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;
3350 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3353 block_rsv = &root->fs_info->delalloc_block_rsv;
3354 space_info = block_rsv->space_info;
3357 reserved = space_info->bytes_reserved;
3362 max_reclaim = min(reserved, to_reclaim);
3364 while (loops < 1024) {
3365 /* have the flusher threads jump in and do some IO */
3367 nr_pages = min_t(unsigned long, nr_pages,
3368 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3369 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3371 spin_lock(&space_info->lock);
3372 if (reserved > space_info->bytes_reserved) {
3374 reclaimed += reserved - space_info->bytes_reserved;
3378 reserved = space_info->bytes_reserved;
3379 spin_unlock(&space_info->lock);
3381 if (reserved == 0 || reclaimed >= max_reclaim)
3384 if (trans && trans->transaction->blocked)
3387 __set_current_state(TASK_INTERRUPTIBLE);
3388 time_left = schedule_timeout(pause);
3390 /* We were interrupted, exit */
3395 if (pause > HZ / 10)
3399 return reclaimed >= to_reclaim;
3403 * Retries tells us how many times we've called reserve_metadata_bytes. The
3404 * idea is if this is the first call (retries == 0) then we will add to our
3405 * reserved count if we can't make the allocation in order to hold our place
3406 * while we go and try and free up space. That way for retries > 1 we don't try
3407 * and add space, we just check to see if the amount of unused space is >= the
3408 * total space, meaning that our reservation is valid.
3410 * However if we don't intend to retry this reservation, pass -1 as retries so
3411 * that it short circuits this logic.
3413 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3414 struct btrfs_root *root,
3415 struct btrfs_block_rsv *block_rsv,
3416 u64 orig_bytes, int flush)
3418 struct btrfs_space_info *space_info = block_rsv->space_info;
3420 u64 num_bytes = orig_bytes;
3423 bool reserved = false;
3424 bool committed = false;
3431 spin_lock(&space_info->lock);
3432 unused = space_info->bytes_used + space_info->bytes_reserved +
3433 space_info->bytes_pinned + space_info->bytes_readonly +
3434 space_info->bytes_may_use;
3437 * The idea here is that we've not already over-reserved the block group
3438 * then we can go ahead and save our reservation first and then start
3439 * flushing if we need to. Otherwise if we've already overcommitted
3440 * lets start flushing stuff first and then come back and try to make
3443 if (unused <= space_info->total_bytes) {
3444 unused = space_info->total_bytes - unused;
3445 if (unused >= num_bytes) {
3447 space_info->bytes_reserved += orig_bytes;
3451 * Ok set num_bytes to orig_bytes since we aren't
3452 * overocmmitted, this way we only try and reclaim what
3455 num_bytes = orig_bytes;
3459 * Ok we're over committed, set num_bytes to the overcommitted
3460 * amount plus the amount of bytes that we need for this
3463 num_bytes = unused - space_info->total_bytes +
3464 (orig_bytes * (retries + 1));
3468 * Couldn't make our reservation, save our place so while we're trying
3469 * to reclaim space we can actually use it instead of somebody else
3470 * stealing it from us.
3472 if (ret && !reserved) {
3473 space_info->bytes_reserved += orig_bytes;
3477 spin_unlock(&space_info->lock);
3486 * We do synchronous shrinking since we don't actually unreserve
3487 * metadata until after the IO is completed.
3489 ret = shrink_delalloc(trans, root, num_bytes, 1);
3496 * So if we were overcommitted it's possible that somebody else flushed
3497 * out enough space and we simply didn't have enough space to reclaim,
3498 * so go back around and try again.
3505 spin_lock(&space_info->lock);
3507 * Not enough space to be reclaimed, don't bother committing the
3510 if (space_info->bytes_pinned < orig_bytes)
3512 spin_unlock(&space_info->lock);
3517 if (trans || committed)
3521 trans = btrfs_join_transaction(root, 1);
3524 ret = btrfs_commit_transaction(trans, root);
3533 spin_lock(&space_info->lock);
3534 space_info->bytes_reserved -= orig_bytes;
3535 spin_unlock(&space_info->lock);
3541 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3542 struct btrfs_root *root)
3544 struct btrfs_block_rsv *block_rsv;
3546 block_rsv = trans->block_rsv;
3548 block_rsv = root->block_rsv;
3551 block_rsv = &root->fs_info->empty_block_rsv;
3556 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3560 spin_lock(&block_rsv->lock);
3561 if (block_rsv->reserved >= num_bytes) {
3562 block_rsv->reserved -= num_bytes;
3563 if (block_rsv->reserved < block_rsv->size)
3564 block_rsv->full = 0;
3567 spin_unlock(&block_rsv->lock);
3571 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3572 u64 num_bytes, int update_size)
3574 spin_lock(&block_rsv->lock);
3575 block_rsv->reserved += num_bytes;
3577 block_rsv->size += num_bytes;
3578 else if (block_rsv->reserved >= block_rsv->size)
3579 block_rsv->full = 1;
3580 spin_unlock(&block_rsv->lock);
3583 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3584 struct btrfs_block_rsv *dest, u64 num_bytes)
3586 struct btrfs_space_info *space_info = block_rsv->space_info;
3588 spin_lock(&block_rsv->lock);
3589 if (num_bytes == (u64)-1)
3590 num_bytes = block_rsv->size;
3591 block_rsv->size -= num_bytes;
3592 if (block_rsv->reserved >= block_rsv->size) {
3593 num_bytes = block_rsv->reserved - block_rsv->size;
3594 block_rsv->reserved = block_rsv->size;
3595 block_rsv->full = 1;
3599 spin_unlock(&block_rsv->lock);
3601 if (num_bytes > 0) {
3603 spin_lock(&dest->lock);
3607 bytes_to_add = dest->size - dest->reserved;
3608 bytes_to_add = min(num_bytes, bytes_to_add);
3609 dest->reserved += bytes_to_add;
3610 if (dest->reserved >= dest->size)
3612 num_bytes -= bytes_to_add;
3614 spin_unlock(&dest->lock);
3617 spin_lock(&space_info->lock);
3618 space_info->bytes_reserved -= num_bytes;
3619 spin_unlock(&space_info->lock);
3624 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3625 struct btrfs_block_rsv *dst, u64 num_bytes)
3629 ret = block_rsv_use_bytes(src, num_bytes);
3633 block_rsv_add_bytes(dst, num_bytes, 1);
3637 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3639 memset(rsv, 0, sizeof(*rsv));
3640 spin_lock_init(&rsv->lock);
3641 atomic_set(&rsv->usage, 1);
3643 INIT_LIST_HEAD(&rsv->list);
3646 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3648 struct btrfs_block_rsv *block_rsv;
3649 struct btrfs_fs_info *fs_info = root->fs_info;
3651 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3655 btrfs_init_block_rsv(block_rsv);
3656 block_rsv->space_info = __find_space_info(fs_info,
3657 BTRFS_BLOCK_GROUP_METADATA);
3661 void btrfs_free_block_rsv(struct btrfs_root *root,
3662 struct btrfs_block_rsv *rsv)
3664 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3665 btrfs_block_rsv_release(root, rsv, (u64)-1);
3672 * make the block_rsv struct be able to capture freed space.
3673 * the captured space will re-add to the the block_rsv struct
3674 * after transaction commit
3676 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3677 struct btrfs_block_rsv *block_rsv)
3679 block_rsv->durable = 1;
3680 mutex_lock(&fs_info->durable_block_rsv_mutex);
3681 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3682 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3685 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3686 struct btrfs_root *root,
3687 struct btrfs_block_rsv *block_rsv,
3695 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3697 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3704 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3705 struct btrfs_root *root,
3706 struct btrfs_block_rsv *block_rsv,
3707 u64 min_reserved, int min_factor)
3710 int commit_trans = 0;
3716 spin_lock(&block_rsv->lock);
3718 num_bytes = div_factor(block_rsv->size, min_factor);
3719 if (min_reserved > num_bytes)
3720 num_bytes = min_reserved;
3722 if (block_rsv->reserved >= num_bytes) {
3725 num_bytes -= block_rsv->reserved;
3726 if (block_rsv->durable &&
3727 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3730 spin_unlock(&block_rsv->lock);
3734 if (block_rsv->refill_used) {
3735 ret = reserve_metadata_bytes(trans, root, block_rsv,
3738 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3747 trans = btrfs_join_transaction(root, 1);
3748 BUG_ON(IS_ERR(trans));
3749 ret = btrfs_commit_transaction(trans, root);
3756 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3757 struct btrfs_block_rsv *dst_rsv,
3760 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3763 void btrfs_block_rsv_release(struct btrfs_root *root,
3764 struct btrfs_block_rsv *block_rsv,
3767 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3768 if (global_rsv->full || global_rsv == block_rsv ||
3769 block_rsv->space_info != global_rsv->space_info)
3771 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3775 * helper to calculate size of global block reservation.
3776 * the desired value is sum of space used by extent tree,
3777 * checksum tree and root tree
3779 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3781 struct btrfs_space_info *sinfo;
3785 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3788 * per tree used space accounting can be inaccuracy, so we
3791 spin_lock(&fs_info->extent_root->accounting_lock);
3792 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3793 spin_unlock(&fs_info->extent_root->accounting_lock);
3795 spin_lock(&fs_info->csum_root->accounting_lock);
3796 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3797 spin_unlock(&fs_info->csum_root->accounting_lock);
3799 spin_lock(&fs_info->tree_root->accounting_lock);
3800 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3801 spin_unlock(&fs_info->tree_root->accounting_lock);
3803 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3804 spin_lock(&sinfo->lock);
3805 data_used = sinfo->bytes_used;
3806 spin_unlock(&sinfo->lock);
3808 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3809 spin_lock(&sinfo->lock);
3810 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3812 meta_used = sinfo->bytes_used;
3813 spin_unlock(&sinfo->lock);
3815 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3817 num_bytes += div64_u64(data_used + meta_used, 50);
3819 if (num_bytes * 3 > meta_used)
3820 num_bytes = div64_u64(meta_used, 3);
3822 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3825 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3827 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3828 struct btrfs_space_info *sinfo = block_rsv->space_info;
3831 num_bytes = calc_global_metadata_size(fs_info);
3833 spin_lock(&block_rsv->lock);
3834 spin_lock(&sinfo->lock);
3836 block_rsv->size = num_bytes;
3838 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3839 sinfo->bytes_reserved + sinfo->bytes_readonly +
3840 sinfo->bytes_may_use;
3842 if (sinfo->total_bytes > num_bytes) {
3843 num_bytes = sinfo->total_bytes - num_bytes;
3844 block_rsv->reserved += num_bytes;
3845 sinfo->bytes_reserved += num_bytes;
3848 if (block_rsv->reserved >= block_rsv->size) {
3849 num_bytes = block_rsv->reserved - block_rsv->size;
3850 sinfo->bytes_reserved -= num_bytes;
3851 block_rsv->reserved = block_rsv->size;
3852 block_rsv->full = 1;
3855 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3856 block_rsv->size, block_rsv->reserved);
3858 spin_unlock(&sinfo->lock);
3859 spin_unlock(&block_rsv->lock);
3862 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3864 struct btrfs_space_info *space_info;
3866 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3867 fs_info->chunk_block_rsv.space_info = space_info;
3868 fs_info->chunk_block_rsv.priority = 10;
3870 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3871 fs_info->global_block_rsv.space_info = space_info;
3872 fs_info->global_block_rsv.priority = 10;
3873 fs_info->global_block_rsv.refill_used = 1;
3874 fs_info->delalloc_block_rsv.space_info = space_info;
3875 fs_info->trans_block_rsv.space_info = space_info;
3876 fs_info->empty_block_rsv.space_info = space_info;
3877 fs_info->empty_block_rsv.priority = 10;
3879 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3880 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3881 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3882 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3883 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3885 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3887 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3889 update_global_block_rsv(fs_info);
3892 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3894 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3895 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3896 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3897 WARN_ON(fs_info->trans_block_rsv.size > 0);
3898 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3899 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3900 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3903 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3905 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3909 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3910 struct btrfs_root *root,
3916 if (num_items == 0 || root->fs_info->chunk_root == root)
3919 num_bytes = calc_trans_metadata_size(root, num_items);
3920 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3923 trans->bytes_reserved += num_bytes;
3924 trans->block_rsv = &root->fs_info->trans_block_rsv;
3929 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3930 struct btrfs_root *root)
3932 if (!trans->bytes_reserved)
3935 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3936 btrfs_block_rsv_release(root, trans->block_rsv,
3937 trans->bytes_reserved);
3938 trans->bytes_reserved = 0;
3941 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3942 struct inode *inode)
3944 struct btrfs_root *root = BTRFS_I(inode)->root;
3945 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3946 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3949 * one for deleting orphan item, one for updating inode and
3950 * two for calling btrfs_truncate_inode_items.
3952 * btrfs_truncate_inode_items is a delete operation, it frees
3953 * more space than it uses in most cases. So two units of
3954 * metadata space should be enough for calling it many times.
3955 * If all of the metadata space is used, we can commit
3956 * transaction and use space it freed.
3958 u64 num_bytes = calc_trans_metadata_size(root, 4);
3959 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3962 void btrfs_orphan_release_metadata(struct inode *inode)
3964 struct btrfs_root *root = BTRFS_I(inode)->root;
3965 u64 num_bytes = calc_trans_metadata_size(root, 4);
3966 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3969 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3970 struct btrfs_pending_snapshot *pending)
3972 struct btrfs_root *root = pending->root;
3973 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3974 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3976 * two for root back/forward refs, two for directory entries
3977 * and one for root of the snapshot.
3979 u64 num_bytes = calc_trans_metadata_size(root, 5);
3980 dst_rsv->space_info = src_rsv->space_info;
3981 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3984 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3986 return num_bytes >>= 3;
3989 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3991 struct btrfs_root *root = BTRFS_I(inode)->root;
3992 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3997 if (btrfs_transaction_in_commit(root->fs_info))
3998 schedule_timeout(1);
4000 num_bytes = ALIGN(num_bytes, root->sectorsize);
4002 spin_lock(&BTRFS_I(inode)->accounting_lock);
4003 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4004 if (nr_extents > BTRFS_I(inode)->reserved_extents) {
4005 nr_extents -= BTRFS_I(inode)->reserved_extents;
4006 to_reserve = calc_trans_metadata_size(root, nr_extents);
4011 spin_unlock(&BTRFS_I(inode)->accounting_lock);
4013 to_reserve += calc_csum_metadata_size(inode, num_bytes);
4014 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4018 spin_lock(&BTRFS_I(inode)->accounting_lock);
4019 BTRFS_I(inode)->reserved_extents += nr_extents;
4020 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4021 spin_unlock(&BTRFS_I(inode)->accounting_lock);
4023 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4025 if (block_rsv->size > 512 * 1024 * 1024)
4026 shrink_delalloc(NULL, root, to_reserve, 0);
4031 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4033 struct btrfs_root *root = BTRFS_I(inode)->root;
4037 num_bytes = ALIGN(num_bytes, root->sectorsize);
4038 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4040 spin_lock(&BTRFS_I(inode)->accounting_lock);
4041 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4042 if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4043 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4044 BTRFS_I(inode)->reserved_extents -= nr_extents;
4048 spin_unlock(&BTRFS_I(inode)->accounting_lock);
4050 to_free = calc_csum_metadata_size(inode, num_bytes);
4052 to_free += calc_trans_metadata_size(root, nr_extents);
4054 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4058 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4062 ret = btrfs_check_data_free_space(inode, num_bytes);
4066 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4068 btrfs_free_reserved_data_space(inode, num_bytes);
4075 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4077 btrfs_delalloc_release_metadata(inode, num_bytes);
4078 btrfs_free_reserved_data_space(inode, num_bytes);
4081 static int update_block_group(struct btrfs_trans_handle *trans,
4082 struct btrfs_root *root,
4083 u64 bytenr, u64 num_bytes, int alloc)
4085 struct btrfs_block_group_cache *cache = NULL;
4086 struct btrfs_fs_info *info = root->fs_info;
4087 u64 total = num_bytes;
4092 /* block accounting for super block */
4093 spin_lock(&info->delalloc_lock);
4094 old_val = btrfs_super_bytes_used(&info->super_copy);
4096 old_val += num_bytes;
4098 old_val -= num_bytes;
4099 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4100 spin_unlock(&info->delalloc_lock);
4103 cache = btrfs_lookup_block_group(info, bytenr);
4106 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4107 BTRFS_BLOCK_GROUP_RAID1 |
4108 BTRFS_BLOCK_GROUP_RAID10))
4113 * If this block group has free space cache written out, we
4114 * need to make sure to load it if we are removing space. This
4115 * is because we need the unpinning stage to actually add the
4116 * space back to the block group, otherwise we will leak space.
4118 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4119 cache_block_group(cache, trans, NULL, 1);
4121 byte_in_group = bytenr - cache->key.objectid;
4122 WARN_ON(byte_in_group > cache->key.offset);
4124 spin_lock(&cache->space_info->lock);
4125 spin_lock(&cache->lock);
4127 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4128 cache->disk_cache_state < BTRFS_DC_CLEAR)
4129 cache->disk_cache_state = BTRFS_DC_CLEAR;
4132 old_val = btrfs_block_group_used(&cache->item);
4133 num_bytes = min(total, cache->key.offset - byte_in_group);
4135 old_val += num_bytes;
4136 btrfs_set_block_group_used(&cache->item, old_val);
4137 cache->reserved -= num_bytes;
4138 cache->space_info->bytes_reserved -= num_bytes;
4139 cache->space_info->bytes_used += num_bytes;
4140 cache->space_info->disk_used += num_bytes * factor;
4141 spin_unlock(&cache->lock);
4142 spin_unlock(&cache->space_info->lock);
4144 old_val -= num_bytes;
4145 btrfs_set_block_group_used(&cache->item, old_val);
4146 cache->pinned += num_bytes;
4147 cache->space_info->bytes_pinned += num_bytes;
4148 cache->space_info->bytes_used -= num_bytes;
4149 cache->space_info->disk_used -= num_bytes * factor;
4150 spin_unlock(&cache->lock);
4151 spin_unlock(&cache->space_info->lock);
4153 set_extent_dirty(info->pinned_extents,
4154 bytenr, bytenr + num_bytes - 1,
4155 GFP_NOFS | __GFP_NOFAIL);
4157 btrfs_put_block_group(cache);
4159 bytenr += num_bytes;
4164 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4166 struct btrfs_block_group_cache *cache;
4169 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4173 bytenr = cache->key.objectid;
4174 btrfs_put_block_group(cache);
4179 static int pin_down_extent(struct btrfs_root *root,
4180 struct btrfs_block_group_cache *cache,
4181 u64 bytenr, u64 num_bytes, int reserved)
4183 spin_lock(&cache->space_info->lock);
4184 spin_lock(&cache->lock);
4185 cache->pinned += num_bytes;
4186 cache->space_info->bytes_pinned += num_bytes;
4188 cache->reserved -= num_bytes;
4189 cache->space_info->bytes_reserved -= num_bytes;
4191 spin_unlock(&cache->lock);
4192 spin_unlock(&cache->space_info->lock);
4194 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4195 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4200 * this function must be called within transaction
4202 int btrfs_pin_extent(struct btrfs_root *root,
4203 u64 bytenr, u64 num_bytes, int reserved)
4205 struct btrfs_block_group_cache *cache;
4207 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4210 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4212 btrfs_put_block_group(cache);
4217 * update size of reserved extents. this function may return -EAGAIN
4218 * if 'reserve' is true or 'sinfo' is false.
4220 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4221 u64 num_bytes, int reserve, int sinfo)
4225 struct btrfs_space_info *space_info = cache->space_info;
4226 spin_lock(&space_info->lock);
4227 spin_lock(&cache->lock);
4232 cache->reserved += num_bytes;
4233 space_info->bytes_reserved += num_bytes;
4237 space_info->bytes_readonly += num_bytes;
4238 cache->reserved -= num_bytes;
4239 space_info->bytes_reserved -= num_bytes;
4241 spin_unlock(&cache->lock);
4242 spin_unlock(&space_info->lock);
4244 spin_lock(&cache->lock);
4249 cache->reserved += num_bytes;
4251 cache->reserved -= num_bytes;
4253 spin_unlock(&cache->lock);
4258 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4259 struct btrfs_root *root)
4261 struct btrfs_fs_info *fs_info = root->fs_info;
4262 struct btrfs_caching_control *next;
4263 struct btrfs_caching_control *caching_ctl;
4264 struct btrfs_block_group_cache *cache;
4266 down_write(&fs_info->extent_commit_sem);
4268 list_for_each_entry_safe(caching_ctl, next,
4269 &fs_info->caching_block_groups, list) {
4270 cache = caching_ctl->block_group;
4271 if (block_group_cache_done(cache)) {
4272 cache->last_byte_to_unpin = (u64)-1;
4273 list_del_init(&caching_ctl->list);
4274 put_caching_control(caching_ctl);
4276 cache->last_byte_to_unpin = caching_ctl->progress;
4280 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4281 fs_info->pinned_extents = &fs_info->freed_extents[1];
4283 fs_info->pinned_extents = &fs_info->freed_extents[0];
4285 up_write(&fs_info->extent_commit_sem);
4287 update_global_block_rsv(fs_info);
4291 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4293 struct btrfs_fs_info *fs_info = root->fs_info;
4294 struct btrfs_block_group_cache *cache = NULL;
4297 while (start <= end) {
4299 start >= cache->key.objectid + cache->key.offset) {
4301 btrfs_put_block_group(cache);
4302 cache = btrfs_lookup_block_group(fs_info, start);
4306 len = cache->key.objectid + cache->key.offset - start;
4307 len = min(len, end + 1 - start);
4309 if (start < cache->last_byte_to_unpin) {
4310 len = min(len, cache->last_byte_to_unpin - start);
4311 btrfs_add_free_space(cache, start, len);
4316 spin_lock(&cache->space_info->lock);
4317 spin_lock(&cache->lock);
4318 cache->pinned -= len;
4319 cache->space_info->bytes_pinned -= len;
4321 cache->space_info->bytes_readonly += len;
4322 } else if (cache->reserved_pinned > 0) {
4323 len = min(len, cache->reserved_pinned);
4324 cache->reserved_pinned -= len;
4325 cache->space_info->bytes_reserved += len;
4327 spin_unlock(&cache->lock);
4328 spin_unlock(&cache->space_info->lock);
4332 btrfs_put_block_group(cache);
4336 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4337 struct btrfs_root *root)
4339 struct btrfs_fs_info *fs_info = root->fs_info;
4340 struct extent_io_tree *unpin;
4341 struct btrfs_block_rsv *block_rsv;
4342 struct btrfs_block_rsv *next_rsv;
4348 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4349 unpin = &fs_info->freed_extents[1];
4351 unpin = &fs_info->freed_extents[0];
4354 ret = find_first_extent_bit(unpin, 0, &start, &end,
4359 ret = btrfs_discard_extent(root, start, end + 1 - start);
4361 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4362 unpin_extent_range(root, start, end);
4366 mutex_lock(&fs_info->durable_block_rsv_mutex);
4367 list_for_each_entry_safe(block_rsv, next_rsv,
4368 &fs_info->durable_block_rsv_list, list) {
4370 idx = trans->transid & 0x1;
4371 if (block_rsv->freed[idx] > 0) {
4372 block_rsv_add_bytes(block_rsv,
4373 block_rsv->freed[idx], 0);
4374 block_rsv->freed[idx] = 0;
4376 if (atomic_read(&block_rsv->usage) == 0) {
4377 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4379 if (block_rsv->freed[0] == 0 &&
4380 block_rsv->freed[1] == 0) {
4381 list_del_init(&block_rsv->list);
4385 btrfs_block_rsv_release(root, block_rsv, 0);
4388 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4393 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4394 struct btrfs_root *root,
4395 u64 bytenr, u64 num_bytes, u64 parent,
4396 u64 root_objectid, u64 owner_objectid,
4397 u64 owner_offset, int refs_to_drop,
4398 struct btrfs_delayed_extent_op *extent_op)
4400 struct btrfs_key key;
4401 struct btrfs_path *path;
4402 struct btrfs_fs_info *info = root->fs_info;
4403 struct btrfs_root *extent_root = info->extent_root;
4404 struct extent_buffer *leaf;
4405 struct btrfs_extent_item *ei;
4406 struct btrfs_extent_inline_ref *iref;
4409 int extent_slot = 0;
4410 int found_extent = 0;
4415 path = btrfs_alloc_path();
4420 path->leave_spinning = 1;
4422 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4423 BUG_ON(!is_data && refs_to_drop != 1);
4425 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4426 bytenr, num_bytes, parent,
4427 root_objectid, owner_objectid,
4430 extent_slot = path->slots[0];
4431 while (extent_slot >= 0) {
4432 btrfs_item_key_to_cpu(path->nodes[0], &key,
4434 if (key.objectid != bytenr)
4436 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4437 key.offset == num_bytes) {
4441 if (path->slots[0] - extent_slot > 5)
4445 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4446 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4447 if (found_extent && item_size < sizeof(*ei))
4450 if (!found_extent) {
4452 ret = remove_extent_backref(trans, extent_root, path,
4456 btrfs_release_path(extent_root, path);
4457 path->leave_spinning = 1;
4459 key.objectid = bytenr;
4460 key.type = BTRFS_EXTENT_ITEM_KEY;
4461 key.offset = num_bytes;
4463 ret = btrfs_search_slot(trans, extent_root,
4466 printk(KERN_ERR "umm, got %d back from search"
4467 ", was looking for %llu\n", ret,
4468 (unsigned long long)bytenr);
4469 btrfs_print_leaf(extent_root, path->nodes[0]);
4472 extent_slot = path->slots[0];
4475 btrfs_print_leaf(extent_root, path->nodes[0]);
4477 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4478 "parent %llu root %llu owner %llu offset %llu\n",
4479 (unsigned long long)bytenr,
4480 (unsigned long long)parent,
4481 (unsigned long long)root_objectid,
4482 (unsigned long long)owner_objectid,
4483 (unsigned long long)owner_offset);
4486 leaf = path->nodes[0];
4487 item_size = btrfs_item_size_nr(leaf, extent_slot);
4488 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4489 if (item_size < sizeof(*ei)) {
4490 BUG_ON(found_extent || extent_slot != path->slots[0]);
4491 ret = convert_extent_item_v0(trans, extent_root, path,
4495 btrfs_release_path(extent_root, path);
4496 path->leave_spinning = 1;
4498 key.objectid = bytenr;
4499 key.type = BTRFS_EXTENT_ITEM_KEY;
4500 key.offset = num_bytes;
4502 ret = btrfs_search_slot(trans, extent_root, &key, path,
4505 printk(KERN_ERR "umm, got %d back from search"
4506 ", was looking for %llu\n", ret,
4507 (unsigned long long)bytenr);
4508 btrfs_print_leaf(extent_root, path->nodes[0]);
4511 extent_slot = path->slots[0];
4512 leaf = path->nodes[0];
4513 item_size = btrfs_item_size_nr(leaf, extent_slot);
4516 BUG_ON(item_size < sizeof(*ei));
4517 ei = btrfs_item_ptr(leaf, extent_slot,
4518 struct btrfs_extent_item);
4519 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4520 struct btrfs_tree_block_info *bi;
4521 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4522 bi = (struct btrfs_tree_block_info *)(ei + 1);
4523 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4526 refs = btrfs_extent_refs(leaf, ei);
4527 BUG_ON(refs < refs_to_drop);
4528 refs -= refs_to_drop;
4532 __run_delayed_extent_op(extent_op, leaf, ei);
4534 * In the case of inline back ref, reference count will
4535 * be updated by remove_extent_backref
4538 BUG_ON(!found_extent);
4540 btrfs_set_extent_refs(leaf, ei, refs);
4541 btrfs_mark_buffer_dirty(leaf);
4544 ret = remove_extent_backref(trans, extent_root, path,
4551 BUG_ON(is_data && refs_to_drop !=
4552 extent_data_ref_count(root, path, iref));
4554 BUG_ON(path->slots[0] != extent_slot);
4556 BUG_ON(path->slots[0] != extent_slot + 1);
4557 path->slots[0] = extent_slot;
4562 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4565 btrfs_release_path(extent_root, path);
4568 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4571 invalidate_mapping_pages(info->btree_inode->i_mapping,
4572 bytenr >> PAGE_CACHE_SHIFT,
4573 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4576 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4579 btrfs_free_path(path);
4584 * when we free an block, it is possible (and likely) that we free the last
4585 * delayed ref for that extent as well. This searches the delayed ref tree for
4586 * a given extent, and if there are no other delayed refs to be processed, it
4587 * removes it from the tree.
4589 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4590 struct btrfs_root *root, u64 bytenr)
4592 struct btrfs_delayed_ref_head *head;
4593 struct btrfs_delayed_ref_root *delayed_refs;
4594 struct btrfs_delayed_ref_node *ref;
4595 struct rb_node *node;
4598 delayed_refs = &trans->transaction->delayed_refs;
4599 spin_lock(&delayed_refs->lock);
4600 head = btrfs_find_delayed_ref_head(trans, bytenr);
4604 node = rb_prev(&head->node.rb_node);
4608 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4610 /* there are still entries for this ref, we can't drop it */
4611 if (ref->bytenr == bytenr)
4614 if (head->extent_op) {
4615 if (!head->must_insert_reserved)
4617 kfree(head->extent_op);
4618 head->extent_op = NULL;
4622 * waiting for the lock here would deadlock. If someone else has it
4623 * locked they are already in the process of dropping it anyway
4625 if (!mutex_trylock(&head->mutex))
4629 * at this point we have a head with no other entries. Go
4630 * ahead and process it.
4632 head->node.in_tree = 0;
4633 rb_erase(&head->node.rb_node, &delayed_refs->root);
4635 delayed_refs->num_entries--;
4638 * we don't take a ref on the node because we're removing it from the
4639 * tree, so we just steal the ref the tree was holding.
4641 delayed_refs->num_heads--;
4642 if (list_empty(&head->cluster))
4643 delayed_refs->num_heads_ready--;
4645 list_del_init(&head->cluster);
4646 spin_unlock(&delayed_refs->lock);
4648 BUG_ON(head->extent_op);
4649 if (head->must_insert_reserved)
4652 mutex_unlock(&head->mutex);
4653 btrfs_put_delayed_ref(&head->node);
4656 spin_unlock(&delayed_refs->lock);
4660 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4661 struct btrfs_root *root,
4662 struct extent_buffer *buf,
4663 u64 parent, int last_ref)
4665 struct btrfs_block_rsv *block_rsv;
4666 struct btrfs_block_group_cache *cache = NULL;
4669 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4670 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4671 parent, root->root_key.objectid,
4672 btrfs_header_level(buf),
4673 BTRFS_DROP_DELAYED_REF, NULL);
4680 block_rsv = get_block_rsv(trans, root);
4681 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4682 if (block_rsv->space_info != cache->space_info)
4685 if (btrfs_header_generation(buf) == trans->transid) {
4686 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4687 ret = check_ref_cleanup(trans, root, buf->start);
4692 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4693 pin_down_extent(root, cache, buf->start, buf->len, 1);
4697 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4699 btrfs_add_free_space(cache, buf->start, buf->len);
4700 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4701 if (ret == -EAGAIN) {
4702 /* block group became read-only */
4703 update_reserved_bytes(cache, buf->len, 0, 1);
4708 spin_lock(&block_rsv->lock);
4709 if (block_rsv->reserved < block_rsv->size) {
4710 block_rsv->reserved += buf->len;
4713 spin_unlock(&block_rsv->lock);
4716 spin_lock(&cache->space_info->lock);
4717 cache->space_info->bytes_reserved -= buf->len;
4718 spin_unlock(&cache->space_info->lock);
4723 if (block_rsv->durable && !cache->ro) {
4725 spin_lock(&cache->lock);
4727 cache->reserved_pinned += buf->len;
4730 spin_unlock(&cache->lock);
4733 spin_lock(&block_rsv->lock);
4734 block_rsv->freed[trans->transid & 0x1] += buf->len;
4735 spin_unlock(&block_rsv->lock);
4739 btrfs_put_block_group(cache);
4742 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4743 struct btrfs_root *root,
4744 u64 bytenr, u64 num_bytes, u64 parent,
4745 u64 root_objectid, u64 owner, u64 offset)
4750 * tree log blocks never actually go into the extent allocation
4751 * tree, just update pinning info and exit early.
4753 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4754 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4755 /* unlocks the pinned mutex */
4756 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4758 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4759 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4760 parent, root_objectid, (int)owner,
4761 BTRFS_DROP_DELAYED_REF, NULL);
4764 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4765 parent, root_objectid, owner,
4766 offset, BTRFS_DROP_DELAYED_REF, NULL);
4772 static u64 stripe_align(struct btrfs_root *root, u64 val)
4774 u64 mask = ((u64)root->stripesize - 1);
4775 u64 ret = (val + mask) & ~mask;
4780 * when we wait for progress in the block group caching, its because
4781 * our allocation attempt failed at least once. So, we must sleep
4782 * and let some progress happen before we try again.
4784 * This function will sleep at least once waiting for new free space to
4785 * show up, and then it will check the block group free space numbers
4786 * for our min num_bytes. Another option is to have it go ahead
4787 * and look in the rbtree for a free extent of a given size, but this
4791 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4794 struct btrfs_caching_control *caching_ctl;
4797 caching_ctl = get_caching_control(cache);
4801 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4802 (cache->free_space >= num_bytes));
4804 put_caching_control(caching_ctl);
4809 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4811 struct btrfs_caching_control *caching_ctl;
4814 caching_ctl = get_caching_control(cache);
4818 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4820 put_caching_control(caching_ctl);
4824 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4827 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4829 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4831 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4833 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4840 enum btrfs_loop_type {
4841 LOOP_FIND_IDEAL = 0,
4842 LOOP_CACHING_NOWAIT = 1,
4843 LOOP_CACHING_WAIT = 2,
4844 LOOP_ALLOC_CHUNK = 3,
4845 LOOP_NO_EMPTY_SIZE = 4,
4849 * walks the btree of allocated extents and find a hole of a given size.
4850 * The key ins is changed to record the hole:
4851 * ins->objectid == block start
4852 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4853 * ins->offset == number of blocks
4854 * Any available blocks before search_start are skipped.
4856 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4857 struct btrfs_root *orig_root,
4858 u64 num_bytes, u64 empty_size,
4859 u64 search_start, u64 search_end,
4860 u64 hint_byte, struct btrfs_key *ins,
4864 struct btrfs_root *root = orig_root->fs_info->extent_root;
4865 struct btrfs_free_cluster *last_ptr = NULL;
4866 struct btrfs_block_group_cache *block_group = NULL;
4867 int empty_cluster = 2 * 1024 * 1024;
4868 int allowed_chunk_alloc = 0;
4869 int done_chunk_alloc = 0;
4870 struct btrfs_space_info *space_info;
4871 int last_ptr_loop = 0;
4874 bool found_uncached_bg = false;
4875 bool failed_cluster_refill = false;
4876 bool failed_alloc = false;
4877 bool use_cluster = true;
4878 u64 ideal_cache_percent = 0;
4879 u64 ideal_cache_offset = 0;
4881 WARN_ON(num_bytes < root->sectorsize);
4882 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4886 space_info = __find_space_info(root->fs_info, data);
4888 printk(KERN_ERR "No space info for %d\n", data);
4893 * If the space info is for both data and metadata it means we have a
4894 * small filesystem and we can't use the clustering stuff.
4896 if (btrfs_mixed_space_info(space_info))
4897 use_cluster = false;
4899 if (orig_root->ref_cows || empty_size)
4900 allowed_chunk_alloc = 1;
4902 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4903 last_ptr = &root->fs_info->meta_alloc_cluster;
4904 if (!btrfs_test_opt(root, SSD))
4905 empty_cluster = 64 * 1024;
4908 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4909 btrfs_test_opt(root, SSD)) {
4910 last_ptr = &root->fs_info->data_alloc_cluster;
4914 spin_lock(&last_ptr->lock);
4915 if (last_ptr->block_group)
4916 hint_byte = last_ptr->window_start;
4917 spin_unlock(&last_ptr->lock);
4920 search_start = max(search_start, first_logical_byte(root, 0));
4921 search_start = max(search_start, hint_byte);
4926 if (search_start == hint_byte) {
4928 block_group = btrfs_lookup_block_group(root->fs_info,
4931 * we don't want to use the block group if it doesn't match our
4932 * allocation bits, or if its not cached.
4934 * However if we are re-searching with an ideal block group
4935 * picked out then we don't care that the block group is cached.
4937 if (block_group && block_group_bits(block_group, data) &&
4938 (block_group->cached != BTRFS_CACHE_NO ||
4939 search_start == ideal_cache_offset)) {
4940 down_read(&space_info->groups_sem);
4941 if (list_empty(&block_group->list) ||
4944 * someone is removing this block group,
4945 * we can't jump into the have_block_group
4946 * target because our list pointers are not
4949 btrfs_put_block_group(block_group);
4950 up_read(&space_info->groups_sem);
4952 index = get_block_group_index(block_group);
4953 goto have_block_group;
4955 } else if (block_group) {
4956 btrfs_put_block_group(block_group);
4960 down_read(&space_info->groups_sem);
4961 list_for_each_entry(block_group, &space_info->block_groups[index],
4966 btrfs_get_block_group(block_group);
4967 search_start = block_group->key.objectid;
4970 * this can happen if we end up cycling through all the
4971 * raid types, but we want to make sure we only allocate
4972 * for the proper type.
4974 if (!block_group_bits(block_group, data)) {
4975 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4976 BTRFS_BLOCK_GROUP_RAID1 |
4977 BTRFS_BLOCK_GROUP_RAID10;
4980 * if they asked for extra copies and this block group
4981 * doesn't provide them, bail. This does allow us to
4982 * fill raid0 from raid1.
4984 if ((data & extra) && !(block_group->flags & extra))
4989 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4992 ret = cache_block_group(block_group, trans,
4994 if (block_group->cached == BTRFS_CACHE_FINISHED)
4995 goto have_block_group;
4997 free_percent = btrfs_block_group_used(&block_group->item);
4998 free_percent *= 100;
4999 free_percent = div64_u64(free_percent,
5000 block_group->key.offset);
5001 free_percent = 100 - free_percent;
5002 if (free_percent > ideal_cache_percent &&
5003 likely(!block_group->ro)) {
5004 ideal_cache_offset = block_group->key.objectid;
5005 ideal_cache_percent = free_percent;
5009 * We only want to start kthread caching if we are at
5010 * the point where we will wait for caching to make
5011 * progress, or if our ideal search is over and we've
5012 * found somebody to start caching.
5014 if (loop > LOOP_CACHING_NOWAIT ||
5015 (loop > LOOP_FIND_IDEAL &&
5016 atomic_read(&space_info->caching_threads) < 2)) {
5017 ret = cache_block_group(block_group, trans,
5021 found_uncached_bg = true;
5024 * If loop is set for cached only, try the next block
5027 if (loop == LOOP_FIND_IDEAL)
5031 cached = block_group_cache_done(block_group);
5032 if (unlikely(!cached))
5033 found_uncached_bg = true;
5035 if (unlikely(block_group->ro))
5039 * Ok we want to try and use the cluster allocator, so lets look
5040 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5041 * have tried the cluster allocator plenty of times at this
5042 * point and not have found anything, so we are likely way too
5043 * fragmented for the clustering stuff to find anything, so lets
5044 * just skip it and let the allocator find whatever block it can
5047 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5049 * the refill lock keeps out other
5050 * people trying to start a new cluster
5052 spin_lock(&last_ptr->refill_lock);
5053 if (last_ptr->block_group &&
5054 (last_ptr->block_group->ro ||
5055 !block_group_bits(last_ptr->block_group, data))) {
5057 goto refill_cluster;
5060 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5061 num_bytes, search_start);
5063 /* we have a block, we're done */
5064 spin_unlock(&last_ptr->refill_lock);
5068 spin_lock(&last_ptr->lock);
5070 * whoops, this cluster doesn't actually point to
5071 * this block group. Get a ref on the block
5072 * group is does point to and try again
5074 if (!last_ptr_loop && last_ptr->block_group &&
5075 last_ptr->block_group != block_group) {
5077 btrfs_put_block_group(block_group);
5078 block_group = last_ptr->block_group;
5079 btrfs_get_block_group(block_group);
5080 spin_unlock(&last_ptr->lock);
5081 spin_unlock(&last_ptr->refill_lock);
5084 search_start = block_group->key.objectid;
5086 * we know this block group is properly
5087 * in the list because
5088 * btrfs_remove_block_group, drops the
5089 * cluster before it removes the block
5090 * group from the list
5092 goto have_block_group;
5094 spin_unlock(&last_ptr->lock);
5097 * this cluster didn't work out, free it and
5100 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5104 /* allocate a cluster in this block group */
5105 ret = btrfs_find_space_cluster(trans, root,
5106 block_group, last_ptr,
5108 empty_cluster + empty_size);
5111 * now pull our allocation out of this
5114 offset = btrfs_alloc_from_cluster(block_group,
5115 last_ptr, num_bytes,
5118 /* we found one, proceed */
5119 spin_unlock(&last_ptr->refill_lock);
5122 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5123 && !failed_cluster_refill) {
5124 spin_unlock(&last_ptr->refill_lock);
5126 failed_cluster_refill = true;
5127 wait_block_group_cache_progress(block_group,
5128 num_bytes + empty_cluster + empty_size);
5129 goto have_block_group;
5133 * at this point we either didn't find a cluster
5134 * or we weren't able to allocate a block from our
5135 * cluster. Free the cluster we've been trying
5136 * to use, and go to the next block group
5138 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5139 spin_unlock(&last_ptr->refill_lock);
5143 offset = btrfs_find_space_for_alloc(block_group, search_start,
5144 num_bytes, empty_size);
5146 * If we didn't find a chunk, and we haven't failed on this
5147 * block group before, and this block group is in the middle of
5148 * caching and we are ok with waiting, then go ahead and wait
5149 * for progress to be made, and set failed_alloc to true.
5151 * If failed_alloc is true then we've already waited on this
5152 * block group once and should move on to the next block group.
5154 if (!offset && !failed_alloc && !cached &&
5155 loop > LOOP_CACHING_NOWAIT) {
5156 wait_block_group_cache_progress(block_group,
5157 num_bytes + empty_size);
5158 failed_alloc = true;
5159 goto have_block_group;
5160 } else if (!offset) {
5164 search_start = stripe_align(root, offset);
5165 /* move on to the next group */
5166 if (search_start + num_bytes >= search_end) {
5167 btrfs_add_free_space(block_group, offset, num_bytes);
5171 /* move on to the next group */
5172 if (search_start + num_bytes >
5173 block_group->key.objectid + block_group->key.offset) {
5174 btrfs_add_free_space(block_group, offset, num_bytes);
5178 ins->objectid = search_start;
5179 ins->offset = num_bytes;
5181 if (offset < search_start)
5182 btrfs_add_free_space(block_group, offset,
5183 search_start - offset);
5184 BUG_ON(offset > search_start);
5186 ret = update_reserved_bytes(block_group, num_bytes, 1,
5187 (data & BTRFS_BLOCK_GROUP_DATA));
5188 if (ret == -EAGAIN) {
5189 btrfs_add_free_space(block_group, offset, num_bytes);
5193 /* we are all good, lets return */
5194 ins->objectid = search_start;
5195 ins->offset = num_bytes;
5197 if (offset < search_start)
5198 btrfs_add_free_space(block_group, offset,
5199 search_start - offset);
5200 BUG_ON(offset > search_start);
5203 failed_cluster_refill = false;
5204 failed_alloc = false;
5205 BUG_ON(index != get_block_group_index(block_group));
5206 btrfs_put_block_group(block_group);
5208 up_read(&space_info->groups_sem);
5210 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5213 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5214 * for them to make caching progress. Also
5215 * determine the best possible bg to cache
5216 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5217 * caching kthreads as we move along
5218 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5219 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5220 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5223 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5224 (found_uncached_bg || empty_size || empty_cluster ||
5225 allowed_chunk_alloc)) {
5227 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5228 found_uncached_bg = false;
5230 if (!ideal_cache_percent &&
5231 atomic_read(&space_info->caching_threads))
5235 * 1 of the following 2 things have happened so far
5237 * 1) We found an ideal block group for caching that
5238 * is mostly full and will cache quickly, so we might
5239 * as well wait for it.
5241 * 2) We searched for cached only and we didn't find
5242 * anything, and we didn't start any caching kthreads
5243 * either, so chances are we will loop through and
5244 * start a couple caching kthreads, and then come back
5245 * around and just wait for them. This will be slower
5246 * because we will have 2 caching kthreads reading at
5247 * the same time when we could have just started one
5248 * and waited for it to get far enough to give us an
5249 * allocation, so go ahead and go to the wait caching
5252 loop = LOOP_CACHING_WAIT;
5253 search_start = ideal_cache_offset;
5254 ideal_cache_percent = 0;
5256 } else if (loop == LOOP_FIND_IDEAL) {
5258 * Didn't find a uncached bg, wait on anything we find
5261 loop = LOOP_CACHING_WAIT;
5265 if (loop < LOOP_CACHING_WAIT) {
5270 if (loop == LOOP_ALLOC_CHUNK) {
5275 if (allowed_chunk_alloc) {
5276 ret = do_chunk_alloc(trans, root, num_bytes +
5277 2 * 1024 * 1024, data, 1);
5278 allowed_chunk_alloc = 0;
5279 done_chunk_alloc = 1;
5280 } else if (!done_chunk_alloc) {
5281 space_info->force_alloc = 1;
5284 if (loop < LOOP_NO_EMPTY_SIZE) {
5289 } else if (!ins->objectid) {
5293 /* we found what we needed */
5294 if (ins->objectid) {
5295 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5296 trans->block_group = block_group->key.objectid;
5298 btrfs_put_block_group(block_group);
5305 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5306 int dump_block_groups)
5308 struct btrfs_block_group_cache *cache;
5311 spin_lock(&info->lock);
5312 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5313 (unsigned long long)(info->total_bytes - info->bytes_used -
5314 info->bytes_pinned - info->bytes_reserved -
5315 info->bytes_readonly),
5316 (info->full) ? "" : "not ");
5317 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5318 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5319 (unsigned long long)info->total_bytes,
5320 (unsigned long long)info->bytes_used,
5321 (unsigned long long)info->bytes_pinned,
5322 (unsigned long long)info->bytes_reserved,
5323 (unsigned long long)info->bytes_may_use,
5324 (unsigned long long)info->bytes_readonly);
5325 spin_unlock(&info->lock);
5327 if (!dump_block_groups)
5330 down_read(&info->groups_sem);
5332 list_for_each_entry(cache, &info->block_groups[index], list) {
5333 spin_lock(&cache->lock);
5334 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5335 "%llu pinned %llu reserved\n",
5336 (unsigned long long)cache->key.objectid,
5337 (unsigned long long)cache->key.offset,
5338 (unsigned long long)btrfs_block_group_used(&cache->item),
5339 (unsigned long long)cache->pinned,
5340 (unsigned long long)cache->reserved);
5341 btrfs_dump_free_space(cache, bytes);
5342 spin_unlock(&cache->lock);
5344 if (++index < BTRFS_NR_RAID_TYPES)
5346 up_read(&info->groups_sem);
5349 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5350 struct btrfs_root *root,
5351 u64 num_bytes, u64 min_alloc_size,
5352 u64 empty_size, u64 hint_byte,
5353 u64 search_end, struct btrfs_key *ins,
5357 u64 search_start = 0;
5359 data = btrfs_get_alloc_profile(root, data);
5362 * the only place that sets empty_size is btrfs_realloc_node, which
5363 * is not called recursively on allocations
5365 if (empty_size || root->ref_cows)
5366 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5367 num_bytes + 2 * 1024 * 1024, data, 0);
5369 WARN_ON(num_bytes < root->sectorsize);
5370 ret = find_free_extent(trans, root, num_bytes, empty_size,
5371 search_start, search_end, hint_byte,
5374 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5375 num_bytes = num_bytes >> 1;
5376 num_bytes = num_bytes & ~(root->sectorsize - 1);
5377 num_bytes = max(num_bytes, min_alloc_size);
5378 do_chunk_alloc(trans, root->fs_info->extent_root,
5379 num_bytes, data, 1);
5382 if (ret == -ENOSPC) {
5383 struct btrfs_space_info *sinfo;
5385 sinfo = __find_space_info(root->fs_info, data);
5386 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5387 "wanted %llu\n", (unsigned long long)data,
5388 (unsigned long long)num_bytes);
5389 dump_space_info(sinfo, num_bytes, 1);
5395 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5397 struct btrfs_block_group_cache *cache;
5400 cache = btrfs_lookup_block_group(root->fs_info, start);
5402 printk(KERN_ERR "Unable to find block group for %llu\n",
5403 (unsigned long long)start);
5407 ret = btrfs_discard_extent(root, start, len);
5409 btrfs_add_free_space(cache, start, len);
5410 update_reserved_bytes(cache, len, 0, 1);
5411 btrfs_put_block_group(cache);
5416 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5417 struct btrfs_root *root,
5418 u64 parent, u64 root_objectid,
5419 u64 flags, u64 owner, u64 offset,
5420 struct btrfs_key *ins, int ref_mod)
5423 struct btrfs_fs_info *fs_info = root->fs_info;
5424 struct btrfs_extent_item *extent_item;
5425 struct btrfs_extent_inline_ref *iref;
5426 struct btrfs_path *path;
5427 struct extent_buffer *leaf;
5432 type = BTRFS_SHARED_DATA_REF_KEY;
5434 type = BTRFS_EXTENT_DATA_REF_KEY;
5436 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5438 path = btrfs_alloc_path();
5441 path->leave_spinning = 1;
5442 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5446 leaf = path->nodes[0];
5447 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5448 struct btrfs_extent_item);
5449 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5450 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5451 btrfs_set_extent_flags(leaf, extent_item,
5452 flags | BTRFS_EXTENT_FLAG_DATA);
5454 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5455 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5457 struct btrfs_shared_data_ref *ref;
5458 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5459 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5460 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5462 struct btrfs_extent_data_ref *ref;
5463 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5464 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5465 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5466 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5467 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5470 btrfs_mark_buffer_dirty(path->nodes[0]);
5471 btrfs_free_path(path);
5473 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5475 printk(KERN_ERR "btrfs update block group failed for %llu "
5476 "%llu\n", (unsigned long long)ins->objectid,
5477 (unsigned long long)ins->offset);
5483 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5484 struct btrfs_root *root,
5485 u64 parent, u64 root_objectid,
5486 u64 flags, struct btrfs_disk_key *key,
5487 int level, struct btrfs_key *ins)
5490 struct btrfs_fs_info *fs_info = root->fs_info;
5491 struct btrfs_extent_item *extent_item;
5492 struct btrfs_tree_block_info *block_info;
5493 struct btrfs_extent_inline_ref *iref;
5494 struct btrfs_path *path;
5495 struct extent_buffer *leaf;
5496 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5498 path = btrfs_alloc_path();
5501 path->leave_spinning = 1;
5502 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5506 leaf = path->nodes[0];
5507 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5508 struct btrfs_extent_item);
5509 btrfs_set_extent_refs(leaf, extent_item, 1);
5510 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5511 btrfs_set_extent_flags(leaf, extent_item,
5512 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5513 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5515 btrfs_set_tree_block_key(leaf, block_info, key);
5516 btrfs_set_tree_block_level(leaf, block_info, level);
5518 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5520 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5521 btrfs_set_extent_inline_ref_type(leaf, iref,
5522 BTRFS_SHARED_BLOCK_REF_KEY);
5523 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5525 btrfs_set_extent_inline_ref_type(leaf, iref,
5526 BTRFS_TREE_BLOCK_REF_KEY);
5527 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5530 btrfs_mark_buffer_dirty(leaf);
5531 btrfs_free_path(path);
5533 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5535 printk(KERN_ERR "btrfs update block group failed for %llu "
5536 "%llu\n", (unsigned long long)ins->objectid,
5537 (unsigned long long)ins->offset);
5543 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5544 struct btrfs_root *root,
5545 u64 root_objectid, u64 owner,
5546 u64 offset, struct btrfs_key *ins)
5550 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5552 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5553 0, root_objectid, owner, offset,
5554 BTRFS_ADD_DELAYED_EXTENT, NULL);
5559 * this is used by the tree logging recovery code. It records that
5560 * an extent has been allocated and makes sure to clear the free
5561 * space cache bits as well
5563 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5564 struct btrfs_root *root,
5565 u64 root_objectid, u64 owner, u64 offset,
5566 struct btrfs_key *ins)
5569 struct btrfs_block_group_cache *block_group;
5570 struct btrfs_caching_control *caching_ctl;
5571 u64 start = ins->objectid;
5572 u64 num_bytes = ins->offset;
5574 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5575 cache_block_group(block_group, trans, NULL, 0);
5576 caching_ctl = get_caching_control(block_group);
5579 BUG_ON(!block_group_cache_done(block_group));
5580 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5583 mutex_lock(&caching_ctl->mutex);
5585 if (start >= caching_ctl->progress) {
5586 ret = add_excluded_extent(root, start, num_bytes);
5588 } else if (start + num_bytes <= caching_ctl->progress) {
5589 ret = btrfs_remove_free_space(block_group,
5593 num_bytes = caching_ctl->progress - start;
5594 ret = btrfs_remove_free_space(block_group,
5598 start = caching_ctl->progress;
5599 num_bytes = ins->objectid + ins->offset -
5600 caching_ctl->progress;
5601 ret = add_excluded_extent(root, start, num_bytes);
5605 mutex_unlock(&caching_ctl->mutex);
5606 put_caching_control(caching_ctl);
5609 ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5611 btrfs_put_block_group(block_group);
5612 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5613 0, owner, offset, ins, 1);
5617 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5618 struct btrfs_root *root,
5619 u64 bytenr, u32 blocksize,
5622 struct extent_buffer *buf;
5624 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5626 return ERR_PTR(-ENOMEM);
5627 btrfs_set_header_generation(buf, trans->transid);
5628 btrfs_set_buffer_lockdep_class(buf, level);
5629 btrfs_tree_lock(buf);
5630 clean_tree_block(trans, root, buf);
5632 btrfs_set_lock_blocking(buf);
5633 btrfs_set_buffer_uptodate(buf);
5635 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5637 * we allow two log transactions at a time, use different
5638 * EXENT bit to differentiate dirty pages.
5640 if (root->log_transid % 2 == 0)
5641 set_extent_dirty(&root->dirty_log_pages, buf->start,
5642 buf->start + buf->len - 1, GFP_NOFS);
5644 set_extent_new(&root->dirty_log_pages, buf->start,
5645 buf->start + buf->len - 1, GFP_NOFS);
5647 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5648 buf->start + buf->len - 1, GFP_NOFS);
5650 trans->blocks_used++;
5651 /* this returns a buffer locked for blocking */
5655 static struct btrfs_block_rsv *
5656 use_block_rsv(struct btrfs_trans_handle *trans,
5657 struct btrfs_root *root, u32 blocksize)
5659 struct btrfs_block_rsv *block_rsv;
5660 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5663 block_rsv = get_block_rsv(trans, root);
5665 if (block_rsv->size == 0) {
5666 ret = reserve_metadata_bytes(trans, root, block_rsv,
5669 * If we couldn't reserve metadata bytes try and use some from
5670 * the global reserve.
5672 if (ret && block_rsv != global_rsv) {
5673 ret = block_rsv_use_bytes(global_rsv, blocksize);
5676 return ERR_PTR(ret);
5678 return ERR_PTR(ret);
5683 ret = block_rsv_use_bytes(block_rsv, blocksize);
5688 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5691 spin_lock(&block_rsv->lock);
5692 block_rsv->size += blocksize;
5693 spin_unlock(&block_rsv->lock);
5695 } else if (ret && block_rsv != global_rsv) {
5696 ret = block_rsv_use_bytes(global_rsv, blocksize);
5702 return ERR_PTR(-ENOSPC);
5705 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5707 block_rsv_add_bytes(block_rsv, blocksize, 0);
5708 block_rsv_release_bytes(block_rsv, NULL, 0);
5712 * finds a free extent and does all the dirty work required for allocation
5713 * returns the key for the extent through ins, and a tree buffer for
5714 * the first block of the extent through buf.
5716 * returns the tree buffer or NULL.
5718 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5719 struct btrfs_root *root, u32 blocksize,
5720 u64 parent, u64 root_objectid,
5721 struct btrfs_disk_key *key, int level,
5722 u64 hint, u64 empty_size)
5724 struct btrfs_key ins;
5725 struct btrfs_block_rsv *block_rsv;
5726 struct extent_buffer *buf;
5731 block_rsv = use_block_rsv(trans, root, blocksize);
5732 if (IS_ERR(block_rsv))
5733 return ERR_CAST(block_rsv);
5735 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5736 empty_size, hint, (u64)-1, &ins, 0);
5738 unuse_block_rsv(block_rsv, blocksize);
5739 return ERR_PTR(ret);
5742 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5744 BUG_ON(IS_ERR(buf));
5746 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5748 parent = ins.objectid;
5749 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5753 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5754 struct btrfs_delayed_extent_op *extent_op;
5755 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5758 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5760 memset(&extent_op->key, 0, sizeof(extent_op->key));
5761 extent_op->flags_to_set = flags;
5762 extent_op->update_key = 1;
5763 extent_op->update_flags = 1;
5764 extent_op->is_data = 0;
5766 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5767 ins.offset, parent, root_objectid,
5768 level, BTRFS_ADD_DELAYED_EXTENT,
5775 struct walk_control {
5776 u64 refs[BTRFS_MAX_LEVEL];
5777 u64 flags[BTRFS_MAX_LEVEL];
5778 struct btrfs_key update_progress;
5788 #define DROP_REFERENCE 1
5789 #define UPDATE_BACKREF 2
5791 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5792 struct btrfs_root *root,
5793 struct walk_control *wc,
5794 struct btrfs_path *path)
5802 struct btrfs_key key;
5803 struct extent_buffer *eb;
5808 if (path->slots[wc->level] < wc->reada_slot) {
5809 wc->reada_count = wc->reada_count * 2 / 3;
5810 wc->reada_count = max(wc->reada_count, 2);
5812 wc->reada_count = wc->reada_count * 3 / 2;
5813 wc->reada_count = min_t(int, wc->reada_count,
5814 BTRFS_NODEPTRS_PER_BLOCK(root));
5817 eb = path->nodes[wc->level];
5818 nritems = btrfs_header_nritems(eb);
5819 blocksize = btrfs_level_size(root, wc->level - 1);
5821 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5822 if (nread >= wc->reada_count)
5826 bytenr = btrfs_node_blockptr(eb, slot);
5827 generation = btrfs_node_ptr_generation(eb, slot);
5829 if (slot == path->slots[wc->level])
5832 if (wc->stage == UPDATE_BACKREF &&
5833 generation <= root->root_key.offset)
5836 /* We don't lock the tree block, it's OK to be racy here */
5837 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5842 if (wc->stage == DROP_REFERENCE) {
5846 if (wc->level == 1 &&
5847 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5849 if (!wc->update_ref ||
5850 generation <= root->root_key.offset)
5852 btrfs_node_key_to_cpu(eb, &key, slot);
5853 ret = btrfs_comp_cpu_keys(&key,
5854 &wc->update_progress);
5858 if (wc->level == 1 &&
5859 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5863 ret = readahead_tree_block(root, bytenr, blocksize,
5869 wc->reada_slot = slot;
5873 * hepler to process tree block while walking down the tree.
5875 * when wc->stage == UPDATE_BACKREF, this function updates
5876 * back refs for pointers in the block.
5878 * NOTE: return value 1 means we should stop walking down.
5880 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5881 struct btrfs_root *root,
5882 struct btrfs_path *path,
5883 struct walk_control *wc, int lookup_info)
5885 int level = wc->level;
5886 struct extent_buffer *eb = path->nodes[level];
5887 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5890 if (wc->stage == UPDATE_BACKREF &&
5891 btrfs_header_owner(eb) != root->root_key.objectid)
5895 * when reference count of tree block is 1, it won't increase
5896 * again. once full backref flag is set, we never clear it.
5899 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5900 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5901 BUG_ON(!path->locks[level]);
5902 ret = btrfs_lookup_extent_info(trans, root,
5907 BUG_ON(wc->refs[level] == 0);
5910 if (wc->stage == DROP_REFERENCE) {
5911 if (wc->refs[level] > 1)
5914 if (path->locks[level] && !wc->keep_locks) {
5915 btrfs_tree_unlock(eb);
5916 path->locks[level] = 0;
5921 /* wc->stage == UPDATE_BACKREF */
5922 if (!(wc->flags[level] & flag)) {
5923 BUG_ON(!path->locks[level]);
5924 ret = btrfs_inc_ref(trans, root, eb, 1);
5926 ret = btrfs_dec_ref(trans, root, eb, 0);
5928 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5931 wc->flags[level] |= flag;
5935 * the block is shared by multiple trees, so it's not good to
5936 * keep the tree lock
5938 if (path->locks[level] && level > 0) {
5939 btrfs_tree_unlock(eb);
5940 path->locks[level] = 0;
5946 * hepler to process tree block pointer.
5948 * when wc->stage == DROP_REFERENCE, this function checks
5949 * reference count of the block pointed to. if the block
5950 * is shared and we need update back refs for the subtree
5951 * rooted at the block, this function changes wc->stage to
5952 * UPDATE_BACKREF. if the block is shared and there is no
5953 * need to update back, this function drops the reference
5956 * NOTE: return value 1 means we should stop walking down.
5958 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5959 struct btrfs_root *root,
5960 struct btrfs_path *path,
5961 struct walk_control *wc, int *lookup_info)
5967 struct btrfs_key key;
5968 struct extent_buffer *next;
5969 int level = wc->level;
5973 generation = btrfs_node_ptr_generation(path->nodes[level],
5974 path->slots[level]);
5976 * if the lower level block was created before the snapshot
5977 * was created, we know there is no need to update back refs
5980 if (wc->stage == UPDATE_BACKREF &&
5981 generation <= root->root_key.offset) {
5986 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5987 blocksize = btrfs_level_size(root, level - 1);
5989 next = btrfs_find_tree_block(root, bytenr, blocksize);
5991 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5996 btrfs_tree_lock(next);
5997 btrfs_set_lock_blocking(next);
5999 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6000 &wc->refs[level - 1],
6001 &wc->flags[level - 1]);
6003 BUG_ON(wc->refs[level - 1] == 0);
6006 if (wc->stage == DROP_REFERENCE) {
6007 if (wc->refs[level - 1] > 1) {
6009 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6012 if (!wc->update_ref ||
6013 generation <= root->root_key.offset)
6016 btrfs_node_key_to_cpu(path->nodes[level], &key,
6017 path->slots[level]);
6018 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6022 wc->stage = UPDATE_BACKREF;
6023 wc->shared_level = level - 1;
6027 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6031 if (!btrfs_buffer_uptodate(next, generation)) {
6032 btrfs_tree_unlock(next);
6033 free_extent_buffer(next);
6039 if (reada && level == 1)
6040 reada_walk_down(trans, root, wc, path);
6041 next = read_tree_block(root, bytenr, blocksize, generation);
6042 btrfs_tree_lock(next);
6043 btrfs_set_lock_blocking(next);
6047 BUG_ON(level != btrfs_header_level(next));
6048 path->nodes[level] = next;
6049 path->slots[level] = 0;
6050 path->locks[level] = 1;
6056 wc->refs[level - 1] = 0;
6057 wc->flags[level - 1] = 0;
6058 if (wc->stage == DROP_REFERENCE) {
6059 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6060 parent = path->nodes[level]->start;
6062 BUG_ON(root->root_key.objectid !=
6063 btrfs_header_owner(path->nodes[level]));
6067 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6068 root->root_key.objectid, level - 1, 0);
6071 btrfs_tree_unlock(next);
6072 free_extent_buffer(next);
6078 * hepler to process tree block while walking up the tree.
6080 * when wc->stage == DROP_REFERENCE, this function drops
6081 * reference count on the block.
6083 * when wc->stage == UPDATE_BACKREF, this function changes
6084 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6085 * to UPDATE_BACKREF previously while processing the block.
6087 * NOTE: return value 1 means we should stop walking up.
6089 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6090 struct btrfs_root *root,
6091 struct btrfs_path *path,
6092 struct walk_control *wc)
6095 int level = wc->level;
6096 struct extent_buffer *eb = path->nodes[level];
6099 if (wc->stage == UPDATE_BACKREF) {
6100 BUG_ON(wc->shared_level < level);
6101 if (level < wc->shared_level)
6104 ret = find_next_key(path, level + 1, &wc->update_progress);
6108 wc->stage = DROP_REFERENCE;
6109 wc->shared_level = -1;
6110 path->slots[level] = 0;
6113 * check reference count again if the block isn't locked.
6114 * we should start walking down the tree again if reference
6117 if (!path->locks[level]) {
6119 btrfs_tree_lock(eb);
6120 btrfs_set_lock_blocking(eb);
6121 path->locks[level] = 1;
6123 ret = btrfs_lookup_extent_info(trans, root,
6128 BUG_ON(wc->refs[level] == 0);
6129 if (wc->refs[level] == 1) {
6130 btrfs_tree_unlock(eb);
6131 path->locks[level] = 0;
6137 /* wc->stage == DROP_REFERENCE */
6138 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6140 if (wc->refs[level] == 1) {
6142 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6143 ret = btrfs_dec_ref(trans, root, eb, 1);
6145 ret = btrfs_dec_ref(trans, root, eb, 0);
6148 /* make block locked assertion in clean_tree_block happy */
6149 if (!path->locks[level] &&
6150 btrfs_header_generation(eb) == trans->transid) {
6151 btrfs_tree_lock(eb);
6152 btrfs_set_lock_blocking(eb);
6153 path->locks[level] = 1;
6155 clean_tree_block(trans, root, eb);
6158 if (eb == root->node) {
6159 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6162 BUG_ON(root->root_key.objectid !=
6163 btrfs_header_owner(eb));
6165 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6166 parent = path->nodes[level + 1]->start;
6168 BUG_ON(root->root_key.objectid !=
6169 btrfs_header_owner(path->nodes[level + 1]));
6172 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6174 wc->refs[level] = 0;
6175 wc->flags[level] = 0;
6179 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6180 struct btrfs_root *root,
6181 struct btrfs_path *path,
6182 struct walk_control *wc)
6184 int level = wc->level;
6185 int lookup_info = 1;
6188 while (level >= 0) {
6189 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6196 if (path->slots[level] >=
6197 btrfs_header_nritems(path->nodes[level]))
6200 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6202 path->slots[level]++;
6211 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6212 struct btrfs_root *root,
6213 struct btrfs_path *path,
6214 struct walk_control *wc, int max_level)
6216 int level = wc->level;
6219 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6220 while (level < max_level && path->nodes[level]) {
6222 if (path->slots[level] + 1 <
6223 btrfs_header_nritems(path->nodes[level])) {
6224 path->slots[level]++;
6227 ret = walk_up_proc(trans, root, path, wc);
6231 if (path->locks[level]) {
6232 btrfs_tree_unlock(path->nodes[level]);
6233 path->locks[level] = 0;
6235 free_extent_buffer(path->nodes[level]);
6236 path->nodes[level] = NULL;
6244 * drop a subvolume tree.
6246 * this function traverses the tree freeing any blocks that only
6247 * referenced by the tree.
6249 * when a shared tree block is found. this function decreases its
6250 * reference count by one. if update_ref is true, this function
6251 * also make sure backrefs for the shared block and all lower level
6252 * blocks are properly updated.
6254 int btrfs_drop_snapshot(struct btrfs_root *root,
6255 struct btrfs_block_rsv *block_rsv, int update_ref)
6257 struct btrfs_path *path;
6258 struct btrfs_trans_handle *trans;
6259 struct btrfs_root *tree_root = root->fs_info->tree_root;
6260 struct btrfs_root_item *root_item = &root->root_item;
6261 struct walk_control *wc;
6262 struct btrfs_key key;
6267 path = btrfs_alloc_path();
6270 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6273 trans = btrfs_start_transaction(tree_root, 0);
6275 trans->block_rsv = block_rsv;
6277 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6278 level = btrfs_header_level(root->node);
6279 path->nodes[level] = btrfs_lock_root_node(root);
6280 btrfs_set_lock_blocking(path->nodes[level]);
6281 path->slots[level] = 0;
6282 path->locks[level] = 1;
6283 memset(&wc->update_progress, 0,
6284 sizeof(wc->update_progress));
6286 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6287 memcpy(&wc->update_progress, &key,
6288 sizeof(wc->update_progress));
6290 level = root_item->drop_level;
6292 path->lowest_level = level;
6293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6294 path->lowest_level = 0;
6302 * unlock our path, this is safe because only this
6303 * function is allowed to delete this snapshot
6305 btrfs_unlock_up_safe(path, 0);
6307 level = btrfs_header_level(root->node);
6309 btrfs_tree_lock(path->nodes[level]);
6310 btrfs_set_lock_blocking(path->nodes[level]);
6312 ret = btrfs_lookup_extent_info(trans, root,
6313 path->nodes[level]->start,
6314 path->nodes[level]->len,
6318 BUG_ON(wc->refs[level] == 0);
6320 if (level == root_item->drop_level)
6323 btrfs_tree_unlock(path->nodes[level]);
6324 WARN_ON(wc->refs[level] != 1);
6330 wc->shared_level = -1;
6331 wc->stage = DROP_REFERENCE;
6332 wc->update_ref = update_ref;
6334 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6337 ret = walk_down_tree(trans, root, path, wc);
6343 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6350 BUG_ON(wc->stage != DROP_REFERENCE);
6354 if (wc->stage == DROP_REFERENCE) {
6356 btrfs_node_key(path->nodes[level],
6357 &root_item->drop_progress,
6358 path->slots[level]);
6359 root_item->drop_level = level;
6362 BUG_ON(wc->level == 0);
6363 if (btrfs_should_end_transaction(trans, tree_root)) {
6364 ret = btrfs_update_root(trans, tree_root,
6369 btrfs_end_transaction_throttle(trans, tree_root);
6370 trans = btrfs_start_transaction(tree_root, 0);
6372 trans->block_rsv = block_rsv;
6375 btrfs_release_path(root, path);
6378 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6381 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6382 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6386 /* if we fail to delete the orphan item this time
6387 * around, it'll get picked up the next time.
6389 * The most common failure here is just -ENOENT.
6391 btrfs_del_orphan_item(trans, tree_root,
6392 root->root_key.objectid);
6396 if (root->in_radix) {
6397 btrfs_free_fs_root(tree_root->fs_info, root);
6399 free_extent_buffer(root->node);
6400 free_extent_buffer(root->commit_root);
6404 btrfs_end_transaction_throttle(trans, tree_root);
6406 btrfs_free_path(path);
6411 * drop subtree rooted at tree block 'node'.
6413 * NOTE: this function will unlock and release tree block 'node'
6415 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6416 struct btrfs_root *root,
6417 struct extent_buffer *node,
6418 struct extent_buffer *parent)
6420 struct btrfs_path *path;
6421 struct walk_control *wc;
6427 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6429 path = btrfs_alloc_path();
6432 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6435 btrfs_assert_tree_locked(parent);
6436 parent_level = btrfs_header_level(parent);
6437 extent_buffer_get(parent);
6438 path->nodes[parent_level] = parent;
6439 path->slots[parent_level] = btrfs_header_nritems(parent);
6441 btrfs_assert_tree_locked(node);
6442 level = btrfs_header_level(node);
6443 path->nodes[level] = node;
6444 path->slots[level] = 0;
6445 path->locks[level] = 1;
6447 wc->refs[parent_level] = 1;
6448 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6450 wc->shared_level = -1;
6451 wc->stage = DROP_REFERENCE;
6454 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6457 wret = walk_down_tree(trans, root, path, wc);
6463 wret = walk_up_tree(trans, root, path, wc, parent_level);
6471 btrfs_free_path(path);
6476 static unsigned long calc_ra(unsigned long start, unsigned long last,
6479 return min(last, start + nr - 1);
6482 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6487 unsigned long first_index;
6488 unsigned long last_index;
6491 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6492 struct file_ra_state *ra;
6493 struct btrfs_ordered_extent *ordered;
6494 unsigned int total_read = 0;
6495 unsigned int total_dirty = 0;
6498 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6500 mutex_lock(&inode->i_mutex);
6501 first_index = start >> PAGE_CACHE_SHIFT;
6502 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6504 /* make sure the dirty trick played by the caller work */
6505 ret = invalidate_inode_pages2_range(inode->i_mapping,
6506 first_index, last_index);
6510 file_ra_state_init(ra, inode->i_mapping);
6512 for (i = first_index ; i <= last_index; i++) {
6513 if (total_read % ra->ra_pages == 0) {
6514 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6515 calc_ra(i, last_index, ra->ra_pages));
6519 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6521 page = grab_cache_page(inode->i_mapping, i);
6526 if (!PageUptodate(page)) {
6527 btrfs_readpage(NULL, page);
6529 if (!PageUptodate(page)) {
6531 page_cache_release(page);
6536 wait_on_page_writeback(page);
6538 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6539 page_end = page_start + PAGE_CACHE_SIZE - 1;
6540 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6542 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6544 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6546 page_cache_release(page);
6547 btrfs_start_ordered_extent(inode, ordered, 1);
6548 btrfs_put_ordered_extent(ordered);
6551 set_page_extent_mapped(page);
6553 if (i == first_index)
6554 set_extent_bits(io_tree, page_start, page_end,
6555 EXTENT_BOUNDARY, GFP_NOFS);
6556 btrfs_set_extent_delalloc(inode, page_start, page_end);
6558 set_page_dirty(page);
6561 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6563 page_cache_release(page);
6568 mutex_unlock(&inode->i_mutex);
6569 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6573 static noinline int relocate_data_extent(struct inode *reloc_inode,
6574 struct btrfs_key *extent_key,
6577 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6578 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6579 struct extent_map *em;
6580 u64 start = extent_key->objectid - offset;
6581 u64 end = start + extent_key->offset - 1;
6583 em = alloc_extent_map(GFP_NOFS);
6584 BUG_ON(!em || IS_ERR(em));
6587 em->len = extent_key->offset;
6588 em->block_len = extent_key->offset;
6589 em->block_start = extent_key->objectid;
6590 em->bdev = root->fs_info->fs_devices->latest_bdev;
6591 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6593 /* setup extent map to cheat btrfs_readpage */
6594 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6597 write_lock(&em_tree->lock);
6598 ret = add_extent_mapping(em_tree, em);
6599 write_unlock(&em_tree->lock);
6600 if (ret != -EEXIST) {
6601 free_extent_map(em);
6604 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6606 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6608 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6611 struct btrfs_ref_path {
6613 u64 nodes[BTRFS_MAX_LEVEL];
6615 u64 root_generation;
6622 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6623 u64 new_nodes[BTRFS_MAX_LEVEL];
6626 struct disk_extent {
6637 static int is_cowonly_root(u64 root_objectid)
6639 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6640 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6641 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6642 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6643 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6644 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6649 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6650 struct btrfs_root *extent_root,
6651 struct btrfs_ref_path *ref_path,
6654 struct extent_buffer *leaf;
6655 struct btrfs_path *path;
6656 struct btrfs_extent_ref *ref;
6657 struct btrfs_key key;
6658 struct btrfs_key found_key;
6664 path = btrfs_alloc_path();
6669 ref_path->lowest_level = -1;
6670 ref_path->current_level = -1;
6671 ref_path->shared_level = -1;
6675 level = ref_path->current_level - 1;
6676 while (level >= -1) {
6678 if (level < ref_path->lowest_level)
6682 bytenr = ref_path->nodes[level];
6684 bytenr = ref_path->extent_start;
6685 BUG_ON(bytenr == 0);
6687 parent = ref_path->nodes[level + 1];
6688 ref_path->nodes[level + 1] = 0;
6689 ref_path->current_level = level;
6690 BUG_ON(parent == 0);
6692 key.objectid = bytenr;
6693 key.offset = parent + 1;
6694 key.type = BTRFS_EXTENT_REF_KEY;
6696 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6701 leaf = path->nodes[0];
6702 nritems = btrfs_header_nritems(leaf);
6703 if (path->slots[0] >= nritems) {
6704 ret = btrfs_next_leaf(extent_root, path);
6709 leaf = path->nodes[0];
6712 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6713 if (found_key.objectid == bytenr &&
6714 found_key.type == BTRFS_EXTENT_REF_KEY) {
6715 if (level < ref_path->shared_level)
6716 ref_path->shared_level = level;
6721 btrfs_release_path(extent_root, path);
6724 /* reached lowest level */
6728 level = ref_path->current_level;
6729 while (level < BTRFS_MAX_LEVEL - 1) {
6733 bytenr = ref_path->nodes[level];
6735 bytenr = ref_path->extent_start;
6737 BUG_ON(bytenr == 0);
6739 key.objectid = bytenr;
6741 key.type = BTRFS_EXTENT_REF_KEY;
6743 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6747 leaf = path->nodes[0];
6748 nritems = btrfs_header_nritems(leaf);
6749 if (path->slots[0] >= nritems) {
6750 ret = btrfs_next_leaf(extent_root, path);
6754 /* the extent was freed by someone */
6755 if (ref_path->lowest_level == level)
6757 btrfs_release_path(extent_root, path);
6760 leaf = path->nodes[0];
6763 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6764 if (found_key.objectid != bytenr ||
6765 found_key.type != BTRFS_EXTENT_REF_KEY) {
6766 /* the extent was freed by someone */
6767 if (ref_path->lowest_level == level) {
6771 btrfs_release_path(extent_root, path);
6775 ref = btrfs_item_ptr(leaf, path->slots[0],
6776 struct btrfs_extent_ref);
6777 ref_objectid = btrfs_ref_objectid(leaf, ref);
6778 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6780 level = (int)ref_objectid;
6781 BUG_ON(level >= BTRFS_MAX_LEVEL);
6782 ref_path->lowest_level = level;
6783 ref_path->current_level = level;
6784 ref_path->nodes[level] = bytenr;
6786 WARN_ON(ref_objectid != level);
6789 WARN_ON(level != -1);
6793 if (ref_path->lowest_level == level) {
6794 ref_path->owner_objectid = ref_objectid;
6795 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6799 * the block is tree root or the block isn't in reference
6802 if (found_key.objectid == found_key.offset ||
6803 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6804 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6805 ref_path->root_generation =
6806 btrfs_ref_generation(leaf, ref);
6808 /* special reference from the tree log */
6809 ref_path->nodes[0] = found_key.offset;
6810 ref_path->current_level = 0;
6817 BUG_ON(ref_path->nodes[level] != 0);
6818 ref_path->nodes[level] = found_key.offset;
6819 ref_path->current_level = level;
6822 * the reference was created in the running transaction,
6823 * no need to continue walking up.
6825 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6826 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6827 ref_path->root_generation =
6828 btrfs_ref_generation(leaf, ref);
6833 btrfs_release_path(extent_root, path);
6836 /* reached max tree level, but no tree root found. */
6839 btrfs_free_path(path);
6843 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6844 struct btrfs_root *extent_root,
6845 struct btrfs_ref_path *ref_path,
6848 memset(ref_path, 0, sizeof(*ref_path));
6849 ref_path->extent_start = extent_start;
6851 return __next_ref_path(trans, extent_root, ref_path, 1);
6854 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6855 struct btrfs_root *extent_root,
6856 struct btrfs_ref_path *ref_path)
6858 return __next_ref_path(trans, extent_root, ref_path, 0);
6861 static noinline int get_new_locations(struct inode *reloc_inode,
6862 struct btrfs_key *extent_key,
6863 u64 offset, int no_fragment,
6864 struct disk_extent **extents,
6867 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6868 struct btrfs_path *path;
6869 struct btrfs_file_extent_item *fi;
6870 struct extent_buffer *leaf;
6871 struct disk_extent *exts = *extents;
6872 struct btrfs_key found_key;
6877 int max = *nr_extents;
6880 WARN_ON(!no_fragment && *extents);
6883 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6888 path = btrfs_alloc_path();
6891 cur_pos = extent_key->objectid - offset;
6892 last_byte = extent_key->objectid + extent_key->offset;
6893 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6903 leaf = path->nodes[0];
6904 nritems = btrfs_header_nritems(leaf);
6905 if (path->slots[0] >= nritems) {
6906 ret = btrfs_next_leaf(root, path);
6911 leaf = path->nodes[0];
6914 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6915 if (found_key.offset != cur_pos ||
6916 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6917 found_key.objectid != reloc_inode->i_ino)
6920 fi = btrfs_item_ptr(leaf, path->slots[0],
6921 struct btrfs_file_extent_item);
6922 if (btrfs_file_extent_type(leaf, fi) !=
6923 BTRFS_FILE_EXTENT_REG ||
6924 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6928 struct disk_extent *old = exts;
6930 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6931 memcpy(exts, old, sizeof(*exts) * nr);
6932 if (old != *extents)
6936 exts[nr].disk_bytenr =
6937 btrfs_file_extent_disk_bytenr(leaf, fi);
6938 exts[nr].disk_num_bytes =
6939 btrfs_file_extent_disk_num_bytes(leaf, fi);
6940 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6941 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6942 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6943 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6944 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6945 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6947 BUG_ON(exts[nr].offset > 0);
6948 BUG_ON(exts[nr].compression || exts[nr].encryption);
6949 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6951 cur_pos += exts[nr].num_bytes;
6954 if (cur_pos + offset >= last_byte)
6964 BUG_ON(cur_pos + offset > last_byte);
6965 if (cur_pos + offset < last_byte) {
6971 btrfs_free_path(path);
6973 if (exts != *extents)
6982 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6983 struct btrfs_root *root,
6984 struct btrfs_path *path,
6985 struct btrfs_key *extent_key,
6986 struct btrfs_key *leaf_key,
6987 struct btrfs_ref_path *ref_path,
6988 struct disk_extent *new_extents,
6991 struct extent_buffer *leaf;
6992 struct btrfs_file_extent_item *fi;
6993 struct inode *inode = NULL;
6994 struct btrfs_key key;
6999 u64 search_end = (u64)-1;
7002 int extent_locked = 0;
7006 memcpy(&key, leaf_key, sizeof(key));
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.objectid = ref_path->owner_objectid;
7012 key.type = BTRFS_EXTENT_DATA_KEY;
7018 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7022 leaf = path->nodes[0];
7023 nritems = btrfs_header_nritems(leaf);
7025 if (extent_locked && ret > 0) {
7027 * the file extent item was modified by someone
7028 * before the extent got locked.
7030 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7031 lock_end, GFP_NOFS);
7035 if (path->slots[0] >= nritems) {
7036 if (++nr_scaned > 2)
7039 BUG_ON(extent_locked);
7040 ret = btrfs_next_leaf(root, path);
7045 leaf = path->nodes[0];
7046 nritems = btrfs_header_nritems(leaf);
7049 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7051 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7052 if ((key.objectid > ref_path->owner_objectid) ||
7053 (key.objectid == ref_path->owner_objectid &&
7054 key.type > BTRFS_EXTENT_DATA_KEY) ||
7055 key.offset >= search_end)
7059 if (inode && key.objectid != inode->i_ino) {
7060 BUG_ON(extent_locked);
7061 btrfs_release_path(root, path);
7062 mutex_unlock(&inode->i_mutex);
7068 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7073 fi = btrfs_item_ptr(leaf, path->slots[0],
7074 struct btrfs_file_extent_item);
7075 extent_type = btrfs_file_extent_type(leaf, fi);
7076 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7077 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7078 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7079 extent_key->objectid)) {
7085 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7086 ext_offset = btrfs_file_extent_offset(leaf, fi);
7088 if (search_end == (u64)-1) {
7089 search_end = key.offset - ext_offset +
7090 btrfs_file_extent_ram_bytes(leaf, fi);
7093 if (!extent_locked) {
7094 lock_start = key.offset;
7095 lock_end = lock_start + num_bytes - 1;
7097 if (lock_start > key.offset ||
7098 lock_end + 1 < key.offset + num_bytes) {
7099 unlock_extent(&BTRFS_I(inode)->io_tree,
7100 lock_start, lock_end, GFP_NOFS);
7106 btrfs_release_path(root, path);
7108 inode = btrfs_iget_locked(root->fs_info->sb,
7109 key.objectid, root);
7110 if (inode->i_state & I_NEW) {
7111 BTRFS_I(inode)->root = root;
7112 BTRFS_I(inode)->location.objectid =
7114 BTRFS_I(inode)->location.type =
7115 BTRFS_INODE_ITEM_KEY;
7116 BTRFS_I(inode)->location.offset = 0;
7117 btrfs_read_locked_inode(inode);
7118 unlock_new_inode(inode);
7121 * some code call btrfs_commit_transaction while
7122 * holding the i_mutex, so we can't use mutex_lock
7125 if (is_bad_inode(inode) ||
7126 !mutex_trylock(&inode->i_mutex)) {
7129 key.offset = (u64)-1;
7134 if (!extent_locked) {
7135 struct btrfs_ordered_extent *ordered;
7137 btrfs_release_path(root, path);
7139 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7140 lock_end, GFP_NOFS);
7141 ordered = btrfs_lookup_first_ordered_extent(inode,
7144 ordered->file_offset <= lock_end &&
7145 ordered->file_offset + ordered->len > lock_start) {
7146 unlock_extent(&BTRFS_I(inode)->io_tree,
7147 lock_start, lock_end, GFP_NOFS);
7148 btrfs_start_ordered_extent(inode, ordered, 1);
7149 btrfs_put_ordered_extent(ordered);
7150 key.offset += num_bytes;
7154 btrfs_put_ordered_extent(ordered);
7160 if (nr_extents == 1) {
7161 /* update extent pointer in place */
7162 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7163 new_extents[0].disk_bytenr);
7164 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7165 new_extents[0].disk_num_bytes);
7166 btrfs_mark_buffer_dirty(leaf);
7168 btrfs_drop_extent_cache(inode, key.offset,
7169 key.offset + num_bytes - 1, 0);
7171 ret = btrfs_inc_extent_ref(trans, root,
7172 new_extents[0].disk_bytenr,
7173 new_extents[0].disk_num_bytes,
7175 root->root_key.objectid,
7180 ret = btrfs_free_extent(trans, root,
7181 extent_key->objectid,
7184 btrfs_header_owner(leaf),
7185 btrfs_header_generation(leaf),
7189 btrfs_release_path(root, path);
7190 key.offset += num_bytes;
7198 * drop old extent pointer at first, then insert the
7199 * new pointers one bye one
7201 btrfs_release_path(root, path);
7202 ret = btrfs_drop_extents(trans, root, inode, key.offset,
7203 key.offset + num_bytes,
7204 key.offset, &alloc_hint);
7207 for (i = 0; i < nr_extents; i++) {
7208 if (ext_offset >= new_extents[i].num_bytes) {
7209 ext_offset -= new_extents[i].num_bytes;
7212 extent_len = min(new_extents[i].num_bytes -
7213 ext_offset, num_bytes);
7215 ret = btrfs_insert_empty_item(trans, root,
7220 leaf = path->nodes[0];
7221 fi = btrfs_item_ptr(leaf, path->slots[0],
7222 struct btrfs_file_extent_item);
7223 btrfs_set_file_extent_generation(leaf, fi,
7225 btrfs_set_file_extent_type(leaf, fi,
7226 BTRFS_FILE_EXTENT_REG);
7227 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7228 new_extents[i].disk_bytenr);
7229 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7230 new_extents[i].disk_num_bytes);
7231 btrfs_set_file_extent_ram_bytes(leaf, fi,
7232 new_extents[i].ram_bytes);
7234 btrfs_set_file_extent_compression(leaf, fi,
7235 new_extents[i].compression);
7236 btrfs_set_file_extent_encryption(leaf, fi,
7237 new_extents[i].encryption);
7238 btrfs_set_file_extent_other_encoding(leaf, fi,
7239 new_extents[i].other_encoding);
7241 btrfs_set_file_extent_num_bytes(leaf, fi,
7243 ext_offset += new_extents[i].offset;
7244 btrfs_set_file_extent_offset(leaf, fi,
7246 btrfs_mark_buffer_dirty(leaf);
7248 btrfs_drop_extent_cache(inode, key.offset,
7249 key.offset + extent_len - 1, 0);
7251 ret = btrfs_inc_extent_ref(trans, root,
7252 new_extents[i].disk_bytenr,
7253 new_extents[i].disk_num_bytes,
7255 root->root_key.objectid,
7256 trans->transid, key.objectid);
7258 btrfs_release_path(root, path);
7260 inode_add_bytes(inode, extent_len);
7263 num_bytes -= extent_len;
7264 key.offset += extent_len;
7269 BUG_ON(i >= nr_extents);
7273 if (extent_locked) {
7274 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7275 lock_end, GFP_NOFS);
7279 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7280 key.offset >= search_end)
7287 btrfs_release_path(root, path);
7289 mutex_unlock(&inode->i_mutex);
7290 if (extent_locked) {
7291 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7292 lock_end, GFP_NOFS);
7299 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7300 struct btrfs_root *root,
7301 struct extent_buffer *buf, u64 orig_start)
7306 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7307 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7309 level = btrfs_header_level(buf);
7311 struct btrfs_leaf_ref *ref;
7312 struct btrfs_leaf_ref *orig_ref;
7314 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7318 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7320 btrfs_free_leaf_ref(root, orig_ref);
7324 ref->nritems = orig_ref->nritems;
7325 memcpy(ref->extents, orig_ref->extents,
7326 sizeof(ref->extents[0]) * ref->nritems);
7328 btrfs_free_leaf_ref(root, orig_ref);
7330 ref->root_gen = trans->transid;
7331 ref->bytenr = buf->start;
7332 ref->owner = btrfs_header_owner(buf);
7333 ref->generation = btrfs_header_generation(buf);
7335 ret = btrfs_add_leaf_ref(root, ref, 0);
7337 btrfs_free_leaf_ref(root, ref);
7342 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7343 struct extent_buffer *leaf,
7344 struct btrfs_block_group_cache *group,
7345 struct btrfs_root *target_root)
7347 struct btrfs_key key;
7348 struct inode *inode = NULL;
7349 struct btrfs_file_extent_item *fi;
7350 struct extent_state *cached_state = NULL;
7352 u64 skip_objectid = 0;
7356 nritems = btrfs_header_nritems(leaf);
7357 for (i = 0; i < nritems; i++) {
7358 btrfs_item_key_to_cpu(leaf, &key, i);
7359 if (key.objectid == skip_objectid ||
7360 key.type != BTRFS_EXTENT_DATA_KEY)
7362 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7363 if (btrfs_file_extent_type(leaf, fi) ==
7364 BTRFS_FILE_EXTENT_INLINE)
7366 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7368 if (!inode || inode->i_ino != key.objectid) {
7370 inode = btrfs_ilookup(target_root->fs_info->sb,
7371 key.objectid, target_root, 1);
7374 skip_objectid = key.objectid;
7377 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7379 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7380 key.offset + num_bytes - 1, 0, &cached_state,
7382 btrfs_drop_extent_cache(inode, key.offset,
7383 key.offset + num_bytes - 1, 1);
7384 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7385 key.offset + num_bytes - 1, &cached_state,
7393 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7394 struct btrfs_root *root,
7395 struct extent_buffer *leaf,
7396 struct btrfs_block_group_cache *group,
7397 struct inode *reloc_inode)
7399 struct btrfs_key key;
7400 struct btrfs_key extent_key;
7401 struct btrfs_file_extent_item *fi;
7402 struct btrfs_leaf_ref *ref;
7403 struct disk_extent *new_extent;
7412 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7413 BUG_ON(!new_extent);
7415 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7419 nritems = btrfs_header_nritems(leaf);
7420 for (i = 0; i < nritems; i++) {
7421 btrfs_item_key_to_cpu(leaf, &key, i);
7422 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7424 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7425 if (btrfs_file_extent_type(leaf, fi) ==
7426 BTRFS_FILE_EXTENT_INLINE)
7428 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7429 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7434 if (bytenr >= group->key.objectid + group->key.offset ||
7435 bytenr + num_bytes <= group->key.objectid)
7438 extent_key.objectid = bytenr;
7439 extent_key.offset = num_bytes;
7440 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7442 ret = get_new_locations(reloc_inode, &extent_key,
7443 group->key.objectid, 1,
7444 &new_extent, &nr_extent);
7449 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7450 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7451 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7452 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7454 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7455 new_extent->disk_bytenr);
7456 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7457 new_extent->disk_num_bytes);
7458 btrfs_mark_buffer_dirty(leaf);
7460 ret = btrfs_inc_extent_ref(trans, root,
7461 new_extent->disk_bytenr,
7462 new_extent->disk_num_bytes,
7464 root->root_key.objectid,
7465 trans->transid, key.objectid);
7468 ret = btrfs_free_extent(trans, root,
7469 bytenr, num_bytes, leaf->start,
7470 btrfs_header_owner(leaf),
7471 btrfs_header_generation(leaf),
7477 BUG_ON(ext_index + 1 != ref->nritems);
7478 btrfs_free_leaf_ref(root, ref);
7482 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7483 struct btrfs_root *root)
7485 struct btrfs_root *reloc_root;
7488 if (root->reloc_root) {
7489 reloc_root = root->reloc_root;
7490 root->reloc_root = NULL;
7491 list_add(&reloc_root->dead_list,
7492 &root->fs_info->dead_reloc_roots);
7494 btrfs_set_root_bytenr(&reloc_root->root_item,
7495 reloc_root->node->start);
7496 btrfs_set_root_level(&root->root_item,
7497 btrfs_header_level(reloc_root->node));
7498 memset(&reloc_root->root_item.drop_progress, 0,
7499 sizeof(struct btrfs_disk_key));
7500 reloc_root->root_item.drop_level = 0;
7502 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7503 &reloc_root->root_key,
7504 &reloc_root->root_item);
7510 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7512 struct btrfs_trans_handle *trans;
7513 struct btrfs_root *reloc_root;
7514 struct btrfs_root *prev_root = NULL;
7515 struct list_head dead_roots;
7519 INIT_LIST_HEAD(&dead_roots);
7520 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7522 while (!list_empty(&dead_roots)) {
7523 reloc_root = list_entry(dead_roots.prev,
7524 struct btrfs_root, dead_list);
7525 list_del_init(&reloc_root->dead_list);
7527 BUG_ON(reloc_root->commit_root != NULL);
7529 trans = btrfs_join_transaction(root, 1);
7530 BUG_ON(IS_ERR(trans));
7532 mutex_lock(&root->fs_info->drop_mutex);
7533 ret = btrfs_drop_snapshot(trans, reloc_root);
7536 mutex_unlock(&root->fs_info->drop_mutex);
7538 nr = trans->blocks_used;
7539 ret = btrfs_end_transaction(trans, root);
7541 btrfs_btree_balance_dirty(root, nr);
7544 free_extent_buffer(reloc_root->node);
7546 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7547 &reloc_root->root_key);
7549 mutex_unlock(&root->fs_info->drop_mutex);
7551 nr = trans->blocks_used;
7552 ret = btrfs_end_transaction(trans, root);
7554 btrfs_btree_balance_dirty(root, nr);
7557 prev_root = reloc_root;
7560 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7566 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7568 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7572 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7574 struct btrfs_root *reloc_root;
7575 struct btrfs_trans_handle *trans;
7576 struct btrfs_key location;
7580 mutex_lock(&root->fs_info->tree_reloc_mutex);
7581 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7583 found = !list_empty(&root->fs_info->dead_reloc_roots);
7584 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7587 trans = btrfs_start_transaction(root, 1);
7589 ret = btrfs_commit_transaction(trans, root);
7593 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7594 location.offset = (u64)-1;
7595 location.type = BTRFS_ROOT_ITEM_KEY;
7597 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7598 BUG_ON(!reloc_root);
7599 btrfs_orphan_cleanup(reloc_root);
7603 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7604 struct btrfs_root *root)
7606 struct btrfs_root *reloc_root;
7607 struct extent_buffer *eb;
7608 struct btrfs_root_item *root_item;
7609 struct btrfs_key root_key;
7612 BUG_ON(!root->ref_cows);
7613 if (root->reloc_root)
7616 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7619 ret = btrfs_copy_root(trans, root, root->commit_root,
7620 &eb, BTRFS_TREE_RELOC_OBJECTID);
7623 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7624 root_key.offset = root->root_key.objectid;
7625 root_key.type = BTRFS_ROOT_ITEM_KEY;
7627 memcpy(root_item, &root->root_item, sizeof(root_item));
7628 btrfs_set_root_refs(root_item, 0);
7629 btrfs_set_root_bytenr(root_item, eb->start);
7630 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7631 btrfs_set_root_generation(root_item, trans->transid);
7633 btrfs_tree_unlock(eb);
7634 free_extent_buffer(eb);
7636 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7637 &root_key, root_item);
7641 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7643 BUG_ON(!reloc_root);
7644 reloc_root->last_trans = trans->transid;
7645 reloc_root->commit_root = NULL;
7646 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7648 root->reloc_root = reloc_root;
7653 * Core function of space balance.
7655 * The idea is using reloc trees to relocate tree blocks in reference
7656 * counted roots. There is one reloc tree for each subvol, and all
7657 * reloc trees share same root key objectid. Reloc trees are snapshots
7658 * of the latest committed roots of subvols (root->commit_root).
7660 * To relocate a tree block referenced by a subvol, there are two steps.
7661 * COW the block through subvol's reloc tree, then update block pointer
7662 * in the subvol to point to the new block. Since all reloc trees share
7663 * same root key objectid, doing special handing for tree blocks owned
7664 * by them is easy. Once a tree block has been COWed in one reloc tree,
7665 * we can use the resulting new block directly when the same block is
7666 * required to COW again through other reloc trees. By this way, relocated
7667 * tree blocks are shared between reloc trees, so they are also shared
7670 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7671 struct btrfs_root *root,
7672 struct btrfs_path *path,
7673 struct btrfs_key *first_key,
7674 struct btrfs_ref_path *ref_path,
7675 struct btrfs_block_group_cache *group,
7676 struct inode *reloc_inode)
7678 struct btrfs_root *reloc_root;
7679 struct extent_buffer *eb = NULL;
7680 struct btrfs_key *keys;
7684 int lowest_level = 0;
7687 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7688 lowest_level = ref_path->owner_objectid;
7690 if (!root->ref_cows) {
7691 path->lowest_level = lowest_level;
7692 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7694 path->lowest_level = 0;
7695 btrfs_release_path(root, path);
7699 mutex_lock(&root->fs_info->tree_reloc_mutex);
7700 ret = init_reloc_tree(trans, root);
7702 reloc_root = root->reloc_root;
7704 shared_level = ref_path->shared_level;
7705 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7707 keys = ref_path->node_keys;
7708 nodes = ref_path->new_nodes;
7709 memset(&keys[shared_level + 1], 0,
7710 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7711 memset(&nodes[shared_level + 1], 0,
7712 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7714 if (nodes[lowest_level] == 0) {
7715 path->lowest_level = lowest_level;
7716 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7719 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7720 eb = path->nodes[level];
7721 if (!eb || eb == reloc_root->node)
7723 nodes[level] = eb->start;
7725 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7727 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7730 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7731 eb = path->nodes[0];
7732 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7733 group, reloc_inode);
7736 btrfs_release_path(reloc_root, path);
7738 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7744 * replace tree blocks in the fs tree with tree blocks in
7747 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7750 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7751 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7754 extent_buffer_get(path->nodes[0]);
7755 eb = path->nodes[0];
7756 btrfs_release_path(reloc_root, path);
7757 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7759 free_extent_buffer(eb);
7762 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7763 path->lowest_level = 0;
7767 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7768 struct btrfs_root *root,
7769 struct btrfs_path *path,
7770 struct btrfs_key *first_key,
7771 struct btrfs_ref_path *ref_path)
7775 ret = relocate_one_path(trans, root, path, first_key,
7776 ref_path, NULL, NULL);
7782 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7783 struct btrfs_root *extent_root,
7784 struct btrfs_path *path,
7785 struct btrfs_key *extent_key)
7789 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7792 ret = btrfs_del_item(trans, extent_root, path);
7794 btrfs_release_path(extent_root, path);
7798 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7799 struct btrfs_ref_path *ref_path)
7801 struct btrfs_key root_key;
7803 root_key.objectid = ref_path->root_objectid;
7804 root_key.type = BTRFS_ROOT_ITEM_KEY;
7805 if (is_cowonly_root(ref_path->root_objectid))
7806 root_key.offset = 0;
7808 root_key.offset = (u64)-1;
7810 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7813 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7814 struct btrfs_path *path,
7815 struct btrfs_key *extent_key,
7816 struct btrfs_block_group_cache *group,
7817 struct inode *reloc_inode, int pass)
7819 struct btrfs_trans_handle *trans;
7820 struct btrfs_root *found_root;
7821 struct btrfs_ref_path *ref_path = NULL;
7822 struct disk_extent *new_extents = NULL;
7827 struct btrfs_key first_key;
7831 trans = btrfs_start_transaction(extent_root, 1);
7834 if (extent_key->objectid == 0) {
7835 ret = del_extent_zero(trans, extent_root, path, extent_key);
7839 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7845 for (loops = 0; ; loops++) {
7847 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7848 extent_key->objectid);
7850 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7857 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7858 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7861 found_root = read_ref_root(extent_root->fs_info, ref_path);
7862 BUG_ON(!found_root);
7864 * for reference counted tree, only process reference paths
7865 * rooted at the latest committed root.
7867 if (found_root->ref_cows &&
7868 ref_path->root_generation != found_root->root_key.offset)
7871 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7874 * copy data extents to new locations
7876 u64 group_start = group->key.objectid;
7877 ret = relocate_data_extent(reloc_inode,
7886 level = ref_path->owner_objectid;
7889 if (prev_block != ref_path->nodes[level]) {
7890 struct extent_buffer *eb;
7891 u64 block_start = ref_path->nodes[level];
7892 u64 block_size = btrfs_level_size(found_root, level);
7894 eb = read_tree_block(found_root, block_start,
7896 btrfs_tree_lock(eb);
7897 BUG_ON(level != btrfs_header_level(eb));
7900 btrfs_item_key_to_cpu(eb, &first_key, 0);
7902 btrfs_node_key_to_cpu(eb, &first_key, 0);
7904 btrfs_tree_unlock(eb);
7905 free_extent_buffer(eb);
7906 prev_block = block_start;
7909 mutex_lock(&extent_root->fs_info->trans_mutex);
7910 btrfs_record_root_in_trans(found_root);
7911 mutex_unlock(&extent_root->fs_info->trans_mutex);
7912 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7914 * try to update data extent references while
7915 * keeping metadata shared between snapshots.
7918 ret = relocate_one_path(trans, found_root,
7919 path, &first_key, ref_path,
7920 group, reloc_inode);
7926 * use fallback method to process the remaining
7930 u64 group_start = group->key.objectid;
7931 new_extents = kmalloc(sizeof(*new_extents),
7934 ret = get_new_locations(reloc_inode,
7942 ret = replace_one_extent(trans, found_root,
7944 &first_key, ref_path,
7945 new_extents, nr_extents);
7947 ret = relocate_tree_block(trans, found_root, path,
7948 &first_key, ref_path);
7955 btrfs_end_transaction(trans, extent_root);
7962 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7965 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7966 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7969 * we add in the count of missing devices because we want
7970 * to make sure that any RAID levels on a degraded FS
7971 * continue to be honored.
7973 num_devices = root->fs_info->fs_devices->rw_devices +
7974 root->fs_info->fs_devices->missing_devices;
7976 if (num_devices == 1) {
7977 stripped |= BTRFS_BLOCK_GROUP_DUP;
7978 stripped = flags & ~stripped;
7980 /* turn raid0 into single device chunks */
7981 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7984 /* turn mirroring into duplication */
7985 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7986 BTRFS_BLOCK_GROUP_RAID10))
7987 return stripped | BTRFS_BLOCK_GROUP_DUP;
7990 /* they already had raid on here, just return */
7991 if (flags & stripped)
7994 stripped |= BTRFS_BLOCK_GROUP_DUP;
7995 stripped = flags & ~stripped;
7997 /* switch duplicated blocks with raid1 */
7998 if (flags & BTRFS_BLOCK_GROUP_DUP)
7999 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8001 /* turn single device chunks into raid0 */
8002 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8007 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8009 struct btrfs_space_info *sinfo = cache->space_info;
8016 spin_lock(&sinfo->lock);
8017 spin_lock(&cache->lock);
8018 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8019 cache->bytes_super - btrfs_block_group_used(&cache->item);
8021 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8022 sinfo->bytes_may_use + sinfo->bytes_readonly +
8023 cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8024 sinfo->bytes_readonly += num_bytes;
8025 sinfo->bytes_reserved += cache->reserved_pinned;
8026 cache->reserved_pinned = 0;
8031 spin_unlock(&cache->lock);
8032 spin_unlock(&sinfo->lock);
8036 int btrfs_set_block_group_ro(struct btrfs_root *root,
8037 struct btrfs_block_group_cache *cache)
8040 struct btrfs_trans_handle *trans;
8046 trans = btrfs_join_transaction(root, 1);
8047 BUG_ON(IS_ERR(trans));
8049 alloc_flags = update_block_group_flags(root, cache->flags);
8050 if (alloc_flags != cache->flags)
8051 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8053 ret = set_block_group_ro(cache);
8056 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8057 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8060 ret = set_block_group_ro(cache);
8062 btrfs_end_transaction(trans, root);
8067 * helper to account the unused space of all the readonly block group in the
8068 * list. takes mirrors into account.
8070 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8072 struct btrfs_block_group_cache *block_group;
8076 list_for_each_entry(block_group, groups_list, list) {
8077 spin_lock(&block_group->lock);
8079 if (!block_group->ro) {
8080 spin_unlock(&block_group->lock);
8084 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8085 BTRFS_BLOCK_GROUP_RAID10 |
8086 BTRFS_BLOCK_GROUP_DUP))
8091 free_bytes += (block_group->key.offset -
8092 btrfs_block_group_used(&block_group->item)) *
8095 spin_unlock(&block_group->lock);
8102 * helper to account the unused space of all the readonly block group in the
8103 * space_info. takes mirrors into account.
8105 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8110 spin_lock(&sinfo->lock);
8112 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8113 if (!list_empty(&sinfo->block_groups[i]))
8114 free_bytes += __btrfs_get_ro_block_group_free_space(
8115 &sinfo->block_groups[i]);
8117 spin_unlock(&sinfo->lock);
8122 int btrfs_set_block_group_rw(struct btrfs_root *root,
8123 struct btrfs_block_group_cache *cache)
8125 struct btrfs_space_info *sinfo = cache->space_info;
8130 spin_lock(&sinfo->lock);
8131 spin_lock(&cache->lock);
8132 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8133 cache->bytes_super - btrfs_block_group_used(&cache->item);
8134 sinfo->bytes_readonly -= num_bytes;
8136 spin_unlock(&cache->lock);
8137 spin_unlock(&sinfo->lock);
8142 * checks to see if its even possible to relocate this block group.
8144 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8145 * ok to go ahead and try.
8147 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8149 struct btrfs_block_group_cache *block_group;
8150 struct btrfs_space_info *space_info;
8151 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8152 struct btrfs_device *device;
8156 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8158 /* odd, couldn't find the block group, leave it alone */
8162 /* no bytes used, we're good */
8163 if (!btrfs_block_group_used(&block_group->item))
8166 space_info = block_group->space_info;
8167 spin_lock(&space_info->lock);
8169 full = space_info->full;
8172 * if this is the last block group we have in this space, we can't
8173 * relocate it unless we're able to allocate a new chunk below.
8175 * Otherwise, we need to make sure we have room in the space to handle
8176 * all of the extents from this block group. If we can, we're good
8178 if ((space_info->total_bytes != block_group->key.offset) &&
8179 (space_info->bytes_used + space_info->bytes_reserved +
8180 space_info->bytes_pinned + space_info->bytes_readonly +
8181 btrfs_block_group_used(&block_group->item) <
8182 space_info->total_bytes)) {
8183 spin_unlock(&space_info->lock);
8186 spin_unlock(&space_info->lock);
8189 * ok we don't have enough space, but maybe we have free space on our
8190 * devices to allocate new chunks for relocation, so loop through our
8191 * alloc devices and guess if we have enough space. However, if we
8192 * were marked as full, then we know there aren't enough chunks, and we
8199 mutex_lock(&root->fs_info->chunk_mutex);
8200 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8201 u64 min_free = btrfs_block_group_used(&block_group->item);
8205 * check to make sure we can actually find a chunk with enough
8206 * space to fit our block group in.
8208 if (device->total_bytes > device->bytes_used + min_free) {
8209 ret = find_free_dev_extent(NULL, device, min_free,
8216 mutex_unlock(&root->fs_info->chunk_mutex);
8218 btrfs_put_block_group(block_group);
8222 static int find_first_block_group(struct btrfs_root *root,
8223 struct btrfs_path *path, struct btrfs_key *key)
8226 struct btrfs_key found_key;
8227 struct extent_buffer *leaf;
8230 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8235 slot = path->slots[0];
8236 leaf = path->nodes[0];
8237 if (slot >= btrfs_header_nritems(leaf)) {
8238 ret = btrfs_next_leaf(root, path);
8245 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8247 if (found_key.objectid >= key->objectid &&
8248 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8258 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8260 struct btrfs_block_group_cache *block_group;
8264 struct inode *inode;
8266 block_group = btrfs_lookup_first_block_group(info, last);
8267 while (block_group) {
8268 spin_lock(&block_group->lock);
8269 if (block_group->iref)
8271 spin_unlock(&block_group->lock);
8272 block_group = next_block_group(info->tree_root,
8282 inode = block_group->inode;
8283 block_group->iref = 0;
8284 block_group->inode = NULL;
8285 spin_unlock(&block_group->lock);
8287 last = block_group->key.objectid + block_group->key.offset;
8288 btrfs_put_block_group(block_group);
8292 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8294 struct btrfs_block_group_cache *block_group;
8295 struct btrfs_space_info *space_info;
8296 struct btrfs_caching_control *caching_ctl;
8299 down_write(&info->extent_commit_sem);
8300 while (!list_empty(&info->caching_block_groups)) {
8301 caching_ctl = list_entry(info->caching_block_groups.next,
8302 struct btrfs_caching_control, list);
8303 list_del(&caching_ctl->list);
8304 put_caching_control(caching_ctl);
8306 up_write(&info->extent_commit_sem);
8308 spin_lock(&info->block_group_cache_lock);
8309 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8310 block_group = rb_entry(n, struct btrfs_block_group_cache,
8312 rb_erase(&block_group->cache_node,
8313 &info->block_group_cache_tree);
8314 spin_unlock(&info->block_group_cache_lock);
8316 down_write(&block_group->space_info->groups_sem);
8317 list_del(&block_group->list);
8318 up_write(&block_group->space_info->groups_sem);
8320 if (block_group->cached == BTRFS_CACHE_STARTED)
8321 wait_block_group_cache_done(block_group);
8323 btrfs_remove_free_space_cache(block_group);
8324 btrfs_put_block_group(block_group);
8326 spin_lock(&info->block_group_cache_lock);
8328 spin_unlock(&info->block_group_cache_lock);
8330 /* now that all the block groups are freed, go through and
8331 * free all the space_info structs. This is only called during
8332 * the final stages of unmount, and so we know nobody is
8333 * using them. We call synchronize_rcu() once before we start,
8334 * just to be on the safe side.
8338 release_global_block_rsv(info);
8340 while(!list_empty(&info->space_info)) {
8341 space_info = list_entry(info->space_info.next,
8342 struct btrfs_space_info,
8344 if (space_info->bytes_pinned > 0 ||
8345 space_info->bytes_reserved > 0) {
8347 dump_space_info(space_info, 0, 0);
8349 list_del(&space_info->list);
8355 static void __link_block_group(struct btrfs_space_info *space_info,
8356 struct btrfs_block_group_cache *cache)
8358 int index = get_block_group_index(cache);
8360 down_write(&space_info->groups_sem);
8361 list_add_tail(&cache->list, &space_info->block_groups[index]);
8362 up_write(&space_info->groups_sem);
8365 int btrfs_read_block_groups(struct btrfs_root *root)
8367 struct btrfs_path *path;
8369 struct btrfs_block_group_cache *cache;
8370 struct btrfs_fs_info *info = root->fs_info;
8371 struct btrfs_space_info *space_info;
8372 struct btrfs_key key;
8373 struct btrfs_key found_key;
8374 struct extent_buffer *leaf;
8378 root = info->extent_root;
8381 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8382 path = btrfs_alloc_path();
8386 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8387 if (cache_gen != 0 &&
8388 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8390 if (btrfs_test_opt(root, CLEAR_CACHE))
8392 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8393 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8396 ret = find_first_block_group(root, path, &key);
8401 leaf = path->nodes[0];
8402 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8403 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8409 atomic_set(&cache->count, 1);
8410 spin_lock_init(&cache->lock);
8411 spin_lock_init(&cache->tree_lock);
8412 cache->fs_info = info;
8413 INIT_LIST_HEAD(&cache->list);
8414 INIT_LIST_HEAD(&cache->cluster_list);
8417 cache->disk_cache_state = BTRFS_DC_CLEAR;
8420 * we only want to have 32k of ram per block group for keeping
8421 * track of free space, and if we pass 1/2 of that we want to
8422 * start converting things over to using bitmaps
8424 cache->extents_thresh = ((1024 * 32) / 2) /
8425 sizeof(struct btrfs_free_space);
8427 read_extent_buffer(leaf, &cache->item,
8428 btrfs_item_ptr_offset(leaf, path->slots[0]),
8429 sizeof(cache->item));
8430 memcpy(&cache->key, &found_key, sizeof(found_key));
8432 key.objectid = found_key.objectid + found_key.offset;
8433 btrfs_release_path(root, path);
8434 cache->flags = btrfs_block_group_flags(&cache->item);
8435 cache->sectorsize = root->sectorsize;
8438 * check for two cases, either we are full, and therefore
8439 * don't need to bother with the caching work since we won't
8440 * find any space, or we are empty, and we can just add all
8441 * the space in and be done with it. This saves us _alot_ of
8442 * time, particularly in the full case.
8444 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8445 exclude_super_stripes(root, cache);
8446 cache->last_byte_to_unpin = (u64)-1;
8447 cache->cached = BTRFS_CACHE_FINISHED;
8448 free_excluded_extents(root, cache);
8449 } else if (btrfs_block_group_used(&cache->item) == 0) {
8450 exclude_super_stripes(root, cache);
8451 cache->last_byte_to_unpin = (u64)-1;
8452 cache->cached = BTRFS_CACHE_FINISHED;
8453 add_new_free_space(cache, root->fs_info,
8455 found_key.objectid +
8457 free_excluded_extents(root, cache);
8460 ret = update_space_info(info, cache->flags, found_key.offset,
8461 btrfs_block_group_used(&cache->item),
8464 cache->space_info = space_info;
8465 spin_lock(&cache->space_info->lock);
8466 cache->space_info->bytes_readonly += cache->bytes_super;
8467 spin_unlock(&cache->space_info->lock);
8469 __link_block_group(space_info, cache);
8471 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8474 set_avail_alloc_bits(root->fs_info, cache->flags);
8475 if (btrfs_chunk_readonly(root, cache->key.objectid))
8476 set_block_group_ro(cache);
8479 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8480 if (!(get_alloc_profile(root, space_info->flags) &
8481 (BTRFS_BLOCK_GROUP_RAID10 |
8482 BTRFS_BLOCK_GROUP_RAID1 |
8483 BTRFS_BLOCK_GROUP_DUP)))
8486 * avoid allocating from un-mirrored block group if there are
8487 * mirrored block groups.
8489 list_for_each_entry(cache, &space_info->block_groups[3], list)
8490 set_block_group_ro(cache);
8491 list_for_each_entry(cache, &space_info->block_groups[4], list)
8492 set_block_group_ro(cache);
8495 init_global_block_rsv(info);
8498 btrfs_free_path(path);
8502 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8503 struct btrfs_root *root, u64 bytes_used,
8504 u64 type, u64 chunk_objectid, u64 chunk_offset,
8508 struct btrfs_root *extent_root;
8509 struct btrfs_block_group_cache *cache;
8511 extent_root = root->fs_info->extent_root;
8513 root->fs_info->last_trans_log_full_commit = trans->transid;
8515 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8519 cache->key.objectid = chunk_offset;
8520 cache->key.offset = size;
8521 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8522 cache->sectorsize = root->sectorsize;
8523 cache->fs_info = root->fs_info;
8526 * we only want to have 32k of ram per block group for keeping track
8527 * of free space, and if we pass 1/2 of that we want to start
8528 * converting things over to using bitmaps
8530 cache->extents_thresh = ((1024 * 32) / 2) /
8531 sizeof(struct btrfs_free_space);
8532 atomic_set(&cache->count, 1);
8533 spin_lock_init(&cache->lock);
8534 spin_lock_init(&cache->tree_lock);
8535 INIT_LIST_HEAD(&cache->list);
8536 INIT_LIST_HEAD(&cache->cluster_list);
8538 btrfs_set_block_group_used(&cache->item, bytes_used);
8539 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8540 cache->flags = type;
8541 btrfs_set_block_group_flags(&cache->item, type);
8543 cache->last_byte_to_unpin = (u64)-1;
8544 cache->cached = BTRFS_CACHE_FINISHED;
8545 exclude_super_stripes(root, cache);
8547 add_new_free_space(cache, root->fs_info, chunk_offset,
8548 chunk_offset + size);
8550 free_excluded_extents(root, cache);
8552 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8553 &cache->space_info);
8556 spin_lock(&cache->space_info->lock);
8557 cache->space_info->bytes_readonly += cache->bytes_super;
8558 spin_unlock(&cache->space_info->lock);
8560 __link_block_group(cache->space_info, cache);
8562 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8565 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8566 sizeof(cache->item));
8569 set_avail_alloc_bits(extent_root->fs_info, type);
8574 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8575 struct btrfs_root *root, u64 group_start)
8577 struct btrfs_path *path;
8578 struct btrfs_block_group_cache *block_group;
8579 struct btrfs_free_cluster *cluster;
8580 struct btrfs_root *tree_root = root->fs_info->tree_root;
8581 struct btrfs_key key;
8582 struct inode *inode;
8586 root = root->fs_info->extent_root;
8588 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8589 BUG_ON(!block_group);
8590 BUG_ON(!block_group->ro);
8592 memcpy(&key, &block_group->key, sizeof(key));
8593 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8594 BTRFS_BLOCK_GROUP_RAID1 |
8595 BTRFS_BLOCK_GROUP_RAID10))
8600 /* make sure this block group isn't part of an allocation cluster */
8601 cluster = &root->fs_info->data_alloc_cluster;
8602 spin_lock(&cluster->refill_lock);
8603 btrfs_return_cluster_to_free_space(block_group, cluster);
8604 spin_unlock(&cluster->refill_lock);
8607 * make sure this block group isn't part of a metadata
8608 * allocation cluster
8610 cluster = &root->fs_info->meta_alloc_cluster;
8611 spin_lock(&cluster->refill_lock);
8612 btrfs_return_cluster_to_free_space(block_group, cluster);
8613 spin_unlock(&cluster->refill_lock);
8615 path = btrfs_alloc_path();
8618 inode = lookup_free_space_inode(root, block_group, path);
8619 if (!IS_ERR(inode)) {
8620 btrfs_orphan_add(trans, inode);
8622 /* One for the block groups ref */
8623 spin_lock(&block_group->lock);
8624 if (block_group->iref) {
8625 block_group->iref = 0;
8626 block_group->inode = NULL;
8627 spin_unlock(&block_group->lock);
8630 spin_unlock(&block_group->lock);
8632 /* One for our lookup ref */
8636 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8637 key.offset = block_group->key.objectid;
8640 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8644 btrfs_release_path(tree_root, path);
8646 ret = btrfs_del_item(trans, tree_root, path);
8649 btrfs_release_path(tree_root, path);
8652 spin_lock(&root->fs_info->block_group_cache_lock);
8653 rb_erase(&block_group->cache_node,
8654 &root->fs_info->block_group_cache_tree);
8655 spin_unlock(&root->fs_info->block_group_cache_lock);
8657 down_write(&block_group->space_info->groups_sem);
8659 * we must use list_del_init so people can check to see if they
8660 * are still on the list after taking the semaphore
8662 list_del_init(&block_group->list);
8663 up_write(&block_group->space_info->groups_sem);
8665 if (block_group->cached == BTRFS_CACHE_STARTED)
8666 wait_block_group_cache_done(block_group);
8668 btrfs_remove_free_space_cache(block_group);
8670 spin_lock(&block_group->space_info->lock);
8671 block_group->space_info->total_bytes -= block_group->key.offset;
8672 block_group->space_info->bytes_readonly -= block_group->key.offset;
8673 block_group->space_info->disk_total -= block_group->key.offset * factor;
8674 spin_unlock(&block_group->space_info->lock);
8676 memcpy(&key, &block_group->key, sizeof(key));
8678 btrfs_clear_space_info_full(root->fs_info);
8680 btrfs_put_block_group(block_group);
8681 btrfs_put_block_group(block_group);
8683 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8689 ret = btrfs_del_item(trans, root, path);
8691 btrfs_free_path(path);
8695 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8697 return unpin_extent_range(root, start, end);
8700 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8703 return btrfs_discard_extent(root, bytenr, num_bytes);