaf, fi,
6344 BTRFS_FILE_EXTENT_REG);
6345 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6346 new_extents[i].disk_bytenr);
6347 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6348 new_extents[i].disk_num_bytes);
6349 btrfs_set_file_extent_ram_bytes(leaf, fi,
6350 new_extents[i].ram_bytes);
6352 btrfs_set_file_extent_compression(leaf, fi,
6353 new_extents[i].compression);
6354 btrfs_set_file_extent_encryption(leaf, fi,
6355 new_extents[i].encryption);
6356 btrfs_set_file_extent_other_encoding(leaf, fi,
6357 new_extents[i].other_encoding);
6359 btrfs_set_file_extent_num_bytes(leaf, fi,
6361 ext_offset += new_extents[i].offset;
6362 btrfs_set_file_extent_offset(leaf, fi,
6364 btrfs_mark_buffer_dirty(leaf);
6366 btrfs_drop_extent_cache(inode, key.offset,
6367 key.offset + extent_len - 1, 0);
6369 ret = btrfs_inc_extent_ref(trans, root,
6370 new_extents[i].disk_bytenr,
6371 new_extents[i].disk_num_bytes,
6373 root->root_key.objectid,
6374 trans->transid, key.objectid);
6376 btrfs_release_path(root, path);
6378 inode_add_bytes(inode, extent_len);
6381 num_bytes -= extent_len;
6382 key.offset += extent_len;
6387 BUG_ON(i >= nr_extents);
6391 if (extent_locked) {
6392 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6393 lock_end, GFP_NOFS);
6397 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6398 key.offset >= search_end)
6405 btrfs_release_path(root, path);
6407 mutex_unlock(&inode->i_mutex);
6408 if (extent_locked) {
6409 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6410 lock_end, GFP_NOFS);
6417 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6418 struct btrfs_root *root,
6419 struct extent_buffer *buf, u64 orig_start)
6424 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6425 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6427 level = btrfs_header_level(buf);
6429 struct btrfs_leaf_ref *ref;
6430 struct btrfs_leaf_ref *orig_ref;
6432 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6436 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6438 btrfs_free_leaf_ref(root, orig_ref);
6442 ref->nritems = orig_ref->nritems;
6443 memcpy(ref->extents, orig_ref->extents,
6444 sizeof(ref->extents[0]) * ref->nritems);
6446 btrfs_free_leaf_ref(root, orig_ref);
6448 ref->root_gen = trans->transid;
6449 ref->bytenr = buf->start;
6450 ref->owner = btrfs_header_owner(buf);
6451 ref->generation = btrfs_header_generation(buf);
6453 ret = btrfs_add_leaf_ref(root, ref, 0);
6455 btrfs_free_leaf_ref(root, ref);
6460 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6461 struct extent_buffer *leaf,
6462 struct btrfs_block_group_cache *group,
6463 struct btrfs_root *target_root)
6465 struct btrfs_key key;
6466 struct inode *inode = NULL;
6467 struct btrfs_file_extent_item *fi;
6469 u64 skip_objectid = 0;
6473 nritems = btrfs_header_nritems(leaf);
6474 for (i = 0; i < nritems; i++) {
6475 btrfs_item_key_to_cpu(leaf, &key, i);
6476 if (key.objectid == skip_objectid ||
6477 key.type != BTRFS_EXTENT_DATA_KEY)
6479 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6480 if (btrfs_file_extent_type(leaf, fi) ==
6481 BTRFS_FILE_EXTENT_INLINE)
6483 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6485 if (!inode || inode->i_ino != key.objectid) {
6487 inode = btrfs_ilookup(target_root->fs_info->sb,
6488 key.objectid, target_root, 1);
6491 skip_objectid = key.objectid;
6494 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6496 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6497 key.offset + num_bytes - 1, GFP_NOFS);
6498 btrfs_drop_extent_cache(inode, key.offset,
6499 key.offset + num_bytes - 1, 1);
6500 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6501 key.offset + num_bytes - 1, GFP_NOFS);
6508 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6509 struct btrfs_root *root,
6510 struct extent_buffer *leaf,
6511 struct btrfs_block_group_cache *group,
6512 struct inode *reloc_inode)
6514 struct btrfs_key key;
6515 struct btrfs_key extent_key;
6516 struct btrfs_file_extent_item *fi;
6517 struct btrfs_leaf_ref *ref;
6518 struct disk_extent *new_extent;
6527 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6528 BUG_ON(!new_extent);
6530 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6534 nritems = btrfs_header_nritems(leaf);
6535 for (i = 0; i < nritems; i++) {
6536 btrfs_item_key_to_cpu(leaf, &key, i);
6537 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6539 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6540 if (btrfs_file_extent_type(leaf, fi) ==
6541 BTRFS_FILE_EXTENT_INLINE)
6543 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6544 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6549 if (bytenr >= group->key.objectid + group->key.offset ||
6550 bytenr + num_bytes <= group->key.objectid)
6553 extent_key.objectid = bytenr;
6554 extent_key.offset = num_bytes;
6555 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6557 ret = get_new_locations(reloc_inode, &extent_key,
6558 group->key.objectid, 1,
6559 &new_extent, &nr_extent);
6564 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6565 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6566 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6567 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6569 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6570 new_extent->disk_bytenr);
6571 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6572 new_extent->disk_num_bytes);
6573 btrfs_mark_buffer_dirty(leaf);
6575 ret = btrfs_inc_extent_ref(trans, root,
6576 new_extent->disk_bytenr,
6577 new_extent->disk_num_bytes,
6579 root->root_key.objectid,
6580 trans->transid, key.objectid);
6583 ret = btrfs_free_extent(trans, root,
6584 bytenr, num_bytes, leaf->start,
6585 btrfs_header_owner(leaf),
6586 btrfs_header_generation(leaf),
6592 BUG_ON(ext_index + 1 != ref->nritems);
6593 btrfs_free_leaf_ref(root, ref);
6597 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6598 struct btrfs_root *root)
6600 struct btrfs_root *reloc_root;
6603 if (root->reloc_root) {
6604 reloc_root = root->reloc_root;
6605 root->reloc_root = NULL;
6606 list_add(&reloc_root->dead_list,
6607 &root->fs_info->dead_reloc_roots);
6609 btrfs_set_root_bytenr(&reloc_root->root_item,
6610 reloc_root->node->start);
6611 btrfs_set_root_level(&root->root_item,
6612 btrfs_header_level(reloc_root->node));
6613 memset(&reloc_root->root_item.drop_progress, 0,
6614 sizeof(struct btrfs_disk_key));
6615 reloc_root->root_item.drop_level = 0;
6617 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6618 &reloc_root->root_key,
6619 &reloc_root->root_item);
6625 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6627 struct btrfs_trans_handle *trans;
6628 struct btrfs_root *reloc_root;
6629 struct btrfs_root *prev_root = NULL;
6630 struct list_head dead_roots;
6634 INIT_LIST_HEAD(&dead_roots);
6635 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6637 while (!list_empty(&dead_roots)) {
6638 reloc_root = list_entry(dead_roots.prev,
6639 struct btrfs_root, dead_list);
6640 list_del_init(&reloc_root->dead_list);
6642 BUG_ON(reloc_root->commit_root != NULL);
6644 trans = btrfs_join_transaction(root, 1);
6647 mutex_lock(&root->fs_info->drop_mutex);
6648 ret = btrfs_drop_snapshot(trans, reloc_root);
6651 mutex_unlock(&root->fs_info->drop_mutex);
6653 nr = trans->blocks_used;
6654 ret = btrfs_end_transaction(trans, root);
6656 btrfs_btree_balance_dirty(root, nr);
6659 free_extent_buffer(reloc_root->node);
6661 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6662 &reloc_root->root_key);
6664 mutex_unlock(&root->fs_info->drop_mutex);
6666 nr = trans->blocks_used;
6667 ret = btrfs_end_transaction(trans, root);
6669 btrfs_btree_balance_dirty(root, nr);
6672 prev_root = reloc_root;
6675 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6681 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6683 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6687 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6689 struct btrfs_root *reloc_root;
6690 struct btrfs_trans_handle *trans;
6691 struct btrfs_key location;
6695 mutex_lock(&root->fs_info->tree_reloc_mutex);
6696 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6698 found = !list_empty(&root->fs_info->dead_reloc_roots);
6699 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6702 trans = btrfs_start_transaction(root, 1);
6704 ret = btrfs_commit_transaction(trans, root);
6708 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6709 location.offset = (u64)-1;
6710 location.type = BTRFS_ROOT_ITEM_KEY;
6712 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6713 BUG_ON(!reloc_root);
6714 btrfs_orphan_cleanup(reloc_root);
6718 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6719 struct btrfs_root *root)
6721 struct btrfs_root *reloc_root;
6722 struct extent_buffer *eb;
6723 struct btrfs_root_item *root_item;
6724 struct btrfs_key root_key;
6727 BUG_ON(!root->ref_cows);
6728 if (root->reloc_root)
6731 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6734 ret = btrfs_copy_root(trans, root, root->commit_root,
6735 &eb, BTRFS_TREE_RELOC_OBJECTID);
6738 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6739 root_key.offset = root->root_key.objectid;
6740 root_key.type = BTRFS_ROOT_ITEM_KEY;
6742 memcpy(root_item, &root->root_item, sizeof(root_item));
6743 btrfs_set_root_refs(root_item, 0);
6744 btrfs_set_root_bytenr(root_item, eb->start);
6745 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6746 btrfs_set_root_generation(root_item, trans->transid);
6748 btrfs_tree_unlock(eb);
6749 free_extent_buffer(eb);
6751 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6752 &root_key, root_item);
6756 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6758 BUG_ON(!reloc_root);
6759 reloc_root->last_trans = trans->transid;
6760 reloc_root->commit_root = NULL;
6761 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6763 root->reloc_root = reloc_root;
6768 * Core function of space balance.
6770 * The idea is using reloc trees to relocate tree blocks in reference
6771 * counted roots. There is one reloc tree for each subvol, and all
6772 * reloc trees share same root key objectid. Reloc trees are snapshots
6773 * of the latest committed roots of subvols (root->commit_root).
6775 * To relocate a tree block referenced by a subvol, there are two steps.
6776 * COW the block through subvol's reloc tree, then update block pointer
6777 * in the subvol to point to the new block. Since all reloc trees share
6778 * same root key objectid, doing special handing for tree blocks owned
6779 * by them is easy. Once a tree block has been COWed in one reloc tree,
6780 * we can use the resulting new block directly when the same block is
6781 * required to COW again through other reloc trees. By this way, relocated
6782 * tree blocks are shared between reloc trees, so they are also shared
6785 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6786 struct btrfs_root *root,
6787 struct btrfs_path *path,
6788 struct btrfs_key *first_key,
6789 struct btrfs_ref_path *ref_path,
6790 struct btrfs_block_group_cache *group,
6791 struct inode *reloc_inode)
6793 struct btrfs_root *reloc_root;
6794 struct extent_buffer *eb = NULL;
6795 struct btrfs_key *keys;
6799 int lowest_level = 0;
6802 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6803 lowest_level = ref_path->owner_objectid;
6805 if (!root->ref_cows) {
6806 path->lowest_level = lowest_level;
6807 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6809 path->lowest_level = 0;
6810 btrfs_release_path(root, path);
6814 mutex_lock(&root->fs_info->tree_reloc_mutex);
6815 ret = init_reloc_tree(trans, root);
6817 reloc_root = root->reloc_root;
6819 shared_level = ref_path->shared_level;
6820 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6822 keys = ref_path->node_keys;
6823 nodes = ref_path->new_nodes;
6824 memset(&keys[shared_level + 1], 0,
6825 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6826 memset(&nodes[shared_level + 1], 0,
6827 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6829 if (nodes[lowest_level] == 0) {
6830 path->lowest_level = lowest_level;
6831 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6834 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6835 eb = path->nodes[level];
6836 if (!eb || eb == reloc_root->node)
6838 nodes[level] = eb->start;
6840 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6842 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6845 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6846 eb = path->nodes[0];
6847 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6848 group, reloc_inode);
6851 btrfs_release_path(reloc_root, path);
6853 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6859 * replace tree blocks in the fs tree with tree blocks in
6862 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6865 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6866 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6869 extent_buffer_get(path->nodes[0]);
6870 eb = path->nodes[0];
6871 btrfs_release_path(reloc_root, path);
6872 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6874 free_extent_buffer(eb);
6877 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6878 path->lowest_level = 0;
6882 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6883 struct btrfs_root *root,
6884 struct btrfs_path *path,
6885 struct btrfs_key *first_key,
6886 struct btrfs_ref_path *ref_path)
6890 ret = relocate_one_path(trans, root, path, first_key,
6891 ref_path, NULL, NULL);
6897 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6898 struct btrfs_root *extent_root,
6899 struct btrfs_path *path,
6900 struct btrfs_key *extent_key)
6904 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6907 ret = btrfs_del_item(trans, extent_root, path);
6909 btrfs_release_path(extent_root, path);
6913 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6914 struct btrfs_ref_path *ref_path)
6916 struct btrfs_key root_key;
6918 root_key.objectid = ref_path->root_objectid;
6919 root_key.type = BTRFS_ROOT_ITEM_KEY;
6920 if (is_cowonly_root(ref_path->root_objectid))
6921 root_key.offset = 0;
6923 root_key.offset = (u64)-1;
6925 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6928 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6929 struct btrfs_path *path,
6930 struct btrfs_key *extent_key,
6931 struct btrfs_block_group_cache *group,
6932 struct inode *reloc_inode, int pass)
6934 struct btrfs_trans_handle *trans;
6935 struct btrfs_root *found_root;
6936 struct btrfs_ref_path *ref_path = NULL;
6937 struct disk_extent *new_extents = NULL;
6942 struct btrfs_key first_key;
6946 trans = btrfs_start_transaction(extent_root, 1);
6949 if (extent_key->objectid == 0) {
6950 ret = del_extent_zero(trans, extent_root, path, extent_key);
6954 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6960 for (loops = 0; ; loops++) {
6962 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6963 extent_key->objectid);
6965 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6972 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6973 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6976 found_root = read_ref_root(extent_root->fs_info, ref_path);
6977 BUG_ON(!found_root);
6979 * for reference counted tree, only process reference paths
6980 * rooted at the latest committed root.
6982 if (found_root->ref_cows &&
6983 ref_path->root_generation != found_root->root_key.offset)
6986 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6989 * copy data extents to new locations
6991 u64 group_start = group->key.objectid;
6992 ret = relocate_data_extent(reloc_inode,
7001 level = ref_path->owner_objectid;
7004 if (prev_block != ref_path->nodes[level]) {
7005 struct extent_buffer *eb;
7006 u64 block_start = ref_path->nodes[level];
7007 u64 block_size = btrfs_level_size(found_root, level);
7009 eb = read_tree_block(found_root, block_start,
7011 btrfs_tree_lock(eb);
7012 BUG_ON(level != btrfs_header_level(eb));
7015 btrfs_item_key_to_cpu(eb, &first_key, 0);
7017 btrfs_node_key_to_cpu(eb, &first_key, 0);
7019 btrfs_tree_unlock(eb);
7020 free_extent_buffer(eb);
7021 prev_block = block_start;
7024 mutex_lock(&extent_root->fs_info->trans_mutex);
7025 btrfs_record_root_in_trans(found_root);
7026 mutex_unlock(&extent_root->fs_info->trans_mutex);
7027 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7029 * try to update data extent references while
7030 * keeping metadata shared between snapshots.
7033 ret = relocate_one_path(trans, found_root,
7034 path, &first_key, ref_path,
7035 group, reloc_inode);
7041 * use fallback method to process the remaining
7045 u64 group_start = group->key.objectid;
7046 new_extents = kmalloc(sizeof(*new_extents),
7049 ret = get_new_locations(reloc_inode,
7057 ret = replace_one_extent(trans, found_root,
7059 &first_key, ref_path,
7060 new_extents, nr_extents);
7062 ret = relocate_tree_block(trans, found_root, path,
7063 &first_key, ref_path);
7070 btrfs_end_transaction(trans, extent_root);
7077 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7080 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7081 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7083 num_devices = root->fs_info->fs_devices->rw_devices;
7084 if (num_devices == 1) {
7085 stripped |= BTRFS_BLOCK_GROUP_DUP;
7086 stripped = flags & ~stripped;
7088 /* turn raid0 into single device chunks */
7089 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7092 /* turn mirroring into duplication */
7093 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7094 BTRFS_BLOCK_GROUP_RAID10))
7095 return stripped | BTRFS_BLOCK_GROUP_DUP;
7098 /* they already had raid on here, just return */
7099 if (flags & stripped)
7102 stripped |= BTRFS_BLOCK_GROUP_DUP;
7103 stripped = flags & ~stripped;
7105 /* switch duplicated blocks with raid1 */
7106 if (flags & BTRFS_BLOCK_GROUP_DUP)
7107 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7109 /* turn single device chunks into raid0 */
7110 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7115 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7116 struct btrfs_block_group_cache *shrink_block_group,
7119 struct btrfs_trans_handle *trans;
7120 u64 new_alloc_flags;
7123 spin_lock(&shrink_block_group->lock);
7124 if (btrfs_block_group_used(&shrink_block_group->item) +
7125 shrink_block_group->reserved > 0) {
7126 spin_unlock(&shrink_block_group->lock);
7128 trans = btrfs_start_transaction(root, 1);
7129 spin_lock(&shrink_block_group->lock);
7131 new_alloc_flags = update_block_group_flags(root,
7132 shrink_block_group->flags);
7133 if (new_alloc_flags != shrink_block_group->flags) {
7135 btrfs_block_group_used(&shrink_block_group->item);
7137 calc = shrink_block_group->key.offset;
7139 spin_unlock(&shrink_block_group->lock);
7141 do_chunk_alloc(trans, root->fs_info->extent_root,
7142 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7144 btrfs_end_transaction(trans, root);
7146 spin_unlock(&shrink_block_group->lock);
7151 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7152 struct btrfs_block_group_cache *group)
7155 __alloc_chunk_for_shrink(root, group, 1);
7156 set_block_group_readonly(group);
7161 * checks to see if its even possible to relocate this block group.
7163 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7164 * ok to go ahead and try.
7166 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7168 struct btrfs_block_group_cache *block_group;
7169 struct btrfs_space_info *space_info;
7170 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7171 struct btrfs_device *device;
7175 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7177 /* odd, couldn't find the block group, leave it alone */
7181 /* no bytes used, we're good */
7182 if (!btrfs_block_group_used(&block_group->item))
7185 space_info = block_group->space_info;
7186 spin_lock(&space_info->lock);
7188 full = space_info->full;
7191 * if this is the last block group we have in this space, we can't
7192 * relocate it unless we're able to allocate a new chunk below.
7194 * Otherwise, we need to make sure we have room in the space to handle
7195 * all of the extents from this block group. If we can, we're good
7197 if ((space_info->total_bytes != block_group->key.offset) &&
7198 (space_info->bytes_used + space_info->bytes_reserved +
7199 space_info->bytes_pinned + space_info->bytes_readonly +
7200 btrfs_block_group_used(&block_group->item) <
7201 space_info->total_bytes)) {
7202 spin_unlock(&space_info->lock);
7205 spin_unlock(&space_info->lock);
7208 * ok we don't have enough space, but maybe we have free space on our
7209 * devices to allocate new chunks for relocation, so loop through our
7210 * alloc devices and guess if we have enough space. However, if we
7211 * were marked as full, then we know there aren't enough chunks, and we
7218 mutex_lock(&root->fs_info->chunk_mutex);
7219 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7220 u64 min_free = btrfs_block_group_used(&block_group->item);
7221 u64 dev_offset, max_avail;
7224 * check to make sure we can actually find a chunk with enough
7225 * space to fit our block group in.
7227 if (device->total_bytes > device->bytes_used + min_free) {
7228 ret = find_free_dev_extent(NULL, device, min_free,
7229 &dev_offset, &max_avail);
7235 mutex_unlock(&root->fs_info->chunk_mutex);
7237 btrfs_put_block_group(block_group);
7241 static int find_first_block_group(struct btrfs_root *root,
7242 struct btrfs_path *path, struct btrfs_key *key)
7245 struct btrfs_key found_key;
7246 struct extent_buffer *leaf;
7249 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7254 slot = path->slots[0];
7255 leaf = path->nodes[0];
7256 if (slot >= btrfs_header_nritems(leaf)) {
7257 ret = btrfs_next_leaf(root, path);
7264 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7266 if (found_key.objectid >= key->objectid &&
7267 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7278 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7280 struct btrfs_block_group_cache *block_group;
7281 struct btrfs_space_info *space_info;
7282 struct btrfs_caching_control *caching_ctl;
7285 down_write(&info->extent_commit_sem);
7286 while (!list_empty(&info->caching_block_groups)) {
7287 caching_ctl = list_entry(info->caching_block_groups.next,
7288 struct btrfs_caching_control, list);
7289 list_del(&caching_ctl->list);
7290 put_caching_control(caching_ctl);
7292 up_write(&info->extent_commit_sem);
7294 spin_lock(&info->block_group_cache_lock);
7295 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7296 block_group = rb_entry(n, struct btrfs_block_group_cache,
7298 rb_erase(&block_group->cache_node,
7299 &info->block_group_cache_tree);
7300 spin_unlock(&info->block_group_cache_lock);
7302 down_write(&block_group->space_info->groups_sem);
7303 list_del(&block_group->list);
7304 up_write(&block_group->space_info->groups_sem);
7306 if (block_group->cached == BTRFS_CACHE_STARTED)
7307 wait_block_group_cache_done(block_group);
7309 btrfs_remove_free_space_cache(block_group);
7311 WARN_ON(atomic_read(&block_group->count) != 1);
7314 spin_lock(&info->block_group_cache_lock);
7316 spin_unlock(&info->block_group_cache_lock);
7318 /* now that all the block groups are freed, go through and
7319 * free all the space_info structs. This is only called during
7320 * the final stages of unmount, and so we know nobody is
7321 * using them. We call synchronize_rcu() once before we start,
7322 * just to be on the safe side.
7326 while(!list_empty(&info->space_info)) {
7327 space_info = list_entry(info->space_info.next,
7328 struct btrfs_space_info,
7331 list_del(&space_info->list);
7337 int btrfs_read_block_groups(struct btrfs_root *root)
7339 struct btrfs_path *path;
7341 struct btrfs_block_group_cache *cache;
7342 struct btrfs_fs_info *info = root->fs_info;
7343 struct btrfs_space_info *space_info;
7344 struct btrfs_key key;
7345 struct btrfs_key found_key;
7346 struct extent_buffer *leaf;
7348 root = info->extent_root;
7351 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7352 path = btrfs_alloc_path();
7357 ret = find_first_block_group(root, path, &key);
7365 leaf = path->nodes[0];
7366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7367 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7373 atomic_set(&cache->count, 1);
7374 spin_lock_init(&cache->lock);
7375 spin_lock_init(&cache->tree_lock);
7376 cache->fs_info = info;
7377 INIT_LIST_HEAD(&cache->list);
7378 INIT_LIST_HEAD(&cache->cluster_list);
7381 * we only want to have 32k of ram per block group for keeping
7382 * track of free space, and if we pass 1/2 of that we want to
7383 * start converting things over to using bitmaps
7385 cache->extents_thresh = ((1024 * 32) / 2) /
7386 sizeof(struct btrfs_free_space);
7388 read_extent_buffer(leaf, &cache->item,
7389 btrfs_item_ptr_offset(leaf, path->slots[0]),
7390 sizeof(cache->item));
7391 memcpy(&cache->key, &found_key, sizeof(found_key));
7393 key.objectid = found_key.objectid + found_key.offset;
7394 btrfs_release_path(root, path);
7395 cache->flags = btrfs_block_group_flags(&cache->item);
7396 cache->sectorsize = root->sectorsize;
7399 * check for two cases, either we are full, and therefore
7400 * don't need to bother with the caching work since we won't
7401 * find any space, or we are empty, and we can just add all
7402 * the space in and be done with it. This saves us _alot_ of
7403 * time, particularly in the full case.
7405 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7406 exclude_super_stripes(root, cache);
7407 cache->last_byte_to_unpin = (u64)-1;
7408 cache->cached = BTRFS_CACHE_FINISHED;
7409 free_excluded_extents(root, cache);
7410 } else if (btrfs_block_group_used(&cache->item) == 0) {
7411 exclude_super_stripes(root, cache);
7412 cache->last_byte_to_unpin = (u64)-1;
7413 cache->cached = BTRFS_CACHE_FINISHED;
7414 add_new_free_space(cache, root->fs_info,
7416 found_key.objectid +
7418 free_excluded_extents(root, cache);
7421 ret = update_space_info(info, cache->flags, found_key.offset,
7422 btrfs_block_group_used(&cache->item),
7425 cache->space_info = space_info;
7426 spin_lock(&cache->space_info->lock);
7427 cache->space_info->bytes_super += cache->bytes_super;
7428 spin_unlock(&cache->space_info->lock);
7430 down_write(&space_info->groups_sem);
7431 list_add_tail(&cache->list, &space_info->block_groups);
7432 up_write(&space_info->groups_sem);
7434 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7437 set_avail_alloc_bits(root->fs_info, cache->flags);
7438 if (btrfs_chunk_readonly(root, cache->key.objectid))
7439 set_block_group_readonly(cache);
7443 btrfs_free_path(path);
7447 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7448 struct btrfs_root *root, u64 bytes_used,
7449 u64 type, u64 chunk_objectid, u64 chunk_offset,
7453 struct btrfs_root *extent_root;
7454 struct btrfs_block_group_cache *cache;
7456 extent_root = root->fs_info->extent_root;
7458 root->fs_info->last_trans_log_full_commit = trans->transid;
7460 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7464 cache->key.objectid = chunk_offset;
7465 cache->key.offset = size;
7466 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7467 cache->sectorsize = root->sectorsize;
7470 * we only want to have 32k of ram per block group for keeping track
7471 * of free space, and if we pass 1/2 of that we want to start
7472 * converting things over to using bitmaps
7474 cache->extents_thresh = ((1024 * 32) / 2) /
7475 sizeof(struct btrfs_free_space);
7476 atomic_set(&cache->count, 1);
7477 spin_lock_init(&cache->lock);
7478 spin_lock_init(&cache->tree_lock);
7479 INIT_LIST_HEAD(&cache->list);
7480 INIT_LIST_HEAD(&cache->cluster_list);
7482 btrfs_set_block_group_used(&cache->item, bytes_used);
7483 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7484 cache->flags = type;
7485 btrfs_set_block_group_flags(&cache->item, type);
7487 cache->last_byte_to_unpin = (u64)-1;
7488 cache->cached = BTRFS_CACHE_FINISHED;
7489 exclude_super_stripes(root, cache);
7491 add_new_free_space(cache, root->fs_info, chunk_offset,
7492 chunk_offset + size);
7494 free_excluded_extents(root, cache);
7496 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7497 &cache->space_info);
7500 spin_lock(&cache->space_info->lock);
7501 cache->space_info->bytes_super += cache->bytes_super;
7502 spin_unlock(&cache->space_info->lock);
7504 down_write(&cache->space_info->groups_sem);
7505 list_add_tail(&cache->list, &cache->space_info->block_groups);
7506 up_write(&cache->space_info->groups_sem);
7508 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7511 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7512 sizeof(cache->item));
7515 set_avail_alloc_bits(extent_root->fs_info, type);
7520 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7521 struct btrfs_root *root, u64 group_start)
7523 struct btrfs_path *path;
7524 struct btrfs_block_group_cache *block_group;
7525 struct btrfs_free_cluster *cluster;
7526 struct btrfs_key key;
7529 root = root->fs_info->extent_root;
7531 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7532 BUG_ON(!block_group);
7533 BUG_ON(!block_group->ro);
7535 memcpy(&key, &block_group->key, sizeof(key));
7537 /* make sure this block group isn't part of an allocation cluster */
7538 cluster = &root->fs_info->data_alloc_cluster;
7539 spin_lock(&cluster->refill_lock);
7540 btrfs_return_cluster_to_free_space(block_group, cluster);
7541 spin_unlock(&cluster->refill_lock);
7544 * make sure this block group isn't part of a metadata
7545 * allocation cluster
7547 cluster = &root->fs_info->meta_alloc_cluster;
7548 spin_lock(&cluster->refill_lock);
7549 btrfs_return_cluster_to_free_space(block_group, cluster);
7550 spin_unlock(&cluster->refill_lock);
7552 path = btrfs_alloc_path();
7555 spin_lock(&root->fs_info->block_group_cache_lock);
7556 rb_erase(&block_group->cache_node,
7557 &root->fs_info->block_group_cache_tree);
7558 spin_unlock(&root->fs_info->block_group_cache_lock);
7560 down_write(&block_group->space_info->groups_sem);
7562 * we must use list_del_init so people can check to see if they
7563 * are still on the list after taking the semaphore
7565 list_del_init(&block_group->list);
7566 up_write(&block_group->space_info->groups_sem);
7568 if (block_group->cached == BTRFS_CACHE_STARTED)
7569 wait_block_group_cache_done(block_group);
7571 btrfs_remove_free_space_cache(block_group);
7573 spin_lock(&block_group->space_info->lock);
7574 block_group->space_info->total_bytes -= block_group->key.offset;
7575 block_group->space_info->bytes_readonly -= block_group->key.offset;
7576 spin_unlock(&block_group->space_info->lock);
7578 btrfs_clear_space_info_full(root->fs_info);
7580 btrfs_put_block_group(block_group);
7581 btrfs_put_block_group(block_group);
7583 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7589 ret = btrfs_del_item(trans, root, path);
7591 btrfs_free_path(path);