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 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
55 static int update_block_group(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 bytenr, u64 num_bytes, int alloc);
58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, u64 parent,
61 u64 root_objectid, u64 owner_objectid,
62 u64 owner_offset, int refs_to_drop,
63 struct btrfs_delayed_extent_op *extra_op);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
65 struct extent_buffer *leaf,
66 struct btrfs_extent_item *ei);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_root *root,
69 u64 parent, u64 root_objectid,
70 u64 flags, u64 owner, u64 offset,
71 struct btrfs_key *ins, int ref_mod);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 parent, u64 root_objectid,
75 u64 flags, struct btrfs_disk_key *key,
76 int level, struct btrfs_key *ins);
77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
78 struct btrfs_root *extent_root, u64 alloc_bytes,
79 u64 flags, int force);
80 static int find_next_key(struct btrfs_path *path, int level,
81 struct btrfs_key *key);
82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
83 int dump_block_groups);
86 block_group_cache_done(struct btrfs_block_group_cache *cache)
89 return cache->cached == BTRFS_CACHE_FINISHED;
92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
94 return (cache->flags & bits) == bits;
97 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
99 atomic_inc(&cache->count);
102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
104 if (atomic_dec_and_test(&cache->count)) {
105 WARN_ON(cache->pinned > 0);
106 WARN_ON(cache->reserved > 0);
107 WARN_ON(cache->reserved_pinned > 0);
108 kfree(cache->free_space_ctl);
114 * this adds the block group to the fs_info rb tree for the block group
117 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
118 struct btrfs_block_group_cache *block_group)
121 struct rb_node *parent = NULL;
122 struct btrfs_block_group_cache *cache;
124 spin_lock(&info->block_group_cache_lock);
125 p = &info->block_group_cache_tree.rb_node;
129 cache = rb_entry(parent, struct btrfs_block_group_cache,
131 if (block_group->key.objectid < cache->key.objectid) {
133 } else if (block_group->key.objectid > cache->key.objectid) {
136 spin_unlock(&info->block_group_cache_lock);
141 rb_link_node(&block_group->cache_node, parent, p);
142 rb_insert_color(&block_group->cache_node,
143 &info->block_group_cache_tree);
144 spin_unlock(&info->block_group_cache_lock);
150 * This will return the block group at or after bytenr if contains is 0, else
151 * it will return the block group that contains the bytenr
153 static struct btrfs_block_group_cache *
154 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
157 struct btrfs_block_group_cache *cache, *ret = NULL;
161 spin_lock(&info->block_group_cache_lock);
162 n = info->block_group_cache_tree.rb_node;
165 cache = rb_entry(n, struct btrfs_block_group_cache,
167 end = cache->key.objectid + cache->key.offset - 1;
168 start = cache->key.objectid;
170 if (bytenr < start) {
171 if (!contains && (!ret || start < ret->key.objectid))
174 } else if (bytenr > start) {
175 if (contains && bytenr <= end) {
186 btrfs_get_block_group(ret);
187 spin_unlock(&info->block_group_cache_lock);
192 static int add_excluded_extent(struct btrfs_root *root,
193 u64 start, u64 num_bytes)
195 u64 end = start + num_bytes - 1;
196 set_extent_bits(&root->fs_info->freed_extents[0],
197 start, end, EXTENT_UPTODATE, GFP_NOFS);
198 set_extent_bits(&root->fs_info->freed_extents[1],
199 start, end, EXTENT_UPTODATE, GFP_NOFS);
203 static void free_excluded_extents(struct btrfs_root *root,
204 struct btrfs_block_group_cache *cache)
208 start = cache->key.objectid;
209 end = start + cache->key.offset - 1;
211 clear_extent_bits(&root->fs_info->freed_extents[0],
212 start, end, EXTENT_UPTODATE, GFP_NOFS);
213 clear_extent_bits(&root->fs_info->freed_extents[1],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
217 static int exclude_super_stripes(struct btrfs_root *root,
218 struct btrfs_block_group_cache *cache)
225 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
226 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
227 cache->bytes_super += stripe_len;
228 ret = add_excluded_extent(root, cache->key.objectid,
233 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
234 bytenr = btrfs_sb_offset(i);
235 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
236 cache->key.objectid, bytenr,
237 0, &logical, &nr, &stripe_len);
241 cache->bytes_super += stripe_len;
242 ret = add_excluded_extent(root, logical[nr],
252 static struct btrfs_caching_control *
253 get_caching_control(struct btrfs_block_group_cache *cache)
255 struct btrfs_caching_control *ctl;
257 spin_lock(&cache->lock);
258 if (cache->cached != BTRFS_CACHE_STARTED) {
259 spin_unlock(&cache->lock);
263 /* We're loading it the fast way, so we don't have a caching_ctl. */
264 if (!cache->caching_ctl) {
265 spin_unlock(&cache->lock);
269 ctl = cache->caching_ctl;
270 atomic_inc(&ctl->count);
271 spin_unlock(&cache->lock);
275 static void put_caching_control(struct btrfs_caching_control *ctl)
277 if (atomic_dec_and_test(&ctl->count))
282 * this is only called by cache_block_group, since we could have freed extents
283 * we need to check the pinned_extents for any extents that can't be used yet
284 * since their free space will be released as soon as the transaction commits.
286 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
287 struct btrfs_fs_info *info, u64 start, u64 end)
289 u64 extent_start, extent_end, size, total_added = 0;
292 while (start < end) {
293 ret = find_first_extent_bit(info->pinned_extents, start,
294 &extent_start, &extent_end,
295 EXTENT_DIRTY | EXTENT_UPTODATE);
299 if (extent_start <= start) {
300 start = extent_end + 1;
301 } else if (extent_start > start && extent_start < end) {
302 size = extent_start - start;
304 ret = btrfs_add_free_space(block_group, start,
307 start = extent_end + 1;
316 ret = btrfs_add_free_space(block_group, start, size);
323 static noinline void caching_thread(struct btrfs_work *work)
325 struct btrfs_block_group_cache *block_group;
326 struct btrfs_fs_info *fs_info;
327 struct btrfs_caching_control *caching_ctl;
328 struct btrfs_root *extent_root;
329 struct btrfs_path *path;
330 struct extent_buffer *leaf;
331 struct btrfs_key key;
337 caching_ctl = container_of(work, struct btrfs_caching_control, work);
338 block_group = caching_ctl->block_group;
339 fs_info = block_group->fs_info;
340 extent_root = fs_info->extent_root;
342 path = btrfs_alloc_path();
346 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
349 * We don't want to deadlock with somebody trying to allocate a new
350 * extent for the extent root while also trying to search the extent
351 * root to add free space. So we skip locking and search the commit
352 * root, since its read-only
354 path->skip_locking = 1;
355 path->search_commit_root = 1;
360 key.type = BTRFS_EXTENT_ITEM_KEY;
362 mutex_lock(&caching_ctl->mutex);
363 /* need to make sure the commit_root doesn't disappear */
364 down_read(&fs_info->extent_commit_sem);
366 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
370 leaf = path->nodes[0];
371 nritems = btrfs_header_nritems(leaf);
374 if (btrfs_fs_closing(fs_info) > 1) {
379 if (path->slots[0] < nritems) {
380 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
382 ret = find_next_key(path, 0, &key);
386 if (need_resched() ||
387 btrfs_next_leaf(extent_root, path)) {
388 caching_ctl->progress = last;
389 btrfs_release_path(path);
390 up_read(&fs_info->extent_commit_sem);
391 mutex_unlock(&caching_ctl->mutex);
395 leaf = path->nodes[0];
396 nritems = btrfs_header_nritems(leaf);
400 if (key.objectid < block_group->key.objectid) {
405 if (key.objectid >= block_group->key.objectid +
406 block_group->key.offset)
409 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
410 total_found += add_new_free_space(block_group,
413 last = key.objectid + key.offset;
415 if (total_found > (1024 * 1024 * 2)) {
417 wake_up(&caching_ctl->wait);
424 total_found += add_new_free_space(block_group, fs_info, last,
425 block_group->key.objectid +
426 block_group->key.offset);
427 caching_ctl->progress = (u64)-1;
429 spin_lock(&block_group->lock);
430 block_group->caching_ctl = NULL;
431 block_group->cached = BTRFS_CACHE_FINISHED;
432 spin_unlock(&block_group->lock);
435 btrfs_free_path(path);
436 up_read(&fs_info->extent_commit_sem);
438 free_excluded_extents(extent_root, block_group);
440 mutex_unlock(&caching_ctl->mutex);
442 wake_up(&caching_ctl->wait);
444 put_caching_control(caching_ctl);
445 btrfs_put_block_group(block_group);
448 static int cache_block_group(struct btrfs_block_group_cache *cache,
449 struct btrfs_trans_handle *trans,
450 struct btrfs_root *root,
453 struct btrfs_fs_info *fs_info = cache->fs_info;
454 struct btrfs_caching_control *caching_ctl;
458 if (cache->cached != BTRFS_CACHE_NO)
462 * We can't do the read from on-disk cache during a commit since we need
463 * to have the normal tree locking. Also if we are currently trying to
464 * allocate blocks for the tree root we can't do the fast caching since
465 * we likely hold important locks.
467 if (trans && (!trans->transaction->in_commit) &&
468 (root && root != root->fs_info->tree_root)) {
469 spin_lock(&cache->lock);
470 if (cache->cached != BTRFS_CACHE_NO) {
471 spin_unlock(&cache->lock);
474 cache->cached = BTRFS_CACHE_STARTED;
475 spin_unlock(&cache->lock);
477 ret = load_free_space_cache(fs_info, cache);
479 spin_lock(&cache->lock);
481 cache->cached = BTRFS_CACHE_FINISHED;
482 cache->last_byte_to_unpin = (u64)-1;
484 cache->cached = BTRFS_CACHE_NO;
486 spin_unlock(&cache->lock);
488 free_excluded_extents(fs_info->extent_root, cache);
496 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
497 BUG_ON(!caching_ctl);
499 INIT_LIST_HEAD(&caching_ctl->list);
500 mutex_init(&caching_ctl->mutex);
501 init_waitqueue_head(&caching_ctl->wait);
502 caching_ctl->block_group = cache;
503 caching_ctl->progress = cache->key.objectid;
504 /* one for caching kthread, one for caching block group list */
505 atomic_set(&caching_ctl->count, 2);
506 caching_ctl->work.func = caching_thread;
508 spin_lock(&cache->lock);
509 if (cache->cached != BTRFS_CACHE_NO) {
510 spin_unlock(&cache->lock);
514 cache->caching_ctl = caching_ctl;
515 cache->cached = BTRFS_CACHE_STARTED;
516 spin_unlock(&cache->lock);
518 down_write(&fs_info->extent_commit_sem);
519 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
520 up_write(&fs_info->extent_commit_sem);
522 btrfs_get_block_group(cache);
524 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
530 * return the block group that starts at or after bytenr
532 static struct btrfs_block_group_cache *
533 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
535 struct btrfs_block_group_cache *cache;
537 cache = block_group_cache_tree_search(info, bytenr, 0);
543 * return the block group that contains the given bytenr
545 struct btrfs_block_group_cache *btrfs_lookup_block_group(
546 struct btrfs_fs_info *info,
549 struct btrfs_block_group_cache *cache;
551 cache = block_group_cache_tree_search(info, bytenr, 1);
556 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
559 struct list_head *head = &info->space_info;
560 struct btrfs_space_info *found;
562 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
563 BTRFS_BLOCK_GROUP_METADATA;
566 list_for_each_entry_rcu(found, head, list) {
567 if (found->flags & flags) {
577 * after adding space to the filesystem, we need to clear the full flags
578 * on all the space infos.
580 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
582 struct list_head *head = &info->space_info;
583 struct btrfs_space_info *found;
586 list_for_each_entry_rcu(found, head, list)
591 static u64 div_factor(u64 num, int factor)
600 static u64 div_factor_fine(u64 num, int factor)
609 u64 btrfs_find_block_group(struct btrfs_root *root,
610 u64 search_start, u64 search_hint, int owner)
612 struct btrfs_block_group_cache *cache;
614 u64 last = max(search_hint, search_start);
621 cache = btrfs_lookup_first_block_group(root->fs_info, last);
625 spin_lock(&cache->lock);
626 last = cache->key.objectid + cache->key.offset;
627 used = btrfs_block_group_used(&cache->item);
629 if ((full_search || !cache->ro) &&
630 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
631 if (used + cache->pinned + cache->reserved <
632 div_factor(cache->key.offset, factor)) {
633 group_start = cache->key.objectid;
634 spin_unlock(&cache->lock);
635 btrfs_put_block_group(cache);
639 spin_unlock(&cache->lock);
640 btrfs_put_block_group(cache);
648 if (!full_search && factor < 10) {
658 /* simple helper to search for an existing extent at a given offset */
659 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
662 struct btrfs_key key;
663 struct btrfs_path *path;
665 path = btrfs_alloc_path();
667 key.objectid = start;
669 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
670 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
672 btrfs_free_path(path);
677 * helper function to lookup reference count and flags of extent.
679 * the head node for delayed ref is used to store the sum of all the
680 * reference count modifications queued up in the rbtree. the head
681 * node may also store the extent flags to set. This way you can check
682 * to see what the reference count and extent flags would be if all of
683 * the delayed refs are not processed.
685 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
686 struct btrfs_root *root, u64 bytenr,
687 u64 num_bytes, u64 *refs, u64 *flags)
689 struct btrfs_delayed_ref_head *head;
690 struct btrfs_delayed_ref_root *delayed_refs;
691 struct btrfs_path *path;
692 struct btrfs_extent_item *ei;
693 struct extent_buffer *leaf;
694 struct btrfs_key key;
700 path = btrfs_alloc_path();
704 key.objectid = bytenr;
705 key.type = BTRFS_EXTENT_ITEM_KEY;
706 key.offset = num_bytes;
708 path->skip_locking = 1;
709 path->search_commit_root = 1;
712 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
718 leaf = path->nodes[0];
719 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
720 if (item_size >= sizeof(*ei)) {
721 ei = btrfs_item_ptr(leaf, path->slots[0],
722 struct btrfs_extent_item);
723 num_refs = btrfs_extent_refs(leaf, ei);
724 extent_flags = btrfs_extent_flags(leaf, ei);
726 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
727 struct btrfs_extent_item_v0 *ei0;
728 BUG_ON(item_size != sizeof(*ei0));
729 ei0 = btrfs_item_ptr(leaf, path->slots[0],
730 struct btrfs_extent_item_v0);
731 num_refs = btrfs_extent_refs_v0(leaf, ei0);
732 /* FIXME: this isn't correct for data */
733 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
738 BUG_ON(num_refs == 0);
748 delayed_refs = &trans->transaction->delayed_refs;
749 spin_lock(&delayed_refs->lock);
750 head = btrfs_find_delayed_ref_head(trans, bytenr);
752 if (!mutex_trylock(&head->mutex)) {
753 atomic_inc(&head->node.refs);
754 spin_unlock(&delayed_refs->lock);
756 btrfs_release_path(path);
759 * Mutex was contended, block until it's released and try
762 mutex_lock(&head->mutex);
763 mutex_unlock(&head->mutex);
764 btrfs_put_delayed_ref(&head->node);
767 if (head->extent_op && head->extent_op->update_flags)
768 extent_flags |= head->extent_op->flags_to_set;
770 BUG_ON(num_refs == 0);
772 num_refs += head->node.ref_mod;
773 mutex_unlock(&head->mutex);
775 spin_unlock(&delayed_refs->lock);
777 WARN_ON(num_refs == 0);
781 *flags = extent_flags;
783 btrfs_free_path(path);
788 * Back reference rules. Back refs have three main goals:
790 * 1) differentiate between all holders of references to an extent so that
791 * when a reference is dropped we can make sure it was a valid reference
792 * before freeing the extent.
794 * 2) Provide enough information to quickly find the holders of an extent
795 * if we notice a given block is corrupted or bad.
797 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
798 * maintenance. This is actually the same as #2, but with a slightly
799 * different use case.
801 * There are two kinds of back refs. The implicit back refs is optimized
802 * for pointers in non-shared tree blocks. For a given pointer in a block,
803 * back refs of this kind provide information about the block's owner tree
804 * and the pointer's key. These information allow us to find the block by
805 * b-tree searching. The full back refs is for pointers in tree blocks not
806 * referenced by their owner trees. The location of tree block is recorded
807 * in the back refs. Actually the full back refs is generic, and can be
808 * used in all cases the implicit back refs is used. The major shortcoming
809 * of the full back refs is its overhead. Every time a tree block gets
810 * COWed, we have to update back refs entry for all pointers in it.
812 * For a newly allocated tree block, we use implicit back refs for
813 * pointers in it. This means most tree related operations only involve
814 * implicit back refs. For a tree block created in old transaction, the
815 * only way to drop a reference to it is COW it. So we can detect the
816 * event that tree block loses its owner tree's reference and do the
817 * back refs conversion.
819 * When a tree block is COW'd through a tree, there are four cases:
821 * The reference count of the block is one and the tree is the block's
822 * owner tree. Nothing to do in this case.
824 * The reference count of the block is one and the tree is not the
825 * block's owner tree. In this case, full back refs is used for pointers
826 * in the block. Remove these full back refs, add implicit back refs for
827 * every pointers in the new block.
829 * The reference count of the block is greater than one and the tree is
830 * the block's owner tree. In this case, implicit back refs is used for
831 * pointers in the block. Add full back refs for every pointers in the
832 * block, increase lower level extents' reference counts. The original
833 * implicit back refs are entailed to the new block.
835 * The reference count of the block is greater than one and the tree is
836 * not the block's owner tree. Add implicit back refs for every pointer in
837 * the new block, increase lower level extents' reference count.
839 * Back Reference Key composing:
841 * The key objectid corresponds to the first byte in the extent,
842 * The key type is used to differentiate between types of back refs.
843 * There are different meanings of the key offset for different types
846 * File extents can be referenced by:
848 * - multiple snapshots, subvolumes, or different generations in one subvol
849 * - different files inside a single subvolume
850 * - different offsets inside a file (bookend extents in file.c)
852 * The extent ref structure for the implicit back refs has fields for:
854 * - Objectid of the subvolume root
855 * - objectid of the file holding the reference
856 * - original offset in the file
857 * - how many bookend extents
859 * The key offset for the implicit back refs is hash of the first
862 * The extent ref structure for the full back refs has field for:
864 * - number of pointers in the tree leaf
866 * The key offset for the implicit back refs is the first byte of
869 * When a file extent is allocated, The implicit back refs is used.
870 * the fields are filled in:
872 * (root_key.objectid, inode objectid, offset in file, 1)
874 * When a file extent is removed file truncation, we find the
875 * corresponding implicit back refs and check the following fields:
877 * (btrfs_header_owner(leaf), inode objectid, offset in file)
879 * Btree extents can be referenced by:
881 * - Different subvolumes
883 * Both the implicit back refs and the full back refs for tree blocks
884 * only consist of key. The key offset for the implicit back refs is
885 * objectid of block's owner tree. The key offset for the full back refs
886 * is the first byte of parent block.
888 * When implicit back refs is used, information about the lowest key and
889 * level of the tree block are required. These information are stored in
890 * tree block info structure.
893 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
894 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
895 struct btrfs_root *root,
896 struct btrfs_path *path,
897 u64 owner, u32 extra_size)
899 struct btrfs_extent_item *item;
900 struct btrfs_extent_item_v0 *ei0;
901 struct btrfs_extent_ref_v0 *ref0;
902 struct btrfs_tree_block_info *bi;
903 struct extent_buffer *leaf;
904 struct btrfs_key key;
905 struct btrfs_key found_key;
906 u32 new_size = sizeof(*item);
910 leaf = path->nodes[0];
911 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
913 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
914 ei0 = btrfs_item_ptr(leaf, path->slots[0],
915 struct btrfs_extent_item_v0);
916 refs = btrfs_extent_refs_v0(leaf, ei0);
918 if (owner == (u64)-1) {
920 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
921 ret = btrfs_next_leaf(root, path);
925 leaf = path->nodes[0];
927 btrfs_item_key_to_cpu(leaf, &found_key,
929 BUG_ON(key.objectid != found_key.objectid);
930 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
934 ref0 = btrfs_item_ptr(leaf, path->slots[0],
935 struct btrfs_extent_ref_v0);
936 owner = btrfs_ref_objectid_v0(leaf, ref0);
940 btrfs_release_path(path);
942 if (owner < BTRFS_FIRST_FREE_OBJECTID)
943 new_size += sizeof(*bi);
945 new_size -= sizeof(*ei0);
946 ret = btrfs_search_slot(trans, root, &key, path,
947 new_size + extra_size, 1);
952 ret = btrfs_extend_item(trans, root, path, new_size);
954 leaf = path->nodes[0];
955 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
956 btrfs_set_extent_refs(leaf, item, refs);
957 /* FIXME: get real generation */
958 btrfs_set_extent_generation(leaf, item, 0);
959 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
960 btrfs_set_extent_flags(leaf, item,
961 BTRFS_EXTENT_FLAG_TREE_BLOCK |
962 BTRFS_BLOCK_FLAG_FULL_BACKREF);
963 bi = (struct btrfs_tree_block_info *)(item + 1);
964 /* FIXME: get first key of the block */
965 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
966 btrfs_set_tree_block_level(leaf, bi, (int)owner);
968 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
970 btrfs_mark_buffer_dirty(leaf);
975 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
977 u32 high_crc = ~(u32)0;
978 u32 low_crc = ~(u32)0;
981 lenum = cpu_to_le64(root_objectid);
982 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
983 lenum = cpu_to_le64(owner);
984 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
985 lenum = cpu_to_le64(offset);
986 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
988 return ((u64)high_crc << 31) ^ (u64)low_crc;
991 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
992 struct btrfs_extent_data_ref *ref)
994 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
995 btrfs_extent_data_ref_objectid(leaf, ref),
996 btrfs_extent_data_ref_offset(leaf, ref));
999 static int match_extent_data_ref(struct extent_buffer *leaf,
1000 struct btrfs_extent_data_ref *ref,
1001 u64 root_objectid, u64 owner, u64 offset)
1003 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1004 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1005 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1010 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1011 struct btrfs_root *root,
1012 struct btrfs_path *path,
1013 u64 bytenr, u64 parent,
1015 u64 owner, u64 offset)
1017 struct btrfs_key key;
1018 struct btrfs_extent_data_ref *ref;
1019 struct extent_buffer *leaf;
1025 key.objectid = bytenr;
1027 key.type = BTRFS_SHARED_DATA_REF_KEY;
1028 key.offset = parent;
1030 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1031 key.offset = hash_extent_data_ref(root_objectid,
1036 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1045 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1046 key.type = BTRFS_EXTENT_REF_V0_KEY;
1047 btrfs_release_path(path);
1048 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1059 leaf = path->nodes[0];
1060 nritems = btrfs_header_nritems(leaf);
1062 if (path->slots[0] >= nritems) {
1063 ret = btrfs_next_leaf(root, path);
1069 leaf = path->nodes[0];
1070 nritems = btrfs_header_nritems(leaf);
1074 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1075 if (key.objectid != bytenr ||
1076 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1079 ref = btrfs_item_ptr(leaf, path->slots[0],
1080 struct btrfs_extent_data_ref);
1082 if (match_extent_data_ref(leaf, ref, root_objectid,
1085 btrfs_release_path(path);
1097 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1098 struct btrfs_root *root,
1099 struct btrfs_path *path,
1100 u64 bytenr, u64 parent,
1101 u64 root_objectid, u64 owner,
1102 u64 offset, int refs_to_add)
1104 struct btrfs_key key;
1105 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1114 size = sizeof(struct btrfs_shared_data_ref);
1116 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 key.offset = hash_extent_data_ref(root_objectid,
1119 size = sizeof(struct btrfs_extent_data_ref);
1122 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1123 if (ret && ret != -EEXIST)
1126 leaf = path->nodes[0];
1128 struct btrfs_shared_data_ref *ref;
1129 ref = btrfs_item_ptr(leaf, path->slots[0],
1130 struct btrfs_shared_data_ref);
1132 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1134 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1135 num_refs += refs_to_add;
1136 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1139 struct btrfs_extent_data_ref *ref;
1140 while (ret == -EEXIST) {
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1143 if (match_extent_data_ref(leaf, ref, root_objectid,
1146 btrfs_release_path(path);
1148 ret = btrfs_insert_empty_item(trans, root, path, &key,
1150 if (ret && ret != -EEXIST)
1153 leaf = path->nodes[0];
1155 ref = btrfs_item_ptr(leaf, path->slots[0],
1156 struct btrfs_extent_data_ref);
1158 btrfs_set_extent_data_ref_root(leaf, ref,
1160 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1161 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1162 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1164 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1165 num_refs += refs_to_add;
1166 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1169 btrfs_mark_buffer_dirty(leaf);
1172 btrfs_release_path(path);
1176 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1177 struct btrfs_root *root,
1178 struct btrfs_path *path,
1181 struct btrfs_key key;
1182 struct btrfs_extent_data_ref *ref1 = NULL;
1183 struct btrfs_shared_data_ref *ref2 = NULL;
1184 struct extent_buffer *leaf;
1188 leaf = path->nodes[0];
1189 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1191 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1192 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_extent_data_ref);
1194 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1195 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1196 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1197 struct btrfs_shared_data_ref);
1198 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1199 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1200 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1201 struct btrfs_extent_ref_v0 *ref0;
1202 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1203 struct btrfs_extent_ref_v0);
1204 num_refs = btrfs_ref_count_v0(leaf, ref0);
1210 BUG_ON(num_refs < refs_to_drop);
1211 num_refs -= refs_to_drop;
1213 if (num_refs == 0) {
1214 ret = btrfs_del_item(trans, root, path);
1216 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1217 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1218 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1219 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1220 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1222 struct btrfs_extent_ref_v0 *ref0;
1223 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_extent_ref_v0);
1225 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1228 btrfs_mark_buffer_dirty(leaf);
1233 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1234 struct btrfs_path *path,
1235 struct btrfs_extent_inline_ref *iref)
1237 struct btrfs_key key;
1238 struct extent_buffer *leaf;
1239 struct btrfs_extent_data_ref *ref1;
1240 struct btrfs_shared_data_ref *ref2;
1243 leaf = path->nodes[0];
1244 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1246 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1247 BTRFS_EXTENT_DATA_REF_KEY) {
1248 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1251 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1252 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1254 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1255 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1256 struct btrfs_extent_data_ref);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1258 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1259 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_shared_data_ref);
1261 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1262 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1263 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1264 struct btrfs_extent_ref_v0 *ref0;
1265 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1266 struct btrfs_extent_ref_v0);
1267 num_refs = btrfs_ref_count_v0(leaf, ref0);
1275 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1276 struct btrfs_root *root,
1277 struct btrfs_path *path,
1278 u64 bytenr, u64 parent,
1281 struct btrfs_key key;
1284 key.objectid = bytenr;
1286 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1287 key.offset = parent;
1289 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1290 key.offset = root_objectid;
1293 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1296 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1297 if (ret == -ENOENT && parent) {
1298 btrfs_release_path(path);
1299 key.type = BTRFS_EXTENT_REF_V0_KEY;
1300 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1308 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1309 struct btrfs_root *root,
1310 struct btrfs_path *path,
1311 u64 bytenr, u64 parent,
1314 struct btrfs_key key;
1317 key.objectid = bytenr;
1319 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1320 key.offset = parent;
1322 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1323 key.offset = root_objectid;
1326 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1327 btrfs_release_path(path);
1331 static inline int extent_ref_type(u64 parent, u64 owner)
1334 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1336 type = BTRFS_SHARED_BLOCK_REF_KEY;
1338 type = BTRFS_TREE_BLOCK_REF_KEY;
1341 type = BTRFS_SHARED_DATA_REF_KEY;
1343 type = BTRFS_EXTENT_DATA_REF_KEY;
1348 static int find_next_key(struct btrfs_path *path, int level,
1349 struct btrfs_key *key)
1352 for (; level < BTRFS_MAX_LEVEL; level++) {
1353 if (!path->nodes[level])
1355 if (path->slots[level] + 1 >=
1356 btrfs_header_nritems(path->nodes[level]))
1359 btrfs_item_key_to_cpu(path->nodes[level], key,
1360 path->slots[level] + 1);
1362 btrfs_node_key_to_cpu(path->nodes[level], key,
1363 path->slots[level] + 1);
1370 * look for inline back ref. if back ref is found, *ref_ret is set
1371 * to the address of inline back ref, and 0 is returned.
1373 * if back ref isn't found, *ref_ret is set to the address where it
1374 * should be inserted, and -ENOENT is returned.
1376 * if insert is true and there are too many inline back refs, the path
1377 * points to the extent item, and -EAGAIN is returned.
1379 * NOTE: inline back refs are ordered in the same way that back ref
1380 * items in the tree are ordered.
1382 static noinline_for_stack
1383 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1384 struct btrfs_root *root,
1385 struct btrfs_path *path,
1386 struct btrfs_extent_inline_ref **ref_ret,
1387 u64 bytenr, u64 num_bytes,
1388 u64 parent, u64 root_objectid,
1389 u64 owner, u64 offset, int insert)
1391 struct btrfs_key key;
1392 struct extent_buffer *leaf;
1393 struct btrfs_extent_item *ei;
1394 struct btrfs_extent_inline_ref *iref;
1405 key.objectid = bytenr;
1406 key.type = BTRFS_EXTENT_ITEM_KEY;
1407 key.offset = num_bytes;
1409 want = extent_ref_type(parent, owner);
1411 extra_size = btrfs_extent_inline_ref_size(want);
1412 path->keep_locks = 1;
1415 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1422 leaf = path->nodes[0];
1423 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1424 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1425 if (item_size < sizeof(*ei)) {
1430 ret = convert_extent_item_v0(trans, root, path, owner,
1436 leaf = path->nodes[0];
1437 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1440 BUG_ON(item_size < sizeof(*ei));
1442 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1443 flags = btrfs_extent_flags(leaf, ei);
1445 ptr = (unsigned long)(ei + 1);
1446 end = (unsigned long)ei + item_size;
1448 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1449 ptr += sizeof(struct btrfs_tree_block_info);
1452 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1461 iref = (struct btrfs_extent_inline_ref *)ptr;
1462 type = btrfs_extent_inline_ref_type(leaf, iref);
1466 ptr += btrfs_extent_inline_ref_size(type);
1470 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1471 struct btrfs_extent_data_ref *dref;
1472 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1473 if (match_extent_data_ref(leaf, dref, root_objectid,
1478 if (hash_extent_data_ref_item(leaf, dref) <
1479 hash_extent_data_ref(root_objectid, owner, offset))
1483 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1485 if (parent == ref_offset) {
1489 if (ref_offset < parent)
1492 if (root_objectid == ref_offset) {
1496 if (ref_offset < root_objectid)
1500 ptr += btrfs_extent_inline_ref_size(type);
1502 if (err == -ENOENT && insert) {
1503 if (item_size + extra_size >=
1504 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1509 * To add new inline back ref, we have to make sure
1510 * there is no corresponding back ref item.
1511 * For simplicity, we just do not add new inline back
1512 * ref if there is any kind of item for this block
1514 if (find_next_key(path, 0, &key) == 0 &&
1515 key.objectid == bytenr &&
1516 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1521 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1524 path->keep_locks = 0;
1525 btrfs_unlock_up_safe(path, 1);
1531 * helper to add new inline back ref
1533 static noinline_for_stack
1534 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1535 struct btrfs_root *root,
1536 struct btrfs_path *path,
1537 struct btrfs_extent_inline_ref *iref,
1538 u64 parent, u64 root_objectid,
1539 u64 owner, u64 offset, int refs_to_add,
1540 struct btrfs_delayed_extent_op *extent_op)
1542 struct extent_buffer *leaf;
1543 struct btrfs_extent_item *ei;
1546 unsigned long item_offset;
1552 leaf = path->nodes[0];
1553 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1554 item_offset = (unsigned long)iref - (unsigned long)ei;
1556 type = extent_ref_type(parent, owner);
1557 size = btrfs_extent_inline_ref_size(type);
1559 ret = btrfs_extend_item(trans, root, path, size);
1561 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1562 refs = btrfs_extent_refs(leaf, ei);
1563 refs += refs_to_add;
1564 btrfs_set_extent_refs(leaf, ei, refs);
1566 __run_delayed_extent_op(extent_op, leaf, ei);
1568 ptr = (unsigned long)ei + item_offset;
1569 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1570 if (ptr < end - size)
1571 memmove_extent_buffer(leaf, ptr + size, ptr,
1574 iref = (struct btrfs_extent_inline_ref *)ptr;
1575 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1576 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1577 struct btrfs_extent_data_ref *dref;
1578 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1579 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1580 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1581 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1582 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1583 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1584 struct btrfs_shared_data_ref *sref;
1585 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1586 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1587 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1588 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1589 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1591 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1593 btrfs_mark_buffer_dirty(leaf);
1597 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 struct btrfs_path *path,
1600 struct btrfs_extent_inline_ref **ref_ret,
1601 u64 bytenr, u64 num_bytes, u64 parent,
1602 u64 root_objectid, u64 owner, u64 offset)
1606 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1607 bytenr, num_bytes, parent,
1608 root_objectid, owner, offset, 0);
1612 btrfs_release_path(path);
1615 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1616 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1619 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1620 root_objectid, owner, offset);
1626 * helper to update/remove inline back ref
1628 static noinline_for_stack
1629 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1630 struct btrfs_root *root,
1631 struct btrfs_path *path,
1632 struct btrfs_extent_inline_ref *iref,
1634 struct btrfs_delayed_extent_op *extent_op)
1636 struct extent_buffer *leaf;
1637 struct btrfs_extent_item *ei;
1638 struct btrfs_extent_data_ref *dref = NULL;
1639 struct btrfs_shared_data_ref *sref = NULL;
1648 leaf = path->nodes[0];
1649 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1650 refs = btrfs_extent_refs(leaf, ei);
1651 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1652 refs += refs_to_mod;
1653 btrfs_set_extent_refs(leaf, ei, refs);
1655 __run_delayed_extent_op(extent_op, leaf, ei);
1657 type = btrfs_extent_inline_ref_type(leaf, iref);
1659 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1660 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1661 refs = btrfs_extent_data_ref_count(leaf, dref);
1662 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1663 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1664 refs = btrfs_shared_data_ref_count(leaf, sref);
1667 BUG_ON(refs_to_mod != -1);
1670 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1671 refs += refs_to_mod;
1674 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1675 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1677 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1679 size = btrfs_extent_inline_ref_size(type);
1680 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1681 ptr = (unsigned long)iref;
1682 end = (unsigned long)ei + item_size;
1683 if (ptr + size < end)
1684 memmove_extent_buffer(leaf, ptr, ptr + size,
1687 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1689 btrfs_mark_buffer_dirty(leaf);
1693 static noinline_for_stack
1694 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct btrfs_path *path,
1697 u64 bytenr, u64 num_bytes, u64 parent,
1698 u64 root_objectid, u64 owner,
1699 u64 offset, int refs_to_add,
1700 struct btrfs_delayed_extent_op *extent_op)
1702 struct btrfs_extent_inline_ref *iref;
1705 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1706 bytenr, num_bytes, parent,
1707 root_objectid, owner, offset, 1);
1709 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1710 ret = update_inline_extent_backref(trans, root, path, iref,
1711 refs_to_add, extent_op);
1712 } else if (ret == -ENOENT) {
1713 ret = setup_inline_extent_backref(trans, root, path, iref,
1714 parent, root_objectid,
1715 owner, offset, refs_to_add,
1721 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1722 struct btrfs_root *root,
1723 struct btrfs_path *path,
1724 u64 bytenr, u64 parent, u64 root_objectid,
1725 u64 owner, u64 offset, int refs_to_add)
1728 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1729 BUG_ON(refs_to_add != 1);
1730 ret = insert_tree_block_ref(trans, root, path, bytenr,
1731 parent, root_objectid);
1733 ret = insert_extent_data_ref(trans, root, path, bytenr,
1734 parent, root_objectid,
1735 owner, offset, refs_to_add);
1740 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 struct btrfs_extent_inline_ref *iref,
1744 int refs_to_drop, int is_data)
1748 BUG_ON(!is_data && refs_to_drop != 1);
1750 ret = update_inline_extent_backref(trans, root, path, iref,
1751 -refs_to_drop, NULL);
1752 } else if (is_data) {
1753 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1755 ret = btrfs_del_item(trans, root, path);
1760 static int btrfs_issue_discard(struct block_device *bdev,
1763 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1766 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1767 u64 num_bytes, u64 *actual_bytes)
1770 u64 discarded_bytes = 0;
1771 struct btrfs_multi_bio *multi = NULL;
1774 /* Tell the block device(s) that the sectors can be discarded */
1775 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1776 bytenr, &num_bytes, &multi, 0);
1778 struct btrfs_bio_stripe *stripe = multi->stripes;
1782 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1783 ret = btrfs_issue_discard(stripe->dev->bdev,
1787 discarded_bytes += stripe->length;
1788 else if (ret != -EOPNOTSUPP)
1793 if (discarded_bytes && ret == -EOPNOTSUPP)
1797 *actual_bytes = discarded_bytes;
1803 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1804 struct btrfs_root *root,
1805 u64 bytenr, u64 num_bytes, u64 parent,
1806 u64 root_objectid, u64 owner, u64 offset)
1809 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1810 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1812 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1813 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1814 parent, root_objectid, (int)owner,
1815 BTRFS_ADD_DELAYED_REF, NULL);
1817 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1818 parent, root_objectid, owner, offset,
1819 BTRFS_ADD_DELAYED_REF, NULL);
1824 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 u64 bytenr, u64 num_bytes,
1827 u64 parent, u64 root_objectid,
1828 u64 owner, u64 offset, int refs_to_add,
1829 struct btrfs_delayed_extent_op *extent_op)
1831 struct btrfs_path *path;
1832 struct extent_buffer *leaf;
1833 struct btrfs_extent_item *item;
1838 path = btrfs_alloc_path();
1843 path->leave_spinning = 1;
1844 /* this will setup the path even if it fails to insert the back ref */
1845 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1846 path, bytenr, num_bytes, parent,
1847 root_objectid, owner, offset,
1848 refs_to_add, extent_op);
1852 if (ret != -EAGAIN) {
1857 leaf = path->nodes[0];
1858 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1859 refs = btrfs_extent_refs(leaf, item);
1860 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1862 __run_delayed_extent_op(extent_op, leaf, item);
1864 btrfs_mark_buffer_dirty(leaf);
1865 btrfs_release_path(path);
1868 path->leave_spinning = 1;
1870 /* now insert the actual backref */
1871 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1872 path, bytenr, parent, root_objectid,
1873 owner, offset, refs_to_add);
1876 btrfs_free_path(path);
1880 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *root,
1882 struct btrfs_delayed_ref_node *node,
1883 struct btrfs_delayed_extent_op *extent_op,
1884 int insert_reserved)
1887 struct btrfs_delayed_data_ref *ref;
1888 struct btrfs_key ins;
1893 ins.objectid = node->bytenr;
1894 ins.offset = node->num_bytes;
1895 ins.type = BTRFS_EXTENT_ITEM_KEY;
1897 ref = btrfs_delayed_node_to_data_ref(node);
1898 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1899 parent = ref->parent;
1901 ref_root = ref->root;
1903 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1905 BUG_ON(extent_op->update_key);
1906 flags |= extent_op->flags_to_set;
1908 ret = alloc_reserved_file_extent(trans, root,
1909 parent, ref_root, flags,
1910 ref->objectid, ref->offset,
1911 &ins, node->ref_mod);
1912 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1913 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1914 node->num_bytes, parent,
1915 ref_root, ref->objectid,
1916 ref->offset, node->ref_mod,
1918 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1919 ret = __btrfs_free_extent(trans, root, node->bytenr,
1920 node->num_bytes, parent,
1921 ref_root, ref->objectid,
1922 ref->offset, node->ref_mod,
1930 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1931 struct extent_buffer *leaf,
1932 struct btrfs_extent_item *ei)
1934 u64 flags = btrfs_extent_flags(leaf, ei);
1935 if (extent_op->update_flags) {
1936 flags |= extent_op->flags_to_set;
1937 btrfs_set_extent_flags(leaf, ei, flags);
1940 if (extent_op->update_key) {
1941 struct btrfs_tree_block_info *bi;
1942 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1943 bi = (struct btrfs_tree_block_info *)(ei + 1);
1944 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1948 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct btrfs_delayed_ref_node *node,
1951 struct btrfs_delayed_extent_op *extent_op)
1953 struct btrfs_key key;
1954 struct btrfs_path *path;
1955 struct btrfs_extent_item *ei;
1956 struct extent_buffer *leaf;
1961 path = btrfs_alloc_path();
1965 key.objectid = node->bytenr;
1966 key.type = BTRFS_EXTENT_ITEM_KEY;
1967 key.offset = node->num_bytes;
1970 path->leave_spinning = 1;
1971 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1982 leaf = path->nodes[0];
1983 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1985 if (item_size < sizeof(*ei)) {
1986 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1992 leaf = path->nodes[0];
1993 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1996 BUG_ON(item_size < sizeof(*ei));
1997 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998 __run_delayed_extent_op(extent_op, leaf, ei);
2000 btrfs_mark_buffer_dirty(leaf);
2002 btrfs_free_path(path);
2006 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2007 struct btrfs_root *root,
2008 struct btrfs_delayed_ref_node *node,
2009 struct btrfs_delayed_extent_op *extent_op,
2010 int insert_reserved)
2013 struct btrfs_delayed_tree_ref *ref;
2014 struct btrfs_key ins;
2018 ins.objectid = node->bytenr;
2019 ins.offset = node->num_bytes;
2020 ins.type = BTRFS_EXTENT_ITEM_KEY;
2022 ref = btrfs_delayed_node_to_tree_ref(node);
2023 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2024 parent = ref->parent;
2026 ref_root = ref->root;
2028 BUG_ON(node->ref_mod != 1);
2029 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2030 BUG_ON(!extent_op || !extent_op->update_flags ||
2031 !extent_op->update_key);
2032 ret = alloc_reserved_tree_block(trans, root,
2034 extent_op->flags_to_set,
2037 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2038 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2039 node->num_bytes, parent, ref_root,
2040 ref->level, 0, 1, extent_op);
2041 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2042 ret = __btrfs_free_extent(trans, root, node->bytenr,
2043 node->num_bytes, parent, ref_root,
2044 ref->level, 0, 1, extent_op);
2051 /* helper function to actually process a single delayed ref entry */
2052 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2053 struct btrfs_root *root,
2054 struct btrfs_delayed_ref_node *node,
2055 struct btrfs_delayed_extent_op *extent_op,
2056 int insert_reserved)
2059 if (btrfs_delayed_ref_is_head(node)) {
2060 struct btrfs_delayed_ref_head *head;
2062 * we've hit the end of the chain and we were supposed
2063 * to insert this extent into the tree. But, it got
2064 * deleted before we ever needed to insert it, so all
2065 * we have to do is clean up the accounting
2068 head = btrfs_delayed_node_to_head(node);
2069 if (insert_reserved) {
2070 btrfs_pin_extent(root, node->bytenr,
2071 node->num_bytes, 1);
2072 if (head->is_data) {
2073 ret = btrfs_del_csums(trans, root,
2079 mutex_unlock(&head->mutex);
2083 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2084 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2085 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2087 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2088 node->type == BTRFS_SHARED_DATA_REF_KEY)
2089 ret = run_delayed_data_ref(trans, root, node, extent_op,
2096 static noinline struct btrfs_delayed_ref_node *
2097 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2099 struct rb_node *node;
2100 struct btrfs_delayed_ref_node *ref;
2101 int action = BTRFS_ADD_DELAYED_REF;
2104 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2105 * this prevents ref count from going down to zero when
2106 * there still are pending delayed ref.
2108 node = rb_prev(&head->node.rb_node);
2112 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2114 if (ref->bytenr != head->node.bytenr)
2116 if (ref->action == action)
2118 node = rb_prev(node);
2120 if (action == BTRFS_ADD_DELAYED_REF) {
2121 action = BTRFS_DROP_DELAYED_REF;
2127 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2128 struct btrfs_root *root,
2129 struct list_head *cluster)
2131 struct btrfs_delayed_ref_root *delayed_refs;
2132 struct btrfs_delayed_ref_node *ref;
2133 struct btrfs_delayed_ref_head *locked_ref = NULL;
2134 struct btrfs_delayed_extent_op *extent_op;
2137 int must_insert_reserved = 0;
2139 delayed_refs = &trans->transaction->delayed_refs;
2142 /* pick a new head ref from the cluster list */
2143 if (list_empty(cluster))
2146 locked_ref = list_entry(cluster->next,
2147 struct btrfs_delayed_ref_head, cluster);
2149 /* grab the lock that says we are going to process
2150 * all the refs for this head */
2151 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2154 * we may have dropped the spin lock to get the head
2155 * mutex lock, and that might have given someone else
2156 * time to free the head. If that's true, it has been
2157 * removed from our list and we can move on.
2159 if (ret == -EAGAIN) {
2167 * record the must insert reserved flag before we
2168 * drop the spin lock.
2170 must_insert_reserved = locked_ref->must_insert_reserved;
2171 locked_ref->must_insert_reserved = 0;
2173 extent_op = locked_ref->extent_op;
2174 locked_ref->extent_op = NULL;
2177 * locked_ref is the head node, so we have to go one
2178 * node back for any delayed ref updates
2180 ref = select_delayed_ref(locked_ref);
2182 /* All delayed refs have been processed, Go ahead
2183 * and send the head node to run_one_delayed_ref,
2184 * so that any accounting fixes can happen
2186 ref = &locked_ref->node;
2188 if (extent_op && must_insert_reserved) {
2194 spin_unlock(&delayed_refs->lock);
2196 ret = run_delayed_extent_op(trans, root,
2202 spin_lock(&delayed_refs->lock);
2206 list_del_init(&locked_ref->cluster);
2211 rb_erase(&ref->rb_node, &delayed_refs->root);
2212 delayed_refs->num_entries--;
2214 spin_unlock(&delayed_refs->lock);
2216 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2217 must_insert_reserved);
2220 btrfs_put_delayed_ref(ref);
2225 spin_lock(&delayed_refs->lock);
2231 * this starts processing the delayed reference count updates and
2232 * extent insertions we have queued up so far. count can be
2233 * 0, which means to process everything in the tree at the start
2234 * of the run (but not newly added entries), or it can be some target
2235 * number you'd like to process.
2237 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2238 struct btrfs_root *root, unsigned long count)
2240 struct rb_node *node;
2241 struct btrfs_delayed_ref_root *delayed_refs;
2242 struct btrfs_delayed_ref_node *ref;
2243 struct list_head cluster;
2245 int run_all = count == (unsigned long)-1;
2248 if (root == root->fs_info->extent_root)
2249 root = root->fs_info->tree_root;
2251 delayed_refs = &trans->transaction->delayed_refs;
2252 INIT_LIST_HEAD(&cluster);
2254 spin_lock(&delayed_refs->lock);
2256 count = delayed_refs->num_entries * 2;
2260 if (!(run_all || run_most) &&
2261 delayed_refs->num_heads_ready < 64)
2265 * go find something we can process in the rbtree. We start at
2266 * the beginning of the tree, and then build a cluster
2267 * of refs to process starting at the first one we are able to
2270 ret = btrfs_find_ref_cluster(trans, &cluster,
2271 delayed_refs->run_delayed_start);
2275 ret = run_clustered_refs(trans, root, &cluster);
2278 count -= min_t(unsigned long, ret, count);
2285 node = rb_first(&delayed_refs->root);
2288 count = (unsigned long)-1;
2291 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2293 if (btrfs_delayed_ref_is_head(ref)) {
2294 struct btrfs_delayed_ref_head *head;
2296 head = btrfs_delayed_node_to_head(ref);
2297 atomic_inc(&ref->refs);
2299 spin_unlock(&delayed_refs->lock);
2301 * Mutex was contended, block until it's
2302 * released and try again
2304 mutex_lock(&head->mutex);
2305 mutex_unlock(&head->mutex);
2307 btrfs_put_delayed_ref(ref);
2311 node = rb_next(node);
2313 spin_unlock(&delayed_refs->lock);
2314 schedule_timeout(1);
2318 spin_unlock(&delayed_refs->lock);
2322 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root,
2324 u64 bytenr, u64 num_bytes, u64 flags,
2327 struct btrfs_delayed_extent_op *extent_op;
2330 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2334 extent_op->flags_to_set = flags;
2335 extent_op->update_flags = 1;
2336 extent_op->update_key = 0;
2337 extent_op->is_data = is_data ? 1 : 0;
2339 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2345 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2346 struct btrfs_root *root,
2347 struct btrfs_path *path,
2348 u64 objectid, u64 offset, u64 bytenr)
2350 struct btrfs_delayed_ref_head *head;
2351 struct btrfs_delayed_ref_node *ref;
2352 struct btrfs_delayed_data_ref *data_ref;
2353 struct btrfs_delayed_ref_root *delayed_refs;
2354 struct rb_node *node;
2358 delayed_refs = &trans->transaction->delayed_refs;
2359 spin_lock(&delayed_refs->lock);
2360 head = btrfs_find_delayed_ref_head(trans, bytenr);
2364 if (!mutex_trylock(&head->mutex)) {
2365 atomic_inc(&head->node.refs);
2366 spin_unlock(&delayed_refs->lock);
2368 btrfs_release_path(path);
2371 * Mutex was contended, block until it's released and let
2374 mutex_lock(&head->mutex);
2375 mutex_unlock(&head->mutex);
2376 btrfs_put_delayed_ref(&head->node);
2380 node = rb_prev(&head->node.rb_node);
2384 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2386 if (ref->bytenr != bytenr)
2390 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2393 data_ref = btrfs_delayed_node_to_data_ref(ref);
2395 node = rb_prev(node);
2397 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2398 if (ref->bytenr == bytenr)
2402 if (data_ref->root != root->root_key.objectid ||
2403 data_ref->objectid != objectid || data_ref->offset != offset)
2408 mutex_unlock(&head->mutex);
2410 spin_unlock(&delayed_refs->lock);
2414 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2415 struct btrfs_root *root,
2416 struct btrfs_path *path,
2417 u64 objectid, u64 offset, u64 bytenr)
2419 struct btrfs_root *extent_root = root->fs_info->extent_root;
2420 struct extent_buffer *leaf;
2421 struct btrfs_extent_data_ref *ref;
2422 struct btrfs_extent_inline_ref *iref;
2423 struct btrfs_extent_item *ei;
2424 struct btrfs_key key;
2428 key.objectid = bytenr;
2429 key.offset = (u64)-1;
2430 key.type = BTRFS_EXTENT_ITEM_KEY;
2432 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2438 if (path->slots[0] == 0)
2442 leaf = path->nodes[0];
2443 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2445 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2449 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2450 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2451 if (item_size < sizeof(*ei)) {
2452 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2456 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2458 if (item_size != sizeof(*ei) +
2459 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2462 if (btrfs_extent_generation(leaf, ei) <=
2463 btrfs_root_last_snapshot(&root->root_item))
2466 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2467 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2468 BTRFS_EXTENT_DATA_REF_KEY)
2471 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2472 if (btrfs_extent_refs(leaf, ei) !=
2473 btrfs_extent_data_ref_count(leaf, ref) ||
2474 btrfs_extent_data_ref_root(leaf, ref) !=
2475 root->root_key.objectid ||
2476 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2477 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2485 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2486 struct btrfs_root *root,
2487 u64 objectid, u64 offset, u64 bytenr)
2489 struct btrfs_path *path;
2493 path = btrfs_alloc_path();
2498 ret = check_committed_ref(trans, root, path, objectid,
2500 if (ret && ret != -ENOENT)
2503 ret2 = check_delayed_ref(trans, root, path, objectid,
2505 } while (ret2 == -EAGAIN);
2507 if (ret2 && ret2 != -ENOENT) {
2512 if (ret != -ENOENT || ret2 != -ENOENT)
2515 btrfs_free_path(path);
2516 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2521 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2522 struct btrfs_root *root,
2523 struct extent_buffer *buf,
2524 int full_backref, int inc)
2531 struct btrfs_key key;
2532 struct btrfs_file_extent_item *fi;
2536 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2537 u64, u64, u64, u64, u64, u64);
2539 ref_root = btrfs_header_owner(buf);
2540 nritems = btrfs_header_nritems(buf);
2541 level = btrfs_header_level(buf);
2543 if (!root->ref_cows && level == 0)
2547 process_func = btrfs_inc_extent_ref;
2549 process_func = btrfs_free_extent;
2552 parent = buf->start;
2556 for (i = 0; i < nritems; i++) {
2558 btrfs_item_key_to_cpu(buf, &key, i);
2559 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2561 fi = btrfs_item_ptr(buf, i,
2562 struct btrfs_file_extent_item);
2563 if (btrfs_file_extent_type(buf, fi) ==
2564 BTRFS_FILE_EXTENT_INLINE)
2566 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2570 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2571 key.offset -= btrfs_file_extent_offset(buf, fi);
2572 ret = process_func(trans, root, bytenr, num_bytes,
2573 parent, ref_root, key.objectid,
2578 bytenr = btrfs_node_blockptr(buf, i);
2579 num_bytes = btrfs_level_size(root, level - 1);
2580 ret = process_func(trans, root, bytenr, num_bytes,
2581 parent, ref_root, level - 1, 0);
2592 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2593 struct extent_buffer *buf, int full_backref)
2595 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2598 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2599 struct extent_buffer *buf, int full_backref)
2601 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2604 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *root,
2606 struct btrfs_path *path,
2607 struct btrfs_block_group_cache *cache)
2610 struct btrfs_root *extent_root = root->fs_info->extent_root;
2612 struct extent_buffer *leaf;
2614 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2619 leaf = path->nodes[0];
2620 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2621 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2622 btrfs_mark_buffer_dirty(leaf);
2623 btrfs_release_path(path);
2631 static struct btrfs_block_group_cache *
2632 next_block_group(struct btrfs_root *root,
2633 struct btrfs_block_group_cache *cache)
2635 struct rb_node *node;
2636 spin_lock(&root->fs_info->block_group_cache_lock);
2637 node = rb_next(&cache->cache_node);
2638 btrfs_put_block_group(cache);
2640 cache = rb_entry(node, struct btrfs_block_group_cache,
2642 btrfs_get_block_group(cache);
2645 spin_unlock(&root->fs_info->block_group_cache_lock);
2649 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2650 struct btrfs_trans_handle *trans,
2651 struct btrfs_path *path)
2653 struct btrfs_root *root = block_group->fs_info->tree_root;
2654 struct inode *inode = NULL;
2656 int dcs = BTRFS_DC_ERROR;
2662 * If this block group is smaller than 100 megs don't bother caching the
2665 if (block_group->key.offset < (100 * 1024 * 1024)) {
2666 spin_lock(&block_group->lock);
2667 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2668 spin_unlock(&block_group->lock);
2673 inode = lookup_free_space_inode(root, block_group, path);
2674 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2675 ret = PTR_ERR(inode);
2676 btrfs_release_path(path);
2680 if (IS_ERR(inode)) {
2684 if (block_group->ro)
2687 ret = create_free_space_inode(root, trans, block_group, path);
2694 * We want to set the generation to 0, that way if anything goes wrong
2695 * from here on out we know not to trust this cache when we load up next
2698 BTRFS_I(inode)->generation = 0;
2699 ret = btrfs_update_inode(trans, root, inode);
2702 if (i_size_read(inode) > 0) {
2703 ret = btrfs_truncate_free_space_cache(root, trans, path,
2709 spin_lock(&block_group->lock);
2710 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2711 /* We're not cached, don't bother trying to write stuff out */
2712 dcs = BTRFS_DC_WRITTEN;
2713 spin_unlock(&block_group->lock);
2716 spin_unlock(&block_group->lock);
2718 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2723 * Just to make absolutely sure we have enough space, we're going to
2724 * preallocate 12 pages worth of space for each block group. In
2725 * practice we ought to use at most 8, but we need extra space so we can
2726 * add our header and have a terminator between the extents and the
2730 num_pages *= PAGE_CACHE_SIZE;
2732 ret = btrfs_check_data_free_space(inode, num_pages);
2736 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2737 num_pages, num_pages,
2740 dcs = BTRFS_DC_SETUP;
2741 btrfs_free_reserved_data_space(inode, num_pages);
2745 btrfs_release_path(path);
2747 spin_lock(&block_group->lock);
2748 block_group->disk_cache_state = dcs;
2749 spin_unlock(&block_group->lock);
2754 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2755 struct btrfs_root *root)
2757 struct btrfs_block_group_cache *cache;
2759 struct btrfs_path *path;
2762 path = btrfs_alloc_path();
2768 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2770 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2772 cache = next_block_group(root, cache);
2780 err = cache_save_setup(cache, trans, path);
2781 last = cache->key.objectid + cache->key.offset;
2782 btrfs_put_block_group(cache);
2787 err = btrfs_run_delayed_refs(trans, root,
2792 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2794 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2795 btrfs_put_block_group(cache);
2801 cache = next_block_group(root, cache);
2810 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2811 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2813 last = cache->key.objectid + cache->key.offset;
2815 err = write_one_cache_group(trans, root, path, cache);
2817 btrfs_put_block_group(cache);
2822 * I don't think this is needed since we're just marking our
2823 * preallocated extent as written, but just in case it can't
2827 err = btrfs_run_delayed_refs(trans, root,
2832 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2835 * Really this shouldn't happen, but it could if we
2836 * couldn't write the entire preallocated extent and
2837 * splitting the extent resulted in a new block.
2840 btrfs_put_block_group(cache);
2843 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2845 cache = next_block_group(root, cache);
2854 btrfs_write_out_cache(root, trans, cache, path);
2857 * If we didn't have an error then the cache state is still
2858 * NEED_WRITE, so we can set it to WRITTEN.
2860 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2861 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2862 last = cache->key.objectid + cache->key.offset;
2863 btrfs_put_block_group(cache);
2866 btrfs_free_path(path);
2870 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2872 struct btrfs_block_group_cache *block_group;
2875 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2876 if (!block_group || block_group->ro)
2879 btrfs_put_block_group(block_group);
2883 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2884 u64 total_bytes, u64 bytes_used,
2885 struct btrfs_space_info **space_info)
2887 struct btrfs_space_info *found;
2891 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2892 BTRFS_BLOCK_GROUP_RAID10))
2897 found = __find_space_info(info, flags);
2899 spin_lock(&found->lock);
2900 found->total_bytes += total_bytes;
2901 found->disk_total += total_bytes * factor;
2902 found->bytes_used += bytes_used;
2903 found->disk_used += bytes_used * factor;
2905 spin_unlock(&found->lock);
2906 *space_info = found;
2909 found = kzalloc(sizeof(*found), GFP_NOFS);
2913 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2914 INIT_LIST_HEAD(&found->block_groups[i]);
2915 init_rwsem(&found->groups_sem);
2916 spin_lock_init(&found->lock);
2917 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2918 BTRFS_BLOCK_GROUP_SYSTEM |
2919 BTRFS_BLOCK_GROUP_METADATA);
2920 found->total_bytes = total_bytes;
2921 found->disk_total = total_bytes * factor;
2922 found->bytes_used = bytes_used;
2923 found->disk_used = bytes_used * factor;
2924 found->bytes_pinned = 0;
2925 found->bytes_reserved = 0;
2926 found->bytes_readonly = 0;
2927 found->bytes_may_use = 0;
2929 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2930 found->chunk_alloc = 0;
2932 init_waitqueue_head(&found->wait);
2933 *space_info = found;
2934 list_add_rcu(&found->list, &info->space_info);
2938 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2940 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2941 BTRFS_BLOCK_GROUP_RAID1 |
2942 BTRFS_BLOCK_GROUP_RAID10 |
2943 BTRFS_BLOCK_GROUP_DUP);
2945 if (flags & BTRFS_BLOCK_GROUP_DATA)
2946 fs_info->avail_data_alloc_bits |= extra_flags;
2947 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2948 fs_info->avail_metadata_alloc_bits |= extra_flags;
2949 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2950 fs_info->avail_system_alloc_bits |= extra_flags;
2954 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2957 * we add in the count of missing devices because we want
2958 * to make sure that any RAID levels on a degraded FS
2959 * continue to be honored.
2961 u64 num_devices = root->fs_info->fs_devices->rw_devices +
2962 root->fs_info->fs_devices->missing_devices;
2964 if (num_devices == 1)
2965 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2966 if (num_devices < 4)
2967 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2969 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2970 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2971 BTRFS_BLOCK_GROUP_RAID10))) {
2972 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2975 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2976 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2977 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2980 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2981 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2982 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2983 (flags & BTRFS_BLOCK_GROUP_DUP)))
2984 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2988 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2990 if (flags & BTRFS_BLOCK_GROUP_DATA)
2991 flags |= root->fs_info->avail_data_alloc_bits &
2992 root->fs_info->data_alloc_profile;
2993 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2994 flags |= root->fs_info->avail_system_alloc_bits &
2995 root->fs_info->system_alloc_profile;
2996 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2997 flags |= root->fs_info->avail_metadata_alloc_bits &
2998 root->fs_info->metadata_alloc_profile;
2999 return btrfs_reduce_alloc_profile(root, flags);
3002 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3007 flags = BTRFS_BLOCK_GROUP_DATA;
3008 else if (root == root->fs_info->chunk_root)
3009 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3011 flags = BTRFS_BLOCK_GROUP_METADATA;
3013 return get_alloc_profile(root, flags);
3016 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3018 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3019 BTRFS_BLOCK_GROUP_DATA);
3023 * This will check the space that the inode allocates from to make sure we have
3024 * enough space for bytes.
3026 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3028 struct btrfs_space_info *data_sinfo;
3029 struct btrfs_root *root = BTRFS_I(inode)->root;
3031 int ret = 0, committed = 0, alloc_chunk = 1;
3033 /* make sure bytes are sectorsize aligned */
3034 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3036 if (root == root->fs_info->tree_root ||
3037 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3042 data_sinfo = BTRFS_I(inode)->space_info;
3047 /* make sure we have enough space to handle the data first */
3048 spin_lock(&data_sinfo->lock);
3049 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3050 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3051 data_sinfo->bytes_may_use;
3053 if (used + bytes > data_sinfo->total_bytes) {
3054 struct btrfs_trans_handle *trans;
3057 * if we don't have enough free bytes in this space then we need
3058 * to alloc a new chunk.
3060 if (!data_sinfo->full && alloc_chunk) {
3063 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3064 spin_unlock(&data_sinfo->lock);
3066 alloc_target = btrfs_get_alloc_profile(root, 1);
3067 trans = btrfs_join_transaction(root);
3069 return PTR_ERR(trans);
3071 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3072 bytes + 2 * 1024 * 1024,
3074 CHUNK_ALLOC_NO_FORCE);
3075 btrfs_end_transaction(trans, root);
3084 btrfs_set_inode_space_info(root, inode);
3085 data_sinfo = BTRFS_I(inode)->space_info;
3091 * If we have less pinned bytes than we want to allocate then
3092 * don't bother committing the transaction, it won't help us.
3094 if (data_sinfo->bytes_pinned < bytes)
3096 spin_unlock(&data_sinfo->lock);
3098 /* commit the current transaction and try again */
3101 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3103 trans = btrfs_join_transaction(root);
3105 return PTR_ERR(trans);
3106 ret = btrfs_commit_transaction(trans, root);
3114 data_sinfo->bytes_may_use += bytes;
3115 BTRFS_I(inode)->reserved_bytes += bytes;
3116 spin_unlock(&data_sinfo->lock);
3122 * called when we are clearing an delalloc extent from the
3123 * inode's io_tree or there was an error for whatever reason
3124 * after calling btrfs_check_data_free_space
3126 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3128 struct btrfs_root *root = BTRFS_I(inode)->root;
3129 struct btrfs_space_info *data_sinfo;
3131 /* make sure bytes are sectorsize aligned */
3132 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3134 data_sinfo = BTRFS_I(inode)->space_info;
3135 spin_lock(&data_sinfo->lock);
3136 data_sinfo->bytes_may_use -= bytes;
3137 BTRFS_I(inode)->reserved_bytes -= bytes;
3138 spin_unlock(&data_sinfo->lock);
3141 static void force_metadata_allocation(struct btrfs_fs_info *info)
3143 struct list_head *head = &info->space_info;
3144 struct btrfs_space_info *found;
3147 list_for_each_entry_rcu(found, head, list) {
3148 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3149 found->force_alloc = CHUNK_ALLOC_FORCE;
3154 static int should_alloc_chunk(struct btrfs_root *root,
3155 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3158 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3159 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3162 if (force == CHUNK_ALLOC_FORCE)
3166 * in limited mode, we want to have some free space up to
3167 * about 1% of the FS size.
3169 if (force == CHUNK_ALLOC_LIMITED) {
3170 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3171 thresh = max_t(u64, 64 * 1024 * 1024,
3172 div_factor_fine(thresh, 1));
3174 if (num_bytes - num_allocated < thresh)
3179 * we have two similar checks here, one based on percentage
3180 * and once based on a hard number of 256MB. The idea
3181 * is that if we have a good amount of free
3182 * room, don't allocate a chunk. A good mount is
3183 * less than 80% utilized of the chunks we have allocated,
3184 * or more than 256MB free
3186 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3189 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3192 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3194 /* 256MB or 5% of the FS */
3195 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3197 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3202 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3203 struct btrfs_root *extent_root, u64 alloc_bytes,
3204 u64 flags, int force)
3206 struct btrfs_space_info *space_info;
3207 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3208 int wait_for_alloc = 0;
3211 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3213 space_info = __find_space_info(extent_root->fs_info, flags);
3215 ret = update_space_info(extent_root->fs_info, flags,
3219 BUG_ON(!space_info);
3222 spin_lock(&space_info->lock);
3223 if (space_info->force_alloc)
3224 force = space_info->force_alloc;
3225 if (space_info->full) {
3226 spin_unlock(&space_info->lock);
3230 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3231 spin_unlock(&space_info->lock);
3233 } else if (space_info->chunk_alloc) {
3236 space_info->chunk_alloc = 1;
3239 spin_unlock(&space_info->lock);
3241 mutex_lock(&fs_info->chunk_mutex);
3244 * The chunk_mutex is held throughout the entirety of a chunk
3245 * allocation, so once we've acquired the chunk_mutex we know that the
3246 * other guy is done and we need to recheck and see if we should
3249 if (wait_for_alloc) {
3250 mutex_unlock(&fs_info->chunk_mutex);
3256 * If we have mixed data/metadata chunks we want to make sure we keep
3257 * allocating mixed chunks instead of individual chunks.
3259 if (btrfs_mixed_space_info(space_info))
3260 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3263 * if we're doing a data chunk, go ahead and make sure that
3264 * we keep a reasonable number of metadata chunks allocated in the
3267 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3268 fs_info->data_chunk_allocations++;
3269 if (!(fs_info->data_chunk_allocations %
3270 fs_info->metadata_ratio))
3271 force_metadata_allocation(fs_info);
3274 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3275 spin_lock(&space_info->lock);
3277 space_info->full = 1;
3281 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3282 space_info->chunk_alloc = 0;
3283 spin_unlock(&space_info->lock);
3284 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3289 * shrink metadata reservation for delalloc
3291 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3292 struct btrfs_root *root, u64 to_reclaim, int sync)
3294 struct btrfs_block_rsv *block_rsv;
3295 struct btrfs_space_info *space_info;
3300 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3302 unsigned long progress;
3304 block_rsv = &root->fs_info->delalloc_block_rsv;
3305 space_info = block_rsv->space_info;
3308 reserved = space_info->bytes_reserved;
3309 progress = space_info->reservation_progress;
3315 if (root->fs_info->delalloc_bytes == 0) {
3318 btrfs_wait_ordered_extents(root, 0, 0);
3322 max_reclaim = min(reserved, to_reclaim);
3324 while (loops < 1024) {
3325 /* have the flusher threads jump in and do some IO */
3327 nr_pages = min_t(unsigned long, nr_pages,
3328 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3329 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3331 spin_lock(&space_info->lock);
3332 if (reserved > space_info->bytes_reserved)
3333 reclaimed += reserved - space_info->bytes_reserved;
3334 reserved = space_info->bytes_reserved;
3335 spin_unlock(&space_info->lock);
3339 if (reserved == 0 || reclaimed >= max_reclaim)
3342 if (trans && trans->transaction->blocked)
3345 time_left = schedule_timeout_interruptible(1);
3347 /* We were interrupted, exit */
3351 /* we've kicked the IO a few times, if anything has been freed,
3352 * exit. There is no sense in looping here for a long time
3353 * when we really need to commit the transaction, or there are
3354 * just too many writers without enough free space
3359 if (progress != space_info->reservation_progress)
3364 if (reclaimed >= to_reclaim && !trans)
3365 btrfs_wait_ordered_extents(root, 0, 0);
3366 return reclaimed >= to_reclaim;
3370 * Retries tells us how many times we've called reserve_metadata_bytes. The
3371 * idea is if this is the first call (retries == 0) then we will add to our
3372 * reserved count if we can't make the allocation in order to hold our place
3373 * while we go and try and free up space. That way for retries > 1 we don't try
3374 * and add space, we just check to see if the amount of unused space is >= the
3375 * total space, meaning that our reservation is valid.
3377 * However if we don't intend to retry this reservation, pass -1 as retries so
3378 * that it short circuits this logic.
3380 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3381 struct btrfs_root *root,
3382 struct btrfs_block_rsv *block_rsv,
3383 u64 orig_bytes, int flush)
3385 struct btrfs_space_info *space_info = block_rsv->space_info;
3387 u64 num_bytes = orig_bytes;
3390 bool committed = false;
3391 bool flushing = false;
3395 spin_lock(&space_info->lock);
3397 * We only want to wait if somebody other than us is flushing and we are
3398 * actually alloed to flush.
3400 while (flush && !flushing && space_info->flush) {
3401 spin_unlock(&space_info->lock);
3403 * If we have a trans handle we can't wait because the flusher
3404 * may have to commit the transaction, which would mean we would
3405 * deadlock since we are waiting for the flusher to finish, but
3406 * hold the current transaction open.
3410 ret = wait_event_interruptible(space_info->wait,
3411 !space_info->flush);
3412 /* Must have been interrupted, return */
3416 spin_lock(&space_info->lock);
3420 unused = space_info->bytes_used + space_info->bytes_reserved +
3421 space_info->bytes_pinned + space_info->bytes_readonly +
3422 space_info->bytes_may_use;
3425 * The idea here is that we've not already over-reserved the block group
3426 * then we can go ahead and save our reservation first and then start
3427 * flushing if we need to. Otherwise if we've already overcommitted
3428 * lets start flushing stuff first and then come back and try to make
3431 if (unused <= space_info->total_bytes) {
3432 unused = space_info->total_bytes - unused;
3433 if (unused >= num_bytes) {
3434 space_info->bytes_reserved += orig_bytes;
3438 * Ok set num_bytes to orig_bytes since we aren't
3439 * overocmmitted, this way we only try and reclaim what
3442 num_bytes = orig_bytes;
3446 * Ok we're over committed, set num_bytes to the overcommitted
3447 * amount plus the amount of bytes that we need for this
3450 num_bytes = unused - space_info->total_bytes +
3451 (orig_bytes * (retries + 1));
3455 * Couldn't make our reservation, save our place so while we're trying
3456 * to reclaim space we can actually use it instead of somebody else
3457 * stealing it from us.
3461 space_info->flush = 1;
3464 spin_unlock(&space_info->lock);
3470 * We do synchronous shrinking since we don't actually unreserve
3471 * metadata until after the IO is completed.
3473 ret = shrink_delalloc(trans, root, num_bytes, 1);
3480 * So if we were overcommitted it's possible that somebody else flushed
3481 * out enough space and we simply didn't have enough space to reclaim,
3482 * so go back around and try again.
3490 * Not enough space to be reclaimed, don't bother committing the
3493 spin_lock(&space_info->lock);
3494 if (space_info->bytes_pinned < orig_bytes)
3496 spin_unlock(&space_info->lock);
3508 trans = btrfs_join_transaction(root);
3511 ret = btrfs_commit_transaction(trans, root);
3520 spin_lock(&space_info->lock);
3521 space_info->flush = 0;
3522 wake_up_all(&space_info->wait);
3523 spin_unlock(&space_info->lock);
3528 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3529 struct btrfs_root *root)
3531 struct btrfs_block_rsv *block_rsv;
3533 block_rsv = trans->block_rsv;
3535 block_rsv = root->block_rsv;
3538 block_rsv = &root->fs_info->empty_block_rsv;
3543 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3547 spin_lock(&block_rsv->lock);
3548 if (block_rsv->reserved >= num_bytes) {
3549 block_rsv->reserved -= num_bytes;
3550 if (block_rsv->reserved < block_rsv->size)
3551 block_rsv->full = 0;
3554 spin_unlock(&block_rsv->lock);
3558 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3559 u64 num_bytes, int update_size)
3561 spin_lock(&block_rsv->lock);
3562 block_rsv->reserved += num_bytes;
3564 block_rsv->size += num_bytes;
3565 else if (block_rsv->reserved >= block_rsv->size)
3566 block_rsv->full = 1;
3567 spin_unlock(&block_rsv->lock);
3570 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3571 struct btrfs_block_rsv *dest, u64 num_bytes)
3573 struct btrfs_space_info *space_info = block_rsv->space_info;
3575 spin_lock(&block_rsv->lock);
3576 if (num_bytes == (u64)-1)
3577 num_bytes = block_rsv->size;
3578 block_rsv->size -= num_bytes;
3579 if (block_rsv->reserved >= block_rsv->size) {
3580 num_bytes = block_rsv->reserved - block_rsv->size;
3581 block_rsv->reserved = block_rsv->size;
3582 block_rsv->full = 1;
3586 spin_unlock(&block_rsv->lock);
3588 if (num_bytes > 0) {
3590 spin_lock(&dest->lock);
3594 bytes_to_add = dest->size - dest->reserved;
3595 bytes_to_add = min(num_bytes, bytes_to_add);
3596 dest->reserved += bytes_to_add;
3597 if (dest->reserved >= dest->size)
3599 num_bytes -= bytes_to_add;
3601 spin_unlock(&dest->lock);
3604 spin_lock(&space_info->lock);
3605 space_info->bytes_reserved -= num_bytes;
3606 space_info->reservation_progress++;
3607 spin_unlock(&space_info->lock);
3612 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3613 struct btrfs_block_rsv *dst, u64 num_bytes)
3617 ret = block_rsv_use_bytes(src, num_bytes);
3621 block_rsv_add_bytes(dst, num_bytes, 1);
3625 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3627 memset(rsv, 0, sizeof(*rsv));
3628 spin_lock_init(&rsv->lock);
3629 atomic_set(&rsv->usage, 1);
3631 INIT_LIST_HEAD(&rsv->list);
3634 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3636 struct btrfs_block_rsv *block_rsv;
3637 struct btrfs_fs_info *fs_info = root->fs_info;
3639 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3643 btrfs_init_block_rsv(block_rsv);
3644 block_rsv->space_info = __find_space_info(fs_info,
3645 BTRFS_BLOCK_GROUP_METADATA);
3649 void btrfs_free_block_rsv(struct btrfs_root *root,
3650 struct btrfs_block_rsv *rsv)
3652 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3653 btrfs_block_rsv_release(root, rsv, (u64)-1);
3660 * make the block_rsv struct be able to capture freed space.
3661 * the captured space will re-add to the the block_rsv struct
3662 * after transaction commit
3664 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3665 struct btrfs_block_rsv *block_rsv)
3667 block_rsv->durable = 1;
3668 mutex_lock(&fs_info->durable_block_rsv_mutex);
3669 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3670 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3673 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3674 struct btrfs_root *root,
3675 struct btrfs_block_rsv *block_rsv,
3683 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3685 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3692 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3693 struct btrfs_root *root,
3694 struct btrfs_block_rsv *block_rsv,
3695 u64 min_reserved, int min_factor)
3698 int commit_trans = 0;
3704 spin_lock(&block_rsv->lock);
3706 num_bytes = div_factor(block_rsv->size, min_factor);
3707 if (min_reserved > num_bytes)
3708 num_bytes = min_reserved;
3710 if (block_rsv->reserved >= num_bytes) {
3713 num_bytes -= block_rsv->reserved;
3714 if (block_rsv->durable &&
3715 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3718 spin_unlock(&block_rsv->lock);
3722 if (block_rsv->refill_used) {
3723 ret = reserve_metadata_bytes(trans, root, block_rsv,
3726 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3734 trans = btrfs_join_transaction(root);
3735 BUG_ON(IS_ERR(trans));
3736 ret = btrfs_commit_transaction(trans, root);
3743 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3744 struct btrfs_block_rsv *dst_rsv,
3747 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3750 void btrfs_block_rsv_release(struct btrfs_root *root,
3751 struct btrfs_block_rsv *block_rsv,
3754 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3755 if (global_rsv->full || global_rsv == block_rsv ||
3756 block_rsv->space_info != global_rsv->space_info)
3758 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3762 * helper to calculate size of global block reservation.
3763 * the desired value is sum of space used by extent tree,
3764 * checksum tree and root tree
3766 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3768 struct btrfs_space_info *sinfo;
3772 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3774 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3775 spin_lock(&sinfo->lock);
3776 data_used = sinfo->bytes_used;
3777 spin_unlock(&sinfo->lock);
3779 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3780 spin_lock(&sinfo->lock);
3781 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3783 meta_used = sinfo->bytes_used;
3784 spin_unlock(&sinfo->lock);
3786 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3788 num_bytes += div64_u64(data_used + meta_used, 50);
3790 if (num_bytes * 3 > meta_used)
3791 num_bytes = div64_u64(meta_used, 3);
3793 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3796 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3798 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3799 struct btrfs_space_info *sinfo = block_rsv->space_info;
3802 num_bytes = calc_global_metadata_size(fs_info);
3804 spin_lock(&block_rsv->lock);
3805 spin_lock(&sinfo->lock);
3807 block_rsv->size = num_bytes;
3809 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3810 sinfo->bytes_reserved + sinfo->bytes_readonly +
3811 sinfo->bytes_may_use;
3813 if (sinfo->total_bytes > num_bytes) {
3814 num_bytes = sinfo->total_bytes - num_bytes;
3815 block_rsv->reserved += num_bytes;
3816 sinfo->bytes_reserved += num_bytes;
3819 if (block_rsv->reserved >= block_rsv->size) {
3820 num_bytes = block_rsv->reserved - block_rsv->size;
3821 sinfo->bytes_reserved -= num_bytes;
3822 sinfo->reservation_progress++;
3823 block_rsv->reserved = block_rsv->size;
3824 block_rsv->full = 1;
3827 spin_unlock(&sinfo->lock);
3828 spin_unlock(&block_rsv->lock);
3831 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3833 struct btrfs_space_info *space_info;
3835 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3836 fs_info->chunk_block_rsv.space_info = space_info;
3837 fs_info->chunk_block_rsv.priority = 10;
3839 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3840 fs_info->global_block_rsv.space_info = space_info;
3841 fs_info->global_block_rsv.priority = 10;
3842 fs_info->global_block_rsv.refill_used = 1;
3843 fs_info->delalloc_block_rsv.space_info = space_info;
3844 fs_info->trans_block_rsv.space_info = space_info;
3845 fs_info->empty_block_rsv.space_info = space_info;
3846 fs_info->empty_block_rsv.priority = 10;
3848 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3849 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3850 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3851 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3852 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3854 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3856 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3858 update_global_block_rsv(fs_info);
3861 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3863 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3864 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3865 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3866 WARN_ON(fs_info->trans_block_rsv.size > 0);
3867 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3868 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3869 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3872 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
3873 struct btrfs_root *root,
3874 struct btrfs_block_rsv *rsv)
3876 struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv;
3881 * Truncate should be freeing data, but give us 2 items just in case it
3882 * needs to use some space. We may want to be smarter about this in the
3885 num_bytes = btrfs_calc_trans_metadata_size(root, 2);
3887 /* We already have enough bytes, just return */
3888 if (rsv->reserved >= num_bytes)
3891 num_bytes -= rsv->reserved;
3894 * You should have reserved enough space before hand to do this, so this
3897 ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes);
3903 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3904 struct btrfs_root *root)
3906 if (!trans->bytes_reserved)
3909 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3910 btrfs_block_rsv_release(root, trans->block_rsv,
3911 trans->bytes_reserved);
3912 trans->bytes_reserved = 0;
3915 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3916 struct inode *inode)
3918 struct btrfs_root *root = BTRFS_I(inode)->root;
3919 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3920 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3923 * We need to hold space in order to delete our orphan item once we've
3924 * added it, so this takes the reservation so we can release it later
3925 * when we are truly done with the orphan item.
3927 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3928 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3931 void btrfs_orphan_release_metadata(struct inode *inode)
3933 struct btrfs_root *root = BTRFS_I(inode)->root;
3934 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3935 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3938 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3939 struct btrfs_pending_snapshot *pending)
3941 struct btrfs_root *root = pending->root;
3942 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3943 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3945 * two for root back/forward refs, two for directory entries
3946 * and one for root of the snapshot.
3948 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3949 dst_rsv->space_info = src_rsv->space_info;
3950 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3953 static unsigned drop_outstanding_extent(struct inode *inode)
3955 unsigned dropped_extents = 0;
3957 spin_lock(&BTRFS_I(inode)->lock);
3958 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3959 BTRFS_I(inode)->outstanding_extents--;
3962 * If we have more or the same amount of outsanding extents than we have
3963 * reserved then we need to leave the reserved extents count alone.
3965 if (BTRFS_I(inode)->outstanding_extents >=
3966 BTRFS_I(inode)->reserved_extents)
3969 dropped_extents = BTRFS_I(inode)->reserved_extents -
3970 BTRFS_I(inode)->outstanding_extents;
3971 BTRFS_I(inode)->reserved_extents -= dropped_extents;
3973 spin_unlock(&BTRFS_I(inode)->lock);
3974 return dropped_extents;
3977 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3979 return num_bytes >>= 3;
3982 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3984 struct btrfs_root *root = BTRFS_I(inode)->root;
3985 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3987 unsigned nr_extents = 0;
3990 if (btrfs_transaction_in_commit(root->fs_info))
3991 schedule_timeout(1);
3993 num_bytes = ALIGN(num_bytes, root->sectorsize);
3995 spin_lock(&BTRFS_I(inode)->lock);
3996 BTRFS_I(inode)->outstanding_extents++;
3998 if (BTRFS_I(inode)->outstanding_extents >
3999 BTRFS_I(inode)->reserved_extents) {
4000 nr_extents = BTRFS_I(inode)->outstanding_extents -
4001 BTRFS_I(inode)->reserved_extents;
4002 BTRFS_I(inode)->reserved_extents += nr_extents;
4004 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4006 spin_unlock(&BTRFS_I(inode)->lock);
4008 to_reserve += calc_csum_metadata_size(inode, num_bytes);
4009 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4013 * We don't need the return value since our reservation failed,
4014 * we just need to clean up our counter.
4016 dropped = drop_outstanding_extent(inode);
4017 WARN_ON(dropped > 1);
4021 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4026 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4028 struct btrfs_root *root = BTRFS_I(inode)->root;
4032 num_bytes = ALIGN(num_bytes, root->sectorsize);
4033 dropped = drop_outstanding_extent(inode);
4035 to_free = calc_csum_metadata_size(inode, num_bytes);
4037 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4039 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4043 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4047 ret = btrfs_check_data_free_space(inode, num_bytes);
4051 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4053 btrfs_free_reserved_data_space(inode, num_bytes);
4060 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4062 btrfs_delalloc_release_metadata(inode, num_bytes);
4063 btrfs_free_reserved_data_space(inode, num_bytes);
4066 static int update_block_group(struct btrfs_trans_handle *trans,
4067 struct btrfs_root *root,
4068 u64 bytenr, u64 num_bytes, int alloc)
4070 struct btrfs_block_group_cache *cache = NULL;
4071 struct btrfs_fs_info *info = root->fs_info;
4072 u64 total = num_bytes;
4077 /* block accounting for super block */
4078 spin_lock(&info->delalloc_lock);
4079 old_val = btrfs_super_bytes_used(&info->super_copy);
4081 old_val += num_bytes;
4083 old_val -= num_bytes;
4084 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4085 spin_unlock(&info->delalloc_lock);
4088 cache = btrfs_lookup_block_group(info, bytenr);
4091 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4092 BTRFS_BLOCK_GROUP_RAID1 |
4093 BTRFS_BLOCK_GROUP_RAID10))
4098 * If this block group has free space cache written out, we
4099 * need to make sure to load it if we are removing space. This
4100 * is because we need the unpinning stage to actually add the
4101 * space back to the block group, otherwise we will leak space.
4103 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4104 cache_block_group(cache, trans, NULL, 1);
4106 byte_in_group = bytenr - cache->key.objectid;
4107 WARN_ON(byte_in_group > cache->key.offset);
4109 spin_lock(&cache->space_info->lock);
4110 spin_lock(&cache->lock);
4112 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4113 cache->disk_cache_state < BTRFS_DC_CLEAR)
4114 cache->disk_cache_state = BTRFS_DC_CLEAR;
4117 old_val = btrfs_block_group_used(&cache->item);
4118 num_bytes = min(total, cache->key.offset - byte_in_group);
4120 old_val += num_bytes;
4121 btrfs_set_block_group_used(&cache->item, old_val);
4122 cache->reserved -= num_bytes;
4123 cache->space_info->bytes_reserved -= num_bytes;
4124 cache->space_info->reservation_progress++;
4125 cache->space_info->bytes_used += num_bytes;
4126 cache->space_info->disk_used += num_bytes * factor;
4127 spin_unlock(&cache->lock);
4128 spin_unlock(&cache->space_info->lock);
4130 old_val -= num_bytes;
4131 btrfs_set_block_group_used(&cache->item, old_val);
4132 cache->pinned += num_bytes;
4133 cache->space_info->bytes_pinned += num_bytes;
4134 cache->space_info->bytes_used -= num_bytes;
4135 cache->space_info->disk_used -= num_bytes * factor;
4136 spin_unlock(&cache->lock);
4137 spin_unlock(&cache->space_info->lock);
4139 set_extent_dirty(info->pinned_extents,
4140 bytenr, bytenr + num_bytes - 1,
4141 GFP_NOFS | __GFP_NOFAIL);
4143 btrfs_put_block_group(cache);
4145 bytenr += num_bytes;
4150 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4152 struct btrfs_block_group_cache *cache;
4155 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4159 bytenr = cache->key.objectid;
4160 btrfs_put_block_group(cache);
4165 static int pin_down_extent(struct btrfs_root *root,
4166 struct btrfs_block_group_cache *cache,
4167 u64 bytenr, u64 num_bytes, int reserved)
4169 spin_lock(&cache->space_info->lock);
4170 spin_lock(&cache->lock);
4171 cache->pinned += num_bytes;
4172 cache->space_info->bytes_pinned += num_bytes;
4174 cache->reserved -= num_bytes;
4175 cache->space_info->bytes_reserved -= num_bytes;
4176 cache->space_info->reservation_progress++;
4178 spin_unlock(&cache->lock);
4179 spin_unlock(&cache->space_info->lock);
4181 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4182 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4187 * this function must be called within transaction
4189 int btrfs_pin_extent(struct btrfs_root *root,
4190 u64 bytenr, u64 num_bytes, int reserved)
4192 struct btrfs_block_group_cache *cache;
4194 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4197 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4199 btrfs_put_block_group(cache);
4204 * update size of reserved extents. this function may return -EAGAIN
4205 * if 'reserve' is true or 'sinfo' is false.
4207 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4208 u64 num_bytes, int reserve, int sinfo)
4212 struct btrfs_space_info *space_info = cache->space_info;
4213 spin_lock(&space_info->lock);
4214 spin_lock(&cache->lock);
4219 cache->reserved += num_bytes;
4220 space_info->bytes_reserved += num_bytes;
4224 space_info->bytes_readonly += num_bytes;
4225 cache->reserved -= num_bytes;
4226 space_info->bytes_reserved -= num_bytes;
4227 space_info->reservation_progress++;
4229 spin_unlock(&cache->lock);
4230 spin_unlock(&space_info->lock);
4232 spin_lock(&cache->lock);
4237 cache->reserved += num_bytes;
4239 cache->reserved -= num_bytes;
4241 spin_unlock(&cache->lock);
4246 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4247 struct btrfs_root *root)
4249 struct btrfs_fs_info *fs_info = root->fs_info;
4250 struct btrfs_caching_control *next;
4251 struct btrfs_caching_control *caching_ctl;
4252 struct btrfs_block_group_cache *cache;
4254 down_write(&fs_info->extent_commit_sem);
4256 list_for_each_entry_safe(caching_ctl, next,
4257 &fs_info->caching_block_groups, list) {
4258 cache = caching_ctl->block_group;
4259 if (block_group_cache_done(cache)) {
4260 cache->last_byte_to_unpin = (u64)-1;
4261 list_del_init(&caching_ctl->list);
4262 put_caching_control(caching_ctl);
4264 cache->last_byte_to_unpin = caching_ctl->progress;
4268 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4269 fs_info->pinned_extents = &fs_info->freed_extents[1];
4271 fs_info->pinned_extents = &fs_info->freed_extents[0];
4273 up_write(&fs_info->extent_commit_sem);
4275 update_global_block_rsv(fs_info);
4279 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4281 struct btrfs_fs_info *fs_info = root->fs_info;
4282 struct btrfs_block_group_cache *cache = NULL;
4285 while (start <= end) {
4287 start >= cache->key.objectid + cache->key.offset) {
4289 btrfs_put_block_group(cache);
4290 cache = btrfs_lookup_block_group(fs_info, start);
4294 len = cache->key.objectid + cache->key.offset - start;
4295 len = min(len, end + 1 - start);
4297 if (start < cache->last_byte_to_unpin) {
4298 len = min(len, cache->last_byte_to_unpin - start);
4299 btrfs_add_free_space(cache, start, len);
4304 spin_lock(&cache->space_info->lock);
4305 spin_lock(&cache->lock);
4306 cache->pinned -= len;
4307 cache->space_info->bytes_pinned -= len;
4309 cache->space_info->bytes_readonly += len;
4310 } else if (cache->reserved_pinned > 0) {
4311 len = min(len, cache->reserved_pinned);
4312 cache->reserved_pinned -= len;
4313 cache->space_info->bytes_reserved += len;
4315 spin_unlock(&cache->lock);
4316 spin_unlock(&cache->space_info->lock);
4320 btrfs_put_block_group(cache);
4324 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4325 struct btrfs_root *root)
4327 struct btrfs_fs_info *fs_info = root->fs_info;
4328 struct extent_io_tree *unpin;
4329 struct btrfs_block_rsv *block_rsv;
4330 struct btrfs_block_rsv *next_rsv;
4336 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4337 unpin = &fs_info->freed_extents[1];
4339 unpin = &fs_info->freed_extents[0];
4342 ret = find_first_extent_bit(unpin, 0, &start, &end,
4347 if (btrfs_test_opt(root, DISCARD))
4348 ret = btrfs_discard_extent(root, start,
4349 end + 1 - start, NULL);
4351 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4352 unpin_extent_range(root, start, end);
4356 mutex_lock(&fs_info->durable_block_rsv_mutex);
4357 list_for_each_entry_safe(block_rsv, next_rsv,
4358 &fs_info->durable_block_rsv_list, list) {
4360 idx = trans->transid & 0x1;
4361 if (block_rsv->freed[idx] > 0) {
4362 block_rsv_add_bytes(block_rsv,
4363 block_rsv->freed[idx], 0);
4364 block_rsv->freed[idx] = 0;
4366 if (atomic_read(&block_rsv->usage) == 0) {
4367 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4369 if (block_rsv->freed[0] == 0 &&
4370 block_rsv->freed[1] == 0) {
4371 list_del_init(&block_rsv->list);
4375 btrfs_block_rsv_release(root, block_rsv, 0);
4378 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4383 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4384 struct btrfs_root *root,
4385 u64 bytenr, u64 num_bytes, u64 parent,
4386 u64 root_objectid, u64 owner_objectid,
4387 u64 owner_offset, int refs_to_drop,
4388 struct btrfs_delayed_extent_op *extent_op)
4390 struct btrfs_key key;
4391 struct btrfs_path *path;
4392 struct btrfs_fs_info *info = root->fs_info;
4393 struct btrfs_root *extent_root = info->extent_root;
4394 struct extent_buffer *leaf;
4395 struct btrfs_extent_item *ei;
4396 struct btrfs_extent_inline_ref *iref;
4399 int extent_slot = 0;
4400 int found_extent = 0;
4405 path = btrfs_alloc_path();
4410 path->leave_spinning = 1;
4412 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4413 BUG_ON(!is_data && refs_to_drop != 1);
4415 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4416 bytenr, num_bytes, parent,
4417 root_objectid, owner_objectid,
4420 extent_slot = path->slots[0];
4421 while (extent_slot >= 0) {
4422 btrfs_item_key_to_cpu(path->nodes[0], &key,
4424 if (key.objectid != bytenr)
4426 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4427 key.offset == num_bytes) {
4431 if (path->slots[0] - extent_slot > 5)
4435 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4436 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4437 if (found_extent && item_size < sizeof(*ei))
4440 if (!found_extent) {
4442 ret = remove_extent_backref(trans, extent_root, path,
4446 btrfs_release_path(path);
4447 path->leave_spinning = 1;
4449 key.objectid = bytenr;
4450 key.type = BTRFS_EXTENT_ITEM_KEY;
4451 key.offset = num_bytes;
4453 ret = btrfs_search_slot(trans, extent_root,
4456 printk(KERN_ERR "umm, got %d back from search"
4457 ", was looking for %llu\n", ret,
4458 (unsigned long long)bytenr);
4459 btrfs_print_leaf(extent_root, path->nodes[0]);
4462 extent_slot = path->slots[0];
4465 btrfs_print_leaf(extent_root, path->nodes[0]);
4467 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4468 "parent %llu root %llu owner %llu offset %llu\n",
4469 (unsigned long long)bytenr,
4470 (unsigned long long)parent,
4471 (unsigned long long)root_objectid,
4472 (unsigned long long)owner_objectid,
4473 (unsigned long long)owner_offset);
4476 leaf = path->nodes[0];
4477 item_size = btrfs_item_size_nr(leaf, extent_slot);
4478 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4479 if (item_size < sizeof(*ei)) {
4480 BUG_ON(found_extent || extent_slot != path->slots[0]);
4481 ret = convert_extent_item_v0(trans, extent_root, path,
4485 btrfs_release_path(path);
4486 path->leave_spinning = 1;
4488 key.objectid = bytenr;
4489 key.type = BTRFS_EXTENT_ITEM_KEY;
4490 key.offset = num_bytes;
4492 ret = btrfs_search_slot(trans, extent_root, &key, path,
4495 printk(KERN_ERR "umm, got %d back from search"
4496 ", was looking for %llu\n", ret,
4497 (unsigned long long)bytenr);
4498 btrfs_print_leaf(extent_root, path->nodes[0]);
4501 extent_slot = path->slots[0];
4502 leaf = path->nodes[0];
4503 item_size = btrfs_item_size_nr(leaf, extent_slot);
4506 BUG_ON(item_size < sizeof(*ei));
4507 ei = btrfs_item_ptr(leaf, extent_slot,
4508 struct btrfs_extent_item);
4509 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4510 struct btrfs_tree_block_info *bi;
4511 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4512 bi = (struct btrfs_tree_block_info *)(ei + 1);
4513 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4516 refs = btrfs_extent_refs(leaf, ei);
4517 BUG_ON(refs < refs_to_drop);
4518 refs -= refs_to_drop;
4522 __run_delayed_extent_op(extent_op, leaf, ei);
4524 * In the case of inline back ref, reference count will
4525 * be updated by remove_extent_backref
4528 BUG_ON(!found_extent);
4530 btrfs_set_extent_refs(leaf, ei, refs);
4531 btrfs_mark_buffer_dirty(leaf);
4534 ret = remove_extent_backref(trans, extent_root, path,
4541 BUG_ON(is_data && refs_to_drop !=
4542 extent_data_ref_count(root, path, iref));
4544 BUG_ON(path->slots[0] != extent_slot);
4546 BUG_ON(path->slots[0] != extent_slot + 1);
4547 path->slots[0] = extent_slot;
4552 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4555 btrfs_release_path(path);
4558 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4561 invalidate_mapping_pages(info->btree_inode->i_mapping,
4562 bytenr >> PAGE_CACHE_SHIFT,
4563 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4566 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4569 btrfs_free_path(path);
4574 * when we free an block, it is possible (and likely) that we free the last
4575 * delayed ref for that extent as well. This searches the delayed ref tree for
4576 * a given extent, and if there are no other delayed refs to be processed, it
4577 * removes it from the tree.
4579 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4580 struct btrfs_root *root, u64 bytenr)
4582 struct btrfs_delayed_ref_head *head;
4583 struct btrfs_delayed_ref_root *delayed_refs;
4584 struct btrfs_delayed_ref_node *ref;
4585 struct rb_node *node;
4588 delayed_refs = &trans->transaction->delayed_refs;
4589 spin_lock(&delayed_refs->lock);
4590 head = btrfs_find_delayed_ref_head(trans, bytenr);
4594 node = rb_prev(&head->node.rb_node);
4598 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4600 /* there are still entries for this ref, we can't drop it */
4601 if (ref->bytenr == bytenr)
4604 if (head->extent_op) {
4605 if (!head->must_insert_reserved)
4607 kfree(head->extent_op);
4608 head->extent_op = NULL;
4612 * waiting for the lock here would deadlock. If someone else has it
4613 * locked they are already in the process of dropping it anyway
4615 if (!mutex_trylock(&head->mutex))
4619 * at this point we have a head with no other entries. Go
4620 * ahead and process it.
4622 head->node.in_tree = 0;
4623 rb_erase(&head->node.rb_node, &delayed_refs->root);
4625 delayed_refs->num_entries--;
4628 * we don't take a ref on the node because we're removing it from the
4629 * tree, so we just steal the ref the tree was holding.
4631 delayed_refs->num_heads--;
4632 if (list_empty(&head->cluster))
4633 delayed_refs->num_heads_ready--;
4635 list_del_init(&head->cluster);
4636 spin_unlock(&delayed_refs->lock);
4638 BUG_ON(head->extent_op);
4639 if (head->must_insert_reserved)
4642 mutex_unlock(&head->mutex);
4643 btrfs_put_delayed_ref(&head->node);
4646 spin_unlock(&delayed_refs->lock);
4650 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4651 struct btrfs_root *root,
4652 struct extent_buffer *buf,
4653 u64 parent, int last_ref)
4655 struct btrfs_block_rsv *block_rsv;
4656 struct btrfs_block_group_cache *cache = NULL;
4659 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4660 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4661 parent, root->root_key.objectid,
4662 btrfs_header_level(buf),
4663 BTRFS_DROP_DELAYED_REF, NULL);
4670 block_rsv = get_block_rsv(trans, root);
4671 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4672 if (block_rsv->space_info != cache->space_info)
4675 if (btrfs_header_generation(buf) == trans->transid) {
4676 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4677 ret = check_ref_cleanup(trans, root, buf->start);
4682 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4683 pin_down_extent(root, cache, buf->start, buf->len, 1);
4687 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4689 btrfs_add_free_space(cache, buf->start, buf->len);
4690 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4691 if (ret == -EAGAIN) {
4692 /* block group became read-only */
4693 btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4698 spin_lock(&block_rsv->lock);
4699 if (block_rsv->reserved < block_rsv->size) {
4700 block_rsv->reserved += buf->len;
4703 spin_unlock(&block_rsv->lock);
4706 spin_lock(&cache->space_info->lock);
4707 cache->space_info->bytes_reserved -= buf->len;
4708 cache->space_info->reservation_progress++;
4709 spin_unlock(&cache->space_info->lock);
4714 if (block_rsv->durable && !cache->ro) {
4716 spin_lock(&cache->lock);
4718 cache->reserved_pinned += buf->len;
4721 spin_unlock(&cache->lock);
4724 spin_lock(&block_rsv->lock);
4725 block_rsv->freed[trans->transid & 0x1] += buf->len;
4726 spin_unlock(&block_rsv->lock);
4731 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4734 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4735 btrfs_put_block_group(cache);
4738 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4739 struct btrfs_root *root,
4740 u64 bytenr, u64 num_bytes, u64 parent,
4741 u64 root_objectid, u64 owner, u64 offset)
4746 * tree log blocks never actually go into the extent allocation
4747 * tree, just update pinning info and exit early.
4749 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4750 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4751 /* unlocks the pinned mutex */
4752 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4754 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4755 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4756 parent, root_objectid, (int)owner,
4757 BTRFS_DROP_DELAYED_REF, NULL);
4760 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4761 parent, root_objectid, owner,
4762 offset, BTRFS_DROP_DELAYED_REF, NULL);
4768 static u64 stripe_align(struct btrfs_root *root, u64 val)
4770 u64 mask = ((u64)root->stripesize - 1);
4771 u64 ret = (val + mask) & ~mask;
4776 * when we wait for progress in the block group caching, its because
4777 * our allocation attempt failed at least once. So, we must sleep
4778 * and let some progress happen before we try again.
4780 * This function will sleep at least once waiting for new free space to
4781 * show up, and then it will check the block group free space numbers
4782 * for our min num_bytes. Another option is to have it go ahead
4783 * and look in the rbtree for a free extent of a given size, but this
4787 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4790 struct btrfs_caching_control *caching_ctl;
4793 caching_ctl = get_caching_control(cache);
4797 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4798 (cache->free_space_ctl->free_space >= num_bytes));
4800 put_caching_control(caching_ctl);
4805 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4807 struct btrfs_caching_control *caching_ctl;
4810 caching_ctl = get_caching_control(cache);
4814 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4816 put_caching_control(caching_ctl);
4820 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4823 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4825 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4827 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4829 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4836 enum btrfs_loop_type {
4837 LOOP_FIND_IDEAL = 0,
4838 LOOP_CACHING_NOWAIT = 1,
4839 LOOP_CACHING_WAIT = 2,
4840 LOOP_ALLOC_CHUNK = 3,
4841 LOOP_NO_EMPTY_SIZE = 4,
4845 * walks the btree of allocated extents and find a hole of a given size.
4846 * The key ins is changed to record the hole:
4847 * ins->objectid == block start
4848 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4849 * ins->offset == number of blocks
4850 * Any available blocks before search_start are skipped.
4852 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4853 struct btrfs_root *orig_root,
4854 u64 num_bytes, u64 empty_size,
4855 u64 search_start, u64 search_end,
4856 u64 hint_byte, struct btrfs_key *ins,
4860 struct btrfs_root *root = orig_root->fs_info->extent_root;
4861 struct btrfs_free_cluster *last_ptr = NULL;
4862 struct btrfs_block_group_cache *block_group = NULL;
4863 int empty_cluster = 2 * 1024 * 1024;
4864 int allowed_chunk_alloc = 0;
4865 int done_chunk_alloc = 0;
4866 struct btrfs_space_info *space_info;
4867 int last_ptr_loop = 0;
4870 bool found_uncached_bg = false;
4871 bool failed_cluster_refill = false;
4872 bool failed_alloc = false;
4873 bool use_cluster = true;
4874 u64 ideal_cache_percent = 0;
4875 u64 ideal_cache_offset = 0;
4877 WARN_ON(num_bytes < root->sectorsize);
4878 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4882 space_info = __find_space_info(root->fs_info, data);
4884 printk(KERN_ERR "No space info for %llu\n", data);
4889 * If the space info is for both data and metadata it means we have a
4890 * small filesystem and we can't use the clustering stuff.
4892 if (btrfs_mixed_space_info(space_info))
4893 use_cluster = false;
4895 if (orig_root->ref_cows || empty_size)
4896 allowed_chunk_alloc = 1;
4898 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4899 last_ptr = &root->fs_info->meta_alloc_cluster;
4900 if (!btrfs_test_opt(root, SSD))
4901 empty_cluster = 64 * 1024;
4904 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4905 btrfs_test_opt(root, SSD)) {
4906 last_ptr = &root->fs_info->data_alloc_cluster;
4910 spin_lock(&last_ptr->lock);
4911 if (last_ptr->block_group)
4912 hint_byte = last_ptr->window_start;
4913 spin_unlock(&last_ptr->lock);
4916 search_start = max(search_start, first_logical_byte(root, 0));
4917 search_start = max(search_start, hint_byte);
4922 if (search_start == hint_byte) {
4924 block_group = btrfs_lookup_block_group(root->fs_info,
4927 * we don't want to use the block group if it doesn't match our
4928 * allocation bits, or if its not cached.
4930 * However if we are re-searching with an ideal block group
4931 * picked out then we don't care that the block group is cached.
4933 if (block_group && block_group_bits(block_group, data) &&
4934 (block_group->cached != BTRFS_CACHE_NO ||
4935 search_start == ideal_cache_offset)) {
4936 down_read(&space_info->groups_sem);
4937 if (list_empty(&block_group->list) ||
4940 * someone is removing this block group,
4941 * we can't jump into the have_block_group
4942 * target because our list pointers are not
4945 btrfs_put_block_group(block_group);
4946 up_read(&space_info->groups_sem);
4948 index = get_block_group_index(block_group);
4949 goto have_block_group;
4951 } else if (block_group) {
4952 btrfs_put_block_group(block_group);
4956 down_read(&space_info->groups_sem);
4957 list_for_each_entry(block_group, &space_info->block_groups[index],
4962 btrfs_get_block_group(block_group);
4963 search_start = block_group->key.objectid;
4966 * this can happen if we end up cycling through all the
4967 * raid types, but we want to make sure we only allocate
4968 * for the proper type.
4970 if (!block_group_bits(block_group, data)) {
4971 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4972 BTRFS_BLOCK_GROUP_RAID1 |
4973 BTRFS_BLOCK_GROUP_RAID10;
4976 * if they asked for extra copies and this block group
4977 * doesn't provide them, bail. This does allow us to
4978 * fill raid0 from raid1.
4980 if ((data & extra) && !(block_group->flags & extra))
4985 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4988 ret = cache_block_group(block_group, trans,
4990 if (block_group->cached == BTRFS_CACHE_FINISHED)
4991 goto have_block_group;
4993 free_percent = btrfs_block_group_used(&block_group->item);
4994 free_percent *= 100;
4995 free_percent = div64_u64(free_percent,
4996 block_group->key.offset);
4997 free_percent = 100 - free_percent;
4998 if (free_percent > ideal_cache_percent &&
4999 likely(!block_group->ro)) {
5000 ideal_cache_offset = block_group->key.objectid;
5001 ideal_cache_percent = free_percent;
5005 * The caching workers are limited to 2 threads, so we
5006 * can queue as much work as we care to.
5008 if (loop > LOOP_FIND_IDEAL) {
5009 ret = cache_block_group(block_group, trans,
5013 found_uncached_bg = true;
5016 * If loop is set for cached only, try the next block
5019 if (loop == LOOP_FIND_IDEAL)
5023 cached = block_group_cache_done(block_group);
5024 if (unlikely(!cached))
5025 found_uncached_bg = true;
5027 if (unlikely(block_group->ro))
5030 spin_lock(&block_group->free_space_ctl->tree_lock);
5032 block_group->free_space_ctl->free_space <
5033 num_bytes + empty_size) {
5034 spin_unlock(&block_group->free_space_ctl->tree_lock);
5037 spin_unlock(&block_group->free_space_ctl->tree_lock);
5040 * Ok we want to try and use the cluster allocator, so lets look
5041 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5042 * have tried the cluster allocator plenty of times at this
5043 * point and not have found anything, so we are likely way too
5044 * fragmented for the clustering stuff to find anything, so lets
5045 * just skip it and let the allocator find whatever block it can
5048 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5050 * the refill lock keeps out other
5051 * people trying to start a new cluster
5053 spin_lock(&last_ptr->refill_lock);
5054 if (last_ptr->block_group &&
5055 (last_ptr->block_group->ro ||
5056 !block_group_bits(last_ptr->block_group, data))) {
5058 goto refill_cluster;
5061 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5062 num_bytes, search_start);
5064 /* we have a block, we're done */
5065 spin_unlock(&last_ptr->refill_lock);
5069 spin_lock(&last_ptr->lock);
5071 * whoops, this cluster doesn't actually point to
5072 * this block group. Get a ref on the block
5073 * group is does point to and try again
5075 if (!last_ptr_loop && last_ptr->block_group &&
5076 last_ptr->block_group != block_group) {
5078 btrfs_put_block_group(block_group);
5079 block_group = last_ptr->block_group;
5080 btrfs_get_block_group(block_group);
5081 spin_unlock(&last_ptr->lock);
5082 spin_unlock(&last_ptr->refill_lock);
5085 search_start = block_group->key.objectid;
5087 * we know this block group is properly
5088 * in the list because
5089 * btrfs_remove_block_group, drops the
5090 * cluster before it removes the block
5091 * group from the list
5093 goto have_block_group;
5095 spin_unlock(&last_ptr->lock);
5098 * this cluster didn't work out, free it and
5101 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5105 /* allocate a cluster in this block group */
5106 ret = btrfs_find_space_cluster(trans, root,
5107 block_group, last_ptr,
5109 empty_cluster + empty_size);
5112 * now pull our allocation out of this
5115 offset = btrfs_alloc_from_cluster(block_group,
5116 last_ptr, num_bytes,
5119 /* we found one, proceed */
5120 spin_unlock(&last_ptr->refill_lock);
5123 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5124 && !failed_cluster_refill) {
5125 spin_unlock(&last_ptr->refill_lock);
5127 failed_cluster_refill = true;
5128 wait_block_group_cache_progress(block_group,
5129 num_bytes + empty_cluster + empty_size);
5130 goto have_block_group;
5134 * at this point we either didn't find a cluster
5135 * or we weren't able to allocate a block from our
5136 * cluster. Free the cluster we've been trying
5137 * to use, and go to the next block group
5139 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5140 spin_unlock(&last_ptr->refill_lock);
5144 offset = btrfs_find_space_for_alloc(block_group, search_start,
5145 num_bytes, empty_size);
5147 * If we didn't find a chunk, and we haven't failed on this
5148 * block group before, and this block group is in the middle of
5149 * caching and we are ok with waiting, then go ahead and wait
5150 * for progress to be made, and set failed_alloc to true.
5152 * If failed_alloc is true then we've already waited on this
5153 * block group once and should move on to the next block group.
5155 if (!offset && !failed_alloc && !cached &&
5156 loop > LOOP_CACHING_NOWAIT) {
5157 wait_block_group_cache_progress(block_group,
5158 num_bytes + empty_size);
5159 failed_alloc = true;
5160 goto have_block_group;
5161 } else if (!offset) {
5165 search_start = stripe_align(root, offset);
5166 /* move on to the next group */
5167 if (search_start + num_bytes >= search_end) {
5168 btrfs_add_free_space(block_group, offset, num_bytes);
5172 /* move on to the next group */
5173 if (search_start + num_bytes >
5174 block_group->key.objectid + block_group->key.offset) {
5175 btrfs_add_free_space(block_group, offset, num_bytes);
5179 ins->objectid = search_start;
5180 ins->offset = num_bytes;
5182 if (offset < search_start)
5183 btrfs_add_free_space(block_group, offset,
5184 search_start - offset);
5185 BUG_ON(offset > search_start);
5187 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5188 (data & BTRFS_BLOCK_GROUP_DATA));
5189 if (ret == -EAGAIN) {
5190 btrfs_add_free_space(block_group, offset, num_bytes);
5194 /* we are all good, lets return */
5195 ins->objectid = search_start;
5196 ins->offset = num_bytes;
5198 if (offset < search_start)
5199 btrfs_add_free_space(block_group, offset,
5200 search_start - offset);
5201 BUG_ON(offset > search_start);
5202 btrfs_put_block_group(block_group);
5205 failed_cluster_refill = false;
5206 failed_alloc = false;
5207 BUG_ON(index != get_block_group_index(block_group));
5208 btrfs_put_block_group(block_group);
5210 up_read(&space_info->groups_sem);
5212 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5215 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5216 * for them to make caching progress. Also
5217 * determine the best possible bg to cache
5218 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5219 * caching kthreads as we move along
5220 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5221 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5222 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5225 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5227 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5228 found_uncached_bg = false;
5230 if (!ideal_cache_percent)
5234 * 1 of the following 2 things have happened so far
5236 * 1) We found an ideal block group for caching that
5237 * is mostly full and will cache quickly, so we might
5238 * as well wait for it.
5240 * 2) We searched for cached only and we didn't find
5241 * anything, and we didn't start any caching kthreads
5242 * either, so chances are we will loop through and
5243 * start a couple caching kthreads, and then come back
5244 * around and just wait for them. This will be slower
5245 * because we will have 2 caching kthreads reading at
5246 * the same time when we could have just started one
5247 * and waited for it to get far enough to give us an
5248 * allocation, so go ahead and go to the wait caching
5251 loop = LOOP_CACHING_WAIT;
5252 search_start = ideal_cache_offset;
5253 ideal_cache_percent = 0;
5255 } else if (loop == LOOP_FIND_IDEAL) {
5257 * Didn't find a uncached bg, wait on anything we find
5260 loop = LOOP_CACHING_WAIT;
5266 if (loop == LOOP_ALLOC_CHUNK) {
5267 if (allowed_chunk_alloc) {
5268 ret = do_chunk_alloc(trans, root, num_bytes +
5269 2 * 1024 * 1024, data,
5270 CHUNK_ALLOC_LIMITED);
5271 allowed_chunk_alloc = 0;
5273 done_chunk_alloc = 1;
5274 } else if (!done_chunk_alloc &&
5275 space_info->force_alloc ==
5276 CHUNK_ALLOC_NO_FORCE) {
5277 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5281 * We didn't allocate a chunk, go ahead and drop the
5282 * empty size and loop again.
5284 if (!done_chunk_alloc)
5285 loop = LOOP_NO_EMPTY_SIZE;
5288 if (loop == LOOP_NO_EMPTY_SIZE) {
5294 } else if (!ins->objectid) {
5296 } else if (ins->objectid) {
5303 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5304 int dump_block_groups)
5306 struct btrfs_block_group_cache *cache;
5309 spin_lock(&info->lock);
5310 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5311 (unsigned long long)(info->total_bytes - info->bytes_used -
5312 info->bytes_pinned - info->bytes_reserved -
5313 info->bytes_readonly),
5314 (info->full) ? "" : "not ");
5315 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5316 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5317 (unsigned long long)info->total_bytes,
5318 (unsigned long long)info->bytes_used,
5319 (unsigned long long)info->bytes_pinned,
5320 (unsigned long long)info->bytes_reserved,
5321 (unsigned long long)info->bytes_may_use,
5322 (unsigned long long)info->bytes_readonly);
5323 spin_unlock(&info->lock);
5325 if (!dump_block_groups)
5328 down_read(&info->groups_sem);
5330 list_for_each_entry(cache, &info->block_groups[index], list) {
5331 spin_lock(&cache->lock);
5332 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5333 "%llu pinned %llu reserved\n",
5334 (unsigned long long)cache->key.objectid,
5335 (unsigned long long)cache->key.offset,
5336 (unsigned long long)btrfs_block_group_used(&cache->item),
5337 (unsigned long long)cache->pinned,
5338 (unsigned long long)cache->reserved);
5339 btrfs_dump_free_space(cache, bytes);
5340 spin_unlock(&cache->lock);
5342 if (++index < BTRFS_NR_RAID_TYPES)
5344 up_read(&info->groups_sem);
5347 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5348 struct btrfs_root *root,
5349 u64 num_bytes, u64 min_alloc_size,
5350 u64 empty_size, u64 hint_byte,
5351 u64 search_end, struct btrfs_key *ins,
5355 u64 search_start = 0;
5357 data = btrfs_get_alloc_profile(root, data);
5360 * the only place that sets empty_size is btrfs_realloc_node, which
5361 * is not called recursively on allocations
5363 if (empty_size || root->ref_cows)
5364 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5365 num_bytes + 2 * 1024 * 1024, data,
5366 CHUNK_ALLOC_NO_FORCE);
5368 WARN_ON(num_bytes < root->sectorsize);
5369 ret = find_free_extent(trans, root, num_bytes, empty_size,
5370 search_start, search_end, hint_byte,
5373 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5374 num_bytes = num_bytes >> 1;
5375 num_bytes = num_bytes & ~(root->sectorsize - 1);
5376 num_bytes = max(num_bytes, min_alloc_size);
5377 do_chunk_alloc(trans, root->fs_info->extent_root,
5378 num_bytes, data, CHUNK_ALLOC_FORCE);
5381 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5382 struct btrfs_space_info *sinfo;
5384 sinfo = __find_space_info(root->fs_info, data);
5385 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5386 "wanted %llu\n", (unsigned long long)data,
5387 (unsigned long long)num_bytes);
5388 dump_space_info(sinfo, num_bytes, 1);
5391 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5396 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5398 struct btrfs_block_group_cache *cache;
5401 cache = btrfs_lookup_block_group(root->fs_info, start);
5403 printk(KERN_ERR "Unable to find block group for %llu\n",
5404 (unsigned long long)start);
5408 if (btrfs_test_opt(root, DISCARD))
5409 ret = btrfs_discard_extent(root, start, len, NULL);
5411 btrfs_add_free_space(cache, start, len);
5412 btrfs_update_reserved_bytes(cache, len, 0, 1);
5413 btrfs_put_block_group(cache);
5415 trace_btrfs_reserved_extent_free(root, start, len);
5420 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5421 struct btrfs_root *root,
5422 u64 parent, u64 root_objectid,
5423 u64 flags, u64 owner, u64 offset,
5424 struct btrfs_key *ins, int ref_mod)
5427 struct btrfs_fs_info *fs_info = root->fs_info;
5428 struct btrfs_extent_item *extent_item;
5429 struct btrfs_extent_inline_ref *iref;
5430 struct btrfs_path *path;
5431 struct extent_buffer *leaf;
5436 type = BTRFS_SHARED_DATA_REF_KEY;
5438 type = BTRFS_EXTENT_DATA_REF_KEY;
5440 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5442 path = btrfs_alloc_path();
5446 path->leave_spinning = 1;
5447 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5451 leaf = path->nodes[0];
5452 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5453 struct btrfs_extent_item);
5454 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5455 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5456 btrfs_set_extent_flags(leaf, extent_item,
5457 flags | BTRFS_EXTENT_FLAG_DATA);
5459 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5460 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5462 struct btrfs_shared_data_ref *ref;
5463 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5464 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5465 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5467 struct btrfs_extent_data_ref *ref;
5468 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5469 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5470 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5471 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5472 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5475 btrfs_mark_buffer_dirty(path->nodes[0]);
5476 btrfs_free_path(path);
5478 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5480 printk(KERN_ERR "btrfs update block group failed for %llu "
5481 "%llu\n", (unsigned long long)ins->objectid,
5482 (unsigned long long)ins->offset);
5488 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5489 struct btrfs_root *root,
5490 u64 parent, u64 root_objectid,
5491 u64 flags, struct btrfs_disk_key *key,
5492 int level, struct btrfs_key *ins)
5495 struct btrfs_fs_info *fs_info = root->fs_info;
5496 struct btrfs_extent_item *extent_item;
5497 struct btrfs_tree_block_info *block_info;
5498 struct btrfs_extent_inline_ref *iref;
5499 struct btrfs_path *path;
5500 struct extent_buffer *leaf;
5501 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5503 path = btrfs_alloc_path();
5506 path->leave_spinning = 1;
5507 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5511 leaf = path->nodes[0];
5512 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5513 struct btrfs_extent_item);
5514 btrfs_set_extent_refs(leaf, extent_item, 1);
5515 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5516 btrfs_set_extent_flags(leaf, extent_item,
5517 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5518 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5520 btrfs_set_tree_block_key(leaf, block_info, key);
5521 btrfs_set_tree_block_level(leaf, block_info, level);
5523 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5525 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5526 btrfs_set_extent_inline_ref_type(leaf, iref,
5527 BTRFS_SHARED_BLOCK_REF_KEY);
5528 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5530 btrfs_set_extent_inline_ref_type(leaf, iref,
5531 BTRFS_TREE_BLOCK_REF_KEY);
5532 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5535 btrfs_mark_buffer_dirty(leaf);
5536 btrfs_free_path(path);
5538 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5540 printk(KERN_ERR "btrfs update block group failed for %llu "
5541 "%llu\n", (unsigned long long)ins->objectid,
5542 (unsigned long long)ins->offset);
5548 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5549 struct btrfs_root *root,
5550 u64 root_objectid, u64 owner,
5551 u64 offset, struct btrfs_key *ins)
5555 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5557 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5558 0, root_objectid, owner, offset,
5559 BTRFS_ADD_DELAYED_EXTENT, NULL);
5564 * this is used by the tree logging recovery code. It records that
5565 * an extent has been allocated and makes sure to clear the free
5566 * space cache bits as well
5568 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5569 struct btrfs_root *root,
5570 u64 root_objectid, u64 owner, u64 offset,
5571 struct btrfs_key *ins)
5574 struct btrfs_block_group_cache *block_group;
5575 struct btrfs_caching_control *caching_ctl;
5576 u64 start = ins->objectid;
5577 u64 num_bytes = ins->offset;
5579 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5580 cache_block_group(block_group, trans, NULL, 0);
5581 caching_ctl = get_caching_control(block_group);
5584 BUG_ON(!block_group_cache_done(block_group));
5585 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5588 mutex_lock(&caching_ctl->mutex);
5590 if (start >= caching_ctl->progress) {
5591 ret = add_excluded_extent(root, start, num_bytes);
5593 } else if (start + num_bytes <= caching_ctl->progress) {
5594 ret = btrfs_remove_free_space(block_group,
5598 num_bytes = caching_ctl->progress - start;
5599 ret = btrfs_remove_free_space(block_group,
5603 start = caching_ctl->progress;
5604 num_bytes = ins->objectid + ins->offset -
5605 caching_ctl->progress;
5606 ret = add_excluded_extent(root, start, num_bytes);
5610 mutex_unlock(&caching_ctl->mutex);
5611 put_caching_control(caching_ctl);
5614 ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5616 btrfs_put_block_group(block_group);
5617 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5618 0, owner, offset, ins, 1);
5622 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5623 struct btrfs_root *root,
5624 u64 bytenr, u32 blocksize,
5627 struct extent_buffer *buf;
5629 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5631 return ERR_PTR(-ENOMEM);
5632 btrfs_set_header_generation(buf, trans->transid);
5633 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5634 btrfs_tree_lock(buf);
5635 clean_tree_block(trans, root, buf);
5637 btrfs_set_lock_blocking(buf);
5638 btrfs_set_buffer_uptodate(buf);
5640 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5642 * we allow two log transactions at a time, use different
5643 * EXENT bit to differentiate dirty pages.
5645 if (root->log_transid % 2 == 0)
5646 set_extent_dirty(&root->dirty_log_pages, buf->start,
5647 buf->start + buf->len - 1, GFP_NOFS);
5649 set_extent_new(&root->dirty_log_pages, buf->start,
5650 buf->start + buf->len - 1, GFP_NOFS);
5652 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5653 buf->start + buf->len - 1, GFP_NOFS);
5655 trans->blocks_used++;
5656 /* this returns a buffer locked for blocking */
5660 static struct btrfs_block_rsv *
5661 use_block_rsv(struct btrfs_trans_handle *trans,
5662 struct btrfs_root *root, u32 blocksize)
5664 struct btrfs_block_rsv *block_rsv;
5665 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5668 block_rsv = get_block_rsv(trans, root);
5670 if (block_rsv->size == 0) {
5671 ret = reserve_metadata_bytes(trans, root, block_rsv,
5674 * If we couldn't reserve metadata bytes try and use some from
5675 * the global reserve.
5677 if (ret && block_rsv != global_rsv) {
5678 ret = block_rsv_use_bytes(global_rsv, blocksize);
5681 return ERR_PTR(ret);
5683 return ERR_PTR(ret);
5688 ret = block_rsv_use_bytes(block_rsv, blocksize);
5693 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5696 spin_lock(&block_rsv->lock);
5697 block_rsv->size += blocksize;
5698 spin_unlock(&block_rsv->lock);
5700 } else if (ret && block_rsv != global_rsv) {
5701 ret = block_rsv_use_bytes(global_rsv, blocksize);
5707 return ERR_PTR(-ENOSPC);
5710 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5712 block_rsv_add_bytes(block_rsv, blocksize, 0);
5713 block_rsv_release_bytes(block_rsv, NULL, 0);
5717 * finds a free extent and does all the dirty work required for allocation
5718 * returns the key for the extent through ins, and a tree buffer for
5719 * the first block of the extent through buf.
5721 * returns the tree buffer or NULL.
5723 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5724 struct btrfs_root *root, u32 blocksize,
5725 u64 parent, u64 root_objectid,
5726 struct btrfs_disk_key *key, int level,
5727 u64 hint, u64 empty_size)
5729 struct btrfs_key ins;
5730 struct btrfs_block_rsv *block_rsv;
5731 struct extent_buffer *buf;
5736 block_rsv = use_block_rsv(trans, root, blocksize);
5737 if (IS_ERR(block_rsv))
5738 return ERR_CAST(block_rsv);
5740 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5741 empty_size, hint, (u64)-1, &ins, 0);
5743 unuse_block_rsv(block_rsv, blocksize);
5744 return ERR_PTR(ret);
5747 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5749 BUG_ON(IS_ERR(buf));
5751 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5753 parent = ins.objectid;
5754 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5758 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5759 struct btrfs_delayed_extent_op *extent_op;
5760 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5763 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5765 memset(&extent_op->key, 0, sizeof(extent_op->key));
5766 extent_op->flags_to_set = flags;
5767 extent_op->update_key = 1;
5768 extent_op->update_flags = 1;
5769 extent_op->is_data = 0;
5771 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5772 ins.offset, parent, root_objectid,
5773 level, BTRFS_ADD_DELAYED_EXTENT,
5780 struct walk_control {
5781 u64 refs[BTRFS_MAX_LEVEL];
5782 u64 flags[BTRFS_MAX_LEVEL];
5783 struct btrfs_key update_progress;
5793 #define DROP_REFERENCE 1
5794 #define UPDATE_BACKREF 2
5796 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5797 struct btrfs_root *root,
5798 struct walk_control *wc,
5799 struct btrfs_path *path)
5807 struct btrfs_key key;
5808 struct extent_buffer *eb;
5813 if (path->slots[wc->level] < wc->reada_slot) {
5814 wc->reada_count = wc->reada_count * 2 / 3;
5815 wc->reada_count = max(wc->reada_count, 2);
5817 wc->reada_count = wc->reada_count * 3 / 2;
5818 wc->reada_count = min_t(int, wc->reada_count,
5819 BTRFS_NODEPTRS_PER_BLOCK(root));
5822 eb = path->nodes[wc->level];
5823 nritems = btrfs_header_nritems(eb);
5824 blocksize = btrfs_level_size(root, wc->level - 1);
5826 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5827 if (nread >= wc->reada_count)
5831 bytenr = btrfs_node_blockptr(eb, slot);
5832 generation = btrfs_node_ptr_generation(eb, slot);
5834 if (slot == path->slots[wc->level])
5837 if (wc->stage == UPDATE_BACKREF &&
5838 generation <= root->root_key.offset)
5841 /* We don't lock the tree block, it's OK to be racy here */
5842 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5847 if (wc->stage == DROP_REFERENCE) {
5851 if (wc->level == 1 &&
5852 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5854 if (!wc->update_ref ||
5855 generation <= root->root_key.offset)
5857 btrfs_node_key_to_cpu(eb, &key, slot);
5858 ret = btrfs_comp_cpu_keys(&key,
5859 &wc->update_progress);
5863 if (wc->level == 1 &&
5864 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5868 ret = readahead_tree_block(root, bytenr, blocksize,
5874 wc->reada_slot = slot;
5878 * hepler to process tree block while walking down the tree.
5880 * when wc->stage == UPDATE_BACKREF, this function updates
5881 * back refs for pointers in the block.
5883 * NOTE: return value 1 means we should stop walking down.
5885 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5886 struct btrfs_root *root,
5887 struct btrfs_path *path,
5888 struct walk_control *wc, int lookup_info)
5890 int level = wc->level;
5891 struct extent_buffer *eb = path->nodes[level];
5892 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5895 if (wc->stage == UPDATE_BACKREF &&
5896 btrfs_header_owner(eb) != root->root_key.objectid)
5900 * when reference count of tree block is 1, it won't increase
5901 * again. once full backref flag is set, we never clear it.
5904 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5905 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5906 BUG_ON(!path->locks[level]);
5907 ret = btrfs_lookup_extent_info(trans, root,
5912 BUG_ON(wc->refs[level] == 0);
5915 if (wc->stage == DROP_REFERENCE) {
5916 if (wc->refs[level] > 1)
5919 if (path->locks[level] && !wc->keep_locks) {
5920 btrfs_tree_unlock_rw(eb, path->locks[level]);
5921 path->locks[level] = 0;
5926 /* wc->stage == UPDATE_BACKREF */
5927 if (!(wc->flags[level] & flag)) {
5928 BUG_ON(!path->locks[level]);
5929 ret = btrfs_inc_ref(trans, root, eb, 1);
5931 ret = btrfs_dec_ref(trans, root, eb, 0);
5933 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5936 wc->flags[level] |= flag;
5940 * the block is shared by multiple trees, so it's not good to
5941 * keep the tree lock
5943 if (path->locks[level] && level > 0) {
5944 btrfs_tree_unlock_rw(eb, path->locks[level]);
5945 path->locks[level] = 0;
5951 * hepler to process tree block pointer.
5953 * when wc->stage == DROP_REFERENCE, this function checks
5954 * reference count of the block pointed to. if the block
5955 * is shared and we need update back refs for the subtree
5956 * rooted at the block, this function changes wc->stage to
5957 * UPDATE_BACKREF. if the block is shared and there is no
5958 * need to update back, this function drops the reference
5961 * NOTE: return value 1 means we should stop walking down.
5963 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5964 struct btrfs_root *root,
5965 struct btrfs_path *path,
5966 struct walk_control *wc, int *lookup_info)
5972 struct btrfs_key key;
5973 struct extent_buffer *next;
5974 int level = wc->level;
5978 generation = btrfs_node_ptr_generation(path->nodes[level],
5979 path->slots[level]);
5981 * if the lower level block was created before the snapshot
5982 * was created, we know there is no need to update back refs
5985 if (wc->stage == UPDATE_BACKREF &&
5986 generation <= root->root_key.offset) {
5991 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5992 blocksize = btrfs_level_size(root, level - 1);
5994 next = btrfs_find_tree_block(root, bytenr, blocksize);
5996 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6001 btrfs_tree_lock(next);
6002 btrfs_set_lock_blocking(next);
6004 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6005 &wc->refs[level - 1],
6006 &wc->flags[level - 1]);
6008 BUG_ON(wc->refs[level - 1] == 0);
6011 if (wc->stage == DROP_REFERENCE) {
6012 if (wc->refs[level - 1] > 1) {
6014 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6017 if (!wc->update_ref ||
6018 generation <= root->root_key.offset)
6021 btrfs_node_key_to_cpu(path->nodes[level], &key,
6022 path->slots[level]);
6023 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6027 wc->stage = UPDATE_BACKREF;
6028 wc->shared_level = level - 1;
6032 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6036 if (!btrfs_buffer_uptodate(next, generation)) {
6037 btrfs_tree_unlock(next);
6038 free_extent_buffer(next);
6044 if (reada && level == 1)
6045 reada_walk_down(trans, root, wc, path);
6046 next = read_tree_block(root, bytenr, blocksize, generation);
6049 btrfs_tree_lock(next);
6050 btrfs_set_lock_blocking(next);
6054 BUG_ON(level != btrfs_header_level(next));
6055 path->nodes[level] = next;
6056 path->slots[level] = 0;
6057 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6063 wc->refs[level - 1] = 0;
6064 wc->flags[level - 1] = 0;
6065 if (wc->stage == DROP_REFERENCE) {
6066 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6067 parent = path->nodes[level]->start;
6069 BUG_ON(root->root_key.objectid !=
6070 btrfs_header_owner(path->nodes[level]));
6074 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6075 root->root_key.objectid, level - 1, 0);
6078 btrfs_tree_unlock(next);
6079 free_extent_buffer(next);
6085 * hepler to process tree block while walking up the tree.
6087 * when wc->stage == DROP_REFERENCE, this function drops
6088 * reference count on the block.
6090 * when wc->stage == UPDATE_BACKREF, this function changes
6091 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6092 * to UPDATE_BACKREF previously while processing the block.
6094 * NOTE: return value 1 means we should stop walking up.
6096 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6097 struct btrfs_root *root,
6098 struct btrfs_path *path,
6099 struct walk_control *wc)
6102 int level = wc->level;
6103 struct extent_buffer *eb = path->nodes[level];
6106 if (wc->stage == UPDATE_BACKREF) {
6107 BUG_ON(wc->shared_level < level);
6108 if (level < wc->shared_level)
6111 ret = find_next_key(path, level + 1, &wc->update_progress);
6115 wc->stage = DROP_REFERENCE;
6116 wc->shared_level = -1;
6117 path->slots[level] = 0;
6120 * check reference count again if the block isn't locked.
6121 * we should start walking down the tree again if reference
6124 if (!path->locks[level]) {
6126 btrfs_tree_lock(eb);
6127 btrfs_set_lock_blocking(eb);
6128 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6130 ret = btrfs_lookup_extent_info(trans, root,
6135 BUG_ON(wc->refs[level] == 0);
6136 if (wc->refs[level] == 1) {
6137 btrfs_tree_unlock_rw(eb, path->locks[level]);
6143 /* wc->stage == DROP_REFERENCE */
6144 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6146 if (wc->refs[level] == 1) {
6148 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6149 ret = btrfs_dec_ref(trans, root, eb, 1);
6151 ret = btrfs_dec_ref(trans, root, eb, 0);
6154 /* make block locked assertion in clean_tree_block happy */
6155 if (!path->locks[level] &&
6156 btrfs_header_generation(eb) == trans->transid) {
6157 btrfs_tree_lock(eb);
6158 btrfs_set_lock_blocking(eb);
6159 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6161 clean_tree_block(trans, root, eb);
6164 if (eb == root->node) {
6165 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6168 BUG_ON(root->root_key.objectid !=
6169 btrfs_header_owner(eb));
6171 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6172 parent = path->nodes[level + 1]->start;
6174 BUG_ON(root->root_key.objectid !=
6175 btrfs_header_owner(path->nodes[level + 1]));
6178 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6180 wc->refs[level] = 0;
6181 wc->flags[level] = 0;
6185 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6186 struct btrfs_root *root,
6187 struct btrfs_path *path,
6188 struct walk_control *wc)
6190 int level = wc->level;
6191 int lookup_info = 1;
6194 while (level >= 0) {
6195 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6202 if (path->slots[level] >=
6203 btrfs_header_nritems(path->nodes[level]))
6206 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6208 path->slots[level]++;
6217 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6218 struct btrfs_root *root,
6219 struct btrfs_path *path,
6220 struct walk_control *wc, int max_level)
6222 int level = wc->level;
6225 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6226 while (level < max_level && path->nodes[level]) {
6228 if (path->slots[level] + 1 <
6229 btrfs_header_nritems(path->nodes[level])) {
6230 path->slots[level]++;
6233 ret = walk_up_proc(trans, root, path, wc);
6237 if (path->locks[level]) {
6238 btrfs_tree_unlock_rw(path->nodes[level],
6239 path->locks[level]);
6240 path->locks[level] = 0;
6242 free_extent_buffer(path->nodes[level]);
6243 path->nodes[level] = NULL;
6251 * drop a subvolume tree.
6253 * this function traverses the tree freeing any blocks that only
6254 * referenced by the tree.
6256 * when a shared tree block is found. this function decreases its
6257 * reference count by one. if update_ref is true, this function
6258 * also make sure backrefs for the shared block and all lower level
6259 * blocks are properly updated.
6261 int btrfs_drop_snapshot(struct btrfs_root *root,
6262 struct btrfs_block_rsv *block_rsv, int update_ref)
6264 struct btrfs_path *path;
6265 struct btrfs_trans_handle *trans;
6266 struct btrfs_root *tree_root = root->fs_info->tree_root;
6267 struct btrfs_root_item *root_item = &root->root_item;
6268 struct walk_control *wc;
6269 struct btrfs_key key;
6274 path = btrfs_alloc_path();
6277 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6280 trans = btrfs_start_transaction(tree_root, 0);
6281 BUG_ON(IS_ERR(trans));
6284 trans->block_rsv = block_rsv;
6286 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6287 level = btrfs_header_level(root->node);
6288 path->nodes[level] = btrfs_lock_root_node(root);
6289 btrfs_set_lock_blocking(path->nodes[level]);
6290 path->slots[level] = 0;
6291 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6292 memset(&wc->update_progress, 0,
6293 sizeof(wc->update_progress));
6295 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6296 memcpy(&wc->update_progress, &key,
6297 sizeof(wc->update_progress));
6299 level = root_item->drop_level;
6301 path->lowest_level = level;
6302 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6303 path->lowest_level = 0;
6311 * unlock our path, this is safe because only this
6312 * function is allowed to delete this snapshot
6314 btrfs_unlock_up_safe(path, 0);
6316 level = btrfs_header_level(root->node);
6318 btrfs_tree_lock(path->nodes[level]);
6319 btrfs_set_lock_blocking(path->nodes[level]);
6321 ret = btrfs_lookup_extent_info(trans, root,
6322 path->nodes[level]->start,
6323 path->nodes[level]->len,
6327 BUG_ON(wc->refs[level] == 0);
6329 if (level == root_item->drop_level)
6332 btrfs_tree_unlock(path->nodes[level]);
6333 WARN_ON(wc->refs[level] != 1);
6339 wc->shared_level = -1;
6340 wc->stage = DROP_REFERENCE;
6341 wc->update_ref = update_ref;
6343 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6346 ret = walk_down_tree(trans, root, path, wc);
6352 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6359 BUG_ON(wc->stage != DROP_REFERENCE);
6363 if (wc->stage == DROP_REFERENCE) {
6365 btrfs_node_key(path->nodes[level],
6366 &root_item->drop_progress,
6367 path->slots[level]);
6368 root_item->drop_level = level;
6371 BUG_ON(wc->level == 0);
6372 if (btrfs_should_end_transaction(trans, tree_root)) {
6373 ret = btrfs_update_root(trans, tree_root,
6378 btrfs_end_transaction_throttle(trans, tree_root);
6379 trans = btrfs_start_transaction(tree_root, 0);
6380 BUG_ON(IS_ERR(trans));
6382 trans->block_rsv = block_rsv;
6385 btrfs_release_path(path);
6388 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6391 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6392 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6396 /* if we fail to delete the orphan item this time
6397 * around, it'll get picked up the next time.
6399 * The most common failure here is just -ENOENT.
6401 btrfs_del_orphan_item(trans, tree_root,
6402 root->root_key.objectid);
6406 if (root->in_radix) {
6407 btrfs_free_fs_root(tree_root->fs_info, root);
6409 free_extent_buffer(root->node);
6410 free_extent_buffer(root->commit_root);
6414 btrfs_end_transaction_throttle(trans, tree_root);
6416 btrfs_free_path(path);
6421 * drop subtree rooted at tree block 'node'.
6423 * NOTE: this function will unlock and release tree block 'node'
6425 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6426 struct btrfs_root *root,
6427 struct extent_buffer *node,
6428 struct extent_buffer *parent)
6430 struct btrfs_path *path;
6431 struct walk_control *wc;
6437 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6439 path = btrfs_alloc_path();
6443 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6445 btrfs_free_path(path);
6449 btrfs_assert_tree_locked(parent);
6450 parent_level = btrfs_header_level(parent);
6451 extent_buffer_get(parent);
6452 path->nodes[parent_level] = parent;
6453 path->slots[parent_level] = btrfs_header_nritems(parent);
6455 btrfs_assert_tree_locked(node);
6456 level = btrfs_header_level(node);
6457 path->nodes[level] = node;
6458 path->slots[level] = 0;
6459 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6461 wc->refs[parent_level] = 1;
6462 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6464 wc->shared_level = -1;
6465 wc->stage = DROP_REFERENCE;
6468 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6471 wret = walk_down_tree(trans, root, path, wc);
6477 wret = walk_up_tree(trans, root, path, wc, parent_level);
6485 btrfs_free_path(path);
6489 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6492 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6493 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6496 * we add in the count of missing devices because we want
6497 * to make sure that any RAID levels on a degraded FS
6498 * continue to be honored.
6500 num_devices = root->fs_info->fs_devices->rw_devices +
6501 root->fs_info->fs_devices->missing_devices;
6503 if (num_devices == 1) {
6504 stripped |= BTRFS_BLOCK_GROUP_DUP;
6505 stripped = flags & ~stripped;
6507 /* turn raid0 into single device chunks */
6508 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6511 /* turn mirroring into duplication */
6512 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6513 BTRFS_BLOCK_GROUP_RAID10))
6514 return stripped | BTRFS_BLOCK_GROUP_DUP;
6517 /* they already had raid on here, just return */
6518 if (flags & stripped)
6521 stripped |= BTRFS_BLOCK_GROUP_DUP;
6522 stripped = flags & ~stripped;
6524 /* switch duplicated blocks with raid1 */
6525 if (flags & BTRFS_BLOCK_GROUP_DUP)
6526 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6528 /* turn single device chunks into raid0 */
6529 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6534 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6536 struct btrfs_space_info *sinfo = cache->space_info;
6538 u64 min_allocable_bytes;
6545 * We need some metadata space and system metadata space for
6546 * allocating chunks in some corner cases until we force to set
6547 * it to be readonly.
6550 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6552 min_allocable_bytes = 1 * 1024 * 1024;
6554 min_allocable_bytes = 0;
6556 spin_lock(&sinfo->lock);
6557 spin_lock(&cache->lock);
6558 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6559 cache->bytes_super - btrfs_block_group_used(&cache->item);
6561 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6562 sinfo->bytes_may_use + sinfo->bytes_readonly +
6563 cache->reserved_pinned + num_bytes + min_allocable_bytes <=
6564 sinfo->total_bytes) {
6565 sinfo->bytes_readonly += num_bytes;
6566 sinfo->bytes_reserved += cache->reserved_pinned;
6567 cache->reserved_pinned = 0;
6572 spin_unlock(&cache->lock);
6573 spin_unlock(&sinfo->lock);
6577 int btrfs_set_block_group_ro(struct btrfs_root *root,
6578 struct btrfs_block_group_cache *cache)
6581 struct btrfs_trans_handle *trans;
6587 trans = btrfs_join_transaction(root);
6588 BUG_ON(IS_ERR(trans));
6590 alloc_flags = update_block_group_flags(root, cache->flags);
6591 if (alloc_flags != cache->flags)
6592 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6595 ret = set_block_group_ro(cache, 0);
6598 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6599 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6603 ret = set_block_group_ro(cache, 0);
6605 btrfs_end_transaction(trans, root);
6609 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6610 struct btrfs_root *root, u64 type)
6612 u64 alloc_flags = get_alloc_profile(root, type);
6613 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6618 * helper to account the unused space of all the readonly block group in the
6619 * list. takes mirrors into account.
6621 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6623 struct btrfs_block_group_cache *block_group;
6627 list_for_each_entry(block_group, groups_list, list) {
6628 spin_lock(&block_group->lock);
6630 if (!block_group->ro) {
6631 spin_unlock(&block_group->lock);
6635 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6636 BTRFS_BLOCK_GROUP_RAID10 |
6637 BTRFS_BLOCK_GROUP_DUP))
6642 free_bytes += (block_group->key.offset -
6643 btrfs_block_group_used(&block_group->item)) *
6646 spin_unlock(&block_group->lock);
6653 * helper to account the unused space of all the readonly block group in the
6654 * space_info. takes mirrors into account.
6656 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6661 spin_lock(&sinfo->lock);
6663 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6664 if (!list_empty(&sinfo->block_groups[i]))
6665 free_bytes += __btrfs_get_ro_block_group_free_space(
6666 &sinfo->block_groups[i]);
6668 spin_unlock(&sinfo->lock);
6673 int btrfs_set_block_group_rw(struct btrfs_root *root,
6674 struct btrfs_block_group_cache *cache)
6676 struct btrfs_space_info *sinfo = cache->space_info;
6681 spin_lock(&sinfo->lock);
6682 spin_lock(&cache->lock);
6683 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6684 cache->bytes_super - btrfs_block_group_used(&cache->item);
6685 sinfo->bytes_readonly -= num_bytes;
6687 spin_unlock(&cache->lock);
6688 spin_unlock(&sinfo->lock);
6693 * checks to see if its even possible to relocate this block group.
6695 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6696 * ok to go ahead and try.
6698 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6700 struct btrfs_block_group_cache *block_group;
6701 struct btrfs_space_info *space_info;
6702 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6703 struct btrfs_device *device;
6707 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6709 /* odd, couldn't find the block group, leave it alone */
6713 /* no bytes used, we're good */
6714 if (!btrfs_block_group_used(&block_group->item))
6717 space_info = block_group->space_info;
6718 spin_lock(&space_info->lock);
6720 full = space_info->full;
6723 * if this is the last block group we have in this space, we can't
6724 * relocate it unless we're able to allocate a new chunk below.
6726 * Otherwise, we need to make sure we have room in the space to handle
6727 * all of the extents from this block group. If we can, we're good
6729 if ((space_info->total_bytes != block_group->key.offset) &&
6730 (space_info->bytes_used + space_info->bytes_reserved +
6731 space_info->bytes_pinned + space_info->bytes_readonly +
6732 btrfs_block_group_used(&block_group->item) <
6733 space_info->total_bytes)) {
6734 spin_unlock(&space_info->lock);
6737 spin_unlock(&space_info->lock);
6740 * ok we don't have enough space, but maybe we have free space on our
6741 * devices to allocate new chunks for relocation, so loop through our
6742 * alloc devices and guess if we have enough space. However, if we
6743 * were marked as full, then we know there aren't enough chunks, and we
6750 mutex_lock(&root->fs_info->chunk_mutex);
6751 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6752 u64 min_free = btrfs_block_group_used(&block_group->item);
6756 * check to make sure we can actually find a chunk with enough
6757 * space to fit our block group in.
6759 if (device->total_bytes > device->bytes_used + min_free) {
6760 ret = find_free_dev_extent(NULL, device, min_free,
6767 mutex_unlock(&root->fs_info->chunk_mutex);
6769 btrfs_put_block_group(block_group);
6773 static int find_first_block_group(struct btrfs_root *root,
6774 struct btrfs_path *path, struct btrfs_key *key)
6777 struct btrfs_key found_key;
6778 struct extent_buffer *leaf;
6781 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6786 slot = path->slots[0];
6787 leaf = path->nodes[0];
6788 if (slot >= btrfs_header_nritems(leaf)) {
6789 ret = btrfs_next_leaf(root, path);
6796 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6798 if (found_key.objectid >= key->objectid &&
6799 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6809 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6811 struct btrfs_block_group_cache *block_group;
6815 struct inode *inode;
6817 block_group = btrfs_lookup_first_block_group(info, last);
6818 while (block_group) {
6819 spin_lock(&block_group->lock);
6820 if (block_group->iref)
6822 spin_unlock(&block_group->lock);
6823 block_group = next_block_group(info->tree_root,
6833 inode = block_group->inode;
6834 block_group->iref = 0;
6835 block_group->inode = NULL;
6836 spin_unlock(&block_group->lock);
6838 last = block_group->key.objectid + block_group->key.offset;
6839 btrfs_put_block_group(block_group);
6843 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6845 struct btrfs_block_group_cache *block_group;
6846 struct btrfs_space_info *space_info;
6847 struct btrfs_caching_control *caching_ctl;
6850 down_write(&info->extent_commit_sem);
6851 while (!list_empty(&info->caching_block_groups)) {
6852 caching_ctl = list_entry(info->caching_block_groups.next,
6853 struct btrfs_caching_control, list);
6854 list_del(&caching_ctl->list);
6855 put_caching_control(caching_ctl);
6857 up_write(&info->extent_commit_sem);
6859 spin_lock(&info->block_group_cache_lock);
6860 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6861 block_group = rb_entry(n, struct btrfs_block_group_cache,
6863 rb_erase(&block_group->cache_node,
6864 &info->block_group_cache_tree);
6865 spin_unlock(&info->block_group_cache_lock);
6867 down_write(&block_group->space_info->groups_sem);
6868 list_del(&block_group->list);
6869 up_write(&block_group->space_info->groups_sem);
6871 if (block_group->cached == BTRFS_CACHE_STARTED)
6872 wait_block_group_cache_done(block_group);
6875 * We haven't cached this block group, which means we could
6876 * possibly have excluded extents on this block group.
6878 if (block_group->cached == BTRFS_CACHE_NO)
6879 free_excluded_extents(info->extent_root, block_group);
6881 btrfs_remove_free_space_cache(block_group);
6882 btrfs_put_block_group(block_group);
6884 spin_lock(&info->block_group_cache_lock);
6886 spin_unlock(&info->block_group_cache_lock);
6888 /* now that all the block groups are freed, go through and
6889 * free all the space_info structs. This is only called during
6890 * the final stages of unmount, and so we know nobody is
6891 * using them. We call synchronize_rcu() once before we start,
6892 * just to be on the safe side.
6896 release_global_block_rsv(info);
6898 while(!list_empty(&info->space_info)) {
6899 space_info = list_entry(info->space_info.next,
6900 struct btrfs_space_info,
6902 if (space_info->bytes_pinned > 0 ||
6903 space_info->bytes_reserved > 0) {
6905 dump_space_info(space_info, 0, 0);
6907 list_del(&space_info->list);
6913 static void __link_block_group(struct btrfs_space_info *space_info,
6914 struct btrfs_block_group_cache *cache)
6916 int index = get_block_group_index(cache);
6918 down_write(&space_info->groups_sem);
6919 list_add_tail(&cache->list, &space_info->block_groups[index]);
6920 up_write(&space_info->groups_sem);
6923 int btrfs_read_block_groups(struct btrfs_root *root)
6925 struct btrfs_path *path;
6927 struct btrfs_block_group_cache *cache;
6928 struct btrfs_fs_info *info = root->fs_info;
6929 struct btrfs_space_info *space_info;
6930 struct btrfs_key key;
6931 struct btrfs_key found_key;
6932 struct extent_buffer *leaf;
6936 root = info->extent_root;
6939 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6940 path = btrfs_alloc_path();
6945 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
6946 if (cache_gen != 0 &&
6947 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
6949 if (btrfs_test_opt(root, CLEAR_CACHE))
6951 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
6952 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
6955 ret = find_first_block_group(root, path, &key);
6960 leaf = path->nodes[0];
6961 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6962 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6967 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
6969 if (!cache->free_space_ctl) {
6975 atomic_set(&cache->count, 1);
6976 spin_lock_init(&cache->lock);
6977 cache->fs_info = info;
6978 INIT_LIST_HEAD(&cache->list);
6979 INIT_LIST_HEAD(&cache->cluster_list);
6982 cache->disk_cache_state = BTRFS_DC_CLEAR;
6984 read_extent_buffer(leaf, &cache->item,
6985 btrfs_item_ptr_offset(leaf, path->slots[0]),
6986 sizeof(cache->item));
6987 memcpy(&cache->key, &found_key, sizeof(found_key));
6989 key.objectid = found_key.objectid + found_key.offset;
6990 btrfs_release_path(path);
6991 cache->flags = btrfs_block_group_flags(&cache->item);
6992 cache->sectorsize = root->sectorsize;
6994 btrfs_init_free_space_ctl(cache);
6997 * We need to exclude the super stripes now so that the space
6998 * info has super bytes accounted for, otherwise we'll think
6999 * we have more space than we actually do.
7001 exclude_super_stripes(root, cache);
7004 * check for two cases, either we are full, and therefore
7005 * don't need to bother with the caching work since we won't
7006 * find any space, or we are empty, and we can just add all
7007 * the space in and be done with it. This saves us _alot_ of
7008 * time, particularly in the full case.
7010 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7011 cache->last_byte_to_unpin = (u64)-1;
7012 cache->cached = BTRFS_CACHE_FINISHED;
7013 free_excluded_extents(root, cache);
7014 } else if (btrfs_block_group_used(&cache->item) == 0) {
7015 cache->last_byte_to_unpin = (u64)-1;
7016 cache->cached = BTRFS_CACHE_FINISHED;
7017 add_new_free_space(cache, root->fs_info,
7019 found_key.objectid +
7021 free_excluded_extents(root, cache);
7024 ret = update_space_info(info, cache->flags, found_key.offset,
7025 btrfs_block_group_used(&cache->item),
7028 cache->space_info = space_info;
7029 spin_lock(&cache->space_info->lock);
7030 cache->space_info->bytes_readonly += cache->bytes_super;
7031 spin_unlock(&cache->space_info->lock);
7033 __link_block_group(space_info, cache);
7035 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7038 set_avail_alloc_bits(root->fs_info, cache->flags);
7039 if (btrfs_chunk_readonly(root, cache->key.objectid))
7040 set_block_group_ro(cache, 1);
7043 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7044 if (!(get_alloc_profile(root, space_info->flags) &
7045 (BTRFS_BLOCK_GROUP_RAID10 |
7046 BTRFS_BLOCK_GROUP_RAID1 |
7047 BTRFS_BLOCK_GROUP_DUP)))
7050 * avoid allocating from un-mirrored block group if there are
7051 * mirrored block groups.
7053 list_for_each_entry(cache, &space_info->block_groups[3], list)
7054 set_block_group_ro(cache, 1);
7055 list_for_each_entry(cache, &space_info->block_groups[4], list)
7056 set_block_group_ro(cache, 1);
7059 init_global_block_rsv(info);
7062 btrfs_free_path(path);
7066 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7067 struct btrfs_root *root, u64 bytes_used,
7068 u64 type, u64 chunk_objectid, u64 chunk_offset,
7072 struct btrfs_root *extent_root;
7073 struct btrfs_block_group_cache *cache;
7075 extent_root = root->fs_info->extent_root;
7077 root->fs_info->last_trans_log_full_commit = trans->transid;
7079 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7082 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7084 if (!cache->free_space_ctl) {
7089 cache->key.objectid = chunk_offset;
7090 cache->key.offset = size;
7091 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7092 cache->sectorsize = root->sectorsize;
7093 cache->fs_info = root->fs_info;
7095 atomic_set(&cache->count, 1);
7096 spin_lock_init(&cache->lock);
7097 INIT_LIST_HEAD(&cache->list);
7098 INIT_LIST_HEAD(&cache->cluster_list);
7100 btrfs_init_free_space_ctl(cache);
7102 btrfs_set_block_group_used(&cache->item, bytes_used);
7103 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7104 cache->flags = type;
7105 btrfs_set_block_group_flags(&cache->item, type);
7107 cache->last_byte_to_unpin = (u64)-1;
7108 cache->cached = BTRFS_CACHE_FINISHED;
7109 exclude_super_stripes(root, cache);
7111 add_new_free_space(cache, root->fs_info, chunk_offset,
7112 chunk_offset + size);
7114 free_excluded_extents(root, cache);
7116 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7117 &cache->space_info);
7120 spin_lock(&cache->space_info->lock);
7121 cache->space_info->bytes_readonly += cache->bytes_super;
7122 spin_unlock(&cache->space_info->lock);
7124 __link_block_group(cache->space_info, cache);
7126 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7129 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7130 sizeof(cache->item));
7133 set_avail_alloc_bits(extent_root->fs_info, type);
7138 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7139 struct btrfs_root *root, u64 group_start)
7141 struct btrfs_path *path;
7142 struct btrfs_block_group_cache *block_group;
7143 struct btrfs_free_cluster *cluster;
7144 struct btrfs_root *tree_root = root->fs_info->tree_root;
7145 struct btrfs_key key;
7146 struct inode *inode;
7150 root = root->fs_info->extent_root;
7152 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7153 BUG_ON(!block_group);
7154 BUG_ON(!block_group->ro);
7157 * Free the reserved super bytes from this block group before
7160 free_excluded_extents(root, block_group);
7162 memcpy(&key, &block_group->key, sizeof(key));
7163 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7164 BTRFS_BLOCK_GROUP_RAID1 |
7165 BTRFS_BLOCK_GROUP_RAID10))
7170 /* make sure this block group isn't part of an allocation cluster */
7171 cluster = &root->fs_info->data_alloc_cluster;
7172 spin_lock(&cluster->refill_lock);
7173 btrfs_return_cluster_to_free_space(block_group, cluster);
7174 spin_unlock(&cluster->refill_lock);
7177 * make sure this block group isn't part of a metadata
7178 * allocation cluster
7180 cluster = &root->fs_info->meta_alloc_cluster;
7181 spin_lock(&cluster->refill_lock);
7182 btrfs_return_cluster_to_free_space(block_group, cluster);
7183 spin_unlock(&cluster->refill_lock);
7185 path = btrfs_alloc_path();
7188 inode = lookup_free_space_inode(root, block_group, path);
7189 if (!IS_ERR(inode)) {
7190 btrfs_orphan_add(trans, inode);
7192 /* One for the block groups ref */
7193 spin_lock(&block_group->lock);
7194 if (block_group->iref) {
7195 block_group->iref = 0;
7196 block_group->inode = NULL;
7197 spin_unlock(&block_group->lock);
7200 spin_unlock(&block_group->lock);
7202 /* One for our lookup ref */
7206 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7207 key.offset = block_group->key.objectid;
7210 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7214 btrfs_release_path(path);
7216 ret = btrfs_del_item(trans, tree_root, path);
7219 btrfs_release_path(path);
7222 spin_lock(&root->fs_info->block_group_cache_lock);
7223 rb_erase(&block_group->cache_node,
7224 &root->fs_info->block_group_cache_tree);
7225 spin_unlock(&root->fs_info->block_group_cache_lock);
7227 down_write(&block_group->space_info->groups_sem);
7229 * we must use list_del_init so people can check to see if they
7230 * are still on the list after taking the semaphore
7232 list_del_init(&block_group->list);
7233 up_write(&block_group->space_info->groups_sem);
7235 if (block_group->cached == BTRFS_CACHE_STARTED)
7236 wait_block_group_cache_done(block_group);
7238 btrfs_remove_free_space_cache(block_group);
7240 spin_lock(&block_group->space_info->lock);
7241 block_group->space_info->total_bytes -= block_group->key.offset;
7242 block_group->space_info->bytes_readonly -= block_group->key.offset;
7243 block_group->space_info->disk_total -= block_group->key.offset * factor;
7244 spin_unlock(&block_group->space_info->lock);
7246 memcpy(&key, &block_group->key, sizeof(key));
7248 btrfs_clear_space_info_full(root->fs_info);
7250 btrfs_put_block_group(block_group);
7251 btrfs_put_block_group(block_group);
7253 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7259 ret = btrfs_del_item(trans, root, path);
7261 btrfs_free_path(path);
7265 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7267 struct btrfs_space_info *space_info;
7268 struct btrfs_super_block *disk_super;
7274 disk_super = &fs_info->super_copy;
7275 if (!btrfs_super_root(disk_super))
7278 features = btrfs_super_incompat_flags(disk_super);
7279 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7282 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7283 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7288 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7289 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7291 flags = BTRFS_BLOCK_GROUP_METADATA;
7292 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7296 flags = BTRFS_BLOCK_GROUP_DATA;
7297 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7303 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7305 return unpin_extent_range(root, start, end);
7308 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7309 u64 num_bytes, u64 *actual_bytes)
7311 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7314 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7316 struct btrfs_fs_info *fs_info = root->fs_info;
7317 struct btrfs_block_group_cache *cache = NULL;
7324 cache = btrfs_lookup_block_group(fs_info, range->start);
7327 if (cache->key.objectid >= (range->start + range->len)) {
7328 btrfs_put_block_group(cache);
7332 start = max(range->start, cache->key.objectid);
7333 end = min(range->start + range->len,
7334 cache->key.objectid + cache->key.offset);
7336 if (end - start >= range->minlen) {
7337 if (!block_group_cache_done(cache)) {
7338 ret = cache_block_group(cache, NULL, root, 0);
7340 wait_block_group_cache_done(cache);
7342 ret = btrfs_trim_block_group(cache,
7348 trimmed += group_trimmed;
7350 btrfs_put_block_group(cache);
7355 cache = next_block_group(fs_info->tree_root, cache);
7358 range->len = trimmed;