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>
25 #include <linux/slab.h>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 static int update_block_group(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 u64 bytenr, u64 num_bytes, int alloc);
39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve, int sinfo);
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 find_next_key(struct btrfs_path *path, int level,
64 struct btrfs_key *key);
65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
66 int dump_block_groups);
69 block_group_cache_done(struct btrfs_block_group_cache *cache)
72 return cache->cached == BTRFS_CACHE_FINISHED;
75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
77 return (cache->flags & bits) == bits;
80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
82 atomic_inc(&cache->count);
85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
87 if (atomic_dec_and_test(&cache->count)) {
88 WARN_ON(cache->pinned > 0);
89 WARN_ON(cache->reserved > 0);
90 WARN_ON(cache->reserved_pinned > 0);
96 * this adds the block group to the fs_info rb tree for the block group
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
100 struct btrfs_block_group_cache *block_group)
103 struct rb_node *parent = NULL;
104 struct btrfs_block_group_cache *cache;
106 spin_lock(&info->block_group_cache_lock);
107 p = &info->block_group_cache_tree.rb_node;
111 cache = rb_entry(parent, struct btrfs_block_group_cache,
113 if (block_group->key.objectid < cache->key.objectid) {
115 } else if (block_group->key.objectid > cache->key.objectid) {
118 spin_unlock(&info->block_group_cache_lock);
123 rb_link_node(&block_group->cache_node, parent, p);
124 rb_insert_color(&block_group->cache_node,
125 &info->block_group_cache_tree);
126 spin_unlock(&info->block_group_cache_lock);
132 * This will return the block group at or after bytenr if contains is 0, else
133 * it will return the block group that contains the bytenr
135 static struct btrfs_block_group_cache *
136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
139 struct btrfs_block_group_cache *cache, *ret = NULL;
143 spin_lock(&info->block_group_cache_lock);
144 n = info->block_group_cache_tree.rb_node;
147 cache = rb_entry(n, struct btrfs_block_group_cache,
149 end = cache->key.objectid + cache->key.offset - 1;
150 start = cache->key.objectid;
152 if (bytenr < start) {
153 if (!contains && (!ret || start < ret->key.objectid))
156 } else if (bytenr > start) {
157 if (contains && bytenr <= end) {
168 btrfs_get_block_group(ret);
169 spin_unlock(&info->block_group_cache_lock);
174 static int add_excluded_extent(struct btrfs_root *root,
175 u64 start, u64 num_bytes)
177 u64 end = start + num_bytes - 1;
178 set_extent_bits(&root->fs_info->freed_extents[0],
179 start, end, EXTENT_UPTODATE, GFP_NOFS);
180 set_extent_bits(&root->fs_info->freed_extents[1],
181 start, end, EXTENT_UPTODATE, GFP_NOFS);
185 static void free_excluded_extents(struct btrfs_root *root,
186 struct btrfs_block_group_cache *cache)
190 start = cache->key.objectid;
191 end = start + cache->key.offset - 1;
193 clear_extent_bits(&root->fs_info->freed_extents[0],
194 start, end, EXTENT_UPTODATE, GFP_NOFS);
195 clear_extent_bits(&root->fs_info->freed_extents[1],
196 start, end, EXTENT_UPTODATE, GFP_NOFS);
199 static int exclude_super_stripes(struct btrfs_root *root,
200 struct btrfs_block_group_cache *cache)
207 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
208 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
209 cache->bytes_super += stripe_len;
210 ret = add_excluded_extent(root, cache->key.objectid,
215 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
216 bytenr = btrfs_sb_offset(i);
217 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
218 cache->key.objectid, bytenr,
219 0, &logical, &nr, &stripe_len);
223 cache->bytes_super += stripe_len;
224 ret = add_excluded_extent(root, logical[nr],
234 static struct btrfs_caching_control *
235 get_caching_control(struct btrfs_block_group_cache *cache)
237 struct btrfs_caching_control *ctl;
239 spin_lock(&cache->lock);
240 if (cache->cached != BTRFS_CACHE_STARTED) {
241 spin_unlock(&cache->lock);
245 /* We're loading it the fast way, so we don't have a caching_ctl. */
246 if (!cache->caching_ctl) {
247 spin_unlock(&cache->lock);
251 ctl = cache->caching_ctl;
252 atomic_inc(&ctl->count);
253 spin_unlock(&cache->lock);
257 static void put_caching_control(struct btrfs_caching_control *ctl)
259 if (atomic_dec_and_test(&ctl->count))
264 * this is only called by cache_block_group, since we could have freed extents
265 * we need to check the pinned_extents for any extents that can't be used yet
266 * since their free space will be released as soon as the transaction commits.
268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
269 struct btrfs_fs_info *info, u64 start, u64 end)
271 u64 extent_start, extent_end, size, total_added = 0;
274 while (start < end) {
275 ret = find_first_extent_bit(info->pinned_extents, start,
276 &extent_start, &extent_end,
277 EXTENT_DIRTY | EXTENT_UPTODATE);
281 if (extent_start <= start) {
282 start = extent_end + 1;
283 } else if (extent_start > start && extent_start < end) {
284 size = extent_start - start;
286 ret = btrfs_add_free_space(block_group, start,
289 start = extent_end + 1;
298 ret = btrfs_add_free_space(block_group, start, size);
305 static int caching_kthread(void *data)
307 struct btrfs_block_group_cache *block_group = data;
308 struct btrfs_fs_info *fs_info = block_group->fs_info;
309 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
310 struct btrfs_root *extent_root = fs_info->extent_root;
311 struct btrfs_path *path;
312 struct extent_buffer *leaf;
313 struct btrfs_key key;
319 path = btrfs_alloc_path();
323 exclude_super_stripes(extent_root, block_group);
324 spin_lock(&block_group->space_info->lock);
325 block_group->space_info->bytes_readonly += block_group->bytes_super;
326 spin_unlock(&block_group->space_info->lock);
328 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
331 * We don't want to deadlock with somebody trying to allocate a new
332 * extent for the extent root while also trying to search the extent
333 * root to add free space. So we skip locking and search the commit
334 * root, since its read-only
336 path->skip_locking = 1;
337 path->search_commit_root = 1;
342 key.type = BTRFS_EXTENT_ITEM_KEY;
344 mutex_lock(&caching_ctl->mutex);
345 /* need to make sure the commit_root doesn't disappear */
346 down_read(&fs_info->extent_commit_sem);
348 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
352 leaf = path->nodes[0];
353 nritems = btrfs_header_nritems(leaf);
357 if (fs_info->closing > 1) {
362 if (path->slots[0] < nritems) {
363 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
365 ret = find_next_key(path, 0, &key);
369 caching_ctl->progress = last;
370 btrfs_release_path(extent_root, path);
371 up_read(&fs_info->extent_commit_sem);
372 mutex_unlock(&caching_ctl->mutex);
373 if (btrfs_transaction_in_commit(fs_info))
380 if (key.objectid < block_group->key.objectid) {
385 if (key.objectid >= block_group->key.objectid +
386 block_group->key.offset)
389 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
390 total_found += add_new_free_space(block_group,
393 last = key.objectid + key.offset;
395 if (total_found > (1024 * 1024 * 2)) {
397 wake_up(&caching_ctl->wait);
404 total_found += add_new_free_space(block_group, fs_info, last,
405 block_group->key.objectid +
406 block_group->key.offset);
407 caching_ctl->progress = (u64)-1;
409 spin_lock(&block_group->lock);
410 block_group->caching_ctl = NULL;
411 block_group->cached = BTRFS_CACHE_FINISHED;
412 spin_unlock(&block_group->lock);
415 btrfs_free_path(path);
416 up_read(&fs_info->extent_commit_sem);
418 free_excluded_extents(extent_root, block_group);
420 mutex_unlock(&caching_ctl->mutex);
421 wake_up(&caching_ctl->wait);
423 put_caching_control(caching_ctl);
424 atomic_dec(&block_group->space_info->caching_threads);
425 btrfs_put_block_group(block_group);
430 static int cache_block_group(struct btrfs_block_group_cache *cache,
431 struct btrfs_trans_handle *trans,
432 struct btrfs_root *root,
435 struct btrfs_fs_info *fs_info = cache->fs_info;
436 struct btrfs_caching_control *caching_ctl;
437 struct task_struct *tsk;
441 if (cache->cached != BTRFS_CACHE_NO)
445 * We can't do the read from on-disk cache during a commit since we need
446 * to have the normal tree locking. Also if we are currently trying to
447 * allocate blocks for the tree root we can't do the fast caching since
448 * we likely hold important locks.
450 if (!trans->transaction->in_commit &&
451 (root && root != root->fs_info->tree_root)) {
452 spin_lock(&cache->lock);
453 if (cache->cached != BTRFS_CACHE_NO) {
454 spin_unlock(&cache->lock);
457 cache->cached = BTRFS_CACHE_STARTED;
458 spin_unlock(&cache->lock);
460 ret = load_free_space_cache(fs_info, cache);
462 spin_lock(&cache->lock);
464 cache->cached = BTRFS_CACHE_FINISHED;
465 cache->last_byte_to_unpin = (u64)-1;
467 cache->cached = BTRFS_CACHE_NO;
469 spin_unlock(&cache->lock);
477 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
478 BUG_ON(!caching_ctl);
480 INIT_LIST_HEAD(&caching_ctl->list);
481 mutex_init(&caching_ctl->mutex);
482 init_waitqueue_head(&caching_ctl->wait);
483 caching_ctl->block_group = cache;
484 caching_ctl->progress = cache->key.objectid;
485 /* one for caching kthread, one for caching block group list */
486 atomic_set(&caching_ctl->count, 2);
488 spin_lock(&cache->lock);
489 if (cache->cached != BTRFS_CACHE_NO) {
490 spin_unlock(&cache->lock);
494 cache->caching_ctl = caching_ctl;
495 cache->cached = BTRFS_CACHE_STARTED;
496 spin_unlock(&cache->lock);
498 down_write(&fs_info->extent_commit_sem);
499 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
500 up_write(&fs_info->extent_commit_sem);
502 atomic_inc(&cache->space_info->caching_threads);
503 btrfs_get_block_group(cache);
505 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
506 cache->key.objectid);
509 printk(KERN_ERR "error running thread %d\n", ret);
517 * return the block group that starts at or after bytenr
519 static struct btrfs_block_group_cache *
520 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
522 struct btrfs_block_group_cache *cache;
524 cache = block_group_cache_tree_search(info, bytenr, 0);
530 * return the block group that contains the given bytenr
532 struct btrfs_block_group_cache *btrfs_lookup_block_group(
533 struct btrfs_fs_info *info,
536 struct btrfs_block_group_cache *cache;
538 cache = block_group_cache_tree_search(info, bytenr, 1);
543 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
546 struct list_head *head = &info->space_info;
547 struct btrfs_space_info *found;
549 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
550 BTRFS_BLOCK_GROUP_METADATA;
553 list_for_each_entry_rcu(found, head, list) {
554 if (found->flags & flags) {
564 * after adding space to the filesystem, we need to clear the full flags
565 * on all the space infos.
567 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
569 struct list_head *head = &info->space_info;
570 struct btrfs_space_info *found;
573 list_for_each_entry_rcu(found, head, list)
578 static u64 div_factor(u64 num, int factor)
587 static u64 div_factor_fine(u64 num, int factor)
596 u64 btrfs_find_block_group(struct btrfs_root *root,
597 u64 search_start, u64 search_hint, int owner)
599 struct btrfs_block_group_cache *cache;
601 u64 last = max(search_hint, search_start);
608 cache = btrfs_lookup_first_block_group(root->fs_info, last);
612 spin_lock(&cache->lock);
613 last = cache->key.objectid + cache->key.offset;
614 used = btrfs_block_group_used(&cache->item);
616 if ((full_search || !cache->ro) &&
617 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
618 if (used + cache->pinned + cache->reserved <
619 div_factor(cache->key.offset, factor)) {
620 group_start = cache->key.objectid;
621 spin_unlock(&cache->lock);
622 btrfs_put_block_group(cache);
626 spin_unlock(&cache->lock);
627 btrfs_put_block_group(cache);
635 if (!full_search && factor < 10) {
645 /* simple helper to search for an existing extent at a given offset */
646 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
649 struct btrfs_key key;
650 struct btrfs_path *path;
652 path = btrfs_alloc_path();
654 key.objectid = start;
656 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
657 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
659 btrfs_free_path(path);
664 * helper function to lookup reference count and flags of extent.
666 * the head node for delayed ref is used to store the sum of all the
667 * reference count modifications queued up in the rbtree. the head
668 * node may also store the extent flags to set. This way you can check
669 * to see what the reference count and extent flags would be if all of
670 * the delayed refs are not processed.
672 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
673 struct btrfs_root *root, u64 bytenr,
674 u64 num_bytes, u64 *refs, u64 *flags)
676 struct btrfs_delayed_ref_head *head;
677 struct btrfs_delayed_ref_root *delayed_refs;
678 struct btrfs_path *path;
679 struct btrfs_extent_item *ei;
680 struct extent_buffer *leaf;
681 struct btrfs_key key;
687 path = btrfs_alloc_path();
691 key.objectid = bytenr;
692 key.type = BTRFS_EXTENT_ITEM_KEY;
693 key.offset = num_bytes;
695 path->skip_locking = 1;
696 path->search_commit_root = 1;
699 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
705 leaf = path->nodes[0];
706 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
707 if (item_size >= sizeof(*ei)) {
708 ei = btrfs_item_ptr(leaf, path->slots[0],
709 struct btrfs_extent_item);
710 num_refs = btrfs_extent_refs(leaf, ei);
711 extent_flags = btrfs_extent_flags(leaf, ei);
713 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
714 struct btrfs_extent_item_v0 *ei0;
715 BUG_ON(item_size != sizeof(*ei0));
716 ei0 = btrfs_item_ptr(leaf, path->slots[0],
717 struct btrfs_extent_item_v0);
718 num_refs = btrfs_extent_refs_v0(leaf, ei0);
719 /* FIXME: this isn't correct for data */
720 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
725 BUG_ON(num_refs == 0);
735 delayed_refs = &trans->transaction->delayed_refs;
736 spin_lock(&delayed_refs->lock);
737 head = btrfs_find_delayed_ref_head(trans, bytenr);
739 if (!mutex_trylock(&head->mutex)) {
740 atomic_inc(&head->node.refs);
741 spin_unlock(&delayed_refs->lock);
743 btrfs_release_path(root->fs_info->extent_root, path);
745 mutex_lock(&head->mutex);
746 mutex_unlock(&head->mutex);
747 btrfs_put_delayed_ref(&head->node);
750 if (head->extent_op && head->extent_op->update_flags)
751 extent_flags |= head->extent_op->flags_to_set;
753 BUG_ON(num_refs == 0);
755 num_refs += head->node.ref_mod;
756 mutex_unlock(&head->mutex);
758 spin_unlock(&delayed_refs->lock);
760 WARN_ON(num_refs == 0);
764 *flags = extent_flags;
766 btrfs_free_path(path);
771 * Back reference rules. Back refs have three main goals:
773 * 1) differentiate between all holders of references to an extent so that
774 * when a reference is dropped we can make sure it was a valid reference
775 * before freeing the extent.
777 * 2) Provide enough information to quickly find the holders of an extent
778 * if we notice a given block is corrupted or bad.
780 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
781 * maintenance. This is actually the same as #2, but with a slightly
782 * different use case.
784 * There are two kinds of back refs. The implicit back refs is optimized
785 * for pointers in non-shared tree blocks. For a given pointer in a block,
786 * back refs of this kind provide information about the block's owner tree
787 * and the pointer's key. These information allow us to find the block by
788 * b-tree searching. The full back refs is for pointers in tree blocks not
789 * referenced by their owner trees. The location of tree block is recorded
790 * in the back refs. Actually the full back refs is generic, and can be
791 * used in all cases the implicit back refs is used. The major shortcoming
792 * of the full back refs is its overhead. Every time a tree block gets
793 * COWed, we have to update back refs entry for all pointers in it.
795 * For a newly allocated tree block, we use implicit back refs for
796 * pointers in it. This means most tree related operations only involve
797 * implicit back refs. For a tree block created in old transaction, the
798 * only way to drop a reference to it is COW it. So we can detect the
799 * event that tree block loses its owner tree's reference and do the
800 * back refs conversion.
802 * When a tree block is COW'd through a tree, there are four cases:
804 * The reference count of the block is one and the tree is the block's
805 * owner tree. Nothing to do in this case.
807 * The reference count of the block is one and the tree is not the
808 * block's owner tree. In this case, full back refs is used for pointers
809 * in the block. Remove these full back refs, add implicit back refs for
810 * every pointers in the new block.
812 * The reference count of the block is greater than one and the tree is
813 * the block's owner tree. In this case, implicit back refs is used for
814 * pointers in the block. Add full back refs for every pointers in the
815 * block, increase lower level extents' reference counts. The original
816 * implicit back refs are entailed to the new block.
818 * The reference count of the block is greater than one and the tree is
819 * not the block's owner tree. Add implicit back refs for every pointer in
820 * the new block, increase lower level extents' reference count.
822 * Back Reference Key composing:
824 * The key objectid corresponds to the first byte in the extent,
825 * The key type is used to differentiate between types of back refs.
826 * There are different meanings of the key offset for different types
829 * File extents can be referenced by:
831 * - multiple snapshots, subvolumes, or different generations in one subvol
832 * - different files inside a single subvolume
833 * - different offsets inside a file (bookend extents in file.c)
835 * The extent ref structure for the implicit back refs has fields for:
837 * - Objectid of the subvolume root
838 * - objectid of the file holding the reference
839 * - original offset in the file
840 * - how many bookend extents
842 * The key offset for the implicit back refs is hash of the first
845 * The extent ref structure for the full back refs has field for:
847 * - number of pointers in the tree leaf
849 * The key offset for the implicit back refs is the first byte of
852 * When a file extent is allocated, The implicit back refs is used.
853 * the fields are filled in:
855 * (root_key.objectid, inode objectid, offset in file, 1)
857 * When a file extent is removed file truncation, we find the
858 * corresponding implicit back refs and check the following fields:
860 * (btrfs_header_owner(leaf), inode objectid, offset in file)
862 * Btree extents can be referenced by:
864 * - Different subvolumes
866 * Both the implicit back refs and the full back refs for tree blocks
867 * only consist of key. The key offset for the implicit back refs is
868 * objectid of block's owner tree. The key offset for the full back refs
869 * is the first byte of parent block.
871 * When implicit back refs is used, information about the lowest key and
872 * level of the tree block are required. These information are stored in
873 * tree block info structure.
876 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
877 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
878 struct btrfs_root *root,
879 struct btrfs_path *path,
880 u64 owner, u32 extra_size)
882 struct btrfs_extent_item *item;
883 struct btrfs_extent_item_v0 *ei0;
884 struct btrfs_extent_ref_v0 *ref0;
885 struct btrfs_tree_block_info *bi;
886 struct extent_buffer *leaf;
887 struct btrfs_key key;
888 struct btrfs_key found_key;
889 u32 new_size = sizeof(*item);
893 leaf = path->nodes[0];
894 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
896 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
897 ei0 = btrfs_item_ptr(leaf, path->slots[0],
898 struct btrfs_extent_item_v0);
899 refs = btrfs_extent_refs_v0(leaf, ei0);
901 if (owner == (u64)-1) {
903 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
904 ret = btrfs_next_leaf(root, path);
908 leaf = path->nodes[0];
910 btrfs_item_key_to_cpu(leaf, &found_key,
912 BUG_ON(key.objectid != found_key.objectid);
913 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
917 ref0 = btrfs_item_ptr(leaf, path->slots[0],
918 struct btrfs_extent_ref_v0);
919 owner = btrfs_ref_objectid_v0(leaf, ref0);
923 btrfs_release_path(root, path);
925 if (owner < BTRFS_FIRST_FREE_OBJECTID)
926 new_size += sizeof(*bi);
928 new_size -= sizeof(*ei0);
929 ret = btrfs_search_slot(trans, root, &key, path,
930 new_size + extra_size, 1);
935 ret = btrfs_extend_item(trans, root, path, new_size);
938 leaf = path->nodes[0];
939 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
940 btrfs_set_extent_refs(leaf, item, refs);
941 /* FIXME: get real generation */
942 btrfs_set_extent_generation(leaf, item, 0);
943 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
944 btrfs_set_extent_flags(leaf, item,
945 BTRFS_EXTENT_FLAG_TREE_BLOCK |
946 BTRFS_BLOCK_FLAG_FULL_BACKREF);
947 bi = (struct btrfs_tree_block_info *)(item + 1);
948 /* FIXME: get first key of the block */
949 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
950 btrfs_set_tree_block_level(leaf, bi, (int)owner);
952 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
954 btrfs_mark_buffer_dirty(leaf);
959 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
961 u32 high_crc = ~(u32)0;
962 u32 low_crc = ~(u32)0;
965 lenum = cpu_to_le64(root_objectid);
966 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
967 lenum = cpu_to_le64(owner);
968 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
969 lenum = cpu_to_le64(offset);
970 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
972 return ((u64)high_crc << 31) ^ (u64)low_crc;
975 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
976 struct btrfs_extent_data_ref *ref)
978 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
979 btrfs_extent_data_ref_objectid(leaf, ref),
980 btrfs_extent_data_ref_offset(leaf, ref));
983 static int match_extent_data_ref(struct extent_buffer *leaf,
984 struct btrfs_extent_data_ref *ref,
985 u64 root_objectid, u64 owner, u64 offset)
987 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
988 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
989 btrfs_extent_data_ref_offset(leaf, ref) != offset)
994 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
995 struct btrfs_root *root,
996 struct btrfs_path *path,
997 u64 bytenr, u64 parent,
999 u64 owner, u64 offset)
1001 struct btrfs_key key;
1002 struct btrfs_extent_data_ref *ref;
1003 struct extent_buffer *leaf;
1009 key.objectid = bytenr;
1011 key.type = BTRFS_SHARED_DATA_REF_KEY;
1012 key.offset = parent;
1014 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1015 key.offset = hash_extent_data_ref(root_objectid,
1020 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1030 key.type = BTRFS_EXTENT_REF_V0_KEY;
1031 btrfs_release_path(root, path);
1032 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1043 leaf = path->nodes[0];
1044 nritems = btrfs_header_nritems(leaf);
1046 if (path->slots[0] >= nritems) {
1047 ret = btrfs_next_leaf(root, path);
1053 leaf = path->nodes[0];
1054 nritems = btrfs_header_nritems(leaf);
1058 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1059 if (key.objectid != bytenr ||
1060 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1063 ref = btrfs_item_ptr(leaf, path->slots[0],
1064 struct btrfs_extent_data_ref);
1066 if (match_extent_data_ref(leaf, ref, root_objectid,
1069 btrfs_release_path(root, path);
1081 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1082 struct btrfs_root *root,
1083 struct btrfs_path *path,
1084 u64 bytenr, u64 parent,
1085 u64 root_objectid, u64 owner,
1086 u64 offset, int refs_to_add)
1088 struct btrfs_key key;
1089 struct extent_buffer *leaf;
1094 key.objectid = bytenr;
1096 key.type = BTRFS_SHARED_DATA_REF_KEY;
1097 key.offset = parent;
1098 size = sizeof(struct btrfs_shared_data_ref);
1100 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1101 key.offset = hash_extent_data_ref(root_objectid,
1103 size = sizeof(struct btrfs_extent_data_ref);
1106 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1107 if (ret && ret != -EEXIST)
1110 leaf = path->nodes[0];
1112 struct btrfs_shared_data_ref *ref;
1113 ref = btrfs_item_ptr(leaf, path->slots[0],
1114 struct btrfs_shared_data_ref);
1116 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1118 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1119 num_refs += refs_to_add;
1120 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1123 struct btrfs_extent_data_ref *ref;
1124 while (ret == -EEXIST) {
1125 ref = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_extent_data_ref);
1127 if (match_extent_data_ref(leaf, ref, root_objectid,
1130 btrfs_release_path(root, path);
1132 ret = btrfs_insert_empty_item(trans, root, path, &key,
1134 if (ret && ret != -EEXIST)
1137 leaf = path->nodes[0];
1139 ref = btrfs_item_ptr(leaf, path->slots[0],
1140 struct btrfs_extent_data_ref);
1142 btrfs_set_extent_data_ref_root(leaf, ref,
1144 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1145 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1146 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1148 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1149 num_refs += refs_to_add;
1150 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1153 btrfs_mark_buffer_dirty(leaf);
1156 btrfs_release_path(root, path);
1160 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1165 struct btrfs_key key;
1166 struct btrfs_extent_data_ref *ref1 = NULL;
1167 struct btrfs_shared_data_ref *ref2 = NULL;
1168 struct extent_buffer *leaf;
1172 leaf = path->nodes[0];
1173 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1175 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1176 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1177 struct btrfs_extent_data_ref);
1178 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1179 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1180 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1181 struct btrfs_shared_data_ref);
1182 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1183 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1184 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1185 struct btrfs_extent_ref_v0 *ref0;
1186 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1187 struct btrfs_extent_ref_v0);
1188 num_refs = btrfs_ref_count_v0(leaf, ref0);
1194 BUG_ON(num_refs < refs_to_drop);
1195 num_refs -= refs_to_drop;
1197 if (num_refs == 0) {
1198 ret = btrfs_del_item(trans, root, path);
1200 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1201 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1202 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1203 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1204 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1206 struct btrfs_extent_ref_v0 *ref0;
1207 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1208 struct btrfs_extent_ref_v0);
1209 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1212 btrfs_mark_buffer_dirty(leaf);
1217 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1218 struct btrfs_path *path,
1219 struct btrfs_extent_inline_ref *iref)
1221 struct btrfs_key key;
1222 struct extent_buffer *leaf;
1223 struct btrfs_extent_data_ref *ref1;
1224 struct btrfs_shared_data_ref *ref2;
1227 leaf = path->nodes[0];
1228 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1230 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1231 BTRFS_EXTENT_DATA_REF_KEY) {
1232 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1233 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1235 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1236 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1238 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1239 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1240 struct btrfs_extent_data_ref);
1241 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1242 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1243 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_shared_data_ref);
1245 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1246 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1247 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1248 struct btrfs_extent_ref_v0 *ref0;
1249 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1250 struct btrfs_extent_ref_v0);
1251 num_refs = btrfs_ref_count_v0(leaf, ref0);
1259 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1260 struct btrfs_root *root,
1261 struct btrfs_path *path,
1262 u64 bytenr, u64 parent,
1265 struct btrfs_key key;
1268 key.objectid = bytenr;
1270 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1271 key.offset = parent;
1273 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1274 key.offset = root_objectid;
1277 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281 if (ret == -ENOENT && parent) {
1282 btrfs_release_path(root, path);
1283 key.type = BTRFS_EXTENT_REF_V0_KEY;
1284 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1292 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1293 struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 u64 bytenr, u64 parent,
1298 struct btrfs_key key;
1301 key.objectid = bytenr;
1303 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1304 key.offset = parent;
1306 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1307 key.offset = root_objectid;
1310 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1311 btrfs_release_path(root, path);
1315 static inline int extent_ref_type(u64 parent, u64 owner)
1318 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1320 type = BTRFS_SHARED_BLOCK_REF_KEY;
1322 type = BTRFS_TREE_BLOCK_REF_KEY;
1325 type = BTRFS_SHARED_DATA_REF_KEY;
1327 type = BTRFS_EXTENT_DATA_REF_KEY;
1332 static int find_next_key(struct btrfs_path *path, int level,
1333 struct btrfs_key *key)
1336 for (; level < BTRFS_MAX_LEVEL; level++) {
1337 if (!path->nodes[level])
1339 if (path->slots[level] + 1 >=
1340 btrfs_header_nritems(path->nodes[level]))
1343 btrfs_item_key_to_cpu(path->nodes[level], key,
1344 path->slots[level] + 1);
1346 btrfs_node_key_to_cpu(path->nodes[level], key,
1347 path->slots[level] + 1);
1354 * look for inline back ref. if back ref is found, *ref_ret is set
1355 * to the address of inline back ref, and 0 is returned.
1357 * if back ref isn't found, *ref_ret is set to the address where it
1358 * should be inserted, and -ENOENT is returned.
1360 * if insert is true and there are too many inline back refs, the path
1361 * points to the extent item, and -EAGAIN is returned.
1363 * NOTE: inline back refs are ordered in the same way that back ref
1364 * items in the tree are ordered.
1366 static noinline_for_stack
1367 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1368 struct btrfs_root *root,
1369 struct btrfs_path *path,
1370 struct btrfs_extent_inline_ref **ref_ret,
1371 u64 bytenr, u64 num_bytes,
1372 u64 parent, u64 root_objectid,
1373 u64 owner, u64 offset, int insert)
1375 struct btrfs_key key;
1376 struct extent_buffer *leaf;
1377 struct btrfs_extent_item *ei;
1378 struct btrfs_extent_inline_ref *iref;
1389 key.objectid = bytenr;
1390 key.type = BTRFS_EXTENT_ITEM_KEY;
1391 key.offset = num_bytes;
1393 want = extent_ref_type(parent, owner);
1395 extra_size = btrfs_extent_inline_ref_size(want);
1396 path->keep_locks = 1;
1399 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1406 leaf = path->nodes[0];
1407 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1408 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1409 if (item_size < sizeof(*ei)) {
1414 ret = convert_extent_item_v0(trans, root, path, owner,
1420 leaf = path->nodes[0];
1421 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1424 BUG_ON(item_size < sizeof(*ei));
1426 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1427 flags = btrfs_extent_flags(leaf, ei);
1429 ptr = (unsigned long)(ei + 1);
1430 end = (unsigned long)ei + item_size;
1432 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1433 ptr += sizeof(struct btrfs_tree_block_info);
1436 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1445 iref = (struct btrfs_extent_inline_ref *)ptr;
1446 type = btrfs_extent_inline_ref_type(leaf, iref);
1450 ptr += btrfs_extent_inline_ref_size(type);
1454 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1455 struct btrfs_extent_data_ref *dref;
1456 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1457 if (match_extent_data_ref(leaf, dref, root_objectid,
1462 if (hash_extent_data_ref_item(leaf, dref) <
1463 hash_extent_data_ref(root_objectid, owner, offset))
1467 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1469 if (parent == ref_offset) {
1473 if (ref_offset < parent)
1476 if (root_objectid == ref_offset) {
1480 if (ref_offset < root_objectid)
1484 ptr += btrfs_extent_inline_ref_size(type);
1486 if (err == -ENOENT && insert) {
1487 if (item_size + extra_size >=
1488 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1493 * To add new inline back ref, we have to make sure
1494 * there is no corresponding back ref item.
1495 * For simplicity, we just do not add new inline back
1496 * ref if there is any kind of item for this block
1498 if (find_next_key(path, 0, &key) == 0 &&
1499 key.objectid == bytenr &&
1500 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1505 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1508 path->keep_locks = 0;
1509 btrfs_unlock_up_safe(path, 1);
1515 * helper to add new inline back ref
1517 static noinline_for_stack
1518 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1519 struct btrfs_root *root,
1520 struct btrfs_path *path,
1521 struct btrfs_extent_inline_ref *iref,
1522 u64 parent, u64 root_objectid,
1523 u64 owner, u64 offset, int refs_to_add,
1524 struct btrfs_delayed_extent_op *extent_op)
1526 struct extent_buffer *leaf;
1527 struct btrfs_extent_item *ei;
1530 unsigned long item_offset;
1536 leaf = path->nodes[0];
1537 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1538 item_offset = (unsigned long)iref - (unsigned long)ei;
1540 type = extent_ref_type(parent, owner);
1541 size = btrfs_extent_inline_ref_size(type);
1543 ret = btrfs_extend_item(trans, root, path, size);
1546 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1547 refs = btrfs_extent_refs(leaf, ei);
1548 refs += refs_to_add;
1549 btrfs_set_extent_refs(leaf, ei, refs);
1551 __run_delayed_extent_op(extent_op, leaf, ei);
1553 ptr = (unsigned long)ei + item_offset;
1554 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1555 if (ptr < end - size)
1556 memmove_extent_buffer(leaf, ptr + size, ptr,
1559 iref = (struct btrfs_extent_inline_ref *)ptr;
1560 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1561 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1562 struct btrfs_extent_data_ref *dref;
1563 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1564 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1565 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1566 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1567 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1568 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1569 struct btrfs_shared_data_ref *sref;
1570 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1571 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1572 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1573 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1574 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1576 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1578 btrfs_mark_buffer_dirty(leaf);
1582 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1583 struct btrfs_root *root,
1584 struct btrfs_path *path,
1585 struct btrfs_extent_inline_ref **ref_ret,
1586 u64 bytenr, u64 num_bytes, u64 parent,
1587 u64 root_objectid, u64 owner, u64 offset)
1591 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1592 bytenr, num_bytes, parent,
1593 root_objectid, owner, offset, 0);
1597 btrfs_release_path(root, path);
1600 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1601 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1604 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1605 root_objectid, owner, offset);
1611 * helper to update/remove inline back ref
1613 static noinline_for_stack
1614 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1615 struct btrfs_root *root,
1616 struct btrfs_path *path,
1617 struct btrfs_extent_inline_ref *iref,
1619 struct btrfs_delayed_extent_op *extent_op)
1621 struct extent_buffer *leaf;
1622 struct btrfs_extent_item *ei;
1623 struct btrfs_extent_data_ref *dref = NULL;
1624 struct btrfs_shared_data_ref *sref = NULL;
1633 leaf = path->nodes[0];
1634 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1635 refs = btrfs_extent_refs(leaf, ei);
1636 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1637 refs += refs_to_mod;
1638 btrfs_set_extent_refs(leaf, ei, refs);
1640 __run_delayed_extent_op(extent_op, leaf, ei);
1642 type = btrfs_extent_inline_ref_type(leaf, iref);
1644 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1645 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1646 refs = btrfs_extent_data_ref_count(leaf, dref);
1647 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1648 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1649 refs = btrfs_shared_data_ref_count(leaf, sref);
1652 BUG_ON(refs_to_mod != -1);
1655 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1656 refs += refs_to_mod;
1659 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1660 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1662 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1664 size = btrfs_extent_inline_ref_size(type);
1665 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1666 ptr = (unsigned long)iref;
1667 end = (unsigned long)ei + item_size;
1668 if (ptr + size < end)
1669 memmove_extent_buffer(leaf, ptr, ptr + size,
1672 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1675 btrfs_mark_buffer_dirty(leaf);
1679 static noinline_for_stack
1680 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1681 struct btrfs_root *root,
1682 struct btrfs_path *path,
1683 u64 bytenr, u64 num_bytes, u64 parent,
1684 u64 root_objectid, u64 owner,
1685 u64 offset, int refs_to_add,
1686 struct btrfs_delayed_extent_op *extent_op)
1688 struct btrfs_extent_inline_ref *iref;
1691 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1692 bytenr, num_bytes, parent,
1693 root_objectid, owner, offset, 1);
1695 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1696 ret = update_inline_extent_backref(trans, root, path, iref,
1697 refs_to_add, extent_op);
1698 } else if (ret == -ENOENT) {
1699 ret = setup_inline_extent_backref(trans, root, path, iref,
1700 parent, root_objectid,
1701 owner, offset, refs_to_add,
1707 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1708 struct btrfs_root *root,
1709 struct btrfs_path *path,
1710 u64 bytenr, u64 parent, u64 root_objectid,
1711 u64 owner, u64 offset, int refs_to_add)
1714 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1715 BUG_ON(refs_to_add != 1);
1716 ret = insert_tree_block_ref(trans, root, path, bytenr,
1717 parent, root_objectid);
1719 ret = insert_extent_data_ref(trans, root, path, bytenr,
1720 parent, root_objectid,
1721 owner, offset, refs_to_add);
1726 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1727 struct btrfs_root *root,
1728 struct btrfs_path *path,
1729 struct btrfs_extent_inline_ref *iref,
1730 int refs_to_drop, int is_data)
1734 BUG_ON(!is_data && refs_to_drop != 1);
1736 ret = update_inline_extent_backref(trans, root, path, iref,
1737 -refs_to_drop, NULL);
1738 } else if (is_data) {
1739 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1741 ret = btrfs_del_item(trans, root, path);
1746 static void btrfs_issue_discard(struct block_device *bdev,
1749 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1750 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1753 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1757 u64 map_length = num_bytes;
1758 struct btrfs_multi_bio *multi = NULL;
1760 if (!btrfs_test_opt(root, DISCARD))
1763 /* Tell the block device(s) that the sectors can be discarded */
1764 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1765 bytenr, &map_length, &multi, 0);
1767 struct btrfs_bio_stripe *stripe = multi->stripes;
1770 if (map_length > num_bytes)
1771 map_length = num_bytes;
1773 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1774 btrfs_issue_discard(stripe->dev->bdev,
1784 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 u64 bytenr, u64 num_bytes, u64 parent,
1787 u64 root_objectid, u64 owner, u64 offset)
1790 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1791 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1793 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1794 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1795 parent, root_objectid, (int)owner,
1796 BTRFS_ADD_DELAYED_REF, NULL);
1798 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1799 parent, root_objectid, owner, offset,
1800 BTRFS_ADD_DELAYED_REF, NULL);
1805 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1806 struct btrfs_root *root,
1807 u64 bytenr, u64 num_bytes,
1808 u64 parent, u64 root_objectid,
1809 u64 owner, u64 offset, int refs_to_add,
1810 struct btrfs_delayed_extent_op *extent_op)
1812 struct btrfs_path *path;
1813 struct extent_buffer *leaf;
1814 struct btrfs_extent_item *item;
1819 path = btrfs_alloc_path();
1824 path->leave_spinning = 1;
1825 /* this will setup the path even if it fails to insert the back ref */
1826 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1827 path, bytenr, num_bytes, parent,
1828 root_objectid, owner, offset,
1829 refs_to_add, extent_op);
1833 if (ret != -EAGAIN) {
1838 leaf = path->nodes[0];
1839 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1840 refs = btrfs_extent_refs(leaf, item);
1841 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1843 __run_delayed_extent_op(extent_op, leaf, item);
1845 btrfs_mark_buffer_dirty(leaf);
1846 btrfs_release_path(root->fs_info->extent_root, path);
1849 path->leave_spinning = 1;
1851 /* now insert the actual backref */
1852 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1853 path, bytenr, parent, root_objectid,
1854 owner, offset, refs_to_add);
1857 btrfs_free_path(path);
1861 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1862 struct btrfs_root *root,
1863 struct btrfs_delayed_ref_node *node,
1864 struct btrfs_delayed_extent_op *extent_op,
1865 int insert_reserved)
1868 struct btrfs_delayed_data_ref *ref;
1869 struct btrfs_key ins;
1874 ins.objectid = node->bytenr;
1875 ins.offset = node->num_bytes;
1876 ins.type = BTRFS_EXTENT_ITEM_KEY;
1878 ref = btrfs_delayed_node_to_data_ref(node);
1879 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1880 parent = ref->parent;
1882 ref_root = ref->root;
1884 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1886 BUG_ON(extent_op->update_key);
1887 flags |= extent_op->flags_to_set;
1889 ret = alloc_reserved_file_extent(trans, root,
1890 parent, ref_root, flags,
1891 ref->objectid, ref->offset,
1892 &ins, node->ref_mod);
1893 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1894 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1895 node->num_bytes, parent,
1896 ref_root, ref->objectid,
1897 ref->offset, node->ref_mod,
1899 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1900 ret = __btrfs_free_extent(trans, root, node->bytenr,
1901 node->num_bytes, parent,
1902 ref_root, ref->objectid,
1903 ref->offset, node->ref_mod,
1911 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1912 struct extent_buffer *leaf,
1913 struct btrfs_extent_item *ei)
1915 u64 flags = btrfs_extent_flags(leaf, ei);
1916 if (extent_op->update_flags) {
1917 flags |= extent_op->flags_to_set;
1918 btrfs_set_extent_flags(leaf, ei, flags);
1921 if (extent_op->update_key) {
1922 struct btrfs_tree_block_info *bi;
1923 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1924 bi = (struct btrfs_tree_block_info *)(ei + 1);
1925 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1929 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1930 struct btrfs_root *root,
1931 struct btrfs_delayed_ref_node *node,
1932 struct btrfs_delayed_extent_op *extent_op)
1934 struct btrfs_key key;
1935 struct btrfs_path *path;
1936 struct btrfs_extent_item *ei;
1937 struct extent_buffer *leaf;
1942 path = btrfs_alloc_path();
1946 key.objectid = node->bytenr;
1947 key.type = BTRFS_EXTENT_ITEM_KEY;
1948 key.offset = node->num_bytes;
1951 path->leave_spinning = 1;
1952 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1963 leaf = path->nodes[0];
1964 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1965 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1966 if (item_size < sizeof(*ei)) {
1967 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1973 leaf = path->nodes[0];
1974 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1977 BUG_ON(item_size < sizeof(*ei));
1978 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1979 __run_delayed_extent_op(extent_op, leaf, ei);
1981 btrfs_mark_buffer_dirty(leaf);
1983 btrfs_free_path(path);
1987 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1988 struct btrfs_root *root,
1989 struct btrfs_delayed_ref_node *node,
1990 struct btrfs_delayed_extent_op *extent_op,
1991 int insert_reserved)
1994 struct btrfs_delayed_tree_ref *ref;
1995 struct btrfs_key ins;
1999 ins.objectid = node->bytenr;
2000 ins.offset = node->num_bytes;
2001 ins.type = BTRFS_EXTENT_ITEM_KEY;
2003 ref = btrfs_delayed_node_to_tree_ref(node);
2004 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2005 parent = ref->parent;
2007 ref_root = ref->root;
2009 BUG_ON(node->ref_mod != 1);
2010 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2011 BUG_ON(!extent_op || !extent_op->update_flags ||
2012 !extent_op->update_key);
2013 ret = alloc_reserved_tree_block(trans, root,
2015 extent_op->flags_to_set,
2018 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2019 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2020 node->num_bytes, parent, ref_root,
2021 ref->level, 0, 1, extent_op);
2022 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2023 ret = __btrfs_free_extent(trans, root, node->bytenr,
2024 node->num_bytes, parent, ref_root,
2025 ref->level, 0, 1, extent_op);
2032 /* helper function to actually process a single delayed ref entry */
2033 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2034 struct btrfs_root *root,
2035 struct btrfs_delayed_ref_node *node,
2036 struct btrfs_delayed_extent_op *extent_op,
2037 int insert_reserved)
2040 if (btrfs_delayed_ref_is_head(node)) {
2041 struct btrfs_delayed_ref_head *head;
2043 * we've hit the end of the chain and we were supposed
2044 * to insert this extent into the tree. But, it got
2045 * deleted before we ever needed to insert it, so all
2046 * we have to do is clean up the accounting
2049 head = btrfs_delayed_node_to_head(node);
2050 if (insert_reserved) {
2051 btrfs_pin_extent(root, node->bytenr,
2052 node->num_bytes, 1);
2053 if (head->is_data) {
2054 ret = btrfs_del_csums(trans, root,
2060 mutex_unlock(&head->mutex);
2064 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2065 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2066 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2068 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2069 node->type == BTRFS_SHARED_DATA_REF_KEY)
2070 ret = run_delayed_data_ref(trans, root, node, extent_op,
2077 static noinline struct btrfs_delayed_ref_node *
2078 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2080 struct rb_node *node;
2081 struct btrfs_delayed_ref_node *ref;
2082 int action = BTRFS_ADD_DELAYED_REF;
2085 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2086 * this prevents ref count from going down to zero when
2087 * there still are pending delayed ref.
2089 node = rb_prev(&head->node.rb_node);
2093 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2095 if (ref->bytenr != head->node.bytenr)
2097 if (ref->action == action)
2099 node = rb_prev(node);
2101 if (action == BTRFS_ADD_DELAYED_REF) {
2102 action = BTRFS_DROP_DELAYED_REF;
2108 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2109 struct btrfs_root *root,
2110 struct list_head *cluster)
2112 struct btrfs_delayed_ref_root *delayed_refs;
2113 struct btrfs_delayed_ref_node *ref;
2114 struct btrfs_delayed_ref_head *locked_ref = NULL;
2115 struct btrfs_delayed_extent_op *extent_op;
2118 int must_insert_reserved = 0;
2120 delayed_refs = &trans->transaction->delayed_refs;
2123 /* pick a new head ref from the cluster list */
2124 if (list_empty(cluster))
2127 locked_ref = list_entry(cluster->next,
2128 struct btrfs_delayed_ref_head, cluster);
2130 /* grab the lock that says we are going to process
2131 * all the refs for this head */
2132 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2135 * we may have dropped the spin lock to get the head
2136 * mutex lock, and that might have given someone else
2137 * time to free the head. If that's true, it has been
2138 * removed from our list and we can move on.
2140 if (ret == -EAGAIN) {
2148 * record the must insert reserved flag before we
2149 * drop the spin lock.
2151 must_insert_reserved = locked_ref->must_insert_reserved;
2152 locked_ref->must_insert_reserved = 0;
2154 extent_op = locked_ref->extent_op;
2155 locked_ref->extent_op = NULL;
2158 * locked_ref is the head node, so we have to go one
2159 * node back for any delayed ref updates
2161 ref = select_delayed_ref(locked_ref);
2163 /* All delayed refs have been processed, Go ahead
2164 * and send the head node to run_one_delayed_ref,
2165 * so that any accounting fixes can happen
2167 ref = &locked_ref->node;
2169 if (extent_op && must_insert_reserved) {
2175 spin_unlock(&delayed_refs->lock);
2177 ret = run_delayed_extent_op(trans, root,
2183 spin_lock(&delayed_refs->lock);
2187 list_del_init(&locked_ref->cluster);
2192 rb_erase(&ref->rb_node, &delayed_refs->root);
2193 delayed_refs->num_entries--;
2195 spin_unlock(&delayed_refs->lock);
2197 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2198 must_insert_reserved);
2201 btrfs_put_delayed_ref(ref);
2206 spin_lock(&delayed_refs->lock);
2212 * this starts processing the delayed reference count updates and
2213 * extent insertions we have queued up so far. count can be
2214 * 0, which means to process everything in the tree at the start
2215 * of the run (but not newly added entries), or it can be some target
2216 * number you'd like to process.
2218 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2219 struct btrfs_root *root, unsigned long count)
2221 struct rb_node *node;
2222 struct btrfs_delayed_ref_root *delayed_refs;
2223 struct btrfs_delayed_ref_node *ref;
2224 struct list_head cluster;
2226 int run_all = count == (unsigned long)-1;
2229 if (root == root->fs_info->extent_root)
2230 root = root->fs_info->tree_root;
2232 delayed_refs = &trans->transaction->delayed_refs;
2233 INIT_LIST_HEAD(&cluster);
2235 spin_lock(&delayed_refs->lock);
2237 count = delayed_refs->num_entries * 2;
2241 if (!(run_all || run_most) &&
2242 delayed_refs->num_heads_ready < 64)
2246 * go find something we can process in the rbtree. We start at
2247 * the beginning of the tree, and then build a cluster
2248 * of refs to process starting at the first one we are able to
2251 ret = btrfs_find_ref_cluster(trans, &cluster,
2252 delayed_refs->run_delayed_start);
2256 ret = run_clustered_refs(trans, root, &cluster);
2259 count -= min_t(unsigned long, ret, count);
2266 node = rb_first(&delayed_refs->root);
2269 count = (unsigned long)-1;
2272 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2274 if (btrfs_delayed_ref_is_head(ref)) {
2275 struct btrfs_delayed_ref_head *head;
2277 head = btrfs_delayed_node_to_head(ref);
2278 atomic_inc(&ref->refs);
2280 spin_unlock(&delayed_refs->lock);
2281 mutex_lock(&head->mutex);
2282 mutex_unlock(&head->mutex);
2284 btrfs_put_delayed_ref(ref);
2288 node = rb_next(node);
2290 spin_unlock(&delayed_refs->lock);
2291 schedule_timeout(1);
2295 spin_unlock(&delayed_refs->lock);
2299 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2300 struct btrfs_root *root,
2301 u64 bytenr, u64 num_bytes, u64 flags,
2304 struct btrfs_delayed_extent_op *extent_op;
2307 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2311 extent_op->flags_to_set = flags;
2312 extent_op->update_flags = 1;
2313 extent_op->update_key = 0;
2314 extent_op->is_data = is_data ? 1 : 0;
2316 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2322 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root,
2324 struct btrfs_path *path,
2325 u64 objectid, u64 offset, u64 bytenr)
2327 struct btrfs_delayed_ref_head *head;
2328 struct btrfs_delayed_ref_node *ref;
2329 struct btrfs_delayed_data_ref *data_ref;
2330 struct btrfs_delayed_ref_root *delayed_refs;
2331 struct rb_node *node;
2335 delayed_refs = &trans->transaction->delayed_refs;
2336 spin_lock(&delayed_refs->lock);
2337 head = btrfs_find_delayed_ref_head(trans, bytenr);
2341 if (!mutex_trylock(&head->mutex)) {
2342 atomic_inc(&head->node.refs);
2343 spin_unlock(&delayed_refs->lock);
2345 btrfs_release_path(root->fs_info->extent_root, path);
2347 mutex_lock(&head->mutex);
2348 mutex_unlock(&head->mutex);
2349 btrfs_put_delayed_ref(&head->node);
2353 node = rb_prev(&head->node.rb_node);
2357 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2359 if (ref->bytenr != bytenr)
2363 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2366 data_ref = btrfs_delayed_node_to_data_ref(ref);
2368 node = rb_prev(node);
2370 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2371 if (ref->bytenr == bytenr)
2375 if (data_ref->root != root->root_key.objectid ||
2376 data_ref->objectid != objectid || data_ref->offset != offset)
2381 mutex_unlock(&head->mutex);
2383 spin_unlock(&delayed_refs->lock);
2387 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2388 struct btrfs_root *root,
2389 struct btrfs_path *path,
2390 u64 objectid, u64 offset, u64 bytenr)
2392 struct btrfs_root *extent_root = root->fs_info->extent_root;
2393 struct extent_buffer *leaf;
2394 struct btrfs_extent_data_ref *ref;
2395 struct btrfs_extent_inline_ref *iref;
2396 struct btrfs_extent_item *ei;
2397 struct btrfs_key key;
2401 key.objectid = bytenr;
2402 key.offset = (u64)-1;
2403 key.type = BTRFS_EXTENT_ITEM_KEY;
2405 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2411 if (path->slots[0] == 0)
2415 leaf = path->nodes[0];
2416 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2418 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2422 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2424 if (item_size < sizeof(*ei)) {
2425 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2429 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2431 if (item_size != sizeof(*ei) +
2432 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2435 if (btrfs_extent_generation(leaf, ei) <=
2436 btrfs_root_last_snapshot(&root->root_item))
2439 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2440 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2441 BTRFS_EXTENT_DATA_REF_KEY)
2444 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2445 if (btrfs_extent_refs(leaf, ei) !=
2446 btrfs_extent_data_ref_count(leaf, ref) ||
2447 btrfs_extent_data_ref_root(leaf, ref) !=
2448 root->root_key.objectid ||
2449 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2450 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2458 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2459 struct btrfs_root *root,
2460 u64 objectid, u64 offset, u64 bytenr)
2462 struct btrfs_path *path;
2466 path = btrfs_alloc_path();
2471 ret = check_committed_ref(trans, root, path, objectid,
2473 if (ret && ret != -ENOENT)
2476 ret2 = check_delayed_ref(trans, root, path, objectid,
2478 } while (ret2 == -EAGAIN);
2480 if (ret2 && ret2 != -ENOENT) {
2485 if (ret != -ENOENT || ret2 != -ENOENT)
2488 btrfs_free_path(path);
2489 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2495 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2496 struct extent_buffer *buf, u32 nr_extents)
2498 struct btrfs_key key;
2499 struct btrfs_file_extent_item *fi;
2507 if (!root->ref_cows)
2510 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2512 root_gen = root->root_key.offset;
2515 root_gen = trans->transid - 1;
2518 level = btrfs_header_level(buf);
2519 nritems = btrfs_header_nritems(buf);
2522 struct btrfs_leaf_ref *ref;
2523 struct btrfs_extent_info *info;
2525 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2531 ref->root_gen = root_gen;
2532 ref->bytenr = buf->start;
2533 ref->owner = btrfs_header_owner(buf);
2534 ref->generation = btrfs_header_generation(buf);
2535 ref->nritems = nr_extents;
2536 info = ref->extents;
2538 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2540 btrfs_item_key_to_cpu(buf, &key, i);
2541 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2543 fi = btrfs_item_ptr(buf, i,
2544 struct btrfs_file_extent_item);
2545 if (btrfs_file_extent_type(buf, fi) ==
2546 BTRFS_FILE_EXTENT_INLINE)
2548 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2549 if (disk_bytenr == 0)
2552 info->bytenr = disk_bytenr;
2554 btrfs_file_extent_disk_num_bytes(buf, fi);
2555 info->objectid = key.objectid;
2556 info->offset = key.offset;
2560 ret = btrfs_add_leaf_ref(root, ref, shared);
2561 if (ret == -EEXIST && shared) {
2562 struct btrfs_leaf_ref *old;
2563 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2565 btrfs_remove_leaf_ref(root, old);
2566 btrfs_free_leaf_ref(root, old);
2567 ret = btrfs_add_leaf_ref(root, ref, shared);
2570 btrfs_free_leaf_ref(root, ref);
2576 /* when a block goes through cow, we update the reference counts of
2577 * everything that block points to. The internal pointers of the block
2578 * can be in just about any order, and it is likely to have clusters of
2579 * things that are close together and clusters of things that are not.
2581 * To help reduce the seeks that come with updating all of these reference
2582 * counts, sort them by byte number before actual updates are done.
2584 * struct refsort is used to match byte number to slot in the btree block.
2585 * we sort based on the byte number and then use the slot to actually
2588 * struct refsort is smaller than strcut btrfs_item and smaller than
2589 * struct btrfs_key_ptr. Since we're currently limited to the page size
2590 * for a btree block, there's no way for a kmalloc of refsorts for a
2591 * single node to be bigger than a page.
2599 * for passing into sort()
2601 static int refsort_cmp(const void *a_void, const void *b_void)
2603 const struct refsort *a = a_void;
2604 const struct refsort *b = b_void;
2606 if (a->bytenr < b->bytenr)
2608 if (a->bytenr > b->bytenr)
2614 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2615 struct btrfs_root *root,
2616 struct extent_buffer *buf,
2617 int full_backref, int inc)
2624 struct btrfs_key key;
2625 struct btrfs_file_extent_item *fi;
2629 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2630 u64, u64, u64, u64, u64, u64);
2632 ref_root = btrfs_header_owner(buf);
2633 nritems = btrfs_header_nritems(buf);
2634 level = btrfs_header_level(buf);
2636 if (!root->ref_cows && level == 0)
2640 process_func = btrfs_inc_extent_ref;
2642 process_func = btrfs_free_extent;
2645 parent = buf->start;
2649 for (i = 0; i < nritems; i++) {
2651 btrfs_item_key_to_cpu(buf, &key, i);
2652 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2654 fi = btrfs_item_ptr(buf, i,
2655 struct btrfs_file_extent_item);
2656 if (btrfs_file_extent_type(buf, fi) ==
2657 BTRFS_FILE_EXTENT_INLINE)
2659 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2663 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2664 key.offset -= btrfs_file_extent_offset(buf, fi);
2665 ret = process_func(trans, root, bytenr, num_bytes,
2666 parent, ref_root, key.objectid,
2671 bytenr = btrfs_node_blockptr(buf, i);
2672 num_bytes = btrfs_level_size(root, level - 1);
2673 ret = process_func(trans, root, bytenr, num_bytes,
2674 parent, ref_root, level - 1, 0);
2685 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2686 struct extent_buffer *buf, int full_backref)
2688 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2691 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2692 struct extent_buffer *buf, int full_backref)
2694 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2697 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2698 struct btrfs_root *root,
2699 struct btrfs_path *path,
2700 struct btrfs_block_group_cache *cache)
2703 struct btrfs_root *extent_root = root->fs_info->extent_root;
2705 struct extent_buffer *leaf;
2707 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2712 leaf = path->nodes[0];
2713 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2714 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2715 btrfs_mark_buffer_dirty(leaf);
2716 btrfs_release_path(extent_root, path);
2724 static struct btrfs_block_group_cache *
2725 next_block_group(struct btrfs_root *root,
2726 struct btrfs_block_group_cache *cache)
2728 struct rb_node *node;
2729 spin_lock(&root->fs_info->block_group_cache_lock);
2730 node = rb_next(&cache->cache_node);
2731 btrfs_put_block_group(cache);
2733 cache = rb_entry(node, struct btrfs_block_group_cache,
2735 btrfs_get_block_group(cache);
2738 spin_unlock(&root->fs_info->block_group_cache_lock);
2742 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2743 struct btrfs_trans_handle *trans,
2744 struct btrfs_path *path)
2746 struct btrfs_root *root = block_group->fs_info->tree_root;
2747 struct inode *inode = NULL;
2749 int dcs = BTRFS_DC_ERROR;
2755 * If this block group is smaller than 100 megs don't bother caching the
2758 if (block_group->key.offset < (100 * 1024 * 1024)) {
2759 spin_lock(&block_group->lock);
2760 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2761 spin_unlock(&block_group->lock);
2766 inode = lookup_free_space_inode(root, block_group, path);
2767 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2768 ret = PTR_ERR(inode);
2769 btrfs_release_path(root, path);
2773 if (IS_ERR(inode)) {
2777 if (block_group->ro)
2780 ret = create_free_space_inode(root, trans, block_group, path);
2787 * We want to set the generation to 0, that way if anything goes wrong
2788 * from here on out we know not to trust this cache when we load up next
2791 BTRFS_I(inode)->generation = 0;
2792 ret = btrfs_update_inode(trans, root, inode);
2795 if (i_size_read(inode) > 0) {
2796 ret = btrfs_truncate_free_space_cache(root, trans, path,
2802 spin_lock(&block_group->lock);
2803 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2804 /* We're not cached, don't bother trying to write stuff out */
2805 dcs = BTRFS_DC_WRITTEN;
2806 spin_unlock(&block_group->lock);
2809 spin_unlock(&block_group->lock);
2811 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2816 * Just to make absolutely sure we have enough space, we're going to
2817 * preallocate 12 pages worth of space for each block group. In
2818 * practice we ought to use at most 8, but we need extra space so we can
2819 * add our header and have a terminator between the extents and the
2823 num_pages *= PAGE_CACHE_SIZE;
2825 ret = btrfs_check_data_free_space(inode, num_pages);
2829 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2830 num_pages, num_pages,
2833 dcs = BTRFS_DC_SETUP;
2834 btrfs_free_reserved_data_space(inode, num_pages);
2838 btrfs_release_path(root, path);
2840 spin_lock(&block_group->lock);
2841 block_group->disk_cache_state = dcs;
2842 spin_unlock(&block_group->lock);
2847 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2848 struct btrfs_root *root)
2850 struct btrfs_block_group_cache *cache;
2852 struct btrfs_path *path;
2855 path = btrfs_alloc_path();
2861 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2863 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2865 cache = next_block_group(root, cache);
2873 err = cache_save_setup(cache, trans, path);
2874 last = cache->key.objectid + cache->key.offset;
2875 btrfs_put_block_group(cache);
2880 err = btrfs_run_delayed_refs(trans, root,
2885 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2887 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2888 btrfs_put_block_group(cache);
2894 cache = next_block_group(root, cache);
2903 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2904 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2906 last = cache->key.objectid + cache->key.offset;
2908 err = write_one_cache_group(trans, root, path, cache);
2910 btrfs_put_block_group(cache);
2915 * I don't think this is needed since we're just marking our
2916 * preallocated extent as written, but just in case it can't
2920 err = btrfs_run_delayed_refs(trans, root,
2925 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2928 * Really this shouldn't happen, but it could if we
2929 * couldn't write the entire preallocated extent and
2930 * splitting the extent resulted in a new block.
2933 btrfs_put_block_group(cache);
2936 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2938 cache = next_block_group(root, cache);
2947 btrfs_write_out_cache(root, trans, cache, path);
2950 * If we didn't have an error then the cache state is still
2951 * NEED_WRITE, so we can set it to WRITTEN.
2953 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2954 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2955 last = cache->key.objectid + cache->key.offset;
2956 btrfs_put_block_group(cache);
2959 btrfs_free_path(path);
2963 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2965 struct btrfs_block_group_cache *block_group;
2968 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2969 if (!block_group || block_group->ro)
2972 btrfs_put_block_group(block_group);
2976 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2977 u64 total_bytes, u64 bytes_used,
2978 struct btrfs_space_info **space_info)
2980 struct btrfs_space_info *found;
2984 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2985 BTRFS_BLOCK_GROUP_RAID10))
2990 found = __find_space_info(info, flags);
2992 spin_lock(&found->lock);
2993 found->total_bytes += total_bytes;
2994 found->disk_total += total_bytes * factor;
2995 found->bytes_used += bytes_used;
2996 found->disk_used += bytes_used * factor;
2998 spin_unlock(&found->lock);
2999 *space_info = found;
3002 found = kzalloc(sizeof(*found), GFP_NOFS);
3006 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3007 INIT_LIST_HEAD(&found->block_groups[i]);
3008 init_rwsem(&found->groups_sem);
3009 spin_lock_init(&found->lock);
3010 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3011 BTRFS_BLOCK_GROUP_SYSTEM |
3012 BTRFS_BLOCK_GROUP_METADATA);
3013 found->total_bytes = total_bytes;
3014 found->disk_total = total_bytes * factor;
3015 found->bytes_used = bytes_used;
3016 found->disk_used = bytes_used * factor;
3017 found->bytes_pinned = 0;
3018 found->bytes_reserved = 0;
3019 found->bytes_readonly = 0;
3020 found->bytes_may_use = 0;
3022 found->force_alloc = 0;
3023 *space_info = found;
3024 list_add_rcu(&found->list, &info->space_info);
3025 atomic_set(&found->caching_threads, 0);
3029 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3031 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3032 BTRFS_BLOCK_GROUP_RAID1 |
3033 BTRFS_BLOCK_GROUP_RAID10 |
3034 BTRFS_BLOCK_GROUP_DUP);
3036 if (flags & BTRFS_BLOCK_GROUP_DATA)
3037 fs_info->avail_data_alloc_bits |= extra_flags;
3038 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3039 fs_info->avail_metadata_alloc_bits |= extra_flags;
3040 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3041 fs_info->avail_system_alloc_bits |= extra_flags;
3045 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3048 * we add in the count of missing devices because we want
3049 * to make sure that any RAID levels on a degraded FS
3050 * continue to be honored.
3052 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3053 root->fs_info->fs_devices->missing_devices;
3055 if (num_devices == 1)
3056 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3057 if (num_devices < 4)
3058 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3060 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3061 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3062 BTRFS_BLOCK_GROUP_RAID10))) {
3063 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3066 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3067 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3068 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3071 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3072 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3073 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3074 (flags & BTRFS_BLOCK_GROUP_DUP)))
3075 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3079 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3081 if (flags & BTRFS_BLOCK_GROUP_DATA)
3082 flags |= root->fs_info->avail_data_alloc_bits &
3083 root->fs_info->data_alloc_profile;
3084 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3085 flags |= root->fs_info->avail_system_alloc_bits &
3086 root->fs_info->system_alloc_profile;
3087 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3088 flags |= root->fs_info->avail_metadata_alloc_bits &
3089 root->fs_info->metadata_alloc_profile;
3090 return btrfs_reduce_alloc_profile(root, flags);
3093 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3098 flags = BTRFS_BLOCK_GROUP_DATA;
3099 else if (root == root->fs_info->chunk_root)
3100 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3102 flags = BTRFS_BLOCK_GROUP_METADATA;
3104 return get_alloc_profile(root, flags);
3107 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3109 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3110 BTRFS_BLOCK_GROUP_DATA);
3114 * This will check the space that the inode allocates from to make sure we have
3115 * enough space for bytes.
3117 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3119 struct btrfs_space_info *data_sinfo;
3120 struct btrfs_root *root = BTRFS_I(inode)->root;
3122 int ret = 0, committed = 0, alloc_chunk = 1;
3124 /* make sure bytes are sectorsize aligned */
3125 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3127 if (root == root->fs_info->tree_root) {
3132 data_sinfo = BTRFS_I(inode)->space_info;
3137 /* make sure we have enough space to handle the data first */
3138 spin_lock(&data_sinfo->lock);
3139 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3140 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3141 data_sinfo->bytes_may_use;
3143 if (used + bytes > data_sinfo->total_bytes) {
3144 struct btrfs_trans_handle *trans;
3147 * if we don't have enough free bytes in this space then we need
3148 * to alloc a new chunk.
3150 if (!data_sinfo->full && alloc_chunk) {
3153 data_sinfo->force_alloc = 1;
3154 spin_unlock(&data_sinfo->lock);
3156 alloc_target = btrfs_get_alloc_profile(root, 1);
3157 trans = btrfs_join_transaction(root, 1);
3159 return PTR_ERR(trans);
3161 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3162 bytes + 2 * 1024 * 1024,
3164 btrfs_end_transaction(trans, root);
3173 btrfs_set_inode_space_info(root, inode);
3174 data_sinfo = BTRFS_I(inode)->space_info;
3178 spin_unlock(&data_sinfo->lock);
3180 /* commit the current transaction and try again */
3182 if (!committed && !root->fs_info->open_ioctl_trans) {
3184 trans = btrfs_join_transaction(root, 1);
3186 return PTR_ERR(trans);
3187 ret = btrfs_commit_transaction(trans, root);
3193 #if 0 /* I hope we never need this code again, just in case */
3194 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3195 "%llu bytes_reserved, " "%llu bytes_pinned, "
3196 "%llu bytes_readonly, %llu may use %llu total\n",
3197 (unsigned long long)bytes,
3198 (unsigned long long)data_sinfo->bytes_used,
3199 (unsigned long long)data_sinfo->bytes_reserved,
3200 (unsigned long long)data_sinfo->bytes_pinned,
3201 (unsigned long long)data_sinfo->bytes_readonly,
3202 (unsigned long long)data_sinfo->bytes_may_use,
3203 (unsigned long long)data_sinfo->total_bytes);
3207 data_sinfo->bytes_may_use += bytes;
3208 BTRFS_I(inode)->reserved_bytes += bytes;
3209 spin_unlock(&data_sinfo->lock);
3215 * called when we are clearing an delalloc extent from the
3216 * inode's io_tree or there was an error for whatever reason
3217 * after calling btrfs_check_data_free_space
3219 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3221 struct btrfs_root *root = BTRFS_I(inode)->root;
3222 struct btrfs_space_info *data_sinfo;
3224 /* make sure bytes are sectorsize aligned */
3225 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3227 data_sinfo = BTRFS_I(inode)->space_info;
3228 spin_lock(&data_sinfo->lock);
3229 data_sinfo->bytes_may_use -= bytes;
3230 BTRFS_I(inode)->reserved_bytes -= bytes;
3231 spin_unlock(&data_sinfo->lock);
3234 static void force_metadata_allocation(struct btrfs_fs_info *info)
3236 struct list_head *head = &info->space_info;
3237 struct btrfs_space_info *found;
3240 list_for_each_entry_rcu(found, head, list) {
3241 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3242 found->force_alloc = 1;
3247 static int should_alloc_chunk(struct btrfs_root *root,
3248 struct btrfs_space_info *sinfo, u64 alloc_bytes)
3250 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3253 if (sinfo->bytes_used + sinfo->bytes_reserved +
3254 alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3257 if (sinfo->bytes_used + sinfo->bytes_reserved +
3258 alloc_bytes < div_factor(num_bytes, 8))
3261 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3262 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3264 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3270 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3271 struct btrfs_root *extent_root, u64 alloc_bytes,
3272 u64 flags, int force)
3274 struct btrfs_space_info *space_info;
3275 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3278 mutex_lock(&fs_info->chunk_mutex);
3280 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3282 space_info = __find_space_info(extent_root->fs_info, flags);
3284 ret = update_space_info(extent_root->fs_info, flags,
3288 BUG_ON(!space_info);
3290 spin_lock(&space_info->lock);
3291 if (space_info->force_alloc)
3293 if (space_info->full) {
3294 spin_unlock(&space_info->lock);
3298 if (!force && !should_alloc_chunk(extent_root, space_info,
3300 spin_unlock(&space_info->lock);
3303 spin_unlock(&space_info->lock);
3306 * If we have mixed data/metadata chunks we want to make sure we keep
3307 * allocating mixed chunks instead of individual chunks.
3309 if (btrfs_mixed_space_info(space_info))
3310 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3313 * if we're doing a data chunk, go ahead and make sure that
3314 * we keep a reasonable number of metadata chunks allocated in the
3317 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3318 fs_info->data_chunk_allocations++;
3319 if (!(fs_info->data_chunk_allocations %
3320 fs_info->metadata_ratio))
3321 force_metadata_allocation(fs_info);
3324 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3325 spin_lock(&space_info->lock);
3327 space_info->full = 1;
3330 space_info->force_alloc = 0;
3331 spin_unlock(&space_info->lock);
3333 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3338 * shrink metadata reservation for delalloc
3340 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3341 struct btrfs_root *root, u64 to_reclaim, int sync)
3343 struct btrfs_block_rsv *block_rsv;
3344 struct btrfs_space_info *space_info;
3349 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3351 block_rsv = &root->fs_info->delalloc_block_rsv;
3352 space_info = block_rsv->space_info;
3355 reserved = space_info->bytes_reserved;
3360 max_reclaim = min(reserved, to_reclaim);
3363 /* have the flusher threads jump in and do some IO */
3365 nr_pages = min_t(unsigned long, nr_pages,
3366 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3367 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3369 spin_lock(&space_info->lock);
3370 if (reserved > space_info->bytes_reserved)
3371 reclaimed += reserved - space_info->bytes_reserved;
3372 reserved = space_info->bytes_reserved;
3373 spin_unlock(&space_info->lock);
3375 if (reserved == 0 || reclaimed >= max_reclaim)
3378 if (trans && trans->transaction->blocked)
3381 __set_current_state(TASK_INTERRUPTIBLE);
3382 schedule_timeout(pause);
3384 if (pause > HZ / 10)
3388 return reclaimed >= to_reclaim;
3392 * Retries tells us how many times we've called reserve_metadata_bytes. The
3393 * idea is if this is the first call (retries == 0) then we will add to our
3394 * reserved count if we can't make the allocation in order to hold our place
3395 * while we go and try and free up space. That way for retries > 1 we don't try
3396 * and add space, we just check to see if the amount of unused space is >= the
3397 * total space, meaning that our reservation is valid.
3399 * However if we don't intend to retry this reservation, pass -1 as retries so
3400 * that it short circuits this logic.
3402 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3403 struct btrfs_root *root,
3404 struct btrfs_block_rsv *block_rsv,
3405 u64 orig_bytes, int flush)
3407 struct btrfs_space_info *space_info = block_rsv->space_info;
3409 u64 num_bytes = orig_bytes;
3412 bool reserved = false;
3413 bool committed = false;
3420 spin_lock(&space_info->lock);
3421 unused = space_info->bytes_used + space_info->bytes_reserved +
3422 space_info->bytes_pinned + space_info->bytes_readonly +
3423 space_info->bytes_may_use;
3426 * The idea here is that we've not already over-reserved the block group
3427 * then we can go ahead and save our reservation first and then start
3428 * flushing if we need to. Otherwise if we've already overcommitted
3429 * lets start flushing stuff first and then come back and try to make
3432 if (unused <= space_info->total_bytes) {
3433 unused = space_info->total_bytes - unused;
3434 if (unused >= num_bytes) {
3436 space_info->bytes_reserved += orig_bytes;
3440 * Ok set num_bytes to orig_bytes since we aren't
3441 * overocmmitted, this way we only try and reclaim what
3444 num_bytes = orig_bytes;
3448 * Ok we're over committed, set num_bytes to the overcommitted
3449 * amount plus the amount of bytes that we need for this
3452 num_bytes = unused - space_info->total_bytes +
3453 (orig_bytes * (retries + 1));
3457 * Couldn't make our reservation, save our place so while we're trying
3458 * to reclaim space we can actually use it instead of somebody else
3459 * stealing it from us.
3461 if (ret && !reserved) {
3462 space_info->bytes_reserved += orig_bytes;
3466 spin_unlock(&space_info->lock);
3475 * We do synchronous shrinking since we don't actually unreserve
3476 * metadata until after the IO is completed.
3478 ret = shrink_delalloc(trans, root, num_bytes, 1);
3485 * So if we were overcommitted it's possible that somebody else flushed
3486 * out enough space and we simply didn't have enough space to reclaim,
3487 * so go back around and try again.
3494 spin_lock(&space_info->lock);
3496 * Not enough space to be reclaimed, don't bother committing the
3499 if (space_info->bytes_pinned < orig_bytes)
3501 spin_unlock(&space_info->lock);
3506 if (trans || committed)
3510 trans = btrfs_join_transaction(root, 1);
3513 ret = btrfs_commit_transaction(trans, root);
3522 spin_lock(&space_info->lock);
3523 space_info->bytes_reserved -= orig_bytes;
3524 spin_unlock(&space_info->lock);
3530 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3531 struct btrfs_root *root)
3533 struct btrfs_block_rsv *block_rsv;
3535 block_rsv = trans->block_rsv;
3537 block_rsv = root->block_rsv;
3540 block_rsv = &root->fs_info->empty_block_rsv;
3545 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3549 spin_lock(&block_rsv->lock);
3550 if (block_rsv->reserved >= num_bytes) {
3551 block_rsv->reserved -= num_bytes;
3552 if (block_rsv->reserved < block_rsv->size)
3553 block_rsv->full = 0;
3556 spin_unlock(&block_rsv->lock);
3560 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3561 u64 num_bytes, int update_size)
3563 spin_lock(&block_rsv->lock);
3564 block_rsv->reserved += num_bytes;
3566 block_rsv->size += num_bytes;
3567 else if (block_rsv->reserved >= block_rsv->size)
3568 block_rsv->full = 1;
3569 spin_unlock(&block_rsv->lock);
3572 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3573 struct btrfs_block_rsv *dest, u64 num_bytes)
3575 struct btrfs_space_info *space_info = block_rsv->space_info;
3577 spin_lock(&block_rsv->lock);
3578 if (num_bytes == (u64)-1)
3579 num_bytes = block_rsv->size;
3580 block_rsv->size -= num_bytes;
3581 if (block_rsv->reserved >= block_rsv->size) {
3582 num_bytes = block_rsv->reserved - block_rsv->size;
3583 block_rsv->reserved = block_rsv->size;
3584 block_rsv->full = 1;
3588 spin_unlock(&block_rsv->lock);
3590 if (num_bytes > 0) {
3592 block_rsv_add_bytes(dest, num_bytes, 0);
3594 spin_lock(&space_info->lock);
3595 space_info->bytes_reserved -= num_bytes;
3596 spin_unlock(&space_info->lock);
3601 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3602 struct btrfs_block_rsv *dst, u64 num_bytes)
3606 ret = block_rsv_use_bytes(src, num_bytes);
3610 block_rsv_add_bytes(dst, num_bytes, 1);
3614 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3616 memset(rsv, 0, sizeof(*rsv));
3617 spin_lock_init(&rsv->lock);
3618 atomic_set(&rsv->usage, 1);
3620 INIT_LIST_HEAD(&rsv->list);
3623 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3625 struct btrfs_block_rsv *block_rsv;
3626 struct btrfs_fs_info *fs_info = root->fs_info;
3628 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3632 btrfs_init_block_rsv(block_rsv);
3633 block_rsv->space_info = __find_space_info(fs_info,
3634 BTRFS_BLOCK_GROUP_METADATA);
3638 void btrfs_free_block_rsv(struct btrfs_root *root,
3639 struct btrfs_block_rsv *rsv)
3641 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3642 btrfs_block_rsv_release(root, rsv, (u64)-1);
3649 * make the block_rsv struct be able to capture freed space.
3650 * the captured space will re-add to the the block_rsv struct
3651 * after transaction commit
3653 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3654 struct btrfs_block_rsv *block_rsv)
3656 block_rsv->durable = 1;
3657 mutex_lock(&fs_info->durable_block_rsv_mutex);
3658 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3659 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3662 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3663 struct btrfs_root *root,
3664 struct btrfs_block_rsv *block_rsv,
3672 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3674 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3681 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3682 struct btrfs_root *root,
3683 struct btrfs_block_rsv *block_rsv,
3684 u64 min_reserved, int min_factor)
3687 int commit_trans = 0;
3693 spin_lock(&block_rsv->lock);
3695 num_bytes = div_factor(block_rsv->size, min_factor);
3696 if (min_reserved > num_bytes)
3697 num_bytes = min_reserved;
3699 if (block_rsv->reserved >= num_bytes) {
3702 num_bytes -= block_rsv->reserved;
3703 if (block_rsv->durable &&
3704 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3707 spin_unlock(&block_rsv->lock);
3711 if (block_rsv->refill_used) {
3712 ret = reserve_metadata_bytes(trans, root, block_rsv,
3715 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3724 trans = btrfs_join_transaction(root, 1);
3725 BUG_ON(IS_ERR(trans));
3726 ret = btrfs_commit_transaction(trans, root);
3731 printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3732 block_rsv->size, block_rsv->reserved,
3733 block_rsv->freed[0], block_rsv->freed[1]);
3738 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3739 struct btrfs_block_rsv *dst_rsv,
3742 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3745 void btrfs_block_rsv_release(struct btrfs_root *root,
3746 struct btrfs_block_rsv *block_rsv,
3749 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3750 if (global_rsv->full || global_rsv == block_rsv ||
3751 block_rsv->space_info != global_rsv->space_info)
3753 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3757 * helper to calculate size of global block reservation.
3758 * the desired value is sum of space used by extent tree,
3759 * checksum tree and root tree
3761 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3763 struct btrfs_space_info *sinfo;
3767 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3770 * per tree used space accounting can be inaccuracy, so we
3773 spin_lock(&fs_info->extent_root->accounting_lock);
3774 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3775 spin_unlock(&fs_info->extent_root->accounting_lock);
3777 spin_lock(&fs_info->csum_root->accounting_lock);
3778 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3779 spin_unlock(&fs_info->csum_root->accounting_lock);
3781 spin_lock(&fs_info->tree_root->accounting_lock);
3782 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3783 spin_unlock(&fs_info->tree_root->accounting_lock);
3785 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3786 spin_lock(&sinfo->lock);
3787 data_used = sinfo->bytes_used;
3788 spin_unlock(&sinfo->lock);
3790 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3791 spin_lock(&sinfo->lock);
3792 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3794 meta_used = sinfo->bytes_used;
3795 spin_unlock(&sinfo->lock);
3797 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3799 num_bytes += div64_u64(data_used + meta_used, 50);
3801 if (num_bytes * 3 > meta_used)
3802 num_bytes = div64_u64(meta_used, 3);
3804 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3807 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3809 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3810 struct btrfs_space_info *sinfo = block_rsv->space_info;
3813 num_bytes = calc_global_metadata_size(fs_info);
3815 spin_lock(&block_rsv->lock);
3816 spin_lock(&sinfo->lock);
3818 block_rsv->size = num_bytes;
3820 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3821 sinfo->bytes_reserved + sinfo->bytes_readonly +
3822 sinfo->bytes_may_use;
3824 if (sinfo->total_bytes > num_bytes) {
3825 num_bytes = sinfo->total_bytes - num_bytes;
3826 block_rsv->reserved += num_bytes;
3827 sinfo->bytes_reserved += num_bytes;
3830 if (block_rsv->reserved >= block_rsv->size) {
3831 num_bytes = block_rsv->reserved - block_rsv->size;
3832 sinfo->bytes_reserved -= num_bytes;
3833 block_rsv->reserved = block_rsv->size;
3834 block_rsv->full = 1;
3837 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3838 block_rsv->size, block_rsv->reserved);
3840 spin_unlock(&sinfo->lock);
3841 spin_unlock(&block_rsv->lock);
3844 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3846 struct btrfs_space_info *space_info;
3848 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3849 fs_info->chunk_block_rsv.space_info = space_info;
3850 fs_info->chunk_block_rsv.priority = 10;
3852 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3853 fs_info->global_block_rsv.space_info = space_info;
3854 fs_info->global_block_rsv.priority = 10;
3855 fs_info->global_block_rsv.refill_used = 1;
3856 fs_info->delalloc_block_rsv.space_info = space_info;
3857 fs_info->trans_block_rsv.space_info = space_info;
3858 fs_info->empty_block_rsv.space_info = space_info;
3859 fs_info->empty_block_rsv.priority = 10;
3861 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3862 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3863 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3864 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3865 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3867 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3869 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3871 update_global_block_rsv(fs_info);
3874 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3876 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3877 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3878 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3879 WARN_ON(fs_info->trans_block_rsv.size > 0);
3880 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3881 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3882 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3885 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3887 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3891 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3892 struct btrfs_root *root,
3898 if (num_items == 0 || root->fs_info->chunk_root == root)
3901 num_bytes = calc_trans_metadata_size(root, num_items);
3902 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3905 trans->bytes_reserved += num_bytes;
3906 trans->block_rsv = &root->fs_info->trans_block_rsv;
3911 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3912 struct btrfs_root *root)
3914 if (!trans->bytes_reserved)
3917 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3918 btrfs_block_rsv_release(root, trans->block_rsv,
3919 trans->bytes_reserved);
3920 trans->bytes_reserved = 0;
3923 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3924 struct inode *inode)
3926 struct btrfs_root *root = BTRFS_I(inode)->root;
3927 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3928 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3931 * one for deleting orphan item, one for updating inode and
3932 * two for calling btrfs_truncate_inode_items.
3934 * btrfs_truncate_inode_items is a delete operation, it frees
3935 * more space than it uses in most cases. So two units of
3936 * metadata space should be enough for calling it many times.
3937 * If all of the metadata space is used, we can commit
3938 * transaction and use space it freed.
3940 u64 num_bytes = calc_trans_metadata_size(root, 4);
3941 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3944 void btrfs_orphan_release_metadata(struct inode *inode)
3946 struct btrfs_root *root = BTRFS_I(inode)->root;
3947 u64 num_bytes = calc_trans_metadata_size(root, 4);
3948 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3951 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3952 struct btrfs_pending_snapshot *pending)
3954 struct btrfs_root *root = pending->root;
3955 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3956 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3958 * two for root back/forward refs, two for directory entries
3959 * and one for root of the snapshot.
3961 u64 num_bytes = calc_trans_metadata_size(root, 5);
3962 dst_rsv->space_info = src_rsv->space_info;
3963 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3966 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3968 return num_bytes >>= 3;
3971 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3973 struct btrfs_root *root = BTRFS_I(inode)->root;
3974 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3979 if (btrfs_transaction_in_commit(root->fs_info))
3980 schedule_timeout(1);
3982 num_bytes = ALIGN(num_bytes, root->sectorsize);
3984 spin_lock(&BTRFS_I(inode)->accounting_lock);
3985 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3986 if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3987 nr_extents -= BTRFS_I(inode)->reserved_extents;
3988 to_reserve = calc_trans_metadata_size(root, nr_extents);
3993 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3995 to_reserve += calc_csum_metadata_size(inode, num_bytes);
3996 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4000 spin_lock(&BTRFS_I(inode)->accounting_lock);
4001 BTRFS_I(inode)->reserved_extents += nr_extents;
4002 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4003 spin_unlock(&BTRFS_I(inode)->accounting_lock);
4005 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4007 if (block_rsv->size > 512 * 1024 * 1024)
4008 shrink_delalloc(NULL, root, to_reserve, 0);
4013 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4015 struct btrfs_root *root = BTRFS_I(inode)->root;
4019 num_bytes = ALIGN(num_bytes, root->sectorsize);
4020 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4022 spin_lock(&BTRFS_I(inode)->accounting_lock);
4023 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4024 if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4025 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4026 BTRFS_I(inode)->reserved_extents -= nr_extents;
4030 spin_unlock(&BTRFS_I(inode)->accounting_lock);
4032 to_free = calc_csum_metadata_size(inode, num_bytes);
4034 to_free += calc_trans_metadata_size(root, nr_extents);
4036 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4040 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4044 ret = btrfs_check_data_free_space(inode, num_bytes);
4048 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4050 btrfs_free_reserved_data_space(inode, num_bytes);
4057 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4059 btrfs_delalloc_release_metadata(inode, num_bytes);
4060 btrfs_free_reserved_data_space(inode, num_bytes);
4063 static int update_block_group(struct btrfs_trans_handle *trans,
4064 struct btrfs_root *root,
4065 u64 bytenr, u64 num_bytes, int alloc)
4067 struct btrfs_block_group_cache *cache = NULL;
4068 struct btrfs_fs_info *info = root->fs_info;
4069 u64 total = num_bytes;
4074 /* block accounting for super block */
4075 spin_lock(&info->delalloc_lock);
4076 old_val = btrfs_super_bytes_used(&info->super_copy);
4078 old_val += num_bytes;
4080 old_val -= num_bytes;
4081 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4082 spin_unlock(&info->delalloc_lock);
4085 cache = btrfs_lookup_block_group(info, bytenr);
4088 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4089 BTRFS_BLOCK_GROUP_RAID1 |
4090 BTRFS_BLOCK_GROUP_RAID10))
4095 * If this block group has free space cache written out, we
4096 * need to make sure to load it if we are removing space. This
4097 * is because we need the unpinning stage to actually add the
4098 * space back to the block group, otherwise we will leak space.
4100 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4101 cache_block_group(cache, trans, NULL, 1);
4103 byte_in_group = bytenr - cache->key.objectid;
4104 WARN_ON(byte_in_group > cache->key.offset);
4106 spin_lock(&cache->space_info->lock);
4107 spin_lock(&cache->lock);
4109 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4110 cache->disk_cache_state < BTRFS_DC_CLEAR)
4111 cache->disk_cache_state = BTRFS_DC_CLEAR;
4114 old_val = btrfs_block_group_used(&cache->item);
4115 num_bytes = min(total, cache->key.offset - byte_in_group);
4117 old_val += num_bytes;
4118 btrfs_set_block_group_used(&cache->item, old_val);
4119 cache->reserved -= num_bytes;
4120 cache->space_info->bytes_reserved -= num_bytes;
4121 cache->space_info->bytes_used += num_bytes;
4122 cache->space_info->disk_used += num_bytes * factor;
4123 spin_unlock(&cache->lock);
4124 spin_unlock(&cache->space_info->lock);
4126 old_val -= num_bytes;
4127 btrfs_set_block_group_used(&cache->item, old_val);
4128 cache->pinned += num_bytes;
4129 cache->space_info->bytes_pinned += num_bytes;
4130 cache->space_info->bytes_used -= num_bytes;
4131 cache->space_info->disk_used -= num_bytes * factor;
4132 spin_unlock(&cache->lock);
4133 spin_unlock(&cache->space_info->lock);
4135 set_extent_dirty(info->pinned_extents,
4136 bytenr, bytenr + num_bytes - 1,
4137 GFP_NOFS | __GFP_NOFAIL);
4139 btrfs_put_block_group(cache);
4141 bytenr += num_bytes;
4146 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4148 struct btrfs_block_group_cache *cache;
4151 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4155 bytenr = cache->key.objectid;
4156 btrfs_put_block_group(cache);
4161 static int pin_down_extent(struct btrfs_root *root,
4162 struct btrfs_block_group_cache *cache,
4163 u64 bytenr, u64 num_bytes, int reserved)
4165 spin_lock(&cache->space_info->lock);
4166 spin_lock(&cache->lock);
4167 cache->pinned += num_bytes;
4168 cache->space_info->bytes_pinned += num_bytes;
4170 cache->reserved -= num_bytes;
4171 cache->space_info->bytes_reserved -= num_bytes;
4173 spin_unlock(&cache->lock);
4174 spin_unlock(&cache->space_info->lock);
4176 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4177 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4182 * this function must be called within transaction
4184 int btrfs_pin_extent(struct btrfs_root *root,
4185 u64 bytenr, u64 num_bytes, int reserved)
4187 struct btrfs_block_group_cache *cache;
4189 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4192 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4194 btrfs_put_block_group(cache);
4199 * update size of reserved extents. this function may return -EAGAIN
4200 * if 'reserve' is true or 'sinfo' is false.
4202 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4203 u64 num_bytes, int reserve, int sinfo)
4207 struct btrfs_space_info *space_info = cache->space_info;
4208 spin_lock(&space_info->lock);
4209 spin_lock(&cache->lock);
4214 cache->reserved += num_bytes;
4215 space_info->bytes_reserved += num_bytes;
4219 space_info->bytes_readonly += num_bytes;
4220 cache->reserved -= num_bytes;
4221 space_info->bytes_reserved -= num_bytes;
4223 spin_unlock(&cache->lock);
4224 spin_unlock(&space_info->lock);
4226 spin_lock(&cache->lock);
4231 cache->reserved += num_bytes;
4233 cache->reserved -= num_bytes;
4235 spin_unlock(&cache->lock);
4240 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4241 struct btrfs_root *root)
4243 struct btrfs_fs_info *fs_info = root->fs_info;
4244 struct btrfs_caching_control *next;
4245 struct btrfs_caching_control *caching_ctl;
4246 struct btrfs_block_group_cache *cache;
4248 down_write(&fs_info->extent_commit_sem);
4250 list_for_each_entry_safe(caching_ctl, next,
4251 &fs_info->caching_block_groups, list) {
4252 cache = caching_ctl->block_group;
4253 if (block_group_cache_done(cache)) {
4254 cache->last_byte_to_unpin = (u64)-1;
4255 list_del_init(&caching_ctl->list);
4256 put_caching_control(caching_ctl);
4258 cache->last_byte_to_unpin = caching_ctl->progress;
4262 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4263 fs_info->pinned_extents = &fs_info->freed_extents[1];
4265 fs_info->pinned_extents = &fs_info->freed_extents[0];
4267 up_write(&fs_info->extent_commit_sem);
4269 update_global_block_rsv(fs_info);
4273 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4275 struct btrfs_fs_info *fs_info = root->fs_info;
4276 struct btrfs_block_group_cache *cache = NULL;
4279 while (start <= end) {
4281 start >= cache->key.objectid + cache->key.offset) {
4283 btrfs_put_block_group(cache);
4284 cache = btrfs_lookup_block_group(fs_info, start);
4288 len = cache->key.objectid + cache->key.offset - start;
4289 len = min(len, end + 1 - start);
4291 if (start < cache->last_byte_to_unpin) {
4292 len = min(len, cache->last_byte_to_unpin - start);
4293 btrfs_add_free_space(cache, start, len);
4298 spin_lock(&cache->space_info->lock);
4299 spin_lock(&cache->lock);
4300 cache->pinned -= len;
4301 cache->space_info->bytes_pinned -= len;
4303 cache->space_info->bytes_readonly += len;
4304 } else if (cache->reserved_pinned > 0) {
4305 len = min(len, cache->reserved_pinned);
4306 cache->reserved_pinned -= len;
4307 cache->space_info->bytes_reserved += len;
4309 spin_unlock(&cache->lock);
4310 spin_unlock(&cache->space_info->lock);
4314 btrfs_put_block_group(cache);
4318 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4319 struct btrfs_root *root)
4321 struct btrfs_fs_info *fs_info = root->fs_info;
4322 struct extent_io_tree *unpin;
4323 struct btrfs_block_rsv *block_rsv;
4324 struct btrfs_block_rsv *next_rsv;
4330 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4331 unpin = &fs_info->freed_extents[1];
4333 unpin = &fs_info->freed_extents[0];
4336 ret = find_first_extent_bit(unpin, 0, &start, &end,
4341 ret = btrfs_discard_extent(root, start, end + 1 - start);
4343 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4344 unpin_extent_range(root, start, end);
4348 mutex_lock(&fs_info->durable_block_rsv_mutex);
4349 list_for_each_entry_safe(block_rsv, next_rsv,
4350 &fs_info->durable_block_rsv_list, list) {
4352 idx = trans->transid & 0x1;
4353 if (block_rsv->freed[idx] > 0) {
4354 block_rsv_add_bytes(block_rsv,
4355 block_rsv->freed[idx], 0);
4356 block_rsv->freed[idx] = 0;
4358 if (atomic_read(&block_rsv->usage) == 0) {
4359 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4361 if (block_rsv->freed[0] == 0 &&
4362 block_rsv->freed[1] == 0) {
4363 list_del_init(&block_rsv->list);
4367 btrfs_block_rsv_release(root, block_rsv, 0);
4370 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4375 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root,
4377 u64 bytenr, u64 num_bytes, u64 parent,
4378 u64 root_objectid, u64 owner_objectid,
4379 u64 owner_offset, int refs_to_drop,
4380 struct btrfs_delayed_extent_op *extent_op)
4382 struct btrfs_key key;
4383 struct btrfs_path *path;
4384 struct btrfs_fs_info *info = root->fs_info;
4385 struct btrfs_root *extent_root = info->extent_root;
4386 struct extent_buffer *leaf;
4387 struct btrfs_extent_item *ei;
4388 struct btrfs_extent_inline_ref *iref;
4391 int extent_slot = 0;
4392 int found_extent = 0;
4397 path = btrfs_alloc_path();
4402 path->leave_spinning = 1;
4404 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4405 BUG_ON(!is_data && refs_to_drop != 1);
4407 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4408 bytenr, num_bytes, parent,
4409 root_objectid, owner_objectid,
4412 extent_slot = path->slots[0];
4413 while (extent_slot >= 0) {
4414 btrfs_item_key_to_cpu(path->nodes[0], &key,
4416 if (key.objectid != bytenr)
4418 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4419 key.offset == num_bytes) {
4423 if (path->slots[0] - extent_slot > 5)
4427 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4428 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4429 if (found_extent && item_size < sizeof(*ei))
4432 if (!found_extent) {
4434 ret = remove_extent_backref(trans, extent_root, path,
4438 btrfs_release_path(extent_root, path);
4439 path->leave_spinning = 1;
4441 key.objectid = bytenr;
4442 key.type = BTRFS_EXTENT_ITEM_KEY;
4443 key.offset = num_bytes;
4445 ret = btrfs_search_slot(trans, extent_root,
4448 printk(KERN_ERR "umm, got %d back from search"
4449 ", was looking for %llu\n", ret,
4450 (unsigned long long)bytenr);
4451 btrfs_print_leaf(extent_root, path->nodes[0]);
4454 extent_slot = path->slots[0];
4457 btrfs_print_leaf(extent_root, path->nodes[0]);
4459 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4460 "parent %llu root %llu owner %llu offset %llu\n",
4461 (unsigned long long)bytenr,
4462 (unsigned long long)parent,
4463 (unsigned long long)root_objectid,
4464 (unsigned long long)owner_objectid,
4465 (unsigned long long)owner_offset);
4468 leaf = path->nodes[0];
4469 item_size = btrfs_item_size_nr(leaf, extent_slot);
4470 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4471 if (item_size < sizeof(*ei)) {
4472 BUG_ON(found_extent || extent_slot != path->slots[0]);
4473 ret = convert_extent_item_v0(trans, extent_root, path,
4477 btrfs_release_path(extent_root, path);
4478 path->leave_spinning = 1;
4480 key.objectid = bytenr;
4481 key.type = BTRFS_EXTENT_ITEM_KEY;
4482 key.offset = num_bytes;
4484 ret = btrfs_search_slot(trans, extent_root, &key, path,
4487 printk(KERN_ERR "umm, got %d back from search"
4488 ", was looking for %llu\n", ret,
4489 (unsigned long long)bytenr);
4490 btrfs_print_leaf(extent_root, path->nodes[0]);
4493 extent_slot = path->slots[0];
4494 leaf = path->nodes[0];
4495 item_size = btrfs_item_size_nr(leaf, extent_slot);
4498 BUG_ON(item_size < sizeof(*ei));
4499 ei = btrfs_item_ptr(leaf, extent_slot,
4500 struct btrfs_extent_item);
4501 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4502 struct btrfs_tree_block_info *bi;
4503 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4504 bi = (struct btrfs_tree_block_info *)(ei + 1);
4505 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4508 refs = btrfs_extent_refs(leaf, ei);
4509 BUG_ON(refs < refs_to_drop);
4510 refs -= refs_to_drop;
4514 __run_delayed_extent_op(extent_op, leaf, ei);
4516 * In the case of inline back ref, reference count will
4517 * be updated by remove_extent_backref
4520 BUG_ON(!found_extent);
4522 btrfs_set_extent_refs(leaf, ei, refs);
4523 btrfs_mark_buffer_dirty(leaf);
4526 ret = remove_extent_backref(trans, extent_root, path,
4533 BUG_ON(is_data && refs_to_drop !=
4534 extent_data_ref_count(root, path, iref));
4536 BUG_ON(path->slots[0] != extent_slot);
4538 BUG_ON(path->slots[0] != extent_slot + 1);
4539 path->slots[0] = extent_slot;
4544 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4547 btrfs_release_path(extent_root, path);
4550 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4553 invalidate_mapping_pages(info->btree_inode->i_mapping,
4554 bytenr >> PAGE_CACHE_SHIFT,
4555 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4558 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4561 btrfs_free_path(path);
4566 * when we free an block, it is possible (and likely) that we free the last
4567 * delayed ref for that extent as well. This searches the delayed ref tree for
4568 * a given extent, and if there are no other delayed refs to be processed, it
4569 * removes it from the tree.
4571 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4572 struct btrfs_root *root, u64 bytenr)
4574 struct btrfs_delayed_ref_head *head;
4575 struct btrfs_delayed_ref_root *delayed_refs;
4576 struct btrfs_delayed_ref_node *ref;
4577 struct rb_node *node;
4580 delayed_refs = &trans->transaction->delayed_refs;
4581 spin_lock(&delayed_refs->lock);
4582 head = btrfs_find_delayed_ref_head(trans, bytenr);
4586 node = rb_prev(&head->node.rb_node);
4590 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4592 /* there are still entries for this ref, we can't drop it */
4593 if (ref->bytenr == bytenr)
4596 if (head->extent_op) {
4597 if (!head->must_insert_reserved)
4599 kfree(head->extent_op);
4600 head->extent_op = NULL;
4604 * waiting for the lock here would deadlock. If someone else has it
4605 * locked they are already in the process of dropping it anyway
4607 if (!mutex_trylock(&head->mutex))
4611 * at this point we have a head with no other entries. Go
4612 * ahead and process it.
4614 head->node.in_tree = 0;
4615 rb_erase(&head->node.rb_node, &delayed_refs->root);
4617 delayed_refs->num_entries--;
4620 * we don't take a ref on the node because we're removing it from the
4621 * tree, so we just steal the ref the tree was holding.
4623 delayed_refs->num_heads--;
4624 if (list_empty(&head->cluster))
4625 delayed_refs->num_heads_ready--;
4627 list_del_init(&head->cluster);
4628 spin_unlock(&delayed_refs->lock);
4630 BUG_ON(head->extent_op);
4631 if (head->must_insert_reserved)
4634 mutex_unlock(&head->mutex);
4635 btrfs_put_delayed_ref(&head->node);
4638 spin_unlock(&delayed_refs->lock);
4642 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4643 struct btrfs_root *root,
4644 struct extent_buffer *buf,
4645 u64 parent, int last_ref)
4647 struct btrfs_block_rsv *block_rsv;
4648 struct btrfs_block_group_cache *cache = NULL;
4651 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4652 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4653 parent, root->root_key.objectid,
4654 btrfs_header_level(buf),
4655 BTRFS_DROP_DELAYED_REF, NULL);
4662 block_rsv = get_block_rsv(trans, root);
4663 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4664 if (block_rsv->space_info != cache->space_info)
4667 if (btrfs_header_generation(buf) == trans->transid) {
4668 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4669 ret = check_ref_cleanup(trans, root, buf->start);
4674 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4675 pin_down_extent(root, cache, buf->start, buf->len, 1);
4679 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4681 btrfs_add_free_space(cache, buf->start, buf->len);
4682 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4683 if (ret == -EAGAIN) {
4684 /* block group became read-only */
4685 update_reserved_bytes(cache, buf->len, 0, 1);
4690 spin_lock(&block_rsv->lock);
4691 if (block_rsv->reserved < block_rsv->size) {
4692 block_rsv->reserved += buf->len;
4695 spin_unlock(&block_rsv->lock);
4698 spin_lock(&cache->space_info->lock);
4699 cache->space_info->bytes_reserved -= buf->len;
4700 spin_unlock(&cache->space_info->lock);
4705 if (block_rsv->durable && !cache->ro) {
4707 spin_lock(&cache->lock);
4709 cache->reserved_pinned += buf->len;
4712 spin_unlock(&cache->lock);
4715 spin_lock(&block_rsv->lock);
4716 block_rsv->freed[trans->transid & 0x1] += buf->len;
4717 spin_unlock(&block_rsv->lock);
4721 btrfs_put_block_group(cache);
4724 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4725 struct btrfs_root *root,
4726 u64 bytenr, u64 num_bytes, u64 parent,
4727 u64 root_objectid, u64 owner, u64 offset)
4732 * tree log blocks never actually go into the extent allocation
4733 * tree, just update pinning info and exit early.
4735 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4736 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4737 /* unlocks the pinned mutex */
4738 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4740 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4741 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4742 parent, root_objectid, (int)owner,
4743 BTRFS_DROP_DELAYED_REF, NULL);
4746 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4747 parent, root_objectid, owner,
4748 offset, BTRFS_DROP_DELAYED_REF, NULL);
4754 static u64 stripe_align(struct btrfs_root *root, u64 val)
4756 u64 mask = ((u64)root->stripesize - 1);
4757 u64 ret = (val + mask) & ~mask;
4762 * when we wait for progress in the block group caching, its because
4763 * our allocation attempt failed at least once. So, we must sleep
4764 * and let some progress happen before we try again.
4766 * This function will sleep at least once waiting for new free space to
4767 * show up, and then it will check the block group free space numbers
4768 * for our min num_bytes. Another option is to have it go ahead
4769 * and look in the rbtree for a free extent of a given size, but this
4773 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4776 struct btrfs_caching_control *caching_ctl;
4779 caching_ctl = get_caching_control(cache);
4783 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4784 (cache->free_space >= num_bytes));
4786 put_caching_control(caching_ctl);
4791 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4793 struct btrfs_caching_control *caching_ctl;
4796 caching_ctl = get_caching_control(cache);
4800 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4802 put_caching_control(caching_ctl);
4806 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4809 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4811 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4813 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4815 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4822 enum btrfs_loop_type {
4823 LOOP_FIND_IDEAL = 0,
4824 LOOP_CACHING_NOWAIT = 1,
4825 LOOP_CACHING_WAIT = 2,
4826 LOOP_ALLOC_CHUNK = 3,
4827 LOOP_NO_EMPTY_SIZE = 4,
4831 * walks the btree of allocated extents and find a hole of a given size.
4832 * The key ins is changed to record the hole:
4833 * ins->objectid == block start
4834 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4835 * ins->offset == number of blocks
4836 * Any available blocks before search_start are skipped.
4838 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4839 struct btrfs_root *orig_root,
4840 u64 num_bytes, u64 empty_size,
4841 u64 search_start, u64 search_end,
4842 u64 hint_byte, struct btrfs_key *ins,
4846 struct btrfs_root *root = orig_root->fs_info->extent_root;
4847 struct btrfs_free_cluster *last_ptr = NULL;
4848 struct btrfs_block_group_cache *block_group = NULL;
4849 int empty_cluster = 2 * 1024 * 1024;
4850 int allowed_chunk_alloc = 0;
4851 int done_chunk_alloc = 0;
4852 struct btrfs_space_info *space_info;
4853 int last_ptr_loop = 0;
4856 bool found_uncached_bg = false;
4857 bool failed_cluster_refill = false;
4858 bool failed_alloc = false;
4859 bool use_cluster = true;
4860 u64 ideal_cache_percent = 0;
4861 u64 ideal_cache_offset = 0;
4863 WARN_ON(num_bytes < root->sectorsize);
4864 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4868 space_info = __find_space_info(root->fs_info, data);
4870 printk(KERN_ERR "No space info for %d\n", data);
4875 * If the space info is for both data and metadata it means we have a
4876 * small filesystem and we can't use the clustering stuff.
4878 if (btrfs_mixed_space_info(space_info))
4879 use_cluster = false;
4881 if (orig_root->ref_cows || empty_size)
4882 allowed_chunk_alloc = 1;
4884 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4885 last_ptr = &root->fs_info->meta_alloc_cluster;
4886 if (!btrfs_test_opt(root, SSD))
4887 empty_cluster = 64 * 1024;
4890 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4891 btrfs_test_opt(root, SSD)) {
4892 last_ptr = &root->fs_info->data_alloc_cluster;
4896 spin_lock(&last_ptr->lock);
4897 if (last_ptr->block_group)
4898 hint_byte = last_ptr->window_start;
4899 spin_unlock(&last_ptr->lock);
4902 search_start = max(search_start, first_logical_byte(root, 0));
4903 search_start = max(search_start, hint_byte);
4908 if (search_start == hint_byte) {
4910 block_group = btrfs_lookup_block_group(root->fs_info,
4913 * we don't want to use the block group if it doesn't match our
4914 * allocation bits, or if its not cached.
4916 * However if we are re-searching with an ideal block group
4917 * picked out then we don't care that the block group is cached.
4919 if (block_group && block_group_bits(block_group, data) &&
4920 (block_group->cached != BTRFS_CACHE_NO ||
4921 search_start == ideal_cache_offset)) {
4922 down_read(&space_info->groups_sem);
4923 if (list_empty(&block_group->list) ||
4926 * someone is removing this block group,
4927 * we can't jump into the have_block_group
4928 * target because our list pointers are not
4931 btrfs_put_block_group(block_group);
4932 up_read(&space_info->groups_sem);
4934 index = get_block_group_index(block_group);
4935 goto have_block_group;
4937 } else if (block_group) {
4938 btrfs_put_block_group(block_group);
4942 down_read(&space_info->groups_sem);
4943 list_for_each_entry(block_group, &space_info->block_groups[index],
4948 btrfs_get_block_group(block_group);
4949 search_start = block_group->key.objectid;
4952 * this can happen if we end up cycling through all the
4953 * raid types, but we want to make sure we only allocate
4954 * for the proper type.
4956 if (!block_group_bits(block_group, data)) {
4957 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4958 BTRFS_BLOCK_GROUP_RAID1 |
4959 BTRFS_BLOCK_GROUP_RAID10;
4962 * if they asked for extra copies and this block group
4963 * doesn't provide them, bail. This does allow us to
4964 * fill raid0 from raid1.
4966 if ((data & extra) && !(block_group->flags & extra))
4971 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4974 ret = cache_block_group(block_group, trans,
4976 if (block_group->cached == BTRFS_CACHE_FINISHED)
4977 goto have_block_group;
4979 free_percent = btrfs_block_group_used(&block_group->item);
4980 free_percent *= 100;
4981 free_percent = div64_u64(free_percent,
4982 block_group->key.offset);
4983 free_percent = 100 - free_percent;
4984 if (free_percent > ideal_cache_percent &&
4985 likely(!block_group->ro)) {
4986 ideal_cache_offset = block_group->key.objectid;
4987 ideal_cache_percent = free_percent;
4991 * We only want to start kthread caching if we are at
4992 * the point where we will wait for caching to make
4993 * progress, or if our ideal search is over and we've
4994 * found somebody to start caching.
4996 if (loop > LOOP_CACHING_NOWAIT ||
4997 (loop > LOOP_FIND_IDEAL &&
4998 atomic_read(&space_info->caching_threads) < 2)) {
4999 ret = cache_block_group(block_group, trans,
5003 found_uncached_bg = true;
5006 * If loop is set for cached only, try the next block
5009 if (loop == LOOP_FIND_IDEAL)
5013 cached = block_group_cache_done(block_group);
5014 if (unlikely(!cached))
5015 found_uncached_bg = true;
5017 if (unlikely(block_group->ro))
5021 * Ok we want to try and use the cluster allocator, so lets look
5022 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5023 * have tried the cluster allocator plenty of times at this
5024 * point and not have found anything, so we are likely way too
5025 * fragmented for the clustering stuff to find anything, so lets
5026 * just skip it and let the allocator find whatever block it can
5029 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5031 * the refill lock keeps out other
5032 * people trying to start a new cluster
5034 spin_lock(&last_ptr->refill_lock);
5035 if (last_ptr->block_group &&
5036 (last_ptr->block_group->ro ||
5037 !block_group_bits(last_ptr->block_group, data))) {
5039 goto refill_cluster;
5042 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5043 num_bytes, search_start);
5045 /* we have a block, we're done */
5046 spin_unlock(&last_ptr->refill_lock);
5050 spin_lock(&last_ptr->lock);
5052 * whoops, this cluster doesn't actually point to
5053 * this block group. Get a ref on the block
5054 * group is does point to and try again
5056 if (!last_ptr_loop && last_ptr->block_group &&
5057 last_ptr->block_group != block_group) {
5059 btrfs_put_block_group(block_group);
5060 block_group = last_ptr->block_group;
5061 btrfs_get_block_group(block_group);
5062 spin_unlock(&last_ptr->lock);
5063 spin_unlock(&last_ptr->refill_lock);
5066 search_start = block_group->key.objectid;
5068 * we know this block group is properly
5069 * in the list because
5070 * btrfs_remove_block_group, drops the
5071 * cluster before it removes the block
5072 * group from the list
5074 goto have_block_group;
5076 spin_unlock(&last_ptr->lock);
5079 * this cluster didn't work out, free it and
5082 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5086 /* allocate a cluster in this block group */
5087 ret = btrfs_find_space_cluster(trans, root,
5088 block_group, last_ptr,
5090 empty_cluster + empty_size);
5093 * now pull our allocation out of this
5096 offset = btrfs_alloc_from_cluster(block_group,
5097 last_ptr, num_bytes,
5100 /* we found one, proceed */
5101 spin_unlock(&last_ptr->refill_lock);
5104 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5105 && !failed_cluster_refill) {
5106 spin_unlock(&last_ptr->refill_lock);
5108 failed_cluster_refill = true;
5109 wait_block_group_cache_progress(block_group,
5110 num_bytes + empty_cluster + empty_size);
5111 goto have_block_group;
5115 * at this point we either didn't find a cluster
5116 * or we weren't able to allocate a block from our
5117 * cluster. Free the cluster we've been trying
5118 * to use, and go to the next block group
5120 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5121 spin_unlock(&last_ptr->refill_lock);
5125 offset = btrfs_find_space_for_alloc(block_group, search_start,
5126 num_bytes, empty_size);
5128 * If we didn't find a chunk, and we haven't failed on this
5129 * block group before, and this block group is in the middle of
5130 * caching and we are ok with waiting, then go ahead and wait
5131 * for progress to be made, and set failed_alloc to true.
5133 * If failed_alloc is true then we've already waited on this
5134 * block group once and should move on to the next block group.
5136 if (!offset && !failed_alloc && !cached &&
5137 loop > LOOP_CACHING_NOWAIT) {
5138 wait_block_group_cache_progress(block_group,
5139 num_bytes + empty_size);
5140 failed_alloc = true;
5141 goto have_block_group;
5142 } else if (!offset) {
5146 search_start = stripe_align(root, offset);
5147 /* move on to the next group */
5148 if (search_start + num_bytes >= search_end) {
5149 btrfs_add_free_space(block_group, offset, num_bytes);
5153 /* move on to the next group */
5154 if (search_start + num_bytes >
5155 block_group->key.objectid + block_group->key.offset) {
5156 btrfs_add_free_space(block_group, offset, num_bytes);
5160 ins->objectid = search_start;
5161 ins->offset = num_bytes;
5163 if (offset < search_start)
5164 btrfs_add_free_space(block_group, offset,
5165 search_start - offset);
5166 BUG_ON(offset > search_start);
5168 ret = update_reserved_bytes(block_group, num_bytes, 1,
5169 (data & BTRFS_BLOCK_GROUP_DATA));
5170 if (ret == -EAGAIN) {
5171 btrfs_add_free_space(block_group, offset, num_bytes);
5175 /* we are all good, lets return */
5176 ins->objectid = search_start;
5177 ins->offset = num_bytes;
5179 if (offset < search_start)
5180 btrfs_add_free_space(block_group, offset,
5181 search_start - offset);
5182 BUG_ON(offset > search_start);
5185 failed_cluster_refill = false;
5186 failed_alloc = false;
5187 BUG_ON(index != get_block_group_index(block_group));
5188 btrfs_put_block_group(block_group);
5190 up_read(&space_info->groups_sem);
5192 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5195 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5196 * for them to make caching progress. Also
5197 * determine the best possible bg to cache
5198 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5199 * caching kthreads as we move along
5200 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5201 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5202 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5205 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5206 (found_uncached_bg || empty_size || empty_cluster ||
5207 allowed_chunk_alloc)) {
5209 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5210 found_uncached_bg = false;
5212 if (!ideal_cache_percent &&
5213 atomic_read(&space_info->caching_threads))
5217 * 1 of the following 2 things have happened so far
5219 * 1) We found an ideal block group for caching that
5220 * is mostly full and will cache quickly, so we might
5221 * as well wait for it.
5223 * 2) We searched for cached only and we didn't find
5224 * anything, and we didn't start any caching kthreads
5225 * either, so chances are we will loop through and
5226 * start a couple caching kthreads, and then come back
5227 * around and just wait for them. This will be slower
5228 * because we will have 2 caching kthreads reading at
5229 * the same time when we could have just started one
5230 * and waited for it to get far enough to give us an
5231 * allocation, so go ahead and go to the wait caching
5234 loop = LOOP_CACHING_WAIT;
5235 search_start = ideal_cache_offset;
5236 ideal_cache_percent = 0;
5238 } else if (loop == LOOP_FIND_IDEAL) {
5240 * Didn't find a uncached bg, wait on anything we find
5243 loop = LOOP_CACHING_WAIT;
5247 if (loop < LOOP_CACHING_WAIT) {
5252 if (loop == LOOP_ALLOC_CHUNK) {
5257 if (allowed_chunk_alloc) {
5258 ret = do_chunk_alloc(trans, root, num_bytes +
5259 2 * 1024 * 1024, data, 1);
5260 allowed_chunk_alloc = 0;
5261 done_chunk_alloc = 1;
5262 } else if (!done_chunk_alloc) {
5263 space_info->force_alloc = 1;
5266 if (loop < LOOP_NO_EMPTY_SIZE) {
5271 } else if (!ins->objectid) {
5275 /* we found what we needed */
5276 if (ins->objectid) {
5277 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5278 trans->block_group = block_group->key.objectid;
5280 btrfs_put_block_group(block_group);
5287 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5288 int dump_block_groups)
5290 struct btrfs_block_group_cache *cache;
5293 spin_lock(&info->lock);
5294 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5295 (unsigned long long)(info->total_bytes - info->bytes_used -
5296 info->bytes_pinned - info->bytes_reserved -
5297 info->bytes_readonly),
5298 (info->full) ? "" : "not ");
5299 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5300 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5301 (unsigned long long)info->total_bytes,
5302 (unsigned long long)info->bytes_used,
5303 (unsigned long long)info->bytes_pinned,
5304 (unsigned long long)info->bytes_reserved,
5305 (unsigned long long)info->bytes_may_use,
5306 (unsigned long long)info->bytes_readonly);
5307 spin_unlock(&info->lock);
5309 if (!dump_block_groups)
5312 down_read(&info->groups_sem);
5314 list_for_each_entry(cache, &info->block_groups[index], list) {
5315 spin_lock(&cache->lock);
5316 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5317 "%llu pinned %llu reserved\n",
5318 (unsigned long long)cache->key.objectid,
5319 (unsigned long long)cache->key.offset,
5320 (unsigned long long)btrfs_block_group_used(&cache->item),
5321 (unsigned long long)cache->pinned,
5322 (unsigned long long)cache->reserved);
5323 btrfs_dump_free_space(cache, bytes);
5324 spin_unlock(&cache->lock);
5326 if (++index < BTRFS_NR_RAID_TYPES)
5328 up_read(&info->groups_sem);
5331 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5332 struct btrfs_root *root,
5333 u64 num_bytes, u64 min_alloc_size,
5334 u64 empty_size, u64 hint_byte,
5335 u64 search_end, struct btrfs_key *ins,
5339 u64 search_start = 0;
5341 data = btrfs_get_alloc_profile(root, data);
5344 * the only place that sets empty_size is btrfs_realloc_node, which
5345 * is not called recursively on allocations
5347 if (empty_size || root->ref_cows)
5348 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5349 num_bytes + 2 * 1024 * 1024, data, 0);
5351 WARN_ON(num_bytes < root->sectorsize);
5352 ret = find_free_extent(trans, root, num_bytes, empty_size,
5353 search_start, search_end, hint_byte,
5356 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5357 num_bytes = num_bytes >> 1;
5358 num_bytes = num_bytes & ~(root->sectorsize - 1);
5359 num_bytes = max(num_bytes, min_alloc_size);
5360 do_chunk_alloc(trans, root->fs_info->extent_root,
5361 num_bytes, data, 1);
5364 if (ret == -ENOSPC) {
5365 struct btrfs_space_info *sinfo;
5367 sinfo = __find_space_info(root->fs_info, data);
5368 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5369 "wanted %llu\n", (unsigned long long)data,
5370 (unsigned long long)num_bytes);
5371 dump_space_info(sinfo, num_bytes, 1);
5377 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5379 struct btrfs_block_group_cache *cache;
5382 cache = btrfs_lookup_block_group(root->fs_info, start);
5384 printk(KERN_ERR "Unable to find block group for %llu\n",
5385 (unsigned long long)start);
5389 ret = btrfs_discard_extent(root, start, len);
5391 btrfs_add_free_space(cache, start, len);
5392 update_reserved_bytes(cache, len, 0, 1);
5393 btrfs_put_block_group(cache);
5398 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5399 struct btrfs_root *root,
5400 u64 parent, u64 root_objectid,
5401 u64 flags, u64 owner, u64 offset,
5402 struct btrfs_key *ins, int ref_mod)
5405 struct btrfs_fs_info *fs_info = root->fs_info;
5406 struct btrfs_extent_item *extent_item;
5407 struct btrfs_extent_inline_ref *iref;
5408 struct btrfs_path *path;
5409 struct extent_buffer *leaf;
5414 type = BTRFS_SHARED_DATA_REF_KEY;
5416 type = BTRFS_EXTENT_DATA_REF_KEY;
5418 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5420 path = btrfs_alloc_path();
5423 path->leave_spinning = 1;
5424 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5428 leaf = path->nodes[0];
5429 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5430 struct btrfs_extent_item);
5431 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5432 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5433 btrfs_set_extent_flags(leaf, extent_item,
5434 flags | BTRFS_EXTENT_FLAG_DATA);
5436 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5437 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5439 struct btrfs_shared_data_ref *ref;
5440 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5441 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5442 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5444 struct btrfs_extent_data_ref *ref;
5445 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5446 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5447 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5448 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5449 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5452 btrfs_mark_buffer_dirty(path->nodes[0]);
5453 btrfs_free_path(path);
5455 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5457 printk(KERN_ERR "btrfs update block group failed for %llu "
5458 "%llu\n", (unsigned long long)ins->objectid,
5459 (unsigned long long)ins->offset);
5465 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5466 struct btrfs_root *root,
5467 u64 parent, u64 root_objectid,
5468 u64 flags, struct btrfs_disk_key *key,
5469 int level, struct btrfs_key *ins)
5472 struct btrfs_fs_info *fs_info = root->fs_info;
5473 struct btrfs_extent_item *extent_item;
5474 struct btrfs_tree_block_info *block_info;
5475 struct btrfs_extent_inline_ref *iref;
5476 struct btrfs_path *path;
5477 struct extent_buffer *leaf;
5478 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5480 path = btrfs_alloc_path();
5483 path->leave_spinning = 1;
5484 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5488 leaf = path->nodes[0];
5489 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5490 struct btrfs_extent_item);
5491 btrfs_set_extent_refs(leaf, extent_item, 1);
5492 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5493 btrfs_set_extent_flags(leaf, extent_item,
5494 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5495 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5497 btrfs_set_tree_block_key(leaf, block_info, key);
5498 btrfs_set_tree_block_level(leaf, block_info, level);
5500 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5502 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5503 btrfs_set_extent_inline_ref_type(leaf, iref,
5504 BTRFS_SHARED_BLOCK_REF_KEY);
5505 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5507 btrfs_set_extent_inline_ref_type(leaf, iref,
5508 BTRFS_TREE_BLOCK_REF_KEY);
5509 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5512 btrfs_mark_buffer_dirty(leaf);
5513 btrfs_free_path(path);
5515 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5517 printk(KERN_ERR "btrfs update block group failed for %llu "
5518 "%llu\n", (unsigned long long)ins->objectid,
5519 (unsigned long long)ins->offset);
5525 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5526 struct btrfs_root *root,
5527 u64 root_objectid, u64 owner,
5528 u64 offset, struct btrfs_key *ins)
5532 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5534 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5535 0, root_objectid, owner, offset,
5536 BTRFS_ADD_DELAYED_EXTENT, NULL);
5541 * this is used by the tree logging recovery code. It records that
5542 * an extent has been allocated and makes sure to clear the free
5543 * space cache bits as well
5545 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5546 struct btrfs_root *root,
5547 u64 root_objectid, u64 owner, u64 offset,
5548 struct btrfs_key *ins)
5551 struct btrfs_block_group_cache *block_group;
5552 struct btrfs_caching_control *caching_ctl;
5553 u64 start = ins->objectid;
5554 u64 num_bytes = ins->offset;
5556 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5557 cache_block_group(block_group, trans, NULL, 0);
5558 caching_ctl = get_caching_control(block_group);
5561 BUG_ON(!block_group_cache_done(block_group));
5562 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5565 mutex_lock(&caching_ctl->mutex);
5567 if (start >= caching_ctl->progress) {
5568 ret = add_excluded_extent(root, start, num_bytes);
5570 } else if (start + num_bytes <= caching_ctl->progress) {
5571 ret = btrfs_remove_free_space(block_group,
5575 num_bytes = caching_ctl->progress - start;
5576 ret = btrfs_remove_free_space(block_group,
5580 start = caching_ctl->progress;
5581 num_bytes = ins->objectid + ins->offset -
5582 caching_ctl->progress;
5583 ret = add_excluded_extent(root, start, num_bytes);
5587 mutex_unlock(&caching_ctl->mutex);
5588 put_caching_control(caching_ctl);
5591 ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5593 btrfs_put_block_group(block_group);
5594 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5595 0, owner, offset, ins, 1);
5599 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5600 struct btrfs_root *root,
5601 u64 bytenr, u32 blocksize,
5604 struct extent_buffer *buf;
5606 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5608 return ERR_PTR(-ENOMEM);
5609 btrfs_set_header_generation(buf, trans->transid);
5610 btrfs_set_buffer_lockdep_class(buf, level);
5611 btrfs_tree_lock(buf);
5612 clean_tree_block(trans, root, buf);
5614 btrfs_set_lock_blocking(buf);
5615 btrfs_set_buffer_uptodate(buf);
5617 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5619 * we allow two log transactions at a time, use different
5620 * EXENT bit to differentiate dirty pages.
5622 if (root->log_transid % 2 == 0)
5623 set_extent_dirty(&root->dirty_log_pages, buf->start,
5624 buf->start + buf->len - 1, GFP_NOFS);
5626 set_extent_new(&root->dirty_log_pages, buf->start,
5627 buf->start + buf->len - 1, GFP_NOFS);
5629 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5630 buf->start + buf->len - 1, GFP_NOFS);
5632 trans->blocks_used++;
5633 /* this returns a buffer locked for blocking */
5637 static struct btrfs_block_rsv *
5638 use_block_rsv(struct btrfs_trans_handle *trans,
5639 struct btrfs_root *root, u32 blocksize)
5641 struct btrfs_block_rsv *block_rsv;
5644 block_rsv = get_block_rsv(trans, root);
5646 if (block_rsv->size == 0) {
5647 ret = reserve_metadata_bytes(trans, root, block_rsv,
5650 return ERR_PTR(ret);
5654 ret = block_rsv_use_bytes(block_rsv, blocksize);
5658 return ERR_PTR(-ENOSPC);
5661 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5663 block_rsv_add_bytes(block_rsv, blocksize, 0);
5664 block_rsv_release_bytes(block_rsv, NULL, 0);
5668 * finds a free extent and does all the dirty work required for allocation
5669 * returns the key for the extent through ins, and a tree buffer for
5670 * the first block of the extent through buf.
5672 * returns the tree buffer or NULL.
5674 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5675 struct btrfs_root *root, u32 blocksize,
5676 u64 parent, u64 root_objectid,
5677 struct btrfs_disk_key *key, int level,
5678 u64 hint, u64 empty_size)
5680 struct btrfs_key ins;
5681 struct btrfs_block_rsv *block_rsv;
5682 struct extent_buffer *buf;
5687 block_rsv = use_block_rsv(trans, root, blocksize);
5688 if (IS_ERR(block_rsv))
5689 return ERR_CAST(block_rsv);
5691 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5692 empty_size, hint, (u64)-1, &ins, 0);
5694 unuse_block_rsv(block_rsv, blocksize);
5695 return ERR_PTR(ret);
5698 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5700 BUG_ON(IS_ERR(buf));
5702 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5704 parent = ins.objectid;
5705 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5709 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5710 struct btrfs_delayed_extent_op *extent_op;
5711 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5714 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5716 memset(&extent_op->key, 0, sizeof(extent_op->key));
5717 extent_op->flags_to_set = flags;
5718 extent_op->update_key = 1;
5719 extent_op->update_flags = 1;
5720 extent_op->is_data = 0;
5722 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5723 ins.offset, parent, root_objectid,
5724 level, BTRFS_ADD_DELAYED_EXTENT,
5731 struct walk_control {
5732 u64 refs[BTRFS_MAX_LEVEL];
5733 u64 flags[BTRFS_MAX_LEVEL];
5734 struct btrfs_key update_progress;
5744 #define DROP_REFERENCE 1
5745 #define UPDATE_BACKREF 2
5747 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5748 struct btrfs_root *root,
5749 struct walk_control *wc,
5750 struct btrfs_path *path)
5758 struct btrfs_key key;
5759 struct extent_buffer *eb;
5764 if (path->slots[wc->level] < wc->reada_slot) {
5765 wc->reada_count = wc->reada_count * 2 / 3;
5766 wc->reada_count = max(wc->reada_count, 2);
5768 wc->reada_count = wc->reada_count * 3 / 2;
5769 wc->reada_count = min_t(int, wc->reada_count,
5770 BTRFS_NODEPTRS_PER_BLOCK(root));
5773 eb = path->nodes[wc->level];
5774 nritems = btrfs_header_nritems(eb);
5775 blocksize = btrfs_level_size(root, wc->level - 1);
5777 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5778 if (nread >= wc->reada_count)
5782 bytenr = btrfs_node_blockptr(eb, slot);
5783 generation = btrfs_node_ptr_generation(eb, slot);
5785 if (slot == path->slots[wc->level])
5788 if (wc->stage == UPDATE_BACKREF &&
5789 generation <= root->root_key.offset)
5792 /* We don't lock the tree block, it's OK to be racy here */
5793 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5798 if (wc->stage == DROP_REFERENCE) {
5802 if (wc->level == 1 &&
5803 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5805 if (!wc->update_ref ||
5806 generation <= root->root_key.offset)
5808 btrfs_node_key_to_cpu(eb, &key, slot);
5809 ret = btrfs_comp_cpu_keys(&key,
5810 &wc->update_progress);
5814 if (wc->level == 1 &&
5815 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5819 ret = readahead_tree_block(root, bytenr, blocksize,
5825 wc->reada_slot = slot;
5829 * hepler to process tree block while walking down the tree.
5831 * when wc->stage == UPDATE_BACKREF, this function updates
5832 * back refs for pointers in the block.
5834 * NOTE: return value 1 means we should stop walking down.
5836 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5837 struct btrfs_root *root,
5838 struct btrfs_path *path,
5839 struct walk_control *wc, int lookup_info)
5841 int level = wc->level;
5842 struct extent_buffer *eb = path->nodes[level];
5843 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5846 if (wc->stage == UPDATE_BACKREF &&
5847 btrfs_header_owner(eb) != root->root_key.objectid)
5851 * when reference count of tree block is 1, it won't increase
5852 * again. once full backref flag is set, we never clear it.
5855 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5856 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5857 BUG_ON(!path->locks[level]);
5858 ret = btrfs_lookup_extent_info(trans, root,
5863 BUG_ON(wc->refs[level] == 0);
5866 if (wc->stage == DROP_REFERENCE) {
5867 if (wc->refs[level] > 1)
5870 if (path->locks[level] && !wc->keep_locks) {
5871 btrfs_tree_unlock(eb);
5872 path->locks[level] = 0;
5877 /* wc->stage == UPDATE_BACKREF */
5878 if (!(wc->flags[level] & flag)) {
5879 BUG_ON(!path->locks[level]);
5880 ret = btrfs_inc_ref(trans, root, eb, 1);
5882 ret = btrfs_dec_ref(trans, root, eb, 0);
5884 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5887 wc->flags[level] |= flag;
5891 * the block is shared by multiple trees, so it's not good to
5892 * keep the tree lock
5894 if (path->locks[level] && level > 0) {
5895 btrfs_tree_unlock(eb);
5896 path->locks[level] = 0;
5902 * hepler to process tree block pointer.
5904 * when wc->stage == DROP_REFERENCE, this function checks
5905 * reference count of the block pointed to. if the block
5906 * is shared and we need update back refs for the subtree
5907 * rooted at the block, this function changes wc->stage to
5908 * UPDATE_BACKREF. if the block is shared and there is no
5909 * need to update back, this function drops the reference
5912 * NOTE: return value 1 means we should stop walking down.
5914 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5915 struct btrfs_root *root,
5916 struct btrfs_path *path,
5917 struct walk_control *wc, int *lookup_info)
5923 struct btrfs_key key;
5924 struct extent_buffer *next;
5925 int level = wc->level;
5929 generation = btrfs_node_ptr_generation(path->nodes[level],
5930 path->slots[level]);
5932 * if the lower level block was created before the snapshot
5933 * was created, we know there is no need to update back refs
5936 if (wc->stage == UPDATE_BACKREF &&
5937 generation <= root->root_key.offset) {
5942 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5943 blocksize = btrfs_level_size(root, level - 1);
5945 next = btrfs_find_tree_block(root, bytenr, blocksize);
5947 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5952 btrfs_tree_lock(next);
5953 btrfs_set_lock_blocking(next);
5955 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5956 &wc->refs[level - 1],
5957 &wc->flags[level - 1]);
5959 BUG_ON(wc->refs[level - 1] == 0);
5962 if (wc->stage == DROP_REFERENCE) {
5963 if (wc->refs[level - 1] > 1) {
5965 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5968 if (!wc->update_ref ||
5969 generation <= root->root_key.offset)
5972 btrfs_node_key_to_cpu(path->nodes[level], &key,
5973 path->slots[level]);
5974 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5978 wc->stage = UPDATE_BACKREF;
5979 wc->shared_level = level - 1;
5983 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5987 if (!btrfs_buffer_uptodate(next, generation)) {
5988 btrfs_tree_unlock(next);
5989 free_extent_buffer(next);
5995 if (reada && level == 1)
5996 reada_walk_down(trans, root, wc, path);
5997 next = read_tree_block(root, bytenr, blocksize, generation);
5998 btrfs_tree_lock(next);
5999 btrfs_set_lock_blocking(next);
6003 BUG_ON(level != btrfs_header_level(next));
6004 path->nodes[level] = next;
6005 path->slots[level] = 0;
6006 path->locks[level] = 1;
6012 wc->refs[level - 1] = 0;
6013 wc->flags[level - 1] = 0;
6014 if (wc->stage == DROP_REFERENCE) {
6015 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6016 parent = path->nodes[level]->start;
6018 BUG_ON(root->root_key.objectid !=
6019 btrfs_header_owner(path->nodes[level]));
6023 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6024 root->root_key.objectid, level - 1, 0);
6027 btrfs_tree_unlock(next);
6028 free_extent_buffer(next);
6034 * hepler to process tree block while walking up the tree.
6036 * when wc->stage == DROP_REFERENCE, this function drops
6037 * reference count on the block.
6039 * when wc->stage == UPDATE_BACKREF, this function changes
6040 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6041 * to UPDATE_BACKREF previously while processing the block.
6043 * NOTE: return value 1 means we should stop walking up.
6045 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6046 struct btrfs_root *root,
6047 struct btrfs_path *path,
6048 struct walk_control *wc)
6051 int level = wc->level;
6052 struct extent_buffer *eb = path->nodes[level];
6055 if (wc->stage == UPDATE_BACKREF) {
6056 BUG_ON(wc->shared_level < level);
6057 if (level < wc->shared_level)
6060 ret = find_next_key(path, level + 1, &wc->update_progress);
6064 wc->stage = DROP_REFERENCE;
6065 wc->shared_level = -1;
6066 path->slots[level] = 0;
6069 * check reference count again if the block isn't locked.
6070 * we should start walking down the tree again if reference
6073 if (!path->locks[level]) {
6075 btrfs_tree_lock(eb);
6076 btrfs_set_lock_blocking(eb);
6077 path->locks[level] = 1;
6079 ret = btrfs_lookup_extent_info(trans, root,
6084 BUG_ON(wc->refs[level] == 0);
6085 if (wc->refs[level] == 1) {
6086 btrfs_tree_unlock(eb);
6087 path->locks[level] = 0;
6093 /* wc->stage == DROP_REFERENCE */
6094 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6096 if (wc->refs[level] == 1) {
6098 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6099 ret = btrfs_dec_ref(trans, root, eb, 1);
6101 ret = btrfs_dec_ref(trans, root, eb, 0);
6104 /* make block locked assertion in clean_tree_block happy */
6105 if (!path->locks[level] &&
6106 btrfs_header_generation(eb) == trans->transid) {
6107 btrfs_tree_lock(eb);
6108 btrfs_set_lock_blocking(eb);
6109 path->locks[level] = 1;
6111 clean_tree_block(trans, root, eb);
6114 if (eb == root->node) {
6115 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6118 BUG_ON(root->root_key.objectid !=
6119 btrfs_header_owner(eb));
6121 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6122 parent = path->nodes[level + 1]->start;
6124 BUG_ON(root->root_key.objectid !=
6125 btrfs_header_owner(path->nodes[level + 1]));
6128 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6130 wc->refs[level] = 0;
6131 wc->flags[level] = 0;
6135 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6136 struct btrfs_root *root,
6137 struct btrfs_path *path,
6138 struct walk_control *wc)
6140 int level = wc->level;
6141 int lookup_info = 1;
6144 while (level >= 0) {
6145 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6152 if (path->slots[level] >=
6153 btrfs_header_nritems(path->nodes[level]))
6156 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6158 path->slots[level]++;
6167 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6168 struct btrfs_root *root,
6169 struct btrfs_path *path,
6170 struct walk_control *wc, int max_level)
6172 int level = wc->level;
6175 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6176 while (level < max_level && path->nodes[level]) {
6178 if (path->slots[level] + 1 <
6179 btrfs_header_nritems(path->nodes[level])) {
6180 path->slots[level]++;
6183 ret = walk_up_proc(trans, root, path, wc);
6187 if (path->locks[level]) {
6188 btrfs_tree_unlock(path->nodes[level]);
6189 path->locks[level] = 0;
6191 free_extent_buffer(path->nodes[level]);
6192 path->nodes[level] = NULL;
6200 * drop a subvolume tree.
6202 * this function traverses the tree freeing any blocks that only
6203 * referenced by the tree.
6205 * when a shared tree block is found. this function decreases its
6206 * reference count by one. if update_ref is true, this function
6207 * also make sure backrefs for the shared block and all lower level
6208 * blocks are properly updated.
6210 int btrfs_drop_snapshot(struct btrfs_root *root,
6211 struct btrfs_block_rsv *block_rsv, int update_ref)
6213 struct btrfs_path *path;
6214 struct btrfs_trans_handle *trans;
6215 struct btrfs_root *tree_root = root->fs_info->tree_root;
6216 struct btrfs_root_item *root_item = &root->root_item;
6217 struct walk_control *wc;
6218 struct btrfs_key key;
6223 path = btrfs_alloc_path();
6226 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6229 trans = btrfs_start_transaction(tree_root, 0);
6231 trans->block_rsv = block_rsv;
6233 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6234 level = btrfs_header_level(root->node);
6235 path->nodes[level] = btrfs_lock_root_node(root);
6236 btrfs_set_lock_blocking(path->nodes[level]);
6237 path->slots[level] = 0;
6238 path->locks[level] = 1;
6239 memset(&wc->update_progress, 0,
6240 sizeof(wc->update_progress));
6242 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6243 memcpy(&wc->update_progress, &key,
6244 sizeof(wc->update_progress));
6246 level = root_item->drop_level;
6248 path->lowest_level = level;
6249 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6250 path->lowest_level = 0;
6258 * unlock our path, this is safe because only this
6259 * function is allowed to delete this snapshot
6261 btrfs_unlock_up_safe(path, 0);
6263 level = btrfs_header_level(root->node);
6265 btrfs_tree_lock(path->nodes[level]);
6266 btrfs_set_lock_blocking(path->nodes[level]);
6268 ret = btrfs_lookup_extent_info(trans, root,
6269 path->nodes[level]->start,
6270 path->nodes[level]->len,
6274 BUG_ON(wc->refs[level] == 0);
6276 if (level == root_item->drop_level)
6279 btrfs_tree_unlock(path->nodes[level]);
6280 WARN_ON(wc->refs[level] != 1);
6286 wc->shared_level = -1;
6287 wc->stage = DROP_REFERENCE;
6288 wc->update_ref = update_ref;
6290 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6293 ret = walk_down_tree(trans, root, path, wc);
6299 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6306 BUG_ON(wc->stage != DROP_REFERENCE);
6310 if (wc->stage == DROP_REFERENCE) {
6312 btrfs_node_key(path->nodes[level],
6313 &root_item->drop_progress,
6314 path->slots[level]);
6315 root_item->drop_level = level;
6318 BUG_ON(wc->level == 0);
6319 if (btrfs_should_end_transaction(trans, tree_root)) {
6320 ret = btrfs_update_root(trans, tree_root,
6325 btrfs_end_transaction_throttle(trans, tree_root);
6326 trans = btrfs_start_transaction(tree_root, 0);
6328 trans->block_rsv = block_rsv;
6331 btrfs_release_path(root, path);
6334 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6337 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6338 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6342 /* if we fail to delete the orphan item this time
6343 * around, it'll get picked up the next time.
6345 * The most common failure here is just -ENOENT.
6347 btrfs_del_orphan_item(trans, tree_root,
6348 root->root_key.objectid);
6352 if (root->in_radix) {
6353 btrfs_free_fs_root(tree_root->fs_info, root);
6355 free_extent_buffer(root->node);
6356 free_extent_buffer(root->commit_root);
6360 btrfs_end_transaction_throttle(trans, tree_root);
6362 btrfs_free_path(path);
6367 * drop subtree rooted at tree block 'node'.
6369 * NOTE: this function will unlock and release tree block 'node'
6371 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6372 struct btrfs_root *root,
6373 struct extent_buffer *node,
6374 struct extent_buffer *parent)
6376 struct btrfs_path *path;
6377 struct walk_control *wc;
6383 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6385 path = btrfs_alloc_path();
6388 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6391 btrfs_assert_tree_locked(parent);
6392 parent_level = btrfs_header_level(parent);
6393 extent_buffer_get(parent);
6394 path->nodes[parent_level] = parent;
6395 path->slots[parent_level] = btrfs_header_nritems(parent);
6397 btrfs_assert_tree_locked(node);
6398 level = btrfs_header_level(node);
6399 path->nodes[level] = node;
6400 path->slots[level] = 0;
6401 path->locks[level] = 1;
6403 wc->refs[parent_level] = 1;
6404 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6406 wc->shared_level = -1;
6407 wc->stage = DROP_REFERENCE;
6410 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6413 wret = walk_down_tree(trans, root, path, wc);
6419 wret = walk_up_tree(trans, root, path, wc, parent_level);
6427 btrfs_free_path(path);
6432 static unsigned long calc_ra(unsigned long start, unsigned long last,
6435 return min(last, start + nr - 1);
6438 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6443 unsigned long first_index;
6444 unsigned long last_index;
6447 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6448 struct file_ra_state *ra;
6449 struct btrfs_ordered_extent *ordered;
6450 unsigned int total_read = 0;
6451 unsigned int total_dirty = 0;
6454 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6456 mutex_lock(&inode->i_mutex);
6457 first_index = start >> PAGE_CACHE_SHIFT;
6458 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6460 /* make sure the dirty trick played by the caller work */
6461 ret = invalidate_inode_pages2_range(inode->i_mapping,
6462 first_index, last_index);
6466 file_ra_state_init(ra, inode->i_mapping);
6468 for (i = first_index ; i <= last_index; i++) {
6469 if (total_read % ra->ra_pages == 0) {
6470 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6471 calc_ra(i, last_index, ra->ra_pages));
6475 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6477 page = grab_cache_page(inode->i_mapping, i);
6482 if (!PageUptodate(page)) {
6483 btrfs_readpage(NULL, page);
6485 if (!PageUptodate(page)) {
6487 page_cache_release(page);
6492 wait_on_page_writeback(page);
6494 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6495 page_end = page_start + PAGE_CACHE_SIZE - 1;
6496 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6498 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6500 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6502 page_cache_release(page);
6503 btrfs_start_ordered_extent(inode, ordered, 1);
6504 btrfs_put_ordered_extent(ordered);
6507 set_page_extent_mapped(page);
6509 if (i == first_index)
6510 set_extent_bits(io_tree, page_start, page_end,
6511 EXTENT_BOUNDARY, GFP_NOFS);
6512 btrfs_set_extent_delalloc(inode, page_start, page_end);
6514 set_page_dirty(page);
6517 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6519 page_cache_release(page);
6524 mutex_unlock(&inode->i_mutex);
6525 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6529 static noinline int relocate_data_extent(struct inode *reloc_inode,
6530 struct btrfs_key *extent_key,
6533 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6534 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6535 struct extent_map *em;
6536 u64 start = extent_key->objectid - offset;
6537 u64 end = start + extent_key->offset - 1;
6539 em = alloc_extent_map(GFP_NOFS);
6540 BUG_ON(!em || IS_ERR(em));
6543 em->len = extent_key->offset;
6544 em->block_len = extent_key->offset;
6545 em->block_start = extent_key->objectid;
6546 em->bdev = root->fs_info->fs_devices->latest_bdev;
6547 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6549 /* setup extent map to cheat btrfs_readpage */
6550 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6553 write_lock(&em_tree->lock);
6554 ret = add_extent_mapping(em_tree, em);
6555 write_unlock(&em_tree->lock);
6556 if (ret != -EEXIST) {
6557 free_extent_map(em);
6560 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6562 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6564 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6567 struct btrfs_ref_path {
6569 u64 nodes[BTRFS_MAX_LEVEL];
6571 u64 root_generation;
6578 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6579 u64 new_nodes[BTRFS_MAX_LEVEL];
6582 struct disk_extent {
6593 static int is_cowonly_root(u64 root_objectid)
6595 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6596 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6597 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6598 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6599 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6600 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6605 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6606 struct btrfs_root *extent_root,
6607 struct btrfs_ref_path *ref_path,
6610 struct extent_buffer *leaf;
6611 struct btrfs_path *path;
6612 struct btrfs_extent_ref *ref;
6613 struct btrfs_key key;
6614 struct btrfs_key found_key;
6620 path = btrfs_alloc_path();
6625 ref_path->lowest_level = -1;
6626 ref_path->current_level = -1;
6627 ref_path->shared_level = -1;
6631 level = ref_path->current_level - 1;
6632 while (level >= -1) {
6634 if (level < ref_path->lowest_level)
6638 bytenr = ref_path->nodes[level];
6640 bytenr = ref_path->extent_start;
6641 BUG_ON(bytenr == 0);
6643 parent = ref_path->nodes[level + 1];
6644 ref_path->nodes[level + 1] = 0;
6645 ref_path->current_level = level;
6646 BUG_ON(parent == 0);
6648 key.objectid = bytenr;
6649 key.offset = parent + 1;
6650 key.type = BTRFS_EXTENT_REF_KEY;
6652 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6657 leaf = path->nodes[0];
6658 nritems = btrfs_header_nritems(leaf);
6659 if (path->slots[0] >= nritems) {
6660 ret = btrfs_next_leaf(extent_root, path);
6665 leaf = path->nodes[0];
6668 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6669 if (found_key.objectid == bytenr &&
6670 found_key.type == BTRFS_EXTENT_REF_KEY) {
6671 if (level < ref_path->shared_level)
6672 ref_path->shared_level = level;
6677 btrfs_release_path(extent_root, path);
6680 /* reached lowest level */
6684 level = ref_path->current_level;
6685 while (level < BTRFS_MAX_LEVEL - 1) {
6689 bytenr = ref_path->nodes[level];
6691 bytenr = ref_path->extent_start;
6693 BUG_ON(bytenr == 0);
6695 key.objectid = bytenr;
6697 key.type = BTRFS_EXTENT_REF_KEY;
6699 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6703 leaf = path->nodes[0];
6704 nritems = btrfs_header_nritems(leaf);
6705 if (path->slots[0] >= nritems) {
6706 ret = btrfs_next_leaf(extent_root, path);
6710 /* the extent was freed by someone */
6711 if (ref_path->lowest_level == level)
6713 btrfs_release_path(extent_root, path);
6716 leaf = path->nodes[0];
6719 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6720 if (found_key.objectid != bytenr ||
6721 found_key.type != BTRFS_EXTENT_REF_KEY) {
6722 /* the extent was freed by someone */
6723 if (ref_path->lowest_level == level) {
6727 btrfs_release_path(extent_root, path);
6731 ref = btrfs_item_ptr(leaf, path->slots[0],
6732 struct btrfs_extent_ref);
6733 ref_objectid = btrfs_ref_objectid(leaf, ref);
6734 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6736 level = (int)ref_objectid;
6737 BUG_ON(level >= BTRFS_MAX_LEVEL);
6738 ref_path->lowest_level = level;
6739 ref_path->current_level = level;
6740 ref_path->nodes[level] = bytenr;
6742 WARN_ON(ref_objectid != level);
6745 WARN_ON(level != -1);
6749 if (ref_path->lowest_level == level) {
6750 ref_path->owner_objectid = ref_objectid;
6751 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6755 * the block is tree root or the block isn't in reference
6758 if (found_key.objectid == found_key.offset ||
6759 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6760 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6761 ref_path->root_generation =
6762 btrfs_ref_generation(leaf, ref);
6764 /* special reference from the tree log */
6765 ref_path->nodes[0] = found_key.offset;
6766 ref_path->current_level = 0;
6773 BUG_ON(ref_path->nodes[level] != 0);
6774 ref_path->nodes[level] = found_key.offset;
6775 ref_path->current_level = level;
6778 * the reference was created in the running transaction,
6779 * no need to continue walking up.
6781 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6782 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6783 ref_path->root_generation =
6784 btrfs_ref_generation(leaf, ref);
6789 btrfs_release_path(extent_root, path);
6792 /* reached max tree level, but no tree root found. */
6795 btrfs_free_path(path);
6799 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6800 struct btrfs_root *extent_root,
6801 struct btrfs_ref_path *ref_path,
6804 memset(ref_path, 0, sizeof(*ref_path));
6805 ref_path->extent_start = extent_start;
6807 return __next_ref_path(trans, extent_root, ref_path, 1);
6810 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6811 struct btrfs_root *extent_root,
6812 struct btrfs_ref_path *ref_path)
6814 return __next_ref_path(trans, extent_root, ref_path, 0);
6817 static noinline int get_new_locations(struct inode *reloc_inode,
6818 struct btrfs_key *extent_key,
6819 u64 offset, int no_fragment,
6820 struct disk_extent **extents,
6823 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6824 struct btrfs_path *path;
6825 struct btrfs_file_extent_item *fi;
6826 struct extent_buffer *leaf;
6827 struct disk_extent *exts = *extents;
6828 struct btrfs_key found_key;
6833 int max = *nr_extents;
6836 WARN_ON(!no_fragment && *extents);
6839 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6844 path = btrfs_alloc_path();
6847 cur_pos = extent_key->objectid - offset;
6848 last_byte = extent_key->objectid + extent_key->offset;
6849 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6859 leaf = path->nodes[0];
6860 nritems = btrfs_header_nritems(leaf);
6861 if (path->slots[0] >= nritems) {
6862 ret = btrfs_next_leaf(root, path);
6867 leaf = path->nodes[0];
6870 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6871 if (found_key.offset != cur_pos ||
6872 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6873 found_key.objectid != reloc_inode->i_ino)
6876 fi = btrfs_item_ptr(leaf, path->slots[0],
6877 struct btrfs_file_extent_item);
6878 if (btrfs_file_extent_type(leaf, fi) !=
6879 BTRFS_FILE_EXTENT_REG ||
6880 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6884 struct disk_extent *old = exts;
6886 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6887 memcpy(exts, old, sizeof(*exts) * nr);
6888 if (old != *extents)
6892 exts[nr].disk_bytenr =
6893 btrfs_file_extent_disk_bytenr(leaf, fi);
6894 exts[nr].disk_num_bytes =
6895 btrfs_file_extent_disk_num_bytes(leaf, fi);
6896 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6897 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6898 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6899 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6900 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6901 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6903 BUG_ON(exts[nr].offset > 0);
6904 BUG_ON(exts[nr].compression || exts[nr].encryption);
6905 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6907 cur_pos += exts[nr].num_bytes;
6910 if (cur_pos + offset >= last_byte)
6920 BUG_ON(cur_pos + offset > last_byte);
6921 if (cur_pos + offset < last_byte) {
6927 btrfs_free_path(path);
6929 if (exts != *extents)
6938 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6939 struct btrfs_root *root,
6940 struct btrfs_path *path,
6941 struct btrfs_key *extent_key,
6942 struct btrfs_key *leaf_key,
6943 struct btrfs_ref_path *ref_path,
6944 struct disk_extent *new_extents,
6947 struct extent_buffer *leaf;
6948 struct btrfs_file_extent_item *fi;
6949 struct inode *inode = NULL;
6950 struct btrfs_key key;
6955 u64 search_end = (u64)-1;
6958 int extent_locked = 0;
6962 memcpy(&key, leaf_key, sizeof(key));
6963 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6964 if (key.objectid < ref_path->owner_objectid ||
6965 (key.objectid == ref_path->owner_objectid &&
6966 key.type < BTRFS_EXTENT_DATA_KEY)) {
6967 key.objectid = ref_path->owner_objectid;
6968 key.type = BTRFS_EXTENT_DATA_KEY;
6974 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6978 leaf = path->nodes[0];
6979 nritems = btrfs_header_nritems(leaf);
6981 if (extent_locked && ret > 0) {
6983 * the file extent item was modified by someone
6984 * before the extent got locked.
6986 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6987 lock_end, GFP_NOFS);
6991 if (path->slots[0] >= nritems) {
6992 if (++nr_scaned > 2)
6995 BUG_ON(extent_locked);
6996 ret = btrfs_next_leaf(root, path);
7001 leaf = path->nodes[0];
7002 nritems = btrfs_header_nritems(leaf);
7005 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7007 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7008 if ((key.objectid > ref_path->owner_objectid) ||
7009 (key.objectid == ref_path->owner_objectid &&
7010 key.type > BTRFS_EXTENT_DATA_KEY) ||
7011 key.offset >= search_end)
7015 if (inode && key.objectid != inode->i_ino) {
7016 BUG_ON(extent_locked);
7017 btrfs_release_path(root, path);
7018 mutex_unlock(&inode->i_mutex);
7024 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7029 fi = btrfs_item_ptr(leaf, path->slots[0],
7030 struct btrfs_file_extent_item);
7031 extent_type = btrfs_file_extent_type(leaf, fi);
7032 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7033 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7034 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7035 extent_key->objectid)) {
7041 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7042 ext_offset = btrfs_file_extent_offset(leaf, fi);
7044 if (search_end == (u64)-1) {
7045 search_end = key.offset - ext_offset +
7046 btrfs_file_extent_ram_bytes(leaf, fi);
7049 if (!extent_locked) {
7050 lock_start = key.offset;
7051 lock_end = lock_start + num_bytes - 1;
7053 if (lock_start > key.offset ||
7054 lock_end + 1 < key.offset + num_bytes) {
7055 unlock_extent(&BTRFS_I(inode)->io_tree,
7056 lock_start, lock_end, GFP_NOFS);
7062 btrfs_release_path(root, path);
7064 inode = btrfs_iget_locked(root->fs_info->sb,
7065 key.objectid, root);
7066 if (inode->i_state & I_NEW) {
7067 BTRFS_I(inode)->root = root;
7068 BTRFS_I(inode)->location.objectid =
7070 BTRFS_I(inode)->location.type =
7071 BTRFS_INODE_ITEM_KEY;
7072 BTRFS_I(inode)->location.offset = 0;
7073 btrfs_read_locked_inode(inode);
7074 unlock_new_inode(inode);
7077 * some code call btrfs_commit_transaction while
7078 * holding the i_mutex, so we can't use mutex_lock
7081 if (is_bad_inode(inode) ||
7082 !mutex_trylock(&inode->i_mutex)) {
7085 key.offset = (u64)-1;
7090 if (!extent_locked) {
7091 struct btrfs_ordered_extent *ordered;
7093 btrfs_release_path(root, path);
7095 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7096 lock_end, GFP_NOFS);
7097 ordered = btrfs_lookup_first_ordered_extent(inode,
7100 ordered->file_offset <= lock_end &&
7101 ordered->file_offset + ordered->len > lock_start) {
7102 unlock_extent(&BTRFS_I(inode)->io_tree,
7103 lock_start, lock_end, GFP_NOFS);
7104 btrfs_start_ordered_extent(inode, ordered, 1);
7105 btrfs_put_ordered_extent(ordered);
7106 key.offset += num_bytes;
7110 btrfs_put_ordered_extent(ordered);
7116 if (nr_extents == 1) {
7117 /* update extent pointer in place */
7118 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7119 new_extents[0].disk_bytenr);
7120 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7121 new_extents[0].disk_num_bytes);
7122 btrfs_mark_buffer_dirty(leaf);
7124 btrfs_drop_extent_cache(inode, key.offset,
7125 key.offset + num_bytes - 1, 0);
7127 ret = btrfs_inc_extent_ref(trans, root,
7128 new_extents[0].disk_bytenr,
7129 new_extents[0].disk_num_bytes,
7131 root->root_key.objectid,
7136 ret = btrfs_free_extent(trans, root,
7137 extent_key->objectid,
7140 btrfs_header_owner(leaf),
7141 btrfs_header_generation(leaf),
7145 btrfs_release_path(root, path);
7146 key.offset += num_bytes;
7154 * drop old extent pointer at first, then insert the
7155 * new pointers one bye one
7157 btrfs_release_path(root, path);
7158 ret = btrfs_drop_extents(trans, root, inode, key.offset,
7159 key.offset + num_bytes,
7160 key.offset, &alloc_hint);
7163 for (i = 0; i < nr_extents; i++) {
7164 if (ext_offset >= new_extents[i].num_bytes) {
7165 ext_offset -= new_extents[i].num_bytes;
7168 extent_len = min(new_extents[i].num_bytes -
7169 ext_offset, num_bytes);
7171 ret = btrfs_insert_empty_item(trans, root,
7176 leaf = path->nodes[0];
7177 fi = btrfs_item_ptr(leaf, path->slots[0],
7178 struct btrfs_file_extent_item);
7179 btrfs_set_file_extent_generation(leaf, fi,
7181 btrfs_set_file_extent_type(leaf, fi,
7182 BTRFS_FILE_EXTENT_REG);
7183 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7184 new_extents[i].disk_bytenr);
7185 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7186 new_extents[i].disk_num_bytes);
7187 btrfs_set_file_extent_ram_bytes(leaf, fi,
7188 new_extents[i].ram_bytes);
7190 btrfs_set_file_extent_compression(leaf, fi,
7191 new_extents[i].compression);
7192 btrfs_set_file_extent_encryption(leaf, fi,
7193 new_extents[i].encryption);
7194 btrfs_set_file_extent_other_encoding(leaf, fi,
7195 new_extents[i].other_encoding);
7197 btrfs_set_file_extent_num_bytes(leaf, fi,
7199 ext_offset += new_extents[i].offset;
7200 btrfs_set_file_extent_offset(leaf, fi,
7202 btrfs_mark_buffer_dirty(leaf);
7204 btrfs_drop_extent_cache(inode, key.offset,
7205 key.offset + extent_len - 1, 0);
7207 ret = btrfs_inc_extent_ref(trans, root,
7208 new_extents[i].disk_bytenr,
7209 new_extents[i].disk_num_bytes,
7211 root->root_key.objectid,
7212 trans->transid, key.objectid);
7214 btrfs_release_path(root, path);
7216 inode_add_bytes(inode, extent_len);
7219 num_bytes -= extent_len;
7220 key.offset += extent_len;
7225 BUG_ON(i >= nr_extents);
7229 if (extent_locked) {
7230 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7231 lock_end, GFP_NOFS);
7235 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7236 key.offset >= search_end)
7243 btrfs_release_path(root, path);
7245 mutex_unlock(&inode->i_mutex);
7246 if (extent_locked) {
7247 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7248 lock_end, GFP_NOFS);
7255 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7256 struct btrfs_root *root,
7257 struct extent_buffer *buf, u64 orig_start)
7262 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7263 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7265 level = btrfs_header_level(buf);
7267 struct btrfs_leaf_ref *ref;
7268 struct btrfs_leaf_ref *orig_ref;
7270 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7274 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7276 btrfs_free_leaf_ref(root, orig_ref);
7280 ref->nritems = orig_ref->nritems;
7281 memcpy(ref->extents, orig_ref->extents,
7282 sizeof(ref->extents[0]) * ref->nritems);
7284 btrfs_free_leaf_ref(root, orig_ref);
7286 ref->root_gen = trans->transid;
7287 ref->bytenr = buf->start;
7288 ref->owner = btrfs_header_owner(buf);
7289 ref->generation = btrfs_header_generation(buf);
7291 ret = btrfs_add_leaf_ref(root, ref, 0);
7293 btrfs_free_leaf_ref(root, ref);
7298 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7299 struct extent_buffer *leaf,
7300 struct btrfs_block_group_cache *group,
7301 struct btrfs_root *target_root)
7303 struct btrfs_key key;
7304 struct inode *inode = NULL;
7305 struct btrfs_file_extent_item *fi;
7306 struct extent_state *cached_state = NULL;
7308 u64 skip_objectid = 0;
7312 nritems = btrfs_header_nritems(leaf);
7313 for (i = 0; i < nritems; i++) {
7314 btrfs_item_key_to_cpu(leaf, &key, i);
7315 if (key.objectid == skip_objectid ||
7316 key.type != BTRFS_EXTENT_DATA_KEY)
7318 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7319 if (btrfs_file_extent_type(leaf, fi) ==
7320 BTRFS_FILE_EXTENT_INLINE)
7322 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7324 if (!inode || inode->i_ino != key.objectid) {
7326 inode = btrfs_ilookup(target_root->fs_info->sb,
7327 key.objectid, target_root, 1);
7330 skip_objectid = key.objectid;
7333 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7335 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7336 key.offset + num_bytes - 1, 0, &cached_state,
7338 btrfs_drop_extent_cache(inode, key.offset,
7339 key.offset + num_bytes - 1, 1);
7340 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7341 key.offset + num_bytes - 1, &cached_state,
7349 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7350 struct btrfs_root *root,
7351 struct extent_buffer *leaf,
7352 struct btrfs_block_group_cache *group,
7353 struct inode *reloc_inode)
7355 struct btrfs_key key;
7356 struct btrfs_key extent_key;
7357 struct btrfs_file_extent_item *fi;
7358 struct btrfs_leaf_ref *ref;
7359 struct disk_extent *new_extent;
7368 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7369 BUG_ON(!new_extent);
7371 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7375 nritems = btrfs_header_nritems(leaf);
7376 for (i = 0; i < nritems; i++) {
7377 btrfs_item_key_to_cpu(leaf, &key, i);
7378 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7380 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7381 if (btrfs_file_extent_type(leaf, fi) ==
7382 BTRFS_FILE_EXTENT_INLINE)
7384 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7385 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7390 if (bytenr >= group->key.objectid + group->key.offset ||
7391 bytenr + num_bytes <= group->key.objectid)
7394 extent_key.objectid = bytenr;
7395 extent_key.offset = num_bytes;
7396 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7398 ret = get_new_locations(reloc_inode, &extent_key,
7399 group->key.objectid, 1,
7400 &new_extent, &nr_extent);
7405 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7406 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7407 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7408 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7410 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7411 new_extent->disk_bytenr);
7412 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7413 new_extent->disk_num_bytes);
7414 btrfs_mark_buffer_dirty(leaf);
7416 ret = btrfs_inc_extent_ref(trans, root,
7417 new_extent->disk_bytenr,
7418 new_extent->disk_num_bytes,
7420 root->root_key.objectid,
7421 trans->transid, key.objectid);
7424 ret = btrfs_free_extent(trans, root,
7425 bytenr, num_bytes, leaf->start,
7426 btrfs_header_owner(leaf),
7427 btrfs_header_generation(leaf),
7433 BUG_ON(ext_index + 1 != ref->nritems);
7434 btrfs_free_leaf_ref(root, ref);
7438 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7439 struct btrfs_root *root)
7441 struct btrfs_root *reloc_root;
7444 if (root->reloc_root) {
7445 reloc_root = root->reloc_root;
7446 root->reloc_root = NULL;
7447 list_add(&reloc_root->dead_list,
7448 &root->fs_info->dead_reloc_roots);
7450 btrfs_set_root_bytenr(&reloc_root->root_item,
7451 reloc_root->node->start);
7452 btrfs_set_root_level(&root->root_item,
7453 btrfs_header_level(reloc_root->node));
7454 memset(&reloc_root->root_item.drop_progress, 0,
7455 sizeof(struct btrfs_disk_key));
7456 reloc_root->root_item.drop_level = 0;
7458 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7459 &reloc_root->root_key,
7460 &reloc_root->root_item);
7466 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7468 struct btrfs_trans_handle *trans;
7469 struct btrfs_root *reloc_root;
7470 struct btrfs_root *prev_root = NULL;
7471 struct list_head dead_roots;
7475 INIT_LIST_HEAD(&dead_roots);
7476 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7478 while (!list_empty(&dead_roots)) {
7479 reloc_root = list_entry(dead_roots.prev,
7480 struct btrfs_root, dead_list);
7481 list_del_init(&reloc_root->dead_list);
7483 BUG_ON(reloc_root->commit_root != NULL);
7485 trans = btrfs_join_transaction(root, 1);
7488 mutex_lock(&root->fs_info->drop_mutex);
7489 ret = btrfs_drop_snapshot(trans, reloc_root);
7492 mutex_unlock(&root->fs_info->drop_mutex);
7494 nr = trans->blocks_used;
7495 ret = btrfs_end_transaction(trans, root);
7497 btrfs_btree_balance_dirty(root, nr);
7500 free_extent_buffer(reloc_root->node);
7502 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7503 &reloc_root->root_key);
7505 mutex_unlock(&root->fs_info->drop_mutex);
7507 nr = trans->blocks_used;
7508 ret = btrfs_end_transaction(trans, root);
7510 btrfs_btree_balance_dirty(root, nr);
7513 prev_root = reloc_root;
7516 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7522 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7524 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7528 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7530 struct btrfs_root *reloc_root;
7531 struct btrfs_trans_handle *trans;
7532 struct btrfs_key location;
7536 mutex_lock(&root->fs_info->tree_reloc_mutex);
7537 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7539 found = !list_empty(&root->fs_info->dead_reloc_roots);
7540 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7543 trans = btrfs_start_transaction(root, 1);
7545 ret = btrfs_commit_transaction(trans, root);
7549 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7550 location.offset = (u64)-1;
7551 location.type = BTRFS_ROOT_ITEM_KEY;
7553 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7554 BUG_ON(!reloc_root);
7555 btrfs_orphan_cleanup(reloc_root);
7559 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7560 struct btrfs_root *root)
7562 struct btrfs_root *reloc_root;
7563 struct extent_buffer *eb;
7564 struct btrfs_root_item *root_item;
7565 struct btrfs_key root_key;
7568 BUG_ON(!root->ref_cows);
7569 if (root->reloc_root)
7572 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7575 ret = btrfs_copy_root(trans, root, root->commit_root,
7576 &eb, BTRFS_TREE_RELOC_OBJECTID);
7579 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7580 root_key.offset = root->root_key.objectid;
7581 root_key.type = BTRFS_ROOT_ITEM_KEY;
7583 memcpy(root_item, &root->root_item, sizeof(root_item));
7584 btrfs_set_root_refs(root_item, 0);
7585 btrfs_set_root_bytenr(root_item, eb->start);
7586 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7587 btrfs_set_root_generation(root_item, trans->transid);
7589 btrfs_tree_unlock(eb);
7590 free_extent_buffer(eb);
7592 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7593 &root_key, root_item);
7597 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7599 BUG_ON(!reloc_root);
7600 reloc_root->last_trans = trans->transid;
7601 reloc_root->commit_root = NULL;
7602 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7604 root->reloc_root = reloc_root;
7609 * Core function of space balance.
7611 * The idea is using reloc trees to relocate tree blocks in reference
7612 * counted roots. There is one reloc tree for each subvol, and all
7613 * reloc trees share same root key objectid. Reloc trees are snapshots
7614 * of the latest committed roots of subvols (root->commit_root).
7616 * To relocate a tree block referenced by a subvol, there are two steps.
7617 * COW the block through subvol's reloc tree, then update block pointer
7618 * in the subvol to point to the new block. Since all reloc trees share
7619 * same root key objectid, doing special handing for tree blocks owned
7620 * by them is easy. Once a tree block has been COWed in one reloc tree,
7621 * we can use the resulting new block directly when the same block is
7622 * required to COW again through other reloc trees. By this way, relocated
7623 * tree blocks are shared between reloc trees, so they are also shared
7626 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7627 struct btrfs_root *root,
7628 struct btrfs_path *path,
7629 struct btrfs_key *first_key,
7630 struct btrfs_ref_path *ref_path,
7631 struct btrfs_block_group_cache *group,
7632 struct inode *reloc_inode)
7634 struct btrfs_root *reloc_root;
7635 struct extent_buffer *eb = NULL;
7636 struct btrfs_key *keys;
7640 int lowest_level = 0;
7643 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7644 lowest_level = ref_path->owner_objectid;
7646 if (!root->ref_cows) {
7647 path->lowest_level = lowest_level;
7648 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7650 path->lowest_level = 0;
7651 btrfs_release_path(root, path);
7655 mutex_lock(&root->fs_info->tree_reloc_mutex);
7656 ret = init_reloc_tree(trans, root);
7658 reloc_root = root->reloc_root;
7660 shared_level = ref_path->shared_level;
7661 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7663 keys = ref_path->node_keys;
7664 nodes = ref_path->new_nodes;
7665 memset(&keys[shared_level + 1], 0,
7666 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7667 memset(&nodes[shared_level + 1], 0,
7668 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7670 if (nodes[lowest_level] == 0) {
7671 path->lowest_level = lowest_level;
7672 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7675 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7676 eb = path->nodes[level];
7677 if (!eb || eb == reloc_root->node)
7679 nodes[level] = eb->start;
7681 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7683 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7686 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7687 eb = path->nodes[0];
7688 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7689 group, reloc_inode);
7692 btrfs_release_path(reloc_root, path);
7694 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7700 * replace tree blocks in the fs tree with tree blocks in
7703 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7706 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7707 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7710 extent_buffer_get(path->nodes[0]);
7711 eb = path->nodes[0];
7712 btrfs_release_path(reloc_root, path);
7713 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7715 free_extent_buffer(eb);
7718 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7719 path->lowest_level = 0;
7723 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7724 struct btrfs_root *root,
7725 struct btrfs_path *path,
7726 struct btrfs_key *first_key,
7727 struct btrfs_ref_path *ref_path)
7731 ret = relocate_one_path(trans, root, path, first_key,
7732 ref_path, NULL, NULL);
7738 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7739 struct btrfs_root *extent_root,
7740 struct btrfs_path *path,
7741 struct btrfs_key *extent_key)
7745 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7748 ret = btrfs_del_item(trans, extent_root, path);
7750 btrfs_release_path(extent_root, path);
7754 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7755 struct btrfs_ref_path *ref_path)
7757 struct btrfs_key root_key;
7759 root_key.objectid = ref_path->root_objectid;
7760 root_key.type = BTRFS_ROOT_ITEM_KEY;
7761 if (is_cowonly_root(ref_path->root_objectid))
7762 root_key.offset = 0;
7764 root_key.offset = (u64)-1;
7766 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7769 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7770 struct btrfs_path *path,
7771 struct btrfs_key *extent_key,
7772 struct btrfs_block_group_cache *group,
7773 struct inode *reloc_inode, int pass)
7775 struct btrfs_trans_handle *trans;
7776 struct btrfs_root *found_root;
7777 struct btrfs_ref_path *ref_path = NULL;
7778 struct disk_extent *new_extents = NULL;
7783 struct btrfs_key first_key;
7787 trans = btrfs_start_transaction(extent_root, 1);
7790 if (extent_key->objectid == 0) {
7791 ret = del_extent_zero(trans, extent_root, path, extent_key);
7795 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7801 for (loops = 0; ; loops++) {
7803 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7804 extent_key->objectid);
7806 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7813 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7814 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7817 found_root = read_ref_root(extent_root->fs_info, ref_path);
7818 BUG_ON(!found_root);
7820 * for reference counted tree, only process reference paths
7821 * rooted at the latest committed root.
7823 if (found_root->ref_cows &&
7824 ref_path->root_generation != found_root->root_key.offset)
7827 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7830 * copy data extents to new locations
7832 u64 group_start = group->key.objectid;
7833 ret = relocate_data_extent(reloc_inode,
7842 level = ref_path->owner_objectid;
7845 if (prev_block != ref_path->nodes[level]) {
7846 struct extent_buffer *eb;
7847 u64 block_start = ref_path->nodes[level];
7848 u64 block_size = btrfs_level_size(found_root, level);
7850 eb = read_tree_block(found_root, block_start,
7852 btrfs_tree_lock(eb);
7853 BUG_ON(level != btrfs_header_level(eb));
7856 btrfs_item_key_to_cpu(eb, &first_key, 0);
7858 btrfs_node_key_to_cpu(eb, &first_key, 0);
7860 btrfs_tree_unlock(eb);
7861 free_extent_buffer(eb);
7862 prev_block = block_start;
7865 mutex_lock(&extent_root->fs_info->trans_mutex);
7866 btrfs_record_root_in_trans(found_root);
7867 mutex_unlock(&extent_root->fs_info->trans_mutex);
7868 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7870 * try to update data extent references while
7871 * keeping metadata shared between snapshots.
7874 ret = relocate_one_path(trans, found_root,
7875 path, &first_key, ref_path,
7876 group, reloc_inode);
7882 * use fallback method to process the remaining
7886 u64 group_start = group->key.objectid;
7887 new_extents = kmalloc(sizeof(*new_extents),
7890 ret = get_new_locations(reloc_inode,
7898 ret = replace_one_extent(trans, found_root,
7900 &first_key, ref_path,
7901 new_extents, nr_extents);
7903 ret = relocate_tree_block(trans, found_root, path,
7904 &first_key, ref_path);
7911 btrfs_end_transaction(trans, extent_root);
7918 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7921 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7922 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7925 * we add in the count of missing devices because we want
7926 * to make sure that any RAID levels on a degraded FS
7927 * continue to be honored.
7929 num_devices = root->fs_info->fs_devices->rw_devices +
7930 root->fs_info->fs_devices->missing_devices;
7932 if (num_devices == 1) {
7933 stripped |= BTRFS_BLOCK_GROUP_DUP;
7934 stripped = flags & ~stripped;
7936 /* turn raid0 into single device chunks */
7937 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7940 /* turn mirroring into duplication */
7941 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7942 BTRFS_BLOCK_GROUP_RAID10))
7943 return stripped | BTRFS_BLOCK_GROUP_DUP;
7946 /* they already had raid on here, just return */
7947 if (flags & stripped)
7950 stripped |= BTRFS_BLOCK_GROUP_DUP;
7951 stripped = flags & ~stripped;
7953 /* switch duplicated blocks with raid1 */
7954 if (flags & BTRFS_BLOCK_GROUP_DUP)
7955 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7957 /* turn single device chunks into raid0 */
7958 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7963 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7965 struct btrfs_space_info *sinfo = cache->space_info;
7972 spin_lock(&sinfo->lock);
7973 spin_lock(&cache->lock);
7974 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7975 cache->bytes_super - btrfs_block_group_used(&cache->item);
7977 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7978 sinfo->bytes_may_use + sinfo->bytes_readonly +
7979 cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
7980 sinfo->bytes_readonly += num_bytes;
7981 sinfo->bytes_reserved += cache->reserved_pinned;
7982 cache->reserved_pinned = 0;
7987 spin_unlock(&cache->lock);
7988 spin_unlock(&sinfo->lock);
7992 int btrfs_set_block_group_ro(struct btrfs_root *root,
7993 struct btrfs_block_group_cache *cache)
7996 struct btrfs_trans_handle *trans;
8002 trans = btrfs_join_transaction(root, 1);
8003 BUG_ON(IS_ERR(trans));
8005 alloc_flags = update_block_group_flags(root, cache->flags);
8006 if (alloc_flags != cache->flags)
8007 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8009 ret = set_block_group_ro(cache);
8012 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8013 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8016 ret = set_block_group_ro(cache);
8018 btrfs_end_transaction(trans, root);
8023 * helper to account the unused space of all the readonly block group in the
8024 * list. takes mirrors into account.
8026 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8028 struct btrfs_block_group_cache *block_group;
8032 list_for_each_entry(block_group, groups_list, list) {
8033 spin_lock(&block_group->lock);
8035 if (!block_group->ro) {
8036 spin_unlock(&block_group->lock);
8040 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8041 BTRFS_BLOCK_GROUP_RAID10 |
8042 BTRFS_BLOCK_GROUP_DUP))
8047 free_bytes += (block_group->key.offset -
8048 btrfs_block_group_used(&block_group->item)) *
8051 spin_unlock(&block_group->lock);
8058 * helper to account the unused space of all the readonly block group in the
8059 * space_info. takes mirrors into account.
8061 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8066 spin_lock(&sinfo->lock);
8068 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8069 if (!list_empty(&sinfo->block_groups[i]))
8070 free_bytes += __btrfs_get_ro_block_group_free_space(
8071 &sinfo->block_groups[i]);
8073 spin_unlock(&sinfo->lock);
8078 int btrfs_set_block_group_rw(struct btrfs_root *root,
8079 struct btrfs_block_group_cache *cache)
8081 struct btrfs_space_info *sinfo = cache->space_info;
8086 spin_lock(&sinfo->lock);
8087 spin_lock(&cache->lock);
8088 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8089 cache->bytes_super - btrfs_block_group_used(&cache->item);
8090 sinfo->bytes_readonly -= num_bytes;
8092 spin_unlock(&cache->lock);
8093 spin_unlock(&sinfo->lock);
8098 * checks to see if its even possible to relocate this block group.
8100 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8101 * ok to go ahead and try.
8103 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8105 struct btrfs_block_group_cache *block_group;
8106 struct btrfs_space_info *space_info;
8107 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8108 struct btrfs_device *device;
8112 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8114 /* odd, couldn't find the block group, leave it alone */
8118 /* no bytes used, we're good */
8119 if (!btrfs_block_group_used(&block_group->item))
8122 space_info = block_group->space_info;
8123 spin_lock(&space_info->lock);
8125 full = space_info->full;
8128 * if this is the last block group we have in this space, we can't
8129 * relocate it unless we're able to allocate a new chunk below.
8131 * Otherwise, we need to make sure we have room in the space to handle
8132 * all of the extents from this block group. If we can, we're good
8134 if ((space_info->total_bytes != block_group->key.offset) &&
8135 (space_info->bytes_used + space_info->bytes_reserved +
8136 space_info->bytes_pinned + space_info->bytes_readonly +
8137 btrfs_block_group_used(&block_group->item) <
8138 space_info->total_bytes)) {
8139 spin_unlock(&space_info->lock);
8142 spin_unlock(&space_info->lock);
8145 * ok we don't have enough space, but maybe we have free space on our
8146 * devices to allocate new chunks for relocation, so loop through our
8147 * alloc devices and guess if we have enough space. However, if we
8148 * were marked as full, then we know there aren't enough chunks, and we
8155 mutex_lock(&root->fs_info->chunk_mutex);
8156 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8157 u64 min_free = btrfs_block_group_used(&block_group->item);
8161 * check to make sure we can actually find a chunk with enough
8162 * space to fit our block group in.
8164 if (device->total_bytes > device->bytes_used + min_free) {
8165 ret = find_free_dev_extent(NULL, device, min_free,
8172 mutex_unlock(&root->fs_info->chunk_mutex);
8174 btrfs_put_block_group(block_group);
8178 static int find_first_block_group(struct btrfs_root *root,
8179 struct btrfs_path *path, struct btrfs_key *key)
8182 struct btrfs_key found_key;
8183 struct extent_buffer *leaf;
8186 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8191 slot = path->slots[0];
8192 leaf = path->nodes[0];
8193 if (slot >= btrfs_header_nritems(leaf)) {
8194 ret = btrfs_next_leaf(root, path);
8201 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8203 if (found_key.objectid >= key->objectid &&
8204 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8214 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8216 struct btrfs_block_group_cache *block_group;
8220 struct inode *inode;
8222 block_group = btrfs_lookup_first_block_group(info, last);
8223 while (block_group) {
8224 spin_lock(&block_group->lock);
8225 if (block_group->iref)
8227 spin_unlock(&block_group->lock);
8228 block_group = next_block_group(info->tree_root,
8238 inode = block_group->inode;
8239 block_group->iref = 0;
8240 block_group->inode = NULL;
8241 spin_unlock(&block_group->lock);
8243 last = block_group->key.objectid + block_group->key.offset;
8244 btrfs_put_block_group(block_group);
8248 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8250 struct btrfs_block_group_cache *block_group;
8251 struct btrfs_space_info *space_info;
8252 struct btrfs_caching_control *caching_ctl;
8255 down_write(&info->extent_commit_sem);
8256 while (!list_empty(&info->caching_block_groups)) {
8257 caching_ctl = list_entry(info->caching_block_groups.next,
8258 struct btrfs_caching_control, list);
8259 list_del(&caching_ctl->list);
8260 put_caching_control(caching_ctl);
8262 up_write(&info->extent_commit_sem);
8264 spin_lock(&info->block_group_cache_lock);
8265 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8266 block_group = rb_entry(n, struct btrfs_block_group_cache,
8268 rb_erase(&block_group->cache_node,
8269 &info->block_group_cache_tree);
8270 spin_unlock(&info->block_group_cache_lock);
8272 down_write(&block_group->space_info->groups_sem);
8273 list_del(&block_group->list);
8274 up_write(&block_group->space_info->groups_sem);
8276 if (block_group->cached == BTRFS_CACHE_STARTED)
8277 wait_block_group_cache_done(block_group);
8279 btrfs_remove_free_space_cache(block_group);
8280 btrfs_put_block_group(block_group);
8282 spin_lock(&info->block_group_cache_lock);
8284 spin_unlock(&info->block_group_cache_lock);
8286 /* now that all the block groups are freed, go through and
8287 * free all the space_info structs. This is only called during
8288 * the final stages of unmount, and so we know nobody is
8289 * using them. We call synchronize_rcu() once before we start,
8290 * just to be on the safe side.
8294 release_global_block_rsv(info);
8296 while(!list_empty(&info->space_info)) {
8297 space_info = list_entry(info->space_info.next,
8298 struct btrfs_space_info,
8300 if (space_info->bytes_pinned > 0 ||
8301 space_info->bytes_reserved > 0) {
8303 dump_space_info(space_info, 0, 0);
8305 list_del(&space_info->list);
8311 static void __link_block_group(struct btrfs_space_info *space_info,
8312 struct btrfs_block_group_cache *cache)
8314 int index = get_block_group_index(cache);
8316 down_write(&space_info->groups_sem);
8317 list_add_tail(&cache->list, &space_info->block_groups[index]);
8318 up_write(&space_info->groups_sem);
8321 int btrfs_read_block_groups(struct btrfs_root *root)
8323 struct btrfs_path *path;
8325 struct btrfs_block_group_cache *cache;
8326 struct btrfs_fs_info *info = root->fs_info;
8327 struct btrfs_space_info *space_info;
8328 struct btrfs_key key;
8329 struct btrfs_key found_key;
8330 struct extent_buffer *leaf;
8334 root = info->extent_root;
8337 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8338 path = btrfs_alloc_path();
8342 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8343 if (cache_gen != 0 &&
8344 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8346 if (btrfs_test_opt(root, CLEAR_CACHE))
8348 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8349 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8352 ret = find_first_block_group(root, path, &key);
8357 leaf = path->nodes[0];
8358 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8359 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8365 atomic_set(&cache->count, 1);
8366 spin_lock_init(&cache->lock);
8367 spin_lock_init(&cache->tree_lock);
8368 cache->fs_info = info;
8369 INIT_LIST_HEAD(&cache->list);
8370 INIT_LIST_HEAD(&cache->cluster_list);
8373 cache->disk_cache_state = BTRFS_DC_CLEAR;
8376 * we only want to have 32k of ram per block group for keeping
8377 * track of free space, and if we pass 1/2 of that we want to
8378 * start converting things over to using bitmaps
8380 cache->extents_thresh = ((1024 * 32) / 2) /
8381 sizeof(struct btrfs_free_space);
8383 read_extent_buffer(leaf, &cache->item,
8384 btrfs_item_ptr_offset(leaf, path->slots[0]),
8385 sizeof(cache->item));
8386 memcpy(&cache->key, &found_key, sizeof(found_key));
8388 key.objectid = found_key.objectid + found_key.offset;
8389 btrfs_release_path(root, path);
8390 cache->flags = btrfs_block_group_flags(&cache->item);
8391 cache->sectorsize = root->sectorsize;
8394 * check for two cases, either we are full, and therefore
8395 * don't need to bother with the caching work since we won't
8396 * find any space, or we are empty, and we can just add all
8397 * the space in and be done with it. This saves us _alot_ of
8398 * time, particularly in the full case.
8400 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8401 exclude_super_stripes(root, cache);
8402 cache->last_byte_to_unpin = (u64)-1;
8403 cache->cached = BTRFS_CACHE_FINISHED;
8404 free_excluded_extents(root, cache);
8405 } else if (btrfs_block_group_used(&cache->item) == 0) {
8406 exclude_super_stripes(root, cache);
8407 cache->last_byte_to_unpin = (u64)-1;
8408 cache->cached = BTRFS_CACHE_FINISHED;
8409 add_new_free_space(cache, root->fs_info,
8411 found_key.objectid +
8413 free_excluded_extents(root, cache);
8416 ret = update_space_info(info, cache->flags, found_key.offset,
8417 btrfs_block_group_used(&cache->item),
8420 cache->space_info = space_info;
8421 spin_lock(&cache->space_info->lock);
8422 cache->space_info->bytes_readonly += cache->bytes_super;
8423 spin_unlock(&cache->space_info->lock);
8425 __link_block_group(space_info, cache);
8427 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8430 set_avail_alloc_bits(root->fs_info, cache->flags);
8431 if (btrfs_chunk_readonly(root, cache->key.objectid))
8432 set_block_group_ro(cache);
8435 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8436 if (!(get_alloc_profile(root, space_info->flags) &
8437 (BTRFS_BLOCK_GROUP_RAID10 |
8438 BTRFS_BLOCK_GROUP_RAID1 |
8439 BTRFS_BLOCK_GROUP_DUP)))
8442 * avoid allocating from un-mirrored block group if there are
8443 * mirrored block groups.
8445 list_for_each_entry(cache, &space_info->block_groups[3], list)
8446 set_block_group_ro(cache);
8447 list_for_each_entry(cache, &space_info->block_groups[4], list)
8448 set_block_group_ro(cache);
8451 init_global_block_rsv(info);
8454 btrfs_free_path(path);
8458 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8459 struct btrfs_root *root, u64 bytes_used,
8460 u64 type, u64 chunk_objectid, u64 chunk_offset,
8464 struct btrfs_root *extent_root;
8465 struct btrfs_block_group_cache *cache;
8467 extent_root = root->fs_info->extent_root;
8469 root->fs_info->last_trans_log_full_commit = trans->transid;
8471 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8475 cache->key.objectid = chunk_offset;
8476 cache->key.offset = size;
8477 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8478 cache->sectorsize = root->sectorsize;
8479 cache->fs_info = root->fs_info;
8482 * we only want to have 32k of ram per block group for keeping track
8483 * of free space, and if we pass 1/2 of that we want to start
8484 * converting things over to using bitmaps
8486 cache->extents_thresh = ((1024 * 32) / 2) /
8487 sizeof(struct btrfs_free_space);
8488 atomic_set(&cache->count, 1);
8489 spin_lock_init(&cache->lock);
8490 spin_lock_init(&cache->tree_lock);
8491 INIT_LIST_HEAD(&cache->list);
8492 INIT_LIST_HEAD(&cache->cluster_list);
8494 btrfs_set_block_group_used(&cache->item, bytes_used);
8495 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8496 cache->flags = type;
8497 btrfs_set_block_group_flags(&cache->item, type);
8499 cache->last_byte_to_unpin = (u64)-1;
8500 cache->cached = BTRFS_CACHE_FINISHED;
8501 exclude_super_stripes(root, cache);
8503 add_new_free_space(cache, root->fs_info, chunk_offset,
8504 chunk_offset + size);
8506 free_excluded_extents(root, cache);
8508 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8509 &cache->space_info);
8512 spin_lock(&cache->space_info->lock);
8513 cache->space_info->bytes_readonly += cache->bytes_super;
8514 spin_unlock(&cache->space_info->lock);
8516 __link_block_group(cache->space_info, cache);
8518 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8521 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8522 sizeof(cache->item));
8525 set_avail_alloc_bits(extent_root->fs_info, type);
8530 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8531 struct btrfs_root *root, u64 group_start)
8533 struct btrfs_path *path;
8534 struct btrfs_block_group_cache *block_group;
8535 struct btrfs_free_cluster *cluster;
8536 struct btrfs_root *tree_root = root->fs_info->tree_root;
8537 struct btrfs_key key;
8538 struct inode *inode;
8542 root = root->fs_info->extent_root;
8544 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8545 BUG_ON(!block_group);
8546 BUG_ON(!block_group->ro);
8548 memcpy(&key, &block_group->key, sizeof(key));
8549 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8550 BTRFS_BLOCK_GROUP_RAID1 |
8551 BTRFS_BLOCK_GROUP_RAID10))
8556 /* make sure this block group isn't part of an allocation cluster */
8557 cluster = &root->fs_info->data_alloc_cluster;
8558 spin_lock(&cluster->refill_lock);
8559 btrfs_return_cluster_to_free_space(block_group, cluster);
8560 spin_unlock(&cluster->refill_lock);
8563 * make sure this block group isn't part of a metadata
8564 * allocation cluster
8566 cluster = &root->fs_info->meta_alloc_cluster;
8567 spin_lock(&cluster->refill_lock);
8568 btrfs_return_cluster_to_free_space(block_group, cluster);
8569 spin_unlock(&cluster->refill_lock);
8571 path = btrfs_alloc_path();
8574 inode = lookup_free_space_inode(root, block_group, path);
8575 if (!IS_ERR(inode)) {
8576 btrfs_orphan_add(trans, inode);
8578 /* One for the block groups ref */
8579 spin_lock(&block_group->lock);
8580 if (block_group->iref) {
8581 block_group->iref = 0;
8582 block_group->inode = NULL;
8583 spin_unlock(&block_group->lock);
8586 spin_unlock(&block_group->lock);
8588 /* One for our lookup ref */
8592 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8593 key.offset = block_group->key.objectid;
8596 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8600 btrfs_release_path(tree_root, path);
8602 ret = btrfs_del_item(trans, tree_root, path);
8605 btrfs_release_path(tree_root, path);
8608 spin_lock(&root->fs_info->block_group_cache_lock);
8609 rb_erase(&block_group->cache_node,
8610 &root->fs_info->block_group_cache_tree);
8611 spin_unlock(&root->fs_info->block_group_cache_lock);
8613 down_write(&block_group->space_info->groups_sem);
8615 * we must use list_del_init so people can check to see if they
8616 * are still on the list after taking the semaphore
8618 list_del_init(&block_group->list);
8619 up_write(&block_group->space_info->groups_sem);
8621 if (block_group->cached == BTRFS_CACHE_STARTED)
8622 wait_block_group_cache_done(block_group);
8624 btrfs_remove_free_space_cache(block_group);
8626 spin_lock(&block_group->space_info->lock);
8627 block_group->space_info->total_bytes -= block_group->key.offset;
8628 block_group->space_info->bytes_readonly -= block_group->key.offset;
8629 block_group->space_info->disk_total -= block_group->key.offset * factor;
8630 spin_unlock(&block_group->space_info->lock);
8632 memcpy(&key, &block_group->key, sizeof(key));
8634 btrfs_clear_space_info_full(root->fs_info);
8636 btrfs_put_block_group(block_group);
8637 btrfs_put_block_group(block_group);
8639 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8645 ret = btrfs_del_item(trans, root, path);
8647 btrfs_free_path(path);