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>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE = 0,
57 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 static int update_block_group(struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
110 block_group_cache_done(struct btrfs_block_group_cache *cache)
113 return cache->cached == BTRFS_CACHE_FINISHED;
116 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
118 return (cache->flags & bits) == bits;
121 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
123 atomic_inc(&cache->count);
126 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
128 if (atomic_dec_and_test(&cache->count)) {
129 WARN_ON(cache->pinned > 0);
130 WARN_ON(cache->reserved > 0);
131 kfree(cache->free_space_ctl);
137 * this adds the block group to the fs_info rb tree for the block group
140 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
141 struct btrfs_block_group_cache *block_group)
144 struct rb_node *parent = NULL;
145 struct btrfs_block_group_cache *cache;
147 spin_lock(&info->block_group_cache_lock);
148 p = &info->block_group_cache_tree.rb_node;
152 cache = rb_entry(parent, struct btrfs_block_group_cache,
154 if (block_group->key.objectid < cache->key.objectid) {
156 } else if (block_group->key.objectid > cache->key.objectid) {
159 spin_unlock(&info->block_group_cache_lock);
164 rb_link_node(&block_group->cache_node, parent, p);
165 rb_insert_color(&block_group->cache_node,
166 &info->block_group_cache_tree);
168 if (info->first_logical_byte > block_group->key.objectid)
169 info->first_logical_byte = block_group->key.objectid;
171 spin_unlock(&info->block_group_cache_lock);
177 * This will return the block group at or after bytenr if contains is 0, else
178 * it will return the block group that contains the bytenr
180 static struct btrfs_block_group_cache *
181 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
184 struct btrfs_block_group_cache *cache, *ret = NULL;
188 spin_lock(&info->block_group_cache_lock);
189 n = info->block_group_cache_tree.rb_node;
192 cache = rb_entry(n, struct btrfs_block_group_cache,
194 end = cache->key.objectid + cache->key.offset - 1;
195 start = cache->key.objectid;
197 if (bytenr < start) {
198 if (!contains && (!ret || start < ret->key.objectid))
201 } else if (bytenr > start) {
202 if (contains && bytenr <= end) {
213 btrfs_get_block_group(ret);
214 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
215 info->first_logical_byte = ret->key.objectid;
217 spin_unlock(&info->block_group_cache_lock);
222 static int add_excluded_extent(struct btrfs_root *root,
223 u64 start, u64 num_bytes)
225 u64 end = start + num_bytes - 1;
226 set_extent_bits(&root->fs_info->freed_extents[0],
227 start, end, EXTENT_UPTODATE, GFP_NOFS);
228 set_extent_bits(&root->fs_info->freed_extents[1],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 static void free_excluded_extents(struct btrfs_root *root,
234 struct btrfs_block_group_cache *cache)
238 start = cache->key.objectid;
239 end = start + cache->key.offset - 1;
241 clear_extent_bits(&root->fs_info->freed_extents[0],
242 start, end, EXTENT_UPTODATE, GFP_NOFS);
243 clear_extent_bits(&root->fs_info->freed_extents[1],
244 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 static int exclude_super_stripes(struct btrfs_root *root,
248 struct btrfs_block_group_cache *cache)
255 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
256 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
257 cache->bytes_super += stripe_len;
258 ret = add_excluded_extent(root, cache->key.objectid,
264 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
265 bytenr = btrfs_sb_offset(i);
266 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
267 cache->key.objectid, bytenr,
268 0, &logical, &nr, &stripe_len);
273 cache->bytes_super += stripe_len;
274 ret = add_excluded_extent(root, logical[nr],
287 static struct btrfs_caching_control *
288 get_caching_control(struct btrfs_block_group_cache *cache)
290 struct btrfs_caching_control *ctl;
292 spin_lock(&cache->lock);
293 if (cache->cached != BTRFS_CACHE_STARTED) {
294 spin_unlock(&cache->lock);
298 /* We're loading it the fast way, so we don't have a caching_ctl. */
299 if (!cache->caching_ctl) {
300 spin_unlock(&cache->lock);
304 ctl = cache->caching_ctl;
305 atomic_inc(&ctl->count);
306 spin_unlock(&cache->lock);
310 static void put_caching_control(struct btrfs_caching_control *ctl)
312 if (atomic_dec_and_test(&ctl->count))
317 * this is only called by cache_block_group, since we could have freed extents
318 * we need to check the pinned_extents for any extents that can't be used yet
319 * since their free space will be released as soon as the transaction commits.
321 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
322 struct btrfs_fs_info *info, u64 start, u64 end)
324 u64 extent_start, extent_end, size, total_added = 0;
327 while (start < end) {
328 ret = find_first_extent_bit(info->pinned_extents, start,
329 &extent_start, &extent_end,
330 EXTENT_DIRTY | EXTENT_UPTODATE,
335 if (extent_start <= start) {
336 start = extent_end + 1;
337 } else if (extent_start > start && extent_start < end) {
338 size = extent_start - start;
340 ret = btrfs_add_free_space(block_group, start,
342 BUG_ON(ret); /* -ENOMEM or logic error */
343 start = extent_end + 1;
352 ret = btrfs_add_free_space(block_group, start, size);
353 BUG_ON(ret); /* -ENOMEM or logic error */
359 static noinline void caching_thread(struct btrfs_work *work)
361 struct btrfs_block_group_cache *block_group;
362 struct btrfs_fs_info *fs_info;
363 struct btrfs_caching_control *caching_ctl;
364 struct btrfs_root *extent_root;
365 struct btrfs_path *path;
366 struct extent_buffer *leaf;
367 struct btrfs_key key;
373 caching_ctl = container_of(work, struct btrfs_caching_control, work);
374 block_group = caching_ctl->block_group;
375 fs_info = block_group->fs_info;
376 extent_root = fs_info->extent_root;
378 path = btrfs_alloc_path();
382 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
385 * We don't want to deadlock with somebody trying to allocate a new
386 * extent for the extent root while also trying to search the extent
387 * root to add free space. So we skip locking and search the commit
388 * root, since its read-only
390 path->skip_locking = 1;
391 path->search_commit_root = 1;
396 key.type = BTRFS_EXTENT_ITEM_KEY;
398 mutex_lock(&caching_ctl->mutex);
399 /* need to make sure the commit_root doesn't disappear */
400 down_read(&fs_info->extent_commit_sem);
402 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
406 leaf = path->nodes[0];
407 nritems = btrfs_header_nritems(leaf);
410 if (btrfs_fs_closing(fs_info) > 1) {
415 if (path->slots[0] < nritems) {
416 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
418 ret = find_next_key(path, 0, &key);
422 if (need_resched() ||
423 btrfs_next_leaf(extent_root, path)) {
424 caching_ctl->progress = last;
425 btrfs_release_path(path);
426 up_read(&fs_info->extent_commit_sem);
427 mutex_unlock(&caching_ctl->mutex);
431 leaf = path->nodes[0];
432 nritems = btrfs_header_nritems(leaf);
436 if (key.objectid < block_group->key.objectid) {
441 if (key.objectid >= block_group->key.objectid +
442 block_group->key.offset)
445 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
446 key.type == BTRFS_METADATA_ITEM_KEY) {
447 total_found += add_new_free_space(block_group,
450 if (key.type == BTRFS_METADATA_ITEM_KEY)
451 last = key.objectid +
452 fs_info->tree_root->leafsize;
454 last = key.objectid + key.offset;
456 if (total_found > (1024 * 1024 * 2)) {
458 wake_up(&caching_ctl->wait);
465 total_found += add_new_free_space(block_group, fs_info, last,
466 block_group->key.objectid +
467 block_group->key.offset);
468 caching_ctl->progress = (u64)-1;
470 spin_lock(&block_group->lock);
471 block_group->caching_ctl = NULL;
472 block_group->cached = BTRFS_CACHE_FINISHED;
473 spin_unlock(&block_group->lock);
476 btrfs_free_path(path);
477 up_read(&fs_info->extent_commit_sem);
479 free_excluded_extents(extent_root, block_group);
481 mutex_unlock(&caching_ctl->mutex);
483 wake_up(&caching_ctl->wait);
485 put_caching_control(caching_ctl);
486 btrfs_put_block_group(block_group);
489 static int cache_block_group(struct btrfs_block_group_cache *cache,
493 struct btrfs_fs_info *fs_info = cache->fs_info;
494 struct btrfs_caching_control *caching_ctl;
497 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
501 INIT_LIST_HEAD(&caching_ctl->list);
502 mutex_init(&caching_ctl->mutex);
503 init_waitqueue_head(&caching_ctl->wait);
504 caching_ctl->block_group = cache;
505 caching_ctl->progress = cache->key.objectid;
506 atomic_set(&caching_ctl->count, 1);
507 caching_ctl->work.func = caching_thread;
509 spin_lock(&cache->lock);
511 * This should be a rare occasion, but this could happen I think in the
512 * case where one thread starts to load the space cache info, and then
513 * some other thread starts a transaction commit which tries to do an
514 * allocation while the other thread is still loading the space cache
515 * info. The previous loop should have kept us from choosing this block
516 * group, but if we've moved to the state where we will wait on caching
517 * block groups we need to first check if we're doing a fast load here,
518 * so we can wait for it to finish, otherwise we could end up allocating
519 * from a block group who's cache gets evicted for one reason or
522 while (cache->cached == BTRFS_CACHE_FAST) {
523 struct btrfs_caching_control *ctl;
525 ctl = cache->caching_ctl;
526 atomic_inc(&ctl->count);
527 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
528 spin_unlock(&cache->lock);
532 finish_wait(&ctl->wait, &wait);
533 put_caching_control(ctl);
534 spin_lock(&cache->lock);
537 if (cache->cached != BTRFS_CACHE_NO) {
538 spin_unlock(&cache->lock);
542 WARN_ON(cache->caching_ctl);
543 cache->caching_ctl = caching_ctl;
544 cache->cached = BTRFS_CACHE_FAST;
545 spin_unlock(&cache->lock);
547 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
548 ret = load_free_space_cache(fs_info, cache);
550 spin_lock(&cache->lock);
552 cache->caching_ctl = NULL;
553 cache->cached = BTRFS_CACHE_FINISHED;
554 cache->last_byte_to_unpin = (u64)-1;
556 if (load_cache_only) {
557 cache->caching_ctl = NULL;
558 cache->cached = BTRFS_CACHE_NO;
560 cache->cached = BTRFS_CACHE_STARTED;
563 spin_unlock(&cache->lock);
564 wake_up(&caching_ctl->wait);
566 put_caching_control(caching_ctl);
567 free_excluded_extents(fs_info->extent_root, cache);
572 * We are not going to do the fast caching, set cached to the
573 * appropriate value and wakeup any waiters.
575 spin_lock(&cache->lock);
576 if (load_cache_only) {
577 cache->caching_ctl = NULL;
578 cache->cached = BTRFS_CACHE_NO;
580 cache->cached = BTRFS_CACHE_STARTED;
582 spin_unlock(&cache->lock);
583 wake_up(&caching_ctl->wait);
586 if (load_cache_only) {
587 put_caching_control(caching_ctl);
591 down_write(&fs_info->extent_commit_sem);
592 atomic_inc(&caching_ctl->count);
593 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
594 up_write(&fs_info->extent_commit_sem);
596 btrfs_get_block_group(cache);
598 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
604 * return the block group that starts at or after bytenr
606 static struct btrfs_block_group_cache *
607 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
609 struct btrfs_block_group_cache *cache;
611 cache = block_group_cache_tree_search(info, bytenr, 0);
617 * return the block group that contains the given bytenr
619 struct btrfs_block_group_cache *btrfs_lookup_block_group(
620 struct btrfs_fs_info *info,
623 struct btrfs_block_group_cache *cache;
625 cache = block_group_cache_tree_search(info, bytenr, 1);
630 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
633 struct list_head *head = &info->space_info;
634 struct btrfs_space_info *found;
636 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
639 list_for_each_entry_rcu(found, head, list) {
640 if (found->flags & flags) {
650 * after adding space to the filesystem, we need to clear the full flags
651 * on all the space infos.
653 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
655 struct list_head *head = &info->space_info;
656 struct btrfs_space_info *found;
659 list_for_each_entry_rcu(found, head, list)
664 u64 btrfs_find_block_group(struct btrfs_root *root,
665 u64 search_start, u64 search_hint, int owner)
667 struct btrfs_block_group_cache *cache;
669 u64 last = max(search_hint, search_start);
676 cache = btrfs_lookup_first_block_group(root->fs_info, last);
680 spin_lock(&cache->lock);
681 last = cache->key.objectid + cache->key.offset;
682 used = btrfs_block_group_used(&cache->item);
684 if ((full_search || !cache->ro) &&
685 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
686 if (used + cache->pinned + cache->reserved <
687 div_factor(cache->key.offset, factor)) {
688 group_start = cache->key.objectid;
689 spin_unlock(&cache->lock);
690 btrfs_put_block_group(cache);
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
703 if (!full_search && factor < 10) {
713 /* simple helper to search for an existing extent at a given offset */
714 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
730 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
731 if (key.objectid == start &&
732 key.type == BTRFS_METADATA_ITEM_KEY)
735 btrfs_free_path(path);
740 * helper function to lookup reference count and flags of a tree block.
742 * the head node for delayed ref is used to store the sum of all the
743 * reference count modifications queued up in the rbtree. the head
744 * node may also store the extent flags to set. This way you can check
745 * to see what the reference count and extent flags would be if all of
746 * the delayed refs are not processed.
748 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
749 struct btrfs_root *root, u64 bytenr,
750 u64 offset, int metadata, u64 *refs, u64 *flags)
752 struct btrfs_delayed_ref_head *head;
753 struct btrfs_delayed_ref_root *delayed_refs;
754 struct btrfs_path *path;
755 struct btrfs_extent_item *ei;
756 struct extent_buffer *leaf;
757 struct btrfs_key key;
764 * If we don't have skinny metadata, don't bother doing anything
767 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
768 offset = root->leafsize;
772 path = btrfs_alloc_path();
777 key.objectid = bytenr;
778 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.objectid = bytenr;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
787 path->skip_locking = 1;
788 path->search_commit_root = 1;
791 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
796 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
797 key.type = BTRFS_EXTENT_ITEM_KEY;
798 key.offset = root->leafsize;
799 btrfs_release_path(path);
804 leaf = path->nodes[0];
805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
806 if (item_size >= sizeof(*ei)) {
807 ei = btrfs_item_ptr(leaf, path->slots[0],
808 struct btrfs_extent_item);
809 num_refs = btrfs_extent_refs(leaf, ei);
810 extent_flags = btrfs_extent_flags(leaf, ei);
812 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
813 struct btrfs_extent_item_v0 *ei0;
814 BUG_ON(item_size != sizeof(*ei0));
815 ei0 = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_extent_item_v0);
817 num_refs = btrfs_extent_refs_v0(leaf, ei0);
818 /* FIXME: this isn't correct for data */
819 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
824 BUG_ON(num_refs == 0);
834 delayed_refs = &trans->transaction->delayed_refs;
835 spin_lock(&delayed_refs->lock);
836 head = btrfs_find_delayed_ref_head(trans, bytenr);
838 if (!mutex_trylock(&head->mutex)) {
839 atomic_inc(&head->node.refs);
840 spin_unlock(&delayed_refs->lock);
842 btrfs_release_path(path);
845 * Mutex was contended, block until it's released and try
848 mutex_lock(&head->mutex);
849 mutex_unlock(&head->mutex);
850 btrfs_put_delayed_ref(&head->node);
853 if (head->extent_op && head->extent_op->update_flags)
854 extent_flags |= head->extent_op->flags_to_set;
856 BUG_ON(num_refs == 0);
858 num_refs += head->node.ref_mod;
859 mutex_unlock(&head->mutex);
861 spin_unlock(&delayed_refs->lock);
863 WARN_ON(num_refs == 0);
867 *flags = extent_flags;
869 btrfs_free_path(path);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct btrfs_path *path,
983 u64 owner, u32 extra_size)
985 struct btrfs_extent_item *item;
986 struct btrfs_extent_item_v0 *ei0;
987 struct btrfs_extent_ref_v0 *ref0;
988 struct btrfs_tree_block_info *bi;
989 struct extent_buffer *leaf;
990 struct btrfs_key key;
991 struct btrfs_key found_key;
992 u32 new_size = sizeof(*item);
996 leaf = path->nodes[0];
997 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
999 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_extent_item_v0);
1002 refs = btrfs_extent_refs_v0(leaf, ei0);
1004 if (owner == (u64)-1) {
1006 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007 ret = btrfs_next_leaf(root, path);
1010 BUG_ON(ret > 0); /* Corruption */
1011 leaf = path->nodes[0];
1013 btrfs_item_key_to_cpu(leaf, &found_key,
1015 BUG_ON(key.objectid != found_key.objectid);
1016 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1020 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_extent_ref_v0);
1022 owner = btrfs_ref_objectid_v0(leaf, ref0);
1026 btrfs_release_path(path);
1028 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029 new_size += sizeof(*bi);
1031 new_size -= sizeof(*ei0);
1032 ret = btrfs_search_slot(trans, root, &key, path,
1033 new_size + extra_size, 1);
1036 BUG_ON(ret); /* Corruption */
1038 btrfs_extend_item(trans, root, path, new_size);
1040 leaf = path->nodes[0];
1041 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042 btrfs_set_extent_refs(leaf, item, refs);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf, item, 0);
1045 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046 btrfs_set_extent_flags(leaf, item,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049 bi = (struct btrfs_tree_block_info *)(item + 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1054 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1056 btrfs_mark_buffer_dirty(leaf);
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1063 u32 high_crc = ~(u32)0;
1064 u32 low_crc = ~(u32)0;
1067 lenum = cpu_to_le64(root_objectid);
1068 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1069 lenum = cpu_to_le64(owner);
1070 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1071 lenum = cpu_to_le64(offset);
1072 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1074 return ((u64)high_crc << 31) ^ (u64)low_crc;
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078 struct btrfs_extent_data_ref *ref)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081 btrfs_extent_data_ref_objectid(leaf, ref),
1082 btrfs_extent_data_ref_offset(leaf, ref));
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref,
1087 u64 root_objectid, u64 owner, u64 offset)
1089 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 u64 bytenr, u64 parent,
1101 u64 owner, u64 offset)
1103 struct btrfs_key key;
1104 struct btrfs_extent_data_ref *ref;
1105 struct extent_buffer *leaf;
1111 key.objectid = bytenr;
1113 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 key.offset = parent;
1116 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 key.offset = hash_extent_data_ref(root_objectid,
1122 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133 btrfs_release_path(path);
1134 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1145 leaf = path->nodes[0];
1146 nritems = btrfs_header_nritems(leaf);
1148 if (path->slots[0] >= nritems) {
1149 ret = btrfs_next_leaf(root, path);
1155 leaf = path->nodes[0];
1156 nritems = btrfs_header_nritems(leaf);
1160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161 if (key.objectid != bytenr ||
1162 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1165 ref = btrfs_item_ptr(leaf, path->slots[0],
1166 struct btrfs_extent_data_ref);
1168 if (match_extent_data_ref(leaf, ref, root_objectid,
1171 btrfs_release_path(path);
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 struct btrfs_path *path,
1186 u64 bytenr, u64 parent,
1187 u64 root_objectid, u64 owner,
1188 u64 offset, int refs_to_add)
1190 struct btrfs_key key;
1191 struct extent_buffer *leaf;
1196 key.objectid = bytenr;
1198 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199 key.offset = parent;
1200 size = sizeof(struct btrfs_shared_data_ref);
1202 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203 key.offset = hash_extent_data_ref(root_objectid,
1205 size = sizeof(struct btrfs_extent_data_ref);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209 if (ret && ret != -EEXIST)
1212 leaf = path->nodes[0];
1214 struct btrfs_shared_data_ref *ref;
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1218 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1220 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221 num_refs += refs_to_add;
1222 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1225 struct btrfs_extent_data_ref *ref;
1226 while (ret == -EEXIST) {
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1229 if (match_extent_data_ref(leaf, ref, root_objectid,
1232 btrfs_release_path(path);
1234 ret = btrfs_insert_empty_item(trans, root, path, &key,
1236 if (ret && ret != -EEXIST)
1239 leaf = path->nodes[0];
1241 ref = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_extent_data_ref);
1244 btrfs_set_extent_data_ref_root(leaf, ref,
1246 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1250 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251 num_refs += refs_to_add;
1252 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1255 btrfs_mark_buffer_dirty(leaf);
1258 btrfs_release_path(path);
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root,
1264 struct btrfs_path *path,
1267 struct btrfs_key key;
1268 struct btrfs_extent_data_ref *ref1 = NULL;
1269 struct btrfs_shared_data_ref *ref2 = NULL;
1270 struct extent_buffer *leaf;
1274 leaf = path->nodes[0];
1275 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_extent_data_ref);
1280 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283 struct btrfs_shared_data_ref);
1284 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287 struct btrfs_extent_ref_v0 *ref0;
1288 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_extent_ref_v0);
1290 num_refs = btrfs_ref_count_v0(leaf, ref0);
1296 BUG_ON(num_refs < refs_to_drop);
1297 num_refs -= refs_to_drop;
1299 if (num_refs == 0) {
1300 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1367 struct btrfs_key key;
1370 key.objectid = bytenr;
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1400 struct btrfs_key key;
1403 key.objectid = bytenr;
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1497 want = extent_ref_type(parent, owner);
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1536 key.type = BTRFS_EXTENT_ITEM_KEY;
1537 key.offset = num_bytes;
1538 btrfs_release_path(path);
1543 if (ret && !insert) {
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1570 BUG_ON(item_size < sizeof(*ei));
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1594 ptr += btrfs_extent_inline_ref_size(type);
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 if (parent == ref_offset) {
1617 if (ref_offset < parent)
1620 if (root_objectid == ref_offset) {
1624 if (ref_offset < root_objectid)
1628 ptr += btrfs_extent_inline_ref_size(type);
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_extent_inline_ref *iref,
1666 u64 parent, u64 root_objectid,
1667 u64 owner, u64 offset, int refs_to_add,
1668 struct btrfs_delayed_extent_op *extent_op)
1670 struct extent_buffer *leaf;
1671 struct btrfs_extent_item *ei;
1674 unsigned long item_offset;
1679 leaf = path->nodes[0];
1680 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681 item_offset = (unsigned long)iref - (unsigned long)ei;
1683 type = extent_ref_type(parent, owner);
1684 size = btrfs_extent_inline_ref_size(type);
1686 btrfs_extend_item(trans, root, path, size);
1688 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 refs = btrfs_extent_refs(leaf, ei);
1690 refs += refs_to_add;
1691 btrfs_set_extent_refs(leaf, ei, refs);
1693 __run_delayed_extent_op(extent_op, leaf, ei);
1695 ptr = (unsigned long)ei + item_offset;
1696 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697 if (ptr < end - size)
1698 memmove_extent_buffer(leaf, ptr + size, ptr,
1701 iref = (struct btrfs_extent_inline_ref *)ptr;
1702 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704 struct btrfs_extent_data_ref *dref;
1705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 struct btrfs_shared_data_ref *sref;
1712 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1720 btrfs_mark_buffer_dirty(leaf);
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref **ref_ret,
1727 u64 bytenr, u64 num_bytes, u64 parent,
1728 u64 root_objectid, u64 owner, u64 offset)
1732 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733 bytenr, num_bytes, parent,
1734 root_objectid, owner, offset, 0);
1738 btrfs_release_path(path);
1741 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1745 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746 root_objectid, owner, offset);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 struct btrfs_path *path,
1758 struct btrfs_extent_inline_ref *iref,
1760 struct btrfs_delayed_extent_op *extent_op)
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *ei;
1764 struct btrfs_extent_data_ref *dref = NULL;
1765 struct btrfs_shared_data_ref *sref = NULL;
1773 leaf = path->nodes[0];
1774 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775 refs = btrfs_extent_refs(leaf, ei);
1776 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777 refs += refs_to_mod;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1782 type = btrfs_extent_inline_ref_type(leaf, iref);
1784 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786 refs = btrfs_extent_data_ref_count(leaf, dref);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789 refs = btrfs_shared_data_ref_count(leaf, sref);
1792 BUG_ON(refs_to_mod != -1);
1795 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796 refs += refs_to_mod;
1799 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1802 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1812 btrfs_truncate_item(trans, root, path, item_size, 1);
1814 btrfs_mark_buffer_dirty(leaf);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct btrfs_extent_inline_ref *iref;
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(trans, root, path, iref,
1835 refs_to_add, extent_op);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(trans, root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data)
1872 BUG_ON(!is_data && refs_to_drop != 1);
1874 update_inline_extent_backref(trans, root, path, iref,
1875 -refs_to_drop, NULL);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1879 ret = btrfs_del_item(trans, root, path);
1884 static int btrfs_issue_discard(struct block_device *bdev,
1887 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1890 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1891 u64 num_bytes, u64 *actual_bytes)
1894 u64 discarded_bytes = 0;
1895 struct btrfs_bio *bbio = NULL;
1898 /* Tell the block device(s) that the sectors can be discarded */
1899 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1900 bytenr, &num_bytes, &bbio, 0);
1901 /* Error condition is -ENOMEM */
1903 struct btrfs_bio_stripe *stripe = bbio->stripes;
1907 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1908 if (!stripe->dev->can_discard)
1911 ret = btrfs_issue_discard(stripe->dev->bdev,
1915 discarded_bytes += stripe->length;
1916 else if (ret != -EOPNOTSUPP)
1917 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1920 * Just in case we get back EOPNOTSUPP for some reason,
1921 * just ignore the return value so we don't screw up
1922 * people calling discard_extent.
1930 *actual_bytes = discarded_bytes;
1933 if (ret == -EOPNOTSUPP)
1938 /* Can return -ENOMEM */
1939 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1940 struct btrfs_root *root,
1941 u64 bytenr, u64 num_bytes, u64 parent,
1942 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1945 struct btrfs_fs_info *fs_info = root->fs_info;
1947 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1948 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1950 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1951 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1953 parent, root_objectid, (int)owner,
1954 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1956 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1958 parent, root_objectid, owner, offset,
1959 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1964 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root,
1966 u64 bytenr, u64 num_bytes,
1967 u64 parent, u64 root_objectid,
1968 u64 owner, u64 offset, int refs_to_add,
1969 struct btrfs_delayed_extent_op *extent_op)
1971 struct btrfs_path *path;
1972 struct extent_buffer *leaf;
1973 struct btrfs_extent_item *item;
1978 path = btrfs_alloc_path();
1983 path->leave_spinning = 1;
1984 /* this will setup the path even if it fails to insert the back ref */
1985 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1986 path, bytenr, num_bytes, parent,
1987 root_objectid, owner, offset,
1988 refs_to_add, extent_op);
1992 if (ret != -EAGAIN) {
1997 leaf = path->nodes[0];
1998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 refs = btrfs_extent_refs(leaf, item);
2000 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2002 __run_delayed_extent_op(extent_op, leaf, item);
2004 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_release_path(path);
2008 path->leave_spinning = 1;
2010 /* now insert the actual backref */
2011 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012 path, bytenr, parent, root_objectid,
2013 owner, offset, refs_to_add);
2015 btrfs_abort_transaction(trans, root, ret);
2017 btrfs_free_path(path);
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op,
2025 int insert_reserved)
2028 struct btrfs_delayed_data_ref *ref;
2029 struct btrfs_key ins;
2034 ins.objectid = node->bytenr;
2035 ins.offset = node->num_bytes;
2036 ins.type = BTRFS_EXTENT_ITEM_KEY;
2038 ref = btrfs_delayed_node_to_data_ref(node);
2039 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2040 parent = ref->parent;
2042 ref_root = ref->root;
2044 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2046 flags |= extent_op->flags_to_set;
2047 ret = alloc_reserved_file_extent(trans, root,
2048 parent, ref_root, flags,
2049 ref->objectid, ref->offset,
2050 &ins, node->ref_mod);
2051 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2052 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2053 node->num_bytes, parent,
2054 ref_root, ref->objectid,
2055 ref->offset, node->ref_mod,
2057 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2058 ret = __btrfs_free_extent(trans, root, node->bytenr,
2059 node->num_bytes, parent,
2060 ref_root, ref->objectid,
2061 ref->offset, node->ref_mod,
2069 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2070 struct extent_buffer *leaf,
2071 struct btrfs_extent_item *ei)
2073 u64 flags = btrfs_extent_flags(leaf, ei);
2074 if (extent_op->update_flags) {
2075 flags |= extent_op->flags_to_set;
2076 btrfs_set_extent_flags(leaf, ei, flags);
2079 if (extent_op->update_key) {
2080 struct btrfs_tree_block_info *bi;
2081 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2082 bi = (struct btrfs_tree_block_info *)(ei + 1);
2083 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2087 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct btrfs_delayed_ref_node *node,
2090 struct btrfs_delayed_extent_op *extent_op)
2092 struct btrfs_key key;
2093 struct btrfs_path *path;
2094 struct btrfs_extent_item *ei;
2095 struct extent_buffer *leaf;
2099 int metadata = (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2100 node->type == BTRFS_SHARED_BLOCK_REF_KEY);
2105 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2108 path = btrfs_alloc_path();
2112 key.objectid = node->bytenr;
2115 struct btrfs_delayed_tree_ref *tree_ref;
2117 tree_ref = btrfs_delayed_node_to_tree_ref(node);
2118 key.type = BTRFS_METADATA_ITEM_KEY;
2119 key.offset = tree_ref->level;
2121 key.type = BTRFS_EXTENT_ITEM_KEY;
2122 key.offset = node->num_bytes;
2127 path->leave_spinning = 1;
2128 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2136 btrfs_release_path(path);
2139 key.offset = node->num_bytes;
2140 key.type = BTRFS_EXTENT_ITEM_KEY;
2147 leaf = path->nodes[0];
2148 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2149 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2150 if (item_size < sizeof(*ei)) {
2151 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2157 leaf = path->nodes[0];
2158 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2161 BUG_ON(item_size < sizeof(*ei));
2162 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2163 __run_delayed_extent_op(extent_op, leaf, ei);
2165 btrfs_mark_buffer_dirty(leaf);
2167 btrfs_free_path(path);
2171 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2172 struct btrfs_root *root,
2173 struct btrfs_delayed_ref_node *node,
2174 struct btrfs_delayed_extent_op *extent_op,
2175 int insert_reserved)
2178 struct btrfs_delayed_tree_ref *ref;
2179 struct btrfs_key ins;
2182 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2185 ref = btrfs_delayed_node_to_tree_ref(node);
2186 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2187 parent = ref->parent;
2189 ref_root = ref->root;
2191 ins.objectid = node->bytenr;
2192 if (skinny_metadata) {
2193 ins.offset = ref->level;
2194 ins.type = BTRFS_METADATA_ITEM_KEY;
2196 ins.offset = node->num_bytes;
2197 ins.type = BTRFS_EXTENT_ITEM_KEY;
2200 BUG_ON(node->ref_mod != 1);
2201 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2202 BUG_ON(!extent_op || !extent_op->update_flags);
2203 ret = alloc_reserved_tree_block(trans, root,
2205 extent_op->flags_to_set,
2208 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2209 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2210 node->num_bytes, parent, ref_root,
2211 ref->level, 0, 1, extent_op);
2212 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2213 ret = __btrfs_free_extent(trans, root, node->bytenr,
2214 node->num_bytes, parent, ref_root,
2215 ref->level, 0, 1, extent_op);
2222 /* helper function to actually process a single delayed ref entry */
2223 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2224 struct btrfs_root *root,
2225 struct btrfs_delayed_ref_node *node,
2226 struct btrfs_delayed_extent_op *extent_op,
2227 int insert_reserved)
2234 if (btrfs_delayed_ref_is_head(node)) {
2235 struct btrfs_delayed_ref_head *head;
2237 * we've hit the end of the chain and we were supposed
2238 * to insert this extent into the tree. But, it got
2239 * deleted before we ever needed to insert it, so all
2240 * we have to do is clean up the accounting
2243 head = btrfs_delayed_node_to_head(node);
2244 if (insert_reserved) {
2245 btrfs_pin_extent(root, node->bytenr,
2246 node->num_bytes, 1);
2247 if (head->is_data) {
2248 ret = btrfs_del_csums(trans, root,
2256 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2257 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2258 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2260 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2261 node->type == BTRFS_SHARED_DATA_REF_KEY)
2262 ret = run_delayed_data_ref(trans, root, node, extent_op,
2269 static noinline struct btrfs_delayed_ref_node *
2270 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2272 struct rb_node *node;
2273 struct btrfs_delayed_ref_node *ref;
2274 int action = BTRFS_ADD_DELAYED_REF;
2277 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2278 * this prevents ref count from going down to zero when
2279 * there still are pending delayed ref.
2281 node = rb_prev(&head->node.rb_node);
2285 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2287 if (ref->bytenr != head->node.bytenr)
2289 if (ref->action == action)
2291 node = rb_prev(node);
2293 if (action == BTRFS_ADD_DELAYED_REF) {
2294 action = BTRFS_DROP_DELAYED_REF;
2301 * Returns 0 on success or if called with an already aborted transaction.
2302 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2304 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2305 struct btrfs_root *root,
2306 struct list_head *cluster)
2308 struct btrfs_delayed_ref_root *delayed_refs;
2309 struct btrfs_delayed_ref_node *ref;
2310 struct btrfs_delayed_ref_head *locked_ref = NULL;
2311 struct btrfs_delayed_extent_op *extent_op;
2312 struct btrfs_fs_info *fs_info = root->fs_info;
2315 int must_insert_reserved = 0;
2317 delayed_refs = &trans->transaction->delayed_refs;
2320 /* pick a new head ref from the cluster list */
2321 if (list_empty(cluster))
2324 locked_ref = list_entry(cluster->next,
2325 struct btrfs_delayed_ref_head, cluster);
2327 /* grab the lock that says we are going to process
2328 * all the refs for this head */
2329 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2332 * we may have dropped the spin lock to get the head
2333 * mutex lock, and that might have given someone else
2334 * time to free the head. If that's true, it has been
2335 * removed from our list and we can move on.
2337 if (ret == -EAGAIN) {
2345 * We need to try and merge add/drops of the same ref since we
2346 * can run into issues with relocate dropping the implicit ref
2347 * and then it being added back again before the drop can
2348 * finish. If we merged anything we need to re-loop so we can
2351 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2355 * locked_ref is the head node, so we have to go one
2356 * node back for any delayed ref updates
2358 ref = select_delayed_ref(locked_ref);
2360 if (ref && ref->seq &&
2361 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2363 * there are still refs with lower seq numbers in the
2364 * process of being added. Don't run this ref yet.
2366 list_del_init(&locked_ref->cluster);
2367 btrfs_delayed_ref_unlock(locked_ref);
2369 delayed_refs->num_heads_ready++;
2370 spin_unlock(&delayed_refs->lock);
2372 spin_lock(&delayed_refs->lock);
2377 * record the must insert reserved flag before we
2378 * drop the spin lock.
2380 must_insert_reserved = locked_ref->must_insert_reserved;
2381 locked_ref->must_insert_reserved = 0;
2383 extent_op = locked_ref->extent_op;
2384 locked_ref->extent_op = NULL;
2387 /* All delayed refs have been processed, Go ahead
2388 * and send the head node to run_one_delayed_ref,
2389 * so that any accounting fixes can happen
2391 ref = &locked_ref->node;
2393 if (extent_op && must_insert_reserved) {
2394 btrfs_free_delayed_extent_op(extent_op);
2399 spin_unlock(&delayed_refs->lock);
2401 ret = run_delayed_extent_op(trans, root,
2403 btrfs_free_delayed_extent_op(extent_op);
2407 "btrfs: run_delayed_extent_op "
2408 "returned %d\n", ret);
2409 spin_lock(&delayed_refs->lock);
2410 btrfs_delayed_ref_unlock(locked_ref);
2419 rb_erase(&ref->rb_node, &delayed_refs->root);
2420 delayed_refs->num_entries--;
2421 if (!btrfs_delayed_ref_is_head(ref)) {
2423 * when we play the delayed ref, also correct the
2426 switch (ref->action) {
2427 case BTRFS_ADD_DELAYED_REF:
2428 case BTRFS_ADD_DELAYED_EXTENT:
2429 locked_ref->node.ref_mod -= ref->ref_mod;
2431 case BTRFS_DROP_DELAYED_REF:
2432 locked_ref->node.ref_mod += ref->ref_mod;
2438 spin_unlock(&delayed_refs->lock);
2440 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2441 must_insert_reserved);
2443 btrfs_free_delayed_extent_op(extent_op);
2445 btrfs_delayed_ref_unlock(locked_ref);
2446 btrfs_put_delayed_ref(ref);
2448 "btrfs: run_one_delayed_ref returned %d\n", ret);
2449 spin_lock(&delayed_refs->lock);
2454 * If this node is a head, that means all the refs in this head
2455 * have been dealt with, and we will pick the next head to deal
2456 * with, so we must unlock the head and drop it from the cluster
2457 * list before we release it.
2459 if (btrfs_delayed_ref_is_head(ref)) {
2460 list_del_init(&locked_ref->cluster);
2461 btrfs_delayed_ref_unlock(locked_ref);
2464 btrfs_put_delayed_ref(ref);
2468 spin_lock(&delayed_refs->lock);
2473 #ifdef SCRAMBLE_DELAYED_REFS
2475 * Normally delayed refs get processed in ascending bytenr order. This
2476 * correlates in most cases to the order added. To expose dependencies on this
2477 * order, we start to process the tree in the middle instead of the beginning
2479 static u64 find_middle(struct rb_root *root)
2481 struct rb_node *n = root->rb_node;
2482 struct btrfs_delayed_ref_node *entry;
2485 u64 first = 0, last = 0;
2489 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2490 first = entry->bytenr;
2494 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2495 last = entry->bytenr;
2500 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501 WARN_ON(!entry->in_tree);
2503 middle = entry->bytenr;
2516 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2517 struct btrfs_fs_info *fs_info)
2519 struct qgroup_update *qgroup_update;
2522 if (list_empty(&trans->qgroup_ref_list) !=
2523 !trans->delayed_ref_elem.seq) {
2524 /* list without seq or seq without list */
2525 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2526 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2527 trans->delayed_ref_elem.seq);
2531 if (!trans->delayed_ref_elem.seq)
2534 while (!list_empty(&trans->qgroup_ref_list)) {
2535 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2536 struct qgroup_update, list);
2537 list_del(&qgroup_update->list);
2539 ret = btrfs_qgroup_account_ref(
2540 trans, fs_info, qgroup_update->node,
2541 qgroup_update->extent_op);
2542 kfree(qgroup_update);
2545 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2550 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2553 int val = atomic_read(&delayed_refs->ref_seq);
2555 if (val < seq || val >= seq + count)
2561 * this starts processing the delayed reference count updates and
2562 * extent insertions we have queued up so far. count can be
2563 * 0, which means to process everything in the tree at the start
2564 * of the run (but not newly added entries), or it can be some target
2565 * number you'd like to process.
2567 * Returns 0 on success or if called with an aborted transaction
2568 * Returns <0 on error and aborts the transaction
2570 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2571 struct btrfs_root *root, unsigned long count)
2573 struct rb_node *node;
2574 struct btrfs_delayed_ref_root *delayed_refs;
2575 struct btrfs_delayed_ref_node *ref;
2576 struct list_head cluster;
2579 int run_all = count == (unsigned long)-1;
2583 /* We'll clean this up in btrfs_cleanup_transaction */
2587 if (root == root->fs_info->extent_root)
2588 root = root->fs_info->tree_root;
2590 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2592 delayed_refs = &trans->transaction->delayed_refs;
2593 INIT_LIST_HEAD(&cluster);
2595 count = delayed_refs->num_entries * 2;
2599 if (!run_all && !run_most) {
2601 int seq = atomic_read(&delayed_refs->ref_seq);
2604 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2606 DEFINE_WAIT(__wait);
2607 if (delayed_refs->num_entries < 16348)
2610 prepare_to_wait(&delayed_refs->wait, &__wait,
2611 TASK_UNINTERRUPTIBLE);
2613 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2616 finish_wait(&delayed_refs->wait, &__wait);
2618 if (!refs_newer(delayed_refs, seq, 256))
2623 finish_wait(&delayed_refs->wait, &__wait);
2629 atomic_inc(&delayed_refs->procs_running_refs);
2634 spin_lock(&delayed_refs->lock);
2636 #ifdef SCRAMBLE_DELAYED_REFS
2637 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2641 if (!(run_all || run_most) &&
2642 delayed_refs->num_heads_ready < 64)
2646 * go find something we can process in the rbtree. We start at
2647 * the beginning of the tree, and then build a cluster
2648 * of refs to process starting at the first one we are able to
2651 delayed_start = delayed_refs->run_delayed_start;
2652 ret = btrfs_find_ref_cluster(trans, &cluster,
2653 delayed_refs->run_delayed_start);
2657 ret = run_clustered_refs(trans, root, &cluster);
2659 btrfs_release_ref_cluster(&cluster);
2660 spin_unlock(&delayed_refs->lock);
2661 btrfs_abort_transaction(trans, root, ret);
2662 atomic_dec(&delayed_refs->procs_running_refs);
2666 atomic_add(ret, &delayed_refs->ref_seq);
2668 count -= min_t(unsigned long, ret, count);
2673 if (delayed_start >= delayed_refs->run_delayed_start) {
2676 * btrfs_find_ref_cluster looped. let's do one
2677 * more cycle. if we don't run any delayed ref
2678 * during that cycle (because we can't because
2679 * all of them are blocked), bail out.
2684 * no runnable refs left, stop trying
2691 /* refs were run, let's reset staleness detection */
2697 if (!list_empty(&trans->new_bgs)) {
2698 spin_unlock(&delayed_refs->lock);
2699 btrfs_create_pending_block_groups(trans, root);
2700 spin_lock(&delayed_refs->lock);
2703 node = rb_first(&delayed_refs->root);
2706 count = (unsigned long)-1;
2709 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2711 if (btrfs_delayed_ref_is_head(ref)) {
2712 struct btrfs_delayed_ref_head *head;
2714 head = btrfs_delayed_node_to_head(ref);
2715 atomic_inc(&ref->refs);
2717 spin_unlock(&delayed_refs->lock);
2719 * Mutex was contended, block until it's
2720 * released and try again
2722 mutex_lock(&head->mutex);
2723 mutex_unlock(&head->mutex);
2725 btrfs_put_delayed_ref(ref);
2729 node = rb_next(node);
2731 spin_unlock(&delayed_refs->lock);
2732 schedule_timeout(1);
2736 atomic_dec(&delayed_refs->procs_running_refs);
2738 if (waitqueue_active(&delayed_refs->wait))
2739 wake_up(&delayed_refs->wait);
2741 spin_unlock(&delayed_refs->lock);
2742 assert_qgroups_uptodate(trans);
2746 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2747 struct btrfs_root *root,
2748 u64 bytenr, u64 num_bytes, u64 flags,
2751 struct btrfs_delayed_extent_op *extent_op;
2754 extent_op = btrfs_alloc_delayed_extent_op();
2758 extent_op->flags_to_set = flags;
2759 extent_op->update_flags = 1;
2760 extent_op->update_key = 0;
2761 extent_op->is_data = is_data ? 1 : 0;
2763 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2764 num_bytes, extent_op);
2766 btrfs_free_delayed_extent_op(extent_op);
2770 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2771 struct btrfs_root *root,
2772 struct btrfs_path *path,
2773 u64 objectid, u64 offset, u64 bytenr)
2775 struct btrfs_delayed_ref_head *head;
2776 struct btrfs_delayed_ref_node *ref;
2777 struct btrfs_delayed_data_ref *data_ref;
2778 struct btrfs_delayed_ref_root *delayed_refs;
2779 struct rb_node *node;
2783 delayed_refs = &trans->transaction->delayed_refs;
2784 spin_lock(&delayed_refs->lock);
2785 head = btrfs_find_delayed_ref_head(trans, bytenr);
2789 if (!mutex_trylock(&head->mutex)) {
2790 atomic_inc(&head->node.refs);
2791 spin_unlock(&delayed_refs->lock);
2793 btrfs_release_path(path);
2796 * Mutex was contended, block until it's released and let
2799 mutex_lock(&head->mutex);
2800 mutex_unlock(&head->mutex);
2801 btrfs_put_delayed_ref(&head->node);
2805 node = rb_prev(&head->node.rb_node);
2809 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2811 if (ref->bytenr != bytenr)
2815 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2818 data_ref = btrfs_delayed_node_to_data_ref(ref);
2820 node = rb_prev(node);
2824 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2825 if (ref->bytenr == bytenr && ref->seq == seq)
2829 if (data_ref->root != root->root_key.objectid ||
2830 data_ref->objectid != objectid || data_ref->offset != offset)
2835 mutex_unlock(&head->mutex);
2837 spin_unlock(&delayed_refs->lock);
2841 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2842 struct btrfs_root *root,
2843 struct btrfs_path *path,
2844 u64 objectid, u64 offset, u64 bytenr)
2846 struct btrfs_root *extent_root = root->fs_info->extent_root;
2847 struct extent_buffer *leaf;
2848 struct btrfs_extent_data_ref *ref;
2849 struct btrfs_extent_inline_ref *iref;
2850 struct btrfs_extent_item *ei;
2851 struct btrfs_key key;
2855 key.objectid = bytenr;
2856 key.offset = (u64)-1;
2857 key.type = BTRFS_EXTENT_ITEM_KEY;
2859 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2862 BUG_ON(ret == 0); /* Corruption */
2865 if (path->slots[0] == 0)
2869 leaf = path->nodes[0];
2870 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2872 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2876 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2877 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2878 if (item_size < sizeof(*ei)) {
2879 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2883 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2885 if (item_size != sizeof(*ei) +
2886 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2889 if (btrfs_extent_generation(leaf, ei) <=
2890 btrfs_root_last_snapshot(&root->root_item))
2893 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2894 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2895 BTRFS_EXTENT_DATA_REF_KEY)
2898 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2899 if (btrfs_extent_refs(leaf, ei) !=
2900 btrfs_extent_data_ref_count(leaf, ref) ||
2901 btrfs_extent_data_ref_root(leaf, ref) !=
2902 root->root_key.objectid ||
2903 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2904 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2912 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2913 struct btrfs_root *root,
2914 u64 objectid, u64 offset, u64 bytenr)
2916 struct btrfs_path *path;
2920 path = btrfs_alloc_path();
2925 ret = check_committed_ref(trans, root, path, objectid,
2927 if (ret && ret != -ENOENT)
2930 ret2 = check_delayed_ref(trans, root, path, objectid,
2932 } while (ret2 == -EAGAIN);
2934 if (ret2 && ret2 != -ENOENT) {
2939 if (ret != -ENOENT || ret2 != -ENOENT)
2942 btrfs_free_path(path);
2943 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2948 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2949 struct btrfs_root *root,
2950 struct extent_buffer *buf,
2951 int full_backref, int inc, int for_cow)
2958 struct btrfs_key key;
2959 struct btrfs_file_extent_item *fi;
2963 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2964 u64, u64, u64, u64, u64, u64, int);
2966 ref_root = btrfs_header_owner(buf);
2967 nritems = btrfs_header_nritems(buf);
2968 level = btrfs_header_level(buf);
2970 if (!root->ref_cows && level == 0)
2974 process_func = btrfs_inc_extent_ref;
2976 process_func = btrfs_free_extent;
2979 parent = buf->start;
2983 for (i = 0; i < nritems; i++) {
2985 btrfs_item_key_to_cpu(buf, &key, i);
2986 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2988 fi = btrfs_item_ptr(buf, i,
2989 struct btrfs_file_extent_item);
2990 if (btrfs_file_extent_type(buf, fi) ==
2991 BTRFS_FILE_EXTENT_INLINE)
2993 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2997 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2998 key.offset -= btrfs_file_extent_offset(buf, fi);
2999 ret = process_func(trans, root, bytenr, num_bytes,
3000 parent, ref_root, key.objectid,
3001 key.offset, for_cow);
3005 bytenr = btrfs_node_blockptr(buf, i);
3006 num_bytes = btrfs_level_size(root, level - 1);
3007 ret = process_func(trans, root, bytenr, num_bytes,
3008 parent, ref_root, level - 1, 0,
3019 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3020 struct extent_buffer *buf, int full_backref, int for_cow)
3022 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3025 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3026 struct extent_buffer *buf, int full_backref, int for_cow)
3028 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3031 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3032 struct btrfs_root *root,
3033 struct btrfs_path *path,
3034 struct btrfs_block_group_cache *cache)
3037 struct btrfs_root *extent_root = root->fs_info->extent_root;
3039 struct extent_buffer *leaf;
3041 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3044 BUG_ON(ret); /* Corruption */
3046 leaf = path->nodes[0];
3047 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3048 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3049 btrfs_mark_buffer_dirty(leaf);
3050 btrfs_release_path(path);
3053 btrfs_abort_transaction(trans, root, ret);
3060 static struct btrfs_block_group_cache *
3061 next_block_group(struct btrfs_root *root,
3062 struct btrfs_block_group_cache *cache)
3064 struct rb_node *node;
3065 spin_lock(&root->fs_info->block_group_cache_lock);
3066 node = rb_next(&cache->cache_node);
3067 btrfs_put_block_group(cache);
3069 cache = rb_entry(node, struct btrfs_block_group_cache,
3071 btrfs_get_block_group(cache);
3074 spin_unlock(&root->fs_info->block_group_cache_lock);
3078 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3079 struct btrfs_trans_handle *trans,
3080 struct btrfs_path *path)
3082 struct btrfs_root *root = block_group->fs_info->tree_root;
3083 struct inode *inode = NULL;
3085 int dcs = BTRFS_DC_ERROR;
3091 * If this block group is smaller than 100 megs don't bother caching the
3094 if (block_group->key.offset < (100 * 1024 * 1024)) {
3095 spin_lock(&block_group->lock);
3096 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3097 spin_unlock(&block_group->lock);
3102 inode = lookup_free_space_inode(root, block_group, path);
3103 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3104 ret = PTR_ERR(inode);
3105 btrfs_release_path(path);
3109 if (IS_ERR(inode)) {
3113 if (block_group->ro)
3116 ret = create_free_space_inode(root, trans, block_group, path);
3122 /* We've already setup this transaction, go ahead and exit */
3123 if (block_group->cache_generation == trans->transid &&
3124 i_size_read(inode)) {
3125 dcs = BTRFS_DC_SETUP;
3130 * We want to set the generation to 0, that way if anything goes wrong
3131 * from here on out we know not to trust this cache when we load up next
3134 BTRFS_I(inode)->generation = 0;
3135 ret = btrfs_update_inode(trans, root, inode);
3138 if (i_size_read(inode) > 0) {
3139 ret = btrfs_truncate_free_space_cache(root, trans, path,
3145 spin_lock(&block_group->lock);
3146 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3147 !btrfs_test_opt(root, SPACE_CACHE)) {
3149 * don't bother trying to write stuff out _if_
3150 * a) we're not cached,
3151 * b) we're with nospace_cache mount option.
3153 dcs = BTRFS_DC_WRITTEN;
3154 spin_unlock(&block_group->lock);
3157 spin_unlock(&block_group->lock);
3160 * Try to preallocate enough space based on how big the block group is.
3161 * Keep in mind this has to include any pinned space which could end up
3162 * taking up quite a bit since it's not folded into the other space
3165 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3170 num_pages *= PAGE_CACHE_SIZE;
3172 ret = btrfs_check_data_free_space(inode, num_pages);
3176 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3177 num_pages, num_pages,
3180 dcs = BTRFS_DC_SETUP;
3181 btrfs_free_reserved_data_space(inode, num_pages);
3186 btrfs_release_path(path);
3188 spin_lock(&block_group->lock);
3189 if (!ret && dcs == BTRFS_DC_SETUP)
3190 block_group->cache_generation = trans->transid;
3191 block_group->disk_cache_state = dcs;
3192 spin_unlock(&block_group->lock);
3197 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3198 struct btrfs_root *root)
3200 struct btrfs_block_group_cache *cache;
3202 struct btrfs_path *path;
3205 path = btrfs_alloc_path();
3211 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3213 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3215 cache = next_block_group(root, cache);
3223 err = cache_save_setup(cache, trans, path);
3224 last = cache->key.objectid + cache->key.offset;
3225 btrfs_put_block_group(cache);
3230 err = btrfs_run_delayed_refs(trans, root,
3232 if (err) /* File system offline */
3236 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3238 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3239 btrfs_put_block_group(cache);
3245 cache = next_block_group(root, cache);
3254 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3255 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3257 last = cache->key.objectid + cache->key.offset;
3259 err = write_one_cache_group(trans, root, path, cache);
3260 if (err) /* File system offline */
3263 btrfs_put_block_group(cache);
3268 * I don't think this is needed since we're just marking our
3269 * preallocated extent as written, but just in case it can't
3273 err = btrfs_run_delayed_refs(trans, root,
3275 if (err) /* File system offline */
3279 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3282 * Really this shouldn't happen, but it could if we
3283 * couldn't write the entire preallocated extent and
3284 * splitting the extent resulted in a new block.
3287 btrfs_put_block_group(cache);
3290 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3292 cache = next_block_group(root, cache);
3301 err = btrfs_write_out_cache(root, trans, cache, path);
3304 * If we didn't have an error then the cache state is still
3305 * NEED_WRITE, so we can set it to WRITTEN.
3307 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3308 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3309 last = cache->key.objectid + cache->key.offset;
3310 btrfs_put_block_group(cache);
3314 btrfs_free_path(path);
3318 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3320 struct btrfs_block_group_cache *block_group;
3323 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3324 if (!block_group || block_group->ro)
3327 btrfs_put_block_group(block_group);
3331 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3332 u64 total_bytes, u64 bytes_used,
3333 struct btrfs_space_info **space_info)
3335 struct btrfs_space_info *found;
3339 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3340 BTRFS_BLOCK_GROUP_RAID10))
3345 found = __find_space_info(info, flags);
3347 spin_lock(&found->lock);
3348 found->total_bytes += total_bytes;
3349 found->disk_total += total_bytes * factor;
3350 found->bytes_used += bytes_used;
3351 found->disk_used += bytes_used * factor;
3353 spin_unlock(&found->lock);
3354 *space_info = found;
3357 found = kzalloc(sizeof(*found), GFP_NOFS);
3361 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3362 INIT_LIST_HEAD(&found->block_groups[i]);
3363 init_rwsem(&found->groups_sem);
3364 spin_lock_init(&found->lock);
3365 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3366 found->total_bytes = total_bytes;
3367 found->disk_total = total_bytes * factor;
3368 found->bytes_used = bytes_used;
3369 found->disk_used = bytes_used * factor;
3370 found->bytes_pinned = 0;
3371 found->bytes_reserved = 0;
3372 found->bytes_readonly = 0;
3373 found->bytes_may_use = 0;
3375 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3376 found->chunk_alloc = 0;
3378 init_waitqueue_head(&found->wait);
3379 *space_info = found;
3380 list_add_rcu(&found->list, &info->space_info);
3381 if (flags & BTRFS_BLOCK_GROUP_DATA)
3382 info->data_sinfo = found;
3386 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3388 u64 extra_flags = chunk_to_extended(flags) &
3389 BTRFS_EXTENDED_PROFILE_MASK;
3391 write_seqlock(&fs_info->profiles_lock);
3392 if (flags & BTRFS_BLOCK_GROUP_DATA)
3393 fs_info->avail_data_alloc_bits |= extra_flags;
3394 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3395 fs_info->avail_metadata_alloc_bits |= extra_flags;
3396 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3397 fs_info->avail_system_alloc_bits |= extra_flags;
3398 write_sequnlock(&fs_info->profiles_lock);
3402 * returns target flags in extended format or 0 if restripe for this
3403 * chunk_type is not in progress
3405 * should be called with either volume_mutex or balance_lock held
3407 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3409 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3415 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3416 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3417 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3418 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3419 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3420 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3421 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3422 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3423 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3430 * @flags: available profiles in extended format (see ctree.h)
3432 * Returns reduced profile in chunk format. If profile changing is in
3433 * progress (either running or paused) picks the target profile (if it's
3434 * already available), otherwise falls back to plain reducing.
3436 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3439 * we add in the count of missing devices because we want
3440 * to make sure that any RAID levels on a degraded FS
3441 * continue to be honored.
3443 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3444 root->fs_info->fs_devices->missing_devices;
3449 * see if restripe for this chunk_type is in progress, if so
3450 * try to reduce to the target profile
3452 spin_lock(&root->fs_info->balance_lock);
3453 target = get_restripe_target(root->fs_info, flags);
3455 /* pick target profile only if it's already available */
3456 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3457 spin_unlock(&root->fs_info->balance_lock);
3458 return extended_to_chunk(target);
3461 spin_unlock(&root->fs_info->balance_lock);
3463 /* First, mask out the RAID levels which aren't possible */
3464 if (num_devices == 1)
3465 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3466 BTRFS_BLOCK_GROUP_RAID5);
3467 if (num_devices < 3)
3468 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3469 if (num_devices < 4)
3470 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3472 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3473 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3474 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3477 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3478 tmp = BTRFS_BLOCK_GROUP_RAID6;
3479 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3480 tmp = BTRFS_BLOCK_GROUP_RAID5;
3481 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3482 tmp = BTRFS_BLOCK_GROUP_RAID10;
3483 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3484 tmp = BTRFS_BLOCK_GROUP_RAID1;
3485 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3486 tmp = BTRFS_BLOCK_GROUP_RAID0;
3488 return extended_to_chunk(flags | tmp);
3491 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3496 seq = read_seqbegin(&root->fs_info->profiles_lock);
3498 if (flags & BTRFS_BLOCK_GROUP_DATA)
3499 flags |= root->fs_info->avail_data_alloc_bits;
3500 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3501 flags |= root->fs_info->avail_system_alloc_bits;
3502 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3503 flags |= root->fs_info->avail_metadata_alloc_bits;
3504 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3506 return btrfs_reduce_alloc_profile(root, flags);
3509 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3515 flags = BTRFS_BLOCK_GROUP_DATA;
3516 else if (root == root->fs_info->chunk_root)
3517 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3519 flags = BTRFS_BLOCK_GROUP_METADATA;
3521 ret = get_alloc_profile(root, flags);
3526 * This will check the space that the inode allocates from to make sure we have
3527 * enough space for bytes.
3529 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3531 struct btrfs_space_info *data_sinfo;
3532 struct btrfs_root *root = BTRFS_I(inode)->root;
3533 struct btrfs_fs_info *fs_info = root->fs_info;
3535 int ret = 0, committed = 0, alloc_chunk = 1;
3537 /* make sure bytes are sectorsize aligned */
3538 bytes = ALIGN(bytes, root->sectorsize);
3540 if (root == root->fs_info->tree_root ||
3541 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3546 data_sinfo = fs_info->data_sinfo;
3551 /* make sure we have enough space to handle the data first */
3552 spin_lock(&data_sinfo->lock);
3553 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3554 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3555 data_sinfo->bytes_may_use;
3557 if (used + bytes > data_sinfo->total_bytes) {
3558 struct btrfs_trans_handle *trans;
3561 * if we don't have enough free bytes in this space then we need
3562 * to alloc a new chunk.
3564 if (!data_sinfo->full && alloc_chunk) {
3567 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3568 spin_unlock(&data_sinfo->lock);
3570 alloc_target = btrfs_get_alloc_profile(root, 1);
3571 trans = btrfs_join_transaction(root);
3573 return PTR_ERR(trans);
3575 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3577 CHUNK_ALLOC_NO_FORCE);
3578 btrfs_end_transaction(trans, root);
3587 data_sinfo = fs_info->data_sinfo;
3593 * If we have less pinned bytes than we want to allocate then
3594 * don't bother committing the transaction, it won't help us.
3596 if (data_sinfo->bytes_pinned < bytes)
3598 spin_unlock(&data_sinfo->lock);
3600 /* commit the current transaction and try again */
3603 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3605 trans = btrfs_join_transaction(root);
3607 return PTR_ERR(trans);
3608 ret = btrfs_commit_transaction(trans, root);
3616 data_sinfo->bytes_may_use += bytes;
3617 trace_btrfs_space_reservation(root->fs_info, "space_info",
3618 data_sinfo->flags, bytes, 1);
3619 spin_unlock(&data_sinfo->lock);
3625 * Called if we need to clear a data reservation for this inode.
3627 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3629 struct btrfs_root *root = BTRFS_I(inode)->root;
3630 struct btrfs_space_info *data_sinfo;
3632 /* make sure bytes are sectorsize aligned */
3633 bytes = ALIGN(bytes, root->sectorsize);
3635 data_sinfo = root->fs_info->data_sinfo;
3636 spin_lock(&data_sinfo->lock);
3637 data_sinfo->bytes_may_use -= bytes;
3638 trace_btrfs_space_reservation(root->fs_info, "space_info",
3639 data_sinfo->flags, bytes, 0);
3640 spin_unlock(&data_sinfo->lock);
3643 static void force_metadata_allocation(struct btrfs_fs_info *info)
3645 struct list_head *head = &info->space_info;
3646 struct btrfs_space_info *found;
3649 list_for_each_entry_rcu(found, head, list) {
3650 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3651 found->force_alloc = CHUNK_ALLOC_FORCE;
3656 static int should_alloc_chunk(struct btrfs_root *root,
3657 struct btrfs_space_info *sinfo, int force)
3659 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3660 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3661 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3664 if (force == CHUNK_ALLOC_FORCE)
3668 * We need to take into account the global rsv because for all intents
3669 * and purposes it's used space. Don't worry about locking the
3670 * global_rsv, it doesn't change except when the transaction commits.
3672 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3673 num_allocated += global_rsv->size;
3676 * in limited mode, we want to have some free space up to
3677 * about 1% of the FS size.
3679 if (force == CHUNK_ALLOC_LIMITED) {
3680 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3681 thresh = max_t(u64, 64 * 1024 * 1024,
3682 div_factor_fine(thresh, 1));
3684 if (num_bytes - num_allocated < thresh)
3688 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3693 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3697 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3698 BTRFS_BLOCK_GROUP_RAID0 |
3699 BTRFS_BLOCK_GROUP_RAID5 |
3700 BTRFS_BLOCK_GROUP_RAID6))
3701 num_dev = root->fs_info->fs_devices->rw_devices;
3702 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3705 num_dev = 1; /* DUP or single */
3707 /* metadata for updaing devices and chunk tree */
3708 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3711 static void check_system_chunk(struct btrfs_trans_handle *trans,
3712 struct btrfs_root *root, u64 type)
3714 struct btrfs_space_info *info;
3718 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3719 spin_lock(&info->lock);
3720 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3721 info->bytes_reserved - info->bytes_readonly;
3722 spin_unlock(&info->lock);
3724 thresh = get_system_chunk_thresh(root, type);
3725 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3726 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3727 left, thresh, type);
3728 dump_space_info(info, 0, 0);
3731 if (left < thresh) {
3734 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3735 btrfs_alloc_chunk(trans, root, flags);
3739 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3740 struct btrfs_root *extent_root, u64 flags, int force)
3742 struct btrfs_space_info *space_info;
3743 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3744 int wait_for_alloc = 0;
3747 /* Don't re-enter if we're already allocating a chunk */
3748 if (trans->allocating_chunk)
3751 space_info = __find_space_info(extent_root->fs_info, flags);
3753 ret = update_space_info(extent_root->fs_info, flags,
3755 BUG_ON(ret); /* -ENOMEM */
3757 BUG_ON(!space_info); /* Logic error */
3760 spin_lock(&space_info->lock);
3761 if (force < space_info->force_alloc)
3762 force = space_info->force_alloc;
3763 if (space_info->full) {
3764 spin_unlock(&space_info->lock);
3768 if (!should_alloc_chunk(extent_root, space_info, force)) {
3769 spin_unlock(&space_info->lock);
3771 } else if (space_info->chunk_alloc) {
3774 space_info->chunk_alloc = 1;
3777 spin_unlock(&space_info->lock);
3779 mutex_lock(&fs_info->chunk_mutex);
3782 * The chunk_mutex is held throughout the entirety of a chunk
3783 * allocation, so once we've acquired the chunk_mutex we know that the
3784 * other guy is done and we need to recheck and see if we should
3787 if (wait_for_alloc) {
3788 mutex_unlock(&fs_info->chunk_mutex);
3793 trans->allocating_chunk = true;
3796 * If we have mixed data/metadata chunks we want to make sure we keep
3797 * allocating mixed chunks instead of individual chunks.
3799 if (btrfs_mixed_space_info(space_info))
3800 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3803 * if we're doing a data chunk, go ahead and make sure that
3804 * we keep a reasonable number of metadata chunks allocated in the
3807 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3808 fs_info->data_chunk_allocations++;
3809 if (!(fs_info->data_chunk_allocations %
3810 fs_info->metadata_ratio))
3811 force_metadata_allocation(fs_info);
3815 * Check if we have enough space in SYSTEM chunk because we may need
3816 * to update devices.
3818 check_system_chunk(trans, extent_root, flags);
3820 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3821 trans->allocating_chunk = false;
3823 spin_lock(&space_info->lock);
3824 if (ret < 0 && ret != -ENOSPC)
3827 space_info->full = 1;
3831 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3833 space_info->chunk_alloc = 0;
3834 spin_unlock(&space_info->lock);
3835 mutex_unlock(&fs_info->chunk_mutex);
3839 static int can_overcommit(struct btrfs_root *root,
3840 struct btrfs_space_info *space_info, u64 bytes,
3841 enum btrfs_reserve_flush_enum flush)
3843 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3844 u64 profile = btrfs_get_alloc_profile(root, 0);
3850 used = space_info->bytes_used + space_info->bytes_reserved +
3851 space_info->bytes_pinned + space_info->bytes_readonly;
3853 spin_lock(&global_rsv->lock);
3854 rsv_size = global_rsv->size;
3855 spin_unlock(&global_rsv->lock);
3858 * We only want to allow over committing if we have lots of actual space
3859 * free, but if we don't have enough space to handle the global reserve
3860 * space then we could end up having a real enospc problem when trying
3861 * to allocate a chunk or some other such important allocation.
3864 if (used + rsv_size >= space_info->total_bytes)
3867 used += space_info->bytes_may_use;
3869 spin_lock(&root->fs_info->free_chunk_lock);
3870 avail = root->fs_info->free_chunk_space;
3871 spin_unlock(&root->fs_info->free_chunk_lock);
3874 * If we have dup, raid1 or raid10 then only half of the free
3875 * space is actually useable. For raid56, the space info used
3876 * doesn't include the parity drive, so we don't have to
3879 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3880 BTRFS_BLOCK_GROUP_RAID1 |
3881 BTRFS_BLOCK_GROUP_RAID10))
3884 to_add = space_info->total_bytes;
3887 * If we aren't flushing all things, let us overcommit up to
3888 * 1/2th of the space. If we can flush, don't let us overcommit
3889 * too much, let it overcommit up to 1/8 of the space.
3891 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3897 * Limit the overcommit to the amount of free space we could possibly
3898 * allocate for chunks.
3900 to_add = min(avail, to_add);
3902 if (used + bytes < space_info->total_bytes + to_add)
3907 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3908 unsigned long nr_pages)
3910 struct super_block *sb = root->fs_info->sb;
3913 /* If we can not start writeback, just sync all the delalloc file. */
3914 started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
3915 WB_REASON_FS_FREE_SPACE);
3918 * We needn't worry the filesystem going from r/w to r/o though
3919 * we don't acquire ->s_umount mutex, because the filesystem
3920 * should guarantee the delalloc inodes list be empty after
3921 * the filesystem is readonly(all dirty pages are written to
3924 btrfs_start_delalloc_inodes(root, 0);
3925 btrfs_wait_ordered_extents(root, 0);
3930 * shrink metadata reservation for delalloc
3932 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3935 struct btrfs_block_rsv *block_rsv;
3936 struct btrfs_space_info *space_info;
3937 struct btrfs_trans_handle *trans;
3941 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3943 enum btrfs_reserve_flush_enum flush;
3945 trans = (struct btrfs_trans_handle *)current->journal_info;
3946 block_rsv = &root->fs_info->delalloc_block_rsv;
3947 space_info = block_rsv->space_info;
3950 delalloc_bytes = percpu_counter_sum_positive(
3951 &root->fs_info->delalloc_bytes);
3952 if (delalloc_bytes == 0) {
3955 btrfs_wait_ordered_extents(root, 0);
3959 while (delalloc_bytes && loops < 3) {
3960 max_reclaim = min(delalloc_bytes, to_reclaim);
3961 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3962 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3964 * We need to wait for the async pages to actually start before
3967 wait_event(root->fs_info->async_submit_wait,
3968 !atomic_read(&root->fs_info->async_delalloc_pages));
3971 flush = BTRFS_RESERVE_FLUSH_ALL;
3973 flush = BTRFS_RESERVE_NO_FLUSH;
3974 spin_lock(&space_info->lock);
3975 if (can_overcommit(root, space_info, orig, flush)) {
3976 spin_unlock(&space_info->lock);
3979 spin_unlock(&space_info->lock);
3982 if (wait_ordered && !trans) {
3983 btrfs_wait_ordered_extents(root, 0);
3985 time_left = schedule_timeout_killable(1);
3990 delalloc_bytes = percpu_counter_sum_positive(
3991 &root->fs_info->delalloc_bytes);
3996 * maybe_commit_transaction - possibly commit the transaction if its ok to
3997 * @root - the root we're allocating for
3998 * @bytes - the number of bytes we want to reserve
3999 * @force - force the commit
4001 * This will check to make sure that committing the transaction will actually
4002 * get us somewhere and then commit the transaction if it does. Otherwise it
4003 * will return -ENOSPC.
4005 static int may_commit_transaction(struct btrfs_root *root,
4006 struct btrfs_space_info *space_info,
4007 u64 bytes, int force)
4009 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4010 struct btrfs_trans_handle *trans;
4012 trans = (struct btrfs_trans_handle *)current->journal_info;
4019 /* See if there is enough pinned space to make this reservation */
4020 spin_lock(&space_info->lock);
4021 if (space_info->bytes_pinned >= bytes) {
4022 spin_unlock(&space_info->lock);
4025 spin_unlock(&space_info->lock);
4028 * See if there is some space in the delayed insertion reservation for
4031 if (space_info != delayed_rsv->space_info)
4034 spin_lock(&space_info->lock);
4035 spin_lock(&delayed_rsv->lock);
4036 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
4037 spin_unlock(&delayed_rsv->lock);
4038 spin_unlock(&space_info->lock);
4041 spin_unlock(&delayed_rsv->lock);
4042 spin_unlock(&space_info->lock);
4045 trans = btrfs_join_transaction(root);
4049 return btrfs_commit_transaction(trans, root);
4053 FLUSH_DELAYED_ITEMS_NR = 1,
4054 FLUSH_DELAYED_ITEMS = 2,
4056 FLUSH_DELALLOC_WAIT = 4,
4061 static int flush_space(struct btrfs_root *root,
4062 struct btrfs_space_info *space_info, u64 num_bytes,
4063 u64 orig_bytes, int state)
4065 struct btrfs_trans_handle *trans;
4070 case FLUSH_DELAYED_ITEMS_NR:
4071 case FLUSH_DELAYED_ITEMS:
4072 if (state == FLUSH_DELAYED_ITEMS_NR) {
4073 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4075 nr = (int)div64_u64(num_bytes, bytes);
4082 trans = btrfs_join_transaction(root);
4083 if (IS_ERR(trans)) {
4084 ret = PTR_ERR(trans);
4087 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4088 btrfs_end_transaction(trans, root);
4090 case FLUSH_DELALLOC:
4091 case FLUSH_DELALLOC_WAIT:
4092 shrink_delalloc(root, num_bytes, orig_bytes,
4093 state == FLUSH_DELALLOC_WAIT);
4096 trans = btrfs_join_transaction(root);
4097 if (IS_ERR(trans)) {
4098 ret = PTR_ERR(trans);
4101 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4102 btrfs_get_alloc_profile(root, 0),
4103 CHUNK_ALLOC_NO_FORCE);
4104 btrfs_end_transaction(trans, root);
4109 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4119 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4120 * @root - the root we're allocating for
4121 * @block_rsv - the block_rsv we're allocating for
4122 * @orig_bytes - the number of bytes we want
4123 * @flush - whether or not we can flush to make our reservation
4125 * This will reserve orgi_bytes number of bytes from the space info associated
4126 * with the block_rsv. If there is not enough space it will make an attempt to
4127 * flush out space to make room. It will do this by flushing delalloc if
4128 * possible or committing the transaction. If flush is 0 then no attempts to
4129 * regain reservations will be made and this will fail if there is not enough
4132 static int reserve_metadata_bytes(struct btrfs_root *root,
4133 struct btrfs_block_rsv *block_rsv,
4135 enum btrfs_reserve_flush_enum flush)
4137 struct btrfs_space_info *space_info = block_rsv->space_info;
4139 u64 num_bytes = orig_bytes;
4140 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4142 bool flushing = false;
4146 spin_lock(&space_info->lock);
4148 * We only want to wait if somebody other than us is flushing and we
4149 * are actually allowed to flush all things.
4151 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4152 space_info->flush) {
4153 spin_unlock(&space_info->lock);
4155 * If we have a trans handle we can't wait because the flusher
4156 * may have to commit the transaction, which would mean we would
4157 * deadlock since we are waiting for the flusher to finish, but
4158 * hold the current transaction open.
4160 if (current->journal_info)
4162 ret = wait_event_killable(space_info->wait, !space_info->flush);
4163 /* Must have been killed, return */
4167 spin_lock(&space_info->lock);
4171 used = space_info->bytes_used + space_info->bytes_reserved +
4172 space_info->bytes_pinned + space_info->bytes_readonly +
4173 space_info->bytes_may_use;
4176 * The idea here is that we've not already over-reserved the block group
4177 * then we can go ahead and save our reservation first and then start
4178 * flushing if we need to. Otherwise if we've already overcommitted
4179 * lets start flushing stuff first and then come back and try to make
4182 if (used <= space_info->total_bytes) {
4183 if (used + orig_bytes <= space_info->total_bytes) {
4184 space_info->bytes_may_use += orig_bytes;
4185 trace_btrfs_space_reservation(root->fs_info,
4186 "space_info", space_info->flags, orig_bytes, 1);
4190 * Ok set num_bytes to orig_bytes since we aren't
4191 * overocmmitted, this way we only try and reclaim what
4194 num_bytes = orig_bytes;
4198 * Ok we're over committed, set num_bytes to the overcommitted
4199 * amount plus the amount of bytes that we need for this
4202 num_bytes = used - space_info->total_bytes +
4206 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4207 space_info->bytes_may_use += orig_bytes;
4208 trace_btrfs_space_reservation(root->fs_info, "space_info",
4209 space_info->flags, orig_bytes,
4215 * Couldn't make our reservation, save our place so while we're trying
4216 * to reclaim space we can actually use it instead of somebody else
4217 * stealing it from us.
4219 * We make the other tasks wait for the flush only when we can flush
4222 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4224 space_info->flush = 1;
4227 spin_unlock(&space_info->lock);
4229 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4232 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4237 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4238 * would happen. So skip delalloc flush.
4240 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4241 (flush_state == FLUSH_DELALLOC ||
4242 flush_state == FLUSH_DELALLOC_WAIT))
4243 flush_state = ALLOC_CHUNK;
4247 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4248 flush_state < COMMIT_TRANS)
4250 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4251 flush_state <= COMMIT_TRANS)
4255 if (ret == -ENOSPC &&
4256 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4257 struct btrfs_block_rsv *global_rsv =
4258 &root->fs_info->global_block_rsv;
4260 if (block_rsv != global_rsv &&
4261 !block_rsv_use_bytes(global_rsv, orig_bytes))
4265 spin_lock(&space_info->lock);
4266 space_info->flush = 0;
4267 wake_up_all(&space_info->wait);
4268 spin_unlock(&space_info->lock);
4273 static struct btrfs_block_rsv *get_block_rsv(
4274 const struct btrfs_trans_handle *trans,
4275 const struct btrfs_root *root)
4277 struct btrfs_block_rsv *block_rsv = NULL;
4280 block_rsv = trans->block_rsv;
4282 if (root == root->fs_info->csum_root && trans->adding_csums)
4283 block_rsv = trans->block_rsv;
4286 block_rsv = root->block_rsv;
4289 block_rsv = &root->fs_info->empty_block_rsv;
4294 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4298 spin_lock(&block_rsv->lock);
4299 if (block_rsv->reserved >= num_bytes) {
4300 block_rsv->reserved -= num_bytes;
4301 if (block_rsv->reserved < block_rsv->size)
4302 block_rsv->full = 0;
4305 spin_unlock(&block_rsv->lock);
4309 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4310 u64 num_bytes, int update_size)
4312 spin_lock(&block_rsv->lock);
4313 block_rsv->reserved += num_bytes;
4315 block_rsv->size += num_bytes;
4316 else if (block_rsv->reserved >= block_rsv->size)
4317 block_rsv->full = 1;
4318 spin_unlock(&block_rsv->lock);
4321 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4322 struct btrfs_block_rsv *block_rsv,
4323 struct btrfs_block_rsv *dest, u64 num_bytes)
4325 struct btrfs_space_info *space_info = block_rsv->space_info;
4327 spin_lock(&block_rsv->lock);
4328 if (num_bytes == (u64)-1)
4329 num_bytes = block_rsv->size;
4330 block_rsv->size -= num_bytes;
4331 if (block_rsv->reserved >= block_rsv->size) {
4332 num_bytes = block_rsv->reserved - block_rsv->size;
4333 block_rsv->reserved = block_rsv->size;
4334 block_rsv->full = 1;
4338 spin_unlock(&block_rsv->lock);
4340 if (num_bytes > 0) {
4342 spin_lock(&dest->lock);
4346 bytes_to_add = dest->size - dest->reserved;
4347 bytes_to_add = min(num_bytes, bytes_to_add);
4348 dest->reserved += bytes_to_add;
4349 if (dest->reserved >= dest->size)
4351 num_bytes -= bytes_to_add;
4353 spin_unlock(&dest->lock);
4356 spin_lock(&space_info->lock);
4357 space_info->bytes_may_use -= num_bytes;
4358 trace_btrfs_space_reservation(fs_info, "space_info",
4359 space_info->flags, num_bytes, 0);
4360 space_info->reservation_progress++;
4361 spin_unlock(&space_info->lock);
4366 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4367 struct btrfs_block_rsv *dst, u64 num_bytes)
4371 ret = block_rsv_use_bytes(src, num_bytes);
4375 block_rsv_add_bytes(dst, num_bytes, 1);
4379 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4381 memset(rsv, 0, sizeof(*rsv));
4382 spin_lock_init(&rsv->lock);
4386 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4387 unsigned short type)
4389 struct btrfs_block_rsv *block_rsv;
4390 struct btrfs_fs_info *fs_info = root->fs_info;
4392 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4396 btrfs_init_block_rsv(block_rsv, type);
4397 block_rsv->space_info = __find_space_info(fs_info,
4398 BTRFS_BLOCK_GROUP_METADATA);
4402 void btrfs_free_block_rsv(struct btrfs_root *root,
4403 struct btrfs_block_rsv *rsv)
4407 btrfs_block_rsv_release(root, rsv, (u64)-1);
4411 int btrfs_block_rsv_add(struct btrfs_root *root,
4412 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4413 enum btrfs_reserve_flush_enum flush)
4420 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4422 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4429 int btrfs_block_rsv_check(struct btrfs_root *root,
4430 struct btrfs_block_rsv *block_rsv, int min_factor)
4438 spin_lock(&block_rsv->lock);
4439 num_bytes = div_factor(block_rsv->size, min_factor);
4440 if (block_rsv->reserved >= num_bytes)
4442 spin_unlock(&block_rsv->lock);
4447 int btrfs_block_rsv_refill(struct btrfs_root *root,
4448 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4449 enum btrfs_reserve_flush_enum flush)
4457 spin_lock(&block_rsv->lock);
4458 num_bytes = min_reserved;
4459 if (block_rsv->reserved >= num_bytes)
4462 num_bytes -= block_rsv->reserved;
4463 spin_unlock(&block_rsv->lock);
4468 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4470 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4477 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4478 struct btrfs_block_rsv *dst_rsv,
4481 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4484 void btrfs_block_rsv_release(struct btrfs_root *root,
4485 struct btrfs_block_rsv *block_rsv,
4488 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4489 if (global_rsv->full || global_rsv == block_rsv ||
4490 block_rsv->space_info != global_rsv->space_info)
4492 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4497 * helper to calculate size of global block reservation.
4498 * the desired value is sum of space used by extent tree,
4499 * checksum tree and root tree
4501 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4503 struct btrfs_space_info *sinfo;
4507 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4509 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4510 spin_lock(&sinfo->lock);
4511 data_used = sinfo->bytes_used;
4512 spin_unlock(&sinfo->lock);
4514 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4515 spin_lock(&sinfo->lock);
4516 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4518 meta_used = sinfo->bytes_used;
4519 spin_unlock(&sinfo->lock);
4521 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4523 num_bytes += div64_u64(data_used + meta_used, 50);
4525 if (num_bytes * 3 > meta_used)
4526 num_bytes = div64_u64(meta_used, 3);
4528 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4531 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4533 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4534 struct btrfs_space_info *sinfo = block_rsv->space_info;
4537 num_bytes = calc_global_metadata_size(fs_info);
4539 spin_lock(&sinfo->lock);
4540 spin_lock(&block_rsv->lock);
4542 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4544 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4545 sinfo->bytes_reserved + sinfo->bytes_readonly +
4546 sinfo->bytes_may_use;
4548 if (sinfo->total_bytes > num_bytes) {
4549 num_bytes = sinfo->total_bytes - num_bytes;
4550 block_rsv->reserved += num_bytes;
4551 sinfo->bytes_may_use += num_bytes;
4552 trace_btrfs_space_reservation(fs_info, "space_info",
4553 sinfo->flags, num_bytes, 1);
4556 if (block_rsv->reserved >= block_rsv->size) {
4557 num_bytes = block_rsv->reserved - block_rsv->size;
4558 sinfo->bytes_may_use -= num_bytes;
4559 trace_btrfs_space_reservation(fs_info, "space_info",
4560 sinfo->flags, num_bytes, 0);
4561 sinfo->reservation_progress++;
4562 block_rsv->reserved = block_rsv->size;
4563 block_rsv->full = 1;
4566 spin_unlock(&block_rsv->lock);
4567 spin_unlock(&sinfo->lock);
4570 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4572 struct btrfs_space_info *space_info;
4574 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4575 fs_info->chunk_block_rsv.space_info = space_info;
4577 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4578 fs_info->global_block_rsv.space_info = space_info;
4579 fs_info->delalloc_block_rsv.space_info = space_info;
4580 fs_info->trans_block_rsv.space_info = space_info;
4581 fs_info->empty_block_rsv.space_info = space_info;
4582 fs_info->delayed_block_rsv.space_info = space_info;
4584 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4585 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4586 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4587 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4588 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4590 update_global_block_rsv(fs_info);
4593 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4595 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4597 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4598 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4599 WARN_ON(fs_info->trans_block_rsv.size > 0);
4600 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4601 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4602 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4603 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4604 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4607 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4608 struct btrfs_root *root)
4610 if (!trans->block_rsv)
4613 if (!trans->bytes_reserved)
4616 trace_btrfs_space_reservation(root->fs_info, "transaction",
4617 trans->transid, trans->bytes_reserved, 0);
4618 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4619 trans->bytes_reserved = 0;
4622 /* Can only return 0 or -ENOSPC */
4623 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4624 struct inode *inode)
4626 struct btrfs_root *root = BTRFS_I(inode)->root;
4627 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4628 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4631 * We need to hold space in order to delete our orphan item once we've
4632 * added it, so this takes the reservation so we can release it later
4633 * when we are truly done with the orphan item.
4635 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4636 trace_btrfs_space_reservation(root->fs_info, "orphan",
4637 btrfs_ino(inode), num_bytes, 1);
4638 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4641 void btrfs_orphan_release_metadata(struct inode *inode)
4643 struct btrfs_root *root = BTRFS_I(inode)->root;
4644 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4645 trace_btrfs_space_reservation(root->fs_info, "orphan",
4646 btrfs_ino(inode), num_bytes, 0);
4647 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4651 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4652 * root: the root of the parent directory
4653 * rsv: block reservation
4654 * items: the number of items that we need do reservation
4655 * qgroup_reserved: used to return the reserved size in qgroup
4657 * This function is used to reserve the space for snapshot/subvolume
4658 * creation and deletion. Those operations are different with the
4659 * common file/directory operations, they change two fs/file trees
4660 * and root tree, the number of items that the qgroup reserves is
4661 * different with the free space reservation. So we can not use
4662 * the space reseravtion mechanism in start_transaction().
4664 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4665 struct btrfs_block_rsv *rsv,
4667 u64 *qgroup_reserved)
4672 if (root->fs_info->quota_enabled) {
4673 /* One for parent inode, two for dir entries */
4674 num_bytes = 3 * root->leafsize;
4675 ret = btrfs_qgroup_reserve(root, num_bytes);
4682 *qgroup_reserved = num_bytes;
4684 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4685 rsv->space_info = __find_space_info(root->fs_info,
4686 BTRFS_BLOCK_GROUP_METADATA);
4687 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4688 BTRFS_RESERVE_FLUSH_ALL);
4690 if (*qgroup_reserved)
4691 btrfs_qgroup_free(root, *qgroup_reserved);
4697 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4698 struct btrfs_block_rsv *rsv,
4699 u64 qgroup_reserved)
4701 btrfs_block_rsv_release(root, rsv, (u64)-1);
4702 if (qgroup_reserved)
4703 btrfs_qgroup_free(root, qgroup_reserved);
4707 * drop_outstanding_extent - drop an outstanding extent
4708 * @inode: the inode we're dropping the extent for
4710 * This is called when we are freeing up an outstanding extent, either called
4711 * after an error or after an extent is written. This will return the number of
4712 * reserved extents that need to be freed. This must be called with
4713 * BTRFS_I(inode)->lock held.
4715 static unsigned drop_outstanding_extent(struct inode *inode)
4717 unsigned drop_inode_space = 0;
4718 unsigned dropped_extents = 0;
4720 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4721 BTRFS_I(inode)->outstanding_extents--;
4723 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4724 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4725 &BTRFS_I(inode)->runtime_flags))
4726 drop_inode_space = 1;
4729 * If we have more or the same amount of outsanding extents than we have
4730 * reserved then we need to leave the reserved extents count alone.
4732 if (BTRFS_I(inode)->outstanding_extents >=
4733 BTRFS_I(inode)->reserved_extents)
4734 return drop_inode_space;
4736 dropped_extents = BTRFS_I(inode)->reserved_extents -
4737 BTRFS_I(inode)->outstanding_extents;
4738 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4739 return dropped_extents + drop_inode_space;
4743 * calc_csum_metadata_size - return the amount of metada space that must be
4744 * reserved/free'd for the given bytes.
4745 * @inode: the inode we're manipulating
4746 * @num_bytes: the number of bytes in question
4747 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4749 * This adjusts the number of csum_bytes in the inode and then returns the
4750 * correct amount of metadata that must either be reserved or freed. We
4751 * calculate how many checksums we can fit into one leaf and then divide the
4752 * number of bytes that will need to be checksumed by this value to figure out
4753 * how many checksums will be required. If we are adding bytes then the number
4754 * may go up and we will return the number of additional bytes that must be
4755 * reserved. If it is going down we will return the number of bytes that must
4758 * This must be called with BTRFS_I(inode)->lock held.
4760 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4763 struct btrfs_root *root = BTRFS_I(inode)->root;
4765 int num_csums_per_leaf;
4769 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4770 BTRFS_I(inode)->csum_bytes == 0)
4773 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4775 BTRFS_I(inode)->csum_bytes += num_bytes;
4777 BTRFS_I(inode)->csum_bytes -= num_bytes;
4778 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4779 num_csums_per_leaf = (int)div64_u64(csum_size,
4780 sizeof(struct btrfs_csum_item) +
4781 sizeof(struct btrfs_disk_key));
4782 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4783 num_csums = num_csums + num_csums_per_leaf - 1;
4784 num_csums = num_csums / num_csums_per_leaf;
4786 old_csums = old_csums + num_csums_per_leaf - 1;
4787 old_csums = old_csums / num_csums_per_leaf;
4789 /* No change, no need to reserve more */
4790 if (old_csums == num_csums)
4794 return btrfs_calc_trans_metadata_size(root,
4795 num_csums - old_csums);
4797 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4800 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4802 struct btrfs_root *root = BTRFS_I(inode)->root;
4803 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4806 unsigned nr_extents = 0;
4807 int extra_reserve = 0;
4808 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4810 bool delalloc_lock = true;
4814 /* If we are a free space inode we need to not flush since we will be in
4815 * the middle of a transaction commit. We also don't need the delalloc
4816 * mutex since we won't race with anybody. We need this mostly to make
4817 * lockdep shut its filthy mouth.
4819 if (btrfs_is_free_space_inode(inode)) {
4820 flush = BTRFS_RESERVE_NO_FLUSH;
4821 delalloc_lock = false;
4824 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4825 btrfs_transaction_in_commit(root->fs_info))
4826 schedule_timeout(1);
4829 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4831 num_bytes = ALIGN(num_bytes, root->sectorsize);
4833 spin_lock(&BTRFS_I(inode)->lock);
4834 BTRFS_I(inode)->outstanding_extents++;
4836 if (BTRFS_I(inode)->outstanding_extents >
4837 BTRFS_I(inode)->reserved_extents)
4838 nr_extents = BTRFS_I(inode)->outstanding_extents -
4839 BTRFS_I(inode)->reserved_extents;
4842 * Add an item to reserve for updating the inode when we complete the
4845 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4846 &BTRFS_I(inode)->runtime_flags)) {
4851 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4852 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4853 csum_bytes = BTRFS_I(inode)->csum_bytes;
4854 spin_unlock(&BTRFS_I(inode)->lock);
4856 if (root->fs_info->quota_enabled) {
4857 ret = btrfs_qgroup_reserve(root, num_bytes +
4858 nr_extents * root->leafsize);
4863 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4864 if (unlikely(ret)) {
4865 if (root->fs_info->quota_enabled)
4866 btrfs_qgroup_free(root, num_bytes +
4867 nr_extents * root->leafsize);
4871 spin_lock(&BTRFS_I(inode)->lock);
4872 if (extra_reserve) {
4873 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4874 &BTRFS_I(inode)->runtime_flags);
4877 BTRFS_I(inode)->reserved_extents += nr_extents;
4878 spin_unlock(&BTRFS_I(inode)->lock);
4881 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4884 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4885 btrfs_ino(inode), to_reserve, 1);
4886 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4891 spin_lock(&BTRFS_I(inode)->lock);
4892 dropped = drop_outstanding_extent(inode);
4894 * If the inodes csum_bytes is the same as the original
4895 * csum_bytes then we know we haven't raced with any free()ers
4896 * so we can just reduce our inodes csum bytes and carry on.
4898 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4899 calc_csum_metadata_size(inode, num_bytes, 0);
4901 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4905 * This is tricky, but first we need to figure out how much we
4906 * free'd from any free-ers that occured during this
4907 * reservation, so we reset ->csum_bytes to the csum_bytes
4908 * before we dropped our lock, and then call the free for the
4909 * number of bytes that were freed while we were trying our
4912 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4913 BTRFS_I(inode)->csum_bytes = csum_bytes;
4914 to_free = calc_csum_metadata_size(inode, bytes, 0);
4918 * Now we need to see how much we would have freed had we not
4919 * been making this reservation and our ->csum_bytes were not
4920 * artificially inflated.
4922 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4923 bytes = csum_bytes - orig_csum_bytes;
4924 bytes = calc_csum_metadata_size(inode, bytes, 0);
4927 * Now reset ->csum_bytes to what it should be. If bytes is
4928 * more than to_free then we would have free'd more space had we
4929 * not had an artificially high ->csum_bytes, so we need to free
4930 * the remainder. If bytes is the same or less then we don't
4931 * need to do anything, the other free-ers did the correct
4934 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4935 if (bytes > to_free)
4936 to_free = bytes - to_free;
4940 spin_unlock(&BTRFS_I(inode)->lock);
4942 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4945 btrfs_block_rsv_release(root, block_rsv, to_free);
4946 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4947 btrfs_ino(inode), to_free, 0);
4950 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4955 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4956 * @inode: the inode to release the reservation for
4957 * @num_bytes: the number of bytes we're releasing
4959 * This will release the metadata reservation for an inode. This can be called
4960 * once we complete IO for a given set of bytes to release their metadata
4963 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4965 struct btrfs_root *root = BTRFS_I(inode)->root;
4969 num_bytes = ALIGN(num_bytes, root->sectorsize);
4970 spin_lock(&BTRFS_I(inode)->lock);
4971 dropped = drop_outstanding_extent(inode);
4974 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4975 spin_unlock(&BTRFS_I(inode)->lock);
4977 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4979 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4980 btrfs_ino(inode), to_free, 0);
4981 if (root->fs_info->quota_enabled) {
4982 btrfs_qgroup_free(root, num_bytes +
4983 dropped * root->leafsize);
4986 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4991 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4992 * @inode: inode we're writing to
4993 * @num_bytes: the number of bytes we want to allocate
4995 * This will do the following things
4997 * o reserve space in the data space info for num_bytes
4998 * o reserve space in the metadata space info based on number of outstanding
4999 * extents and how much csums will be needed
5000 * o add to the inodes ->delalloc_bytes
5001 * o add it to the fs_info's delalloc inodes list.
5003 * This will return 0 for success and -ENOSPC if there is no space left.
5005 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5009 ret = btrfs_check_data_free_space(inode, num_bytes);
5013 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5015 btrfs_free_reserved_data_space(inode, num_bytes);
5023 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5024 * @inode: inode we're releasing space for
5025 * @num_bytes: the number of bytes we want to free up
5027 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5028 * called in the case that we don't need the metadata AND data reservations
5029 * anymore. So if there is an error or we insert an inline extent.
5031 * This function will release the metadata space that was not used and will
5032 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5033 * list if there are no delalloc bytes left.
5035 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5037 btrfs_delalloc_release_metadata(inode, num_bytes);
5038 btrfs_free_reserved_data_space(inode, num_bytes);
5041 static int update_block_group(struct btrfs_root *root,
5042 u64 bytenr, u64 num_bytes, int alloc)
5044 struct btrfs_block_group_cache *cache = NULL;
5045 struct btrfs_fs_info *info = root->fs_info;
5046 u64 total = num_bytes;
5051 /* block accounting for super block */
5052 spin_lock(&info->delalloc_lock);
5053 old_val = btrfs_super_bytes_used(info->super_copy);
5055 old_val += num_bytes;
5057 old_val -= num_bytes;
5058 btrfs_set_super_bytes_used(info->super_copy, old_val);
5059 spin_unlock(&info->delalloc_lock);
5062 cache = btrfs_lookup_block_group(info, bytenr);
5065 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5066 BTRFS_BLOCK_GROUP_RAID1 |
5067 BTRFS_BLOCK_GROUP_RAID10))
5072 * If this block group has free space cache written out, we
5073 * need to make sure to load it if we are removing space. This
5074 * is because we need the unpinning stage to actually add the
5075 * space back to the block group, otherwise we will leak space.
5077 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5078 cache_block_group(cache, 1);
5080 byte_in_group = bytenr - cache->key.objectid;
5081 WARN_ON(byte_in_group > cache->key.offset);
5083 spin_lock(&cache->space_info->lock);
5084 spin_lock(&cache->lock);
5086 if (btrfs_test_opt(root, SPACE_CACHE) &&
5087 cache->disk_cache_state < BTRFS_DC_CLEAR)
5088 cache->disk_cache_state = BTRFS_DC_CLEAR;
5091 old_val = btrfs_block_group_used(&cache->item);
5092 num_bytes = min(total, cache->key.offset - byte_in_group);
5094 old_val += num_bytes;
5095 btrfs_set_block_group_used(&cache->item, old_val);
5096 cache->reserved -= num_bytes;
5097 cache->space_info->bytes_reserved -= num_bytes;
5098 cache->space_info->bytes_used += num_bytes;
5099 cache->space_info->disk_used += num_bytes * factor;
5100 spin_unlock(&cache->lock);
5101 spin_unlock(&cache->space_info->lock);
5103 old_val -= num_bytes;
5104 btrfs_set_block_group_used(&cache->item, old_val);
5105 cache->pinned += num_bytes;
5106 cache->space_info->bytes_pinned += num_bytes;
5107 cache->space_info->bytes_used -= num_bytes;
5108 cache->space_info->disk_used -= num_bytes * factor;
5109 spin_unlock(&cache->lock);
5110 spin_unlock(&cache->space_info->lock);
5112 set_extent_dirty(info->pinned_extents,
5113 bytenr, bytenr + num_bytes - 1,
5114 GFP_NOFS | __GFP_NOFAIL);
5116 btrfs_put_block_group(cache);
5118 bytenr += num_bytes;
5123 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5125 struct btrfs_block_group_cache *cache;
5128 spin_lock(&root->fs_info->block_group_cache_lock);
5129 bytenr = root->fs_info->first_logical_byte;
5130 spin_unlock(&root->fs_info->block_group_cache_lock);
5132 if (bytenr < (u64)-1)
5135 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5139 bytenr = cache->key.objectid;
5140 btrfs_put_block_group(cache);
5145 static int pin_down_extent(struct btrfs_root *root,
5146 struct btrfs_block_group_cache *cache,
5147 u64 bytenr, u64 num_bytes, int reserved)
5149 spin_lock(&cache->space_info->lock);
5150 spin_lock(&cache->lock);
5151 cache->pinned += num_bytes;
5152 cache->space_info->bytes_pinned += num_bytes;
5154 cache->reserved -= num_bytes;
5155 cache->space_info->bytes_reserved -= num_bytes;
5157 spin_unlock(&cache->lock);
5158 spin_unlock(&cache->space_info->lock);
5160 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5161 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5166 * this function must be called within transaction
5168 int btrfs_pin_extent(struct btrfs_root *root,
5169 u64 bytenr, u64 num_bytes, int reserved)
5171 struct btrfs_block_group_cache *cache;
5173 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5174 BUG_ON(!cache); /* Logic error */
5176 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5178 btrfs_put_block_group(cache);
5183 * this function must be called within transaction
5185 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5186 u64 bytenr, u64 num_bytes)
5188 struct btrfs_block_group_cache *cache;
5190 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5191 BUG_ON(!cache); /* Logic error */
5194 * pull in the free space cache (if any) so that our pin
5195 * removes the free space from the cache. We have load_only set
5196 * to one because the slow code to read in the free extents does check
5197 * the pinned extents.
5199 cache_block_group(cache, 1);
5201 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5203 /* remove us from the free space cache (if we're there at all) */
5204 btrfs_remove_free_space(cache, bytenr, num_bytes);
5205 btrfs_put_block_group(cache);
5210 * btrfs_update_reserved_bytes - update the block_group and space info counters
5211 * @cache: The cache we are manipulating
5212 * @num_bytes: The number of bytes in question
5213 * @reserve: One of the reservation enums
5215 * This is called by the allocator when it reserves space, or by somebody who is
5216 * freeing space that was never actually used on disk. For example if you
5217 * reserve some space for a new leaf in transaction A and before transaction A
5218 * commits you free that leaf, you call this with reserve set to 0 in order to
5219 * clear the reservation.
5221 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5222 * ENOSPC accounting. For data we handle the reservation through clearing the
5223 * delalloc bits in the io_tree. We have to do this since we could end up
5224 * allocating less disk space for the amount of data we have reserved in the
5225 * case of compression.
5227 * If this is a reservation and the block group has become read only we cannot
5228 * make the reservation and return -EAGAIN, otherwise this function always
5231 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5232 u64 num_bytes, int reserve)
5234 struct btrfs_space_info *space_info = cache->space_info;
5237 spin_lock(&space_info->lock);
5238 spin_lock(&cache->lock);
5239 if (reserve != RESERVE_FREE) {
5243 cache->reserved += num_bytes;
5244 space_info->bytes_reserved += num_bytes;
5245 if (reserve == RESERVE_ALLOC) {
5246 trace_btrfs_space_reservation(cache->fs_info,
5247 "space_info", space_info->flags,
5249 space_info->bytes_may_use -= num_bytes;
5254 space_info->bytes_readonly += num_bytes;
5255 cache->reserved -= num_bytes;
5256 space_info->bytes_reserved -= num_bytes;
5257 space_info->reservation_progress++;
5259 spin_unlock(&cache->lock);
5260 spin_unlock(&space_info->lock);
5264 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5265 struct btrfs_root *root)
5267 struct btrfs_fs_info *fs_info = root->fs_info;
5268 struct btrfs_caching_control *next;
5269 struct btrfs_caching_control *caching_ctl;
5270 struct btrfs_block_group_cache *cache;
5272 down_write(&fs_info->extent_commit_sem);
5274 list_for_each_entry_safe(caching_ctl, next,
5275 &fs_info->caching_block_groups, list) {
5276 cache = caching_ctl->block_group;
5277 if (block_group_cache_done(cache)) {
5278 cache->last_byte_to_unpin = (u64)-1;
5279 list_del_init(&caching_ctl->list);
5280 put_caching_control(caching_ctl);
5282 cache->last_byte_to_unpin = caching_ctl->progress;
5286 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5287 fs_info->pinned_extents = &fs_info->freed_extents[1];
5289 fs_info->pinned_extents = &fs_info->freed_extents[0];
5291 up_write(&fs_info->extent_commit_sem);
5293 update_global_block_rsv(fs_info);
5296 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5298 struct btrfs_fs_info *fs_info = root->fs_info;
5299 struct btrfs_block_group_cache *cache = NULL;
5300 struct btrfs_space_info *space_info;
5301 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5305 while (start <= end) {
5308 start >= cache->key.objectid + cache->key.offset) {
5310 btrfs_put_block_group(cache);
5311 cache = btrfs_lookup_block_group(fs_info, start);
5312 BUG_ON(!cache); /* Logic error */
5315 len = cache->key.objectid + cache->key.offset - start;
5316 len = min(len, end + 1 - start);
5318 if (start < cache->last_byte_to_unpin) {
5319 len = min(len, cache->last_byte_to_unpin - start);
5320 btrfs_add_free_space(cache, start, len);
5324 space_info = cache->space_info;
5326 spin_lock(&space_info->lock);
5327 spin_lock(&cache->lock);
5328 cache->pinned -= len;
5329 space_info->bytes_pinned -= len;
5331 space_info->bytes_readonly += len;
5334 spin_unlock(&cache->lock);
5335 if (!readonly && global_rsv->space_info == space_info) {
5336 spin_lock(&global_rsv->lock);
5337 if (!global_rsv->full) {
5338 len = min(len, global_rsv->size -
5339 global_rsv->reserved);
5340 global_rsv->reserved += len;
5341 space_info->bytes_may_use += len;
5342 if (global_rsv->reserved >= global_rsv->size)
5343 global_rsv->full = 1;
5345 spin_unlock(&global_rsv->lock);
5347 spin_unlock(&space_info->lock);
5351 btrfs_put_block_group(cache);
5355 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5356 struct btrfs_root *root)
5358 struct btrfs_fs_info *fs_info = root->fs_info;
5359 struct extent_io_tree *unpin;
5367 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5368 unpin = &fs_info->freed_extents[1];
5370 unpin = &fs_info->freed_extents[0];
5373 ret = find_first_extent_bit(unpin, 0, &start, &end,
5374 EXTENT_DIRTY, NULL);
5378 if (btrfs_test_opt(root, DISCARD))
5379 ret = btrfs_discard_extent(root, start,
5380 end + 1 - start, NULL);
5382 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5383 unpin_extent_range(root, start, end);
5390 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5391 struct btrfs_root *root,
5392 u64 bytenr, u64 num_bytes, u64 parent,
5393 u64 root_objectid, u64 owner_objectid,
5394 u64 owner_offset, int refs_to_drop,
5395 struct btrfs_delayed_extent_op *extent_op)
5397 struct btrfs_key key;
5398 struct btrfs_path *path;
5399 struct btrfs_fs_info *info = root->fs_info;
5400 struct btrfs_root *extent_root = info->extent_root;
5401 struct extent_buffer *leaf;
5402 struct btrfs_extent_item *ei;
5403 struct btrfs_extent_inline_ref *iref;
5406 int extent_slot = 0;
5407 int found_extent = 0;
5411 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5414 path = btrfs_alloc_path();
5419 path->leave_spinning = 1;
5421 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5422 BUG_ON(!is_data && refs_to_drop != 1);
5425 skinny_metadata = 0;
5427 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5428 bytenr, num_bytes, parent,
5429 root_objectid, owner_objectid,
5432 extent_slot = path->slots[0];
5433 while (extent_slot >= 0) {
5434 btrfs_item_key_to_cpu(path->nodes[0], &key,
5436 if (key.objectid != bytenr)
5438 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5439 key.offset == num_bytes) {
5443 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5444 key.offset == owner_objectid) {
5448 if (path->slots[0] - extent_slot > 5)
5452 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5453 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5454 if (found_extent && item_size < sizeof(*ei))
5457 if (!found_extent) {
5459 ret = remove_extent_backref(trans, extent_root, path,
5463 btrfs_abort_transaction(trans, extent_root, ret);
5466 btrfs_release_path(path);
5467 path->leave_spinning = 1;
5469 key.objectid = bytenr;
5470 key.type = BTRFS_EXTENT_ITEM_KEY;
5471 key.offset = num_bytes;
5473 if (!is_data && skinny_metadata) {
5474 key.type = BTRFS_METADATA_ITEM_KEY;
5475 key.offset = owner_objectid;
5478 ret = btrfs_search_slot(trans, extent_root,
5480 if (ret > 0 && skinny_metadata && path->slots[0]) {
5482 * Couldn't find our skinny metadata item,
5483 * see if we have ye olde extent item.
5486 btrfs_item_key_to_cpu(path->nodes[0], &key,
5488 if (key.objectid == bytenr &&
5489 key.type == BTRFS_EXTENT_ITEM_KEY &&
5490 key.offset == num_bytes)
5494 if (ret > 0 && skinny_metadata) {
5495 skinny_metadata = false;
5496 key.type = BTRFS_EXTENT_ITEM_KEY;
5497 key.offset = num_bytes;
5498 btrfs_release_path(path);
5499 ret = btrfs_search_slot(trans, extent_root,
5504 printk(KERN_ERR "umm, got %d back from search"
5505 ", was looking for %llu\n", ret,
5506 (unsigned long long)bytenr);
5508 btrfs_print_leaf(extent_root,
5512 btrfs_abort_transaction(trans, extent_root, ret);
5515 extent_slot = path->slots[0];
5517 } else if (ret == -ENOENT) {
5518 btrfs_print_leaf(extent_root, path->nodes[0]);
5520 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5521 "parent %llu root %llu owner %llu offset %llu\n",
5522 (unsigned long long)bytenr,
5523 (unsigned long long)parent,
5524 (unsigned long long)root_objectid,
5525 (unsigned long long)owner_objectid,
5526 (unsigned long long)owner_offset);
5528 btrfs_abort_transaction(trans, extent_root, ret);
5532 leaf = path->nodes[0];
5533 item_size = btrfs_item_size_nr(leaf, extent_slot);
5534 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5535 if (item_size < sizeof(*ei)) {
5536 BUG_ON(found_extent || extent_slot != path->slots[0]);
5537 ret = convert_extent_item_v0(trans, extent_root, path,
5540 btrfs_abort_transaction(trans, extent_root, ret);
5544 btrfs_release_path(path);
5545 path->leave_spinning = 1;
5547 key.objectid = bytenr;
5548 key.type = BTRFS_EXTENT_ITEM_KEY;
5549 key.offset = num_bytes;
5551 ret = btrfs_search_slot(trans, extent_root, &key, path,
5554 printk(KERN_ERR "umm, got %d back from search"
5555 ", was looking for %llu\n", ret,
5556 (unsigned long long)bytenr);
5557 btrfs_print_leaf(extent_root, path->nodes[0]);
5560 btrfs_abort_transaction(trans, extent_root, ret);
5564 extent_slot = path->slots[0];
5565 leaf = path->nodes[0];
5566 item_size = btrfs_item_size_nr(leaf, extent_slot);
5569 BUG_ON(item_size < sizeof(*ei));
5570 ei = btrfs_item_ptr(leaf, extent_slot,
5571 struct btrfs_extent_item);
5572 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5573 key.type == BTRFS_EXTENT_ITEM_KEY) {
5574 struct btrfs_tree_block_info *bi;
5575 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5576 bi = (struct btrfs_tree_block_info *)(ei + 1);
5577 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5580 refs = btrfs_extent_refs(leaf, ei);
5581 BUG_ON(refs < refs_to_drop);
5582 refs -= refs_to_drop;
5586 __run_delayed_extent_op(extent_op, leaf, ei);
5588 * In the case of inline back ref, reference count will
5589 * be updated by remove_extent_backref
5592 BUG_ON(!found_extent);
5594 btrfs_set_extent_refs(leaf, ei, refs);
5595 btrfs_mark_buffer_dirty(leaf);
5598 ret = remove_extent_backref(trans, extent_root, path,
5602 btrfs_abort_transaction(trans, extent_root, ret);
5608 BUG_ON(is_data && refs_to_drop !=
5609 extent_data_ref_count(root, path, iref));
5611 BUG_ON(path->slots[0] != extent_slot);
5613 BUG_ON(path->slots[0] != extent_slot + 1);
5614 path->slots[0] = extent_slot;
5619 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5622 btrfs_abort_transaction(trans, extent_root, ret);
5625 btrfs_release_path(path);
5628 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5630 btrfs_abort_transaction(trans, extent_root, ret);
5635 ret = update_block_group(root, bytenr, num_bytes, 0);
5637 btrfs_abort_transaction(trans, extent_root, ret);
5642 btrfs_free_path(path);
5647 * when we free an block, it is possible (and likely) that we free the last
5648 * delayed ref for that extent as well. This searches the delayed ref tree for
5649 * a given extent, and if there are no other delayed refs to be processed, it
5650 * removes it from the tree.
5652 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5653 struct btrfs_root *root, u64 bytenr)
5655 struct btrfs_delayed_ref_head *head;
5656 struct btrfs_delayed_ref_root *delayed_refs;
5657 struct btrfs_delayed_ref_node *ref;
5658 struct rb_node *node;
5661 delayed_refs = &trans->transaction->delayed_refs;
5662 spin_lock(&delayed_refs->lock);
5663 head = btrfs_find_delayed_ref_head(trans, bytenr);
5667 node = rb_prev(&head->node.rb_node);
5671 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5673 /* there are still entries for this ref, we can't drop it */
5674 if (ref->bytenr == bytenr)
5677 if (head->extent_op) {
5678 if (!head->must_insert_reserved)
5680 btrfs_free_delayed_extent_op(head->extent_op);
5681 head->extent_op = NULL;
5685 * waiting for the lock here would deadlock. If someone else has it
5686 * locked they are already in the process of dropping it anyway
5688 if (!mutex_trylock(&head->mutex))
5692 * at this point we have a head with no other entries. Go
5693 * ahead and process it.
5695 head->node.in_tree = 0;
5696 rb_erase(&head->node.rb_node, &delayed_refs->root);
5698 delayed_refs->num_entries--;
5701 * we don't take a ref on the node because we're removing it from the
5702 * tree, so we just steal the ref the tree was holding.
5704 delayed_refs->num_heads--;
5705 if (list_empty(&head->cluster))
5706 delayed_refs->num_heads_ready--;
5708 list_del_init(&head->cluster);
5709 spin_unlock(&delayed_refs->lock);
5711 BUG_ON(head->extent_op);
5712 if (head->must_insert_reserved)
5715 mutex_unlock(&head->mutex);
5716 btrfs_put_delayed_ref(&head->node);
5719 spin_unlock(&delayed_refs->lock);
5723 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5724 struct btrfs_root *root,
5725 struct extent_buffer *buf,
5726 u64 parent, int last_ref)
5728 struct btrfs_block_group_cache *cache = NULL;
5731 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5732 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5733 buf->start, buf->len,
5734 parent, root->root_key.objectid,
5735 btrfs_header_level(buf),
5736 BTRFS_DROP_DELAYED_REF, NULL, 0);
5737 BUG_ON(ret); /* -ENOMEM */
5743 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5745 if (btrfs_header_generation(buf) == trans->transid) {
5746 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5747 ret = check_ref_cleanup(trans, root, buf->start);
5752 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5753 pin_down_extent(root, cache, buf->start, buf->len, 1);
5757 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5759 btrfs_add_free_space(cache, buf->start, buf->len);
5760 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5764 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5767 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5768 btrfs_put_block_group(cache);
5771 /* Can return -ENOMEM */
5772 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5773 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5774 u64 owner, u64 offset, int for_cow)
5777 struct btrfs_fs_info *fs_info = root->fs_info;
5780 * tree log blocks never actually go into the extent allocation
5781 * tree, just update pinning info and exit early.
5783 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5784 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5785 /* unlocks the pinned mutex */
5786 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5788 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5789 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5791 parent, root_objectid, (int)owner,
5792 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5794 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5796 parent, root_objectid, owner,
5797 offset, BTRFS_DROP_DELAYED_REF,
5803 static u64 stripe_align(struct btrfs_root *root,
5804 struct btrfs_block_group_cache *cache,
5805 u64 val, u64 num_bytes)
5807 u64 ret = ALIGN(val, root->stripesize);
5812 * when we wait for progress in the block group caching, its because
5813 * our allocation attempt failed at least once. So, we must sleep
5814 * and let some progress happen before we try again.
5816 * This function will sleep at least once waiting for new free space to
5817 * show up, and then it will check the block group free space numbers
5818 * for our min num_bytes. Another option is to have it go ahead
5819 * and look in the rbtree for a free extent of a given size, but this
5823 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5826 struct btrfs_caching_control *caching_ctl;
5828 caching_ctl = get_caching_control(cache);
5832 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5833 (cache->free_space_ctl->free_space >= num_bytes));
5835 put_caching_control(caching_ctl);
5840 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5842 struct btrfs_caching_control *caching_ctl;
5844 caching_ctl = get_caching_control(cache);
5848 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5850 put_caching_control(caching_ctl);
5854 int __get_raid_index(u64 flags)
5856 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5857 return BTRFS_RAID_RAID10;
5858 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5859 return BTRFS_RAID_RAID1;
5860 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5861 return BTRFS_RAID_DUP;
5862 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5863 return BTRFS_RAID_RAID0;
5864 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5865 return BTRFS_RAID_RAID5;
5866 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5867 return BTRFS_RAID_RAID6;
5869 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
5872 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5874 return __get_raid_index(cache->flags);
5877 enum btrfs_loop_type {
5878 LOOP_CACHING_NOWAIT = 0,
5879 LOOP_CACHING_WAIT = 1,
5880 LOOP_ALLOC_CHUNK = 2,
5881 LOOP_NO_EMPTY_SIZE = 3,
5885 * walks the btree of allocated extents and find a hole of a given size.
5886 * The key ins is changed to record the hole:
5887 * ins->objectid == block start
5888 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5889 * ins->offset == number of blocks
5890 * Any available blocks before search_start are skipped.
5892 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5893 struct btrfs_root *orig_root,
5894 u64 num_bytes, u64 empty_size,
5895 u64 hint_byte, struct btrfs_key *ins,
5899 struct btrfs_root *root = orig_root->fs_info->extent_root;
5900 struct btrfs_free_cluster *last_ptr = NULL;
5901 struct btrfs_block_group_cache *block_group = NULL;
5902 struct btrfs_block_group_cache *used_block_group;
5903 u64 search_start = 0;
5904 int empty_cluster = 2 * 1024 * 1024;
5905 struct btrfs_space_info *space_info;
5907 int index = __get_raid_index(data);
5908 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5909 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5910 bool found_uncached_bg = false;
5911 bool failed_cluster_refill = false;
5912 bool failed_alloc = false;
5913 bool use_cluster = true;
5914 bool have_caching_bg = false;
5916 WARN_ON(num_bytes < root->sectorsize);
5917 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5921 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5923 space_info = __find_space_info(root->fs_info, data);
5925 printk(KERN_ERR "No space info for %llu\n", data);
5930 * If the space info is for both data and metadata it means we have a
5931 * small filesystem and we can't use the clustering stuff.
5933 if (btrfs_mixed_space_info(space_info))
5934 use_cluster = false;
5936 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5937 last_ptr = &root->fs_info->meta_alloc_cluster;
5938 if (!btrfs_test_opt(root, SSD))
5939 empty_cluster = 64 * 1024;
5942 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5943 btrfs_test_opt(root, SSD)) {
5944 last_ptr = &root->fs_info->data_alloc_cluster;
5948 spin_lock(&last_ptr->lock);
5949 if (last_ptr->block_group)
5950 hint_byte = last_ptr->window_start;
5951 spin_unlock(&last_ptr->lock);
5954 search_start = max(search_start, first_logical_byte(root, 0));
5955 search_start = max(search_start, hint_byte);
5960 if (search_start == hint_byte) {
5961 block_group = btrfs_lookup_block_group(root->fs_info,
5963 used_block_group = block_group;
5965 * we don't want to use the block group if it doesn't match our
5966 * allocation bits, or if its not cached.
5968 * However if we are re-searching with an ideal block group
5969 * picked out then we don't care that the block group is cached.
5971 if (block_group && block_group_bits(block_group, data) &&
5972 block_group->cached != BTRFS_CACHE_NO) {
5973 down_read(&space_info->groups_sem);
5974 if (list_empty(&block_group->list) ||
5977 * someone is removing this block group,
5978 * we can't jump into the have_block_group
5979 * target because our list pointers are not
5982 btrfs_put_block_group(block_group);
5983 up_read(&space_info->groups_sem);
5985 index = get_block_group_index(block_group);
5986 goto have_block_group;
5988 } else if (block_group) {
5989 btrfs_put_block_group(block_group);
5993 have_caching_bg = false;
5994 down_read(&space_info->groups_sem);
5995 list_for_each_entry(block_group, &space_info->block_groups[index],
6000 used_block_group = block_group;
6001 btrfs_get_block_group(block_group);
6002 search_start = block_group->key.objectid;
6005 * this can happen if we end up cycling through all the
6006 * raid types, but we want to make sure we only allocate
6007 * for the proper type.
6009 if (!block_group_bits(block_group, data)) {
6010 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6011 BTRFS_BLOCK_GROUP_RAID1 |
6012 BTRFS_BLOCK_GROUP_RAID5 |
6013 BTRFS_BLOCK_GROUP_RAID6 |
6014 BTRFS_BLOCK_GROUP_RAID10;
6017 * if they asked for extra copies and this block group
6018 * doesn't provide them, bail. This does allow us to
6019 * fill raid0 from raid1.
6021 if ((data & extra) && !(block_group->flags & extra))
6026 cached = block_group_cache_done(block_group);
6027 if (unlikely(!cached)) {
6028 found_uncached_bg = true;
6029 ret = cache_block_group(block_group, 0);
6034 if (unlikely(block_group->ro))
6038 * Ok we want to try and use the cluster allocator, so
6042 unsigned long aligned_cluster;
6044 * the refill lock keeps out other
6045 * people trying to start a new cluster
6047 spin_lock(&last_ptr->refill_lock);
6048 used_block_group = last_ptr->block_group;
6049 if (used_block_group != block_group &&
6050 (!used_block_group ||
6051 used_block_group->ro ||
6052 !block_group_bits(used_block_group, data))) {
6053 used_block_group = block_group;
6054 goto refill_cluster;
6057 if (used_block_group != block_group)
6058 btrfs_get_block_group(used_block_group);
6060 offset = btrfs_alloc_from_cluster(used_block_group,
6061 last_ptr, num_bytes, used_block_group->key.objectid);
6063 /* we have a block, we're done */
6064 spin_unlock(&last_ptr->refill_lock);
6065 trace_btrfs_reserve_extent_cluster(root,
6066 block_group, search_start, num_bytes);
6070 WARN_ON(last_ptr->block_group != used_block_group);
6071 if (used_block_group != block_group) {
6072 btrfs_put_block_group(used_block_group);
6073 used_block_group = block_group;
6076 BUG_ON(used_block_group != block_group);
6077 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6078 * set up a new clusters, so lets just skip it
6079 * and let the allocator find whatever block
6080 * it can find. If we reach this point, we
6081 * will have tried the cluster allocator
6082 * plenty of times and not have found
6083 * anything, so we are likely way too
6084 * fragmented for the clustering stuff to find
6087 * However, if the cluster is taken from the
6088 * current block group, release the cluster
6089 * first, so that we stand a better chance of
6090 * succeeding in the unclustered
6092 if (loop >= LOOP_NO_EMPTY_SIZE &&
6093 last_ptr->block_group != block_group) {
6094 spin_unlock(&last_ptr->refill_lock);
6095 goto unclustered_alloc;
6099 * this cluster didn't work out, free it and
6102 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6104 if (loop >= LOOP_NO_EMPTY_SIZE) {
6105 spin_unlock(&last_ptr->refill_lock);
6106 goto unclustered_alloc;
6109 aligned_cluster = max_t(unsigned long,
6110 empty_cluster + empty_size,
6111 block_group->full_stripe_len);
6113 /* allocate a cluster in this block group */
6114 ret = btrfs_find_space_cluster(trans, root,
6115 block_group, last_ptr,
6116 search_start, num_bytes,
6120 * now pull our allocation out of this
6123 offset = btrfs_alloc_from_cluster(block_group,
6124 last_ptr, num_bytes,
6127 /* we found one, proceed */
6128 spin_unlock(&last_ptr->refill_lock);
6129 trace_btrfs_reserve_extent_cluster(root,
6130 block_group, search_start,
6134 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6135 && !failed_cluster_refill) {
6136 spin_unlock(&last_ptr->refill_lock);
6138 failed_cluster_refill = true;
6139 wait_block_group_cache_progress(block_group,
6140 num_bytes + empty_cluster + empty_size);
6141 goto have_block_group;
6145 * at this point we either didn't find a cluster
6146 * or we weren't able to allocate a block from our
6147 * cluster. Free the cluster we've been trying
6148 * to use, and go to the next block group
6150 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6151 spin_unlock(&last_ptr->refill_lock);
6156 spin_lock(&block_group->free_space_ctl->tree_lock);
6158 block_group->free_space_ctl->free_space <
6159 num_bytes + empty_cluster + empty_size) {
6160 spin_unlock(&block_group->free_space_ctl->tree_lock);
6163 spin_unlock(&block_group->free_space_ctl->tree_lock);
6165 offset = btrfs_find_space_for_alloc(block_group, search_start,
6166 num_bytes, empty_size);
6168 * If we didn't find a chunk, and we haven't failed on this
6169 * block group before, and this block group is in the middle of
6170 * caching and we are ok with waiting, then go ahead and wait
6171 * for progress to be made, and set failed_alloc to true.
6173 * If failed_alloc is true then we've already waited on this
6174 * block group once and should move on to the next block group.
6176 if (!offset && !failed_alloc && !cached &&
6177 loop > LOOP_CACHING_NOWAIT) {
6178 wait_block_group_cache_progress(block_group,
6179 num_bytes + empty_size);
6180 failed_alloc = true;
6181 goto have_block_group;
6182 } else if (!offset) {
6184 have_caching_bg = true;
6188 search_start = stripe_align(root, used_block_group,
6191 /* move on to the next group */
6192 if (search_start + num_bytes >
6193 used_block_group->key.objectid + used_block_group->key.offset) {
6194 btrfs_add_free_space(used_block_group, offset, num_bytes);
6198 if (offset < search_start)
6199 btrfs_add_free_space(used_block_group, offset,
6200 search_start - offset);
6201 BUG_ON(offset > search_start);
6203 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6205 if (ret == -EAGAIN) {
6206 btrfs_add_free_space(used_block_group, offset, num_bytes);
6210 /* we are all good, lets return */
6211 ins->objectid = search_start;
6212 ins->offset = num_bytes;
6214 trace_btrfs_reserve_extent(orig_root, block_group,
6215 search_start, num_bytes);
6216 if (used_block_group != block_group)
6217 btrfs_put_block_group(used_block_group);
6218 btrfs_put_block_group(block_group);
6221 failed_cluster_refill = false;
6222 failed_alloc = false;
6223 BUG_ON(index != get_block_group_index(block_group));
6224 if (used_block_group != block_group)
6225 btrfs_put_block_group(used_block_group);
6226 btrfs_put_block_group(block_group);
6228 up_read(&space_info->groups_sem);
6230 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6233 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6237 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6238 * caching kthreads as we move along
6239 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6240 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6241 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6244 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6247 if (loop == LOOP_ALLOC_CHUNK) {
6248 ret = do_chunk_alloc(trans, root, data,
6251 * Do not bail out on ENOSPC since we
6252 * can do more things.
6254 if (ret < 0 && ret != -ENOSPC) {
6255 btrfs_abort_transaction(trans,
6261 if (loop == LOOP_NO_EMPTY_SIZE) {
6267 } else if (!ins->objectid) {
6269 } else if (ins->objectid) {
6277 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6278 int dump_block_groups)
6280 struct btrfs_block_group_cache *cache;
6283 spin_lock(&info->lock);
6284 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6285 (unsigned long long)info->flags,
6286 (unsigned long long)(info->total_bytes - info->bytes_used -
6287 info->bytes_pinned - info->bytes_reserved -
6288 info->bytes_readonly),
6289 (info->full) ? "" : "not ");
6290 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6291 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6292 (unsigned long long)info->total_bytes,
6293 (unsigned long long)info->bytes_used,
6294 (unsigned long long)info->bytes_pinned,
6295 (unsigned long long)info->bytes_reserved,
6296 (unsigned long long)info->bytes_may_use,
6297 (unsigned long long)info->bytes_readonly);
6298 spin_unlock(&info->lock);
6300 if (!dump_block_groups)
6303 down_read(&info->groups_sem);
6305 list_for_each_entry(cache, &info->block_groups[index], list) {
6306 spin_lock(&cache->lock);
6307 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6308 (unsigned long long)cache->key.objectid,
6309 (unsigned long long)cache->key.offset,
6310 (unsigned long long)btrfs_block_group_used(&cache->item),
6311 (unsigned long long)cache->pinned,
6312 (unsigned long long)cache->reserved,
6313 cache->ro ? "[readonly]" : "");
6314 btrfs_dump_free_space(cache, bytes);
6315 spin_unlock(&cache->lock);
6317 if (++index < BTRFS_NR_RAID_TYPES)
6319 up_read(&info->groups_sem);
6322 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6323 struct btrfs_root *root,
6324 u64 num_bytes, u64 min_alloc_size,
6325 u64 empty_size, u64 hint_byte,
6326 struct btrfs_key *ins, u64 data)
6328 bool final_tried = false;
6331 data = btrfs_get_alloc_profile(root, data);
6333 WARN_ON(num_bytes < root->sectorsize);
6334 ret = find_free_extent(trans, root, num_bytes, empty_size,
6335 hint_byte, ins, data);
6337 if (ret == -ENOSPC) {
6339 num_bytes = num_bytes >> 1;
6340 num_bytes = round_down(num_bytes, root->sectorsize);
6341 num_bytes = max(num_bytes, min_alloc_size);
6342 if (num_bytes == min_alloc_size)
6345 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6346 struct btrfs_space_info *sinfo;
6348 sinfo = __find_space_info(root->fs_info, data);
6349 printk(KERN_ERR "btrfs allocation failed flags %llu, "
6350 "wanted %llu\n", (unsigned long long)data,
6351 (unsigned long long)num_bytes);
6353 dump_space_info(sinfo, num_bytes, 1);
6357 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6362 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6363 u64 start, u64 len, int pin)
6365 struct btrfs_block_group_cache *cache;
6368 cache = btrfs_lookup_block_group(root->fs_info, start);
6370 printk(KERN_ERR "Unable to find block group for %llu\n",
6371 (unsigned long long)start);
6375 if (btrfs_test_opt(root, DISCARD))
6376 ret = btrfs_discard_extent(root, start, len, NULL);
6379 pin_down_extent(root, cache, start, len, 1);
6381 btrfs_add_free_space(cache, start, len);
6382 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6384 btrfs_put_block_group(cache);
6386 trace_btrfs_reserved_extent_free(root, start, len);
6391 int btrfs_free_reserved_extent(struct btrfs_root *root,
6394 return __btrfs_free_reserved_extent(root, start, len, 0);
6397 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6400 return __btrfs_free_reserved_extent(root, start, len, 1);
6403 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6404 struct btrfs_root *root,
6405 u64 parent, u64 root_objectid,
6406 u64 flags, u64 owner, u64 offset,
6407 struct btrfs_key *ins, int ref_mod)
6410 struct btrfs_fs_info *fs_info = root->fs_info;
6411 struct btrfs_extent_item *extent_item;
6412 struct btrfs_extent_inline_ref *iref;
6413 struct btrfs_path *path;
6414 struct extent_buffer *leaf;
6419 type = BTRFS_SHARED_DATA_REF_KEY;
6421 type = BTRFS_EXTENT_DATA_REF_KEY;
6423 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6425 path = btrfs_alloc_path();
6429 path->leave_spinning = 1;
6430 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6433 btrfs_free_path(path);
6437 leaf = path->nodes[0];
6438 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6439 struct btrfs_extent_item);
6440 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6441 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6442 btrfs_set_extent_flags(leaf, extent_item,
6443 flags | BTRFS_EXTENT_FLAG_DATA);
6445 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6446 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6448 struct btrfs_shared_data_ref *ref;
6449 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6450 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6451 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6453 struct btrfs_extent_data_ref *ref;
6454 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6455 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6456 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6457 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6458 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6461 btrfs_mark_buffer_dirty(path->nodes[0]);
6462 btrfs_free_path(path);
6464 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6465 if (ret) { /* -ENOENT, logic error */
6466 printk(KERN_ERR "btrfs update block group failed for %llu "
6467 "%llu\n", (unsigned long long)ins->objectid,
6468 (unsigned long long)ins->offset);
6474 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6475 struct btrfs_root *root,
6476 u64 parent, u64 root_objectid,
6477 u64 flags, struct btrfs_disk_key *key,
6478 int level, struct btrfs_key *ins)
6481 struct btrfs_fs_info *fs_info = root->fs_info;
6482 struct btrfs_extent_item *extent_item;
6483 struct btrfs_tree_block_info *block_info;
6484 struct btrfs_extent_inline_ref *iref;
6485 struct btrfs_path *path;
6486 struct extent_buffer *leaf;
6487 u32 size = sizeof(*extent_item) + sizeof(*iref);
6488 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6491 if (!skinny_metadata)
6492 size += sizeof(*block_info);
6494 path = btrfs_alloc_path();
6498 path->leave_spinning = 1;
6499 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6502 btrfs_free_path(path);
6506 leaf = path->nodes[0];
6507 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6508 struct btrfs_extent_item);
6509 btrfs_set_extent_refs(leaf, extent_item, 1);
6510 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6511 btrfs_set_extent_flags(leaf, extent_item,
6512 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6514 if (skinny_metadata) {
6515 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6517 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6518 btrfs_set_tree_block_key(leaf, block_info, key);
6519 btrfs_set_tree_block_level(leaf, block_info, level);
6520 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6524 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6525 btrfs_set_extent_inline_ref_type(leaf, iref,
6526 BTRFS_SHARED_BLOCK_REF_KEY);
6527 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6529 btrfs_set_extent_inline_ref_type(leaf, iref,
6530 BTRFS_TREE_BLOCK_REF_KEY);
6531 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6534 btrfs_mark_buffer_dirty(leaf);
6535 btrfs_free_path(path);
6537 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6538 if (ret) { /* -ENOENT, logic error */
6539 printk(KERN_ERR "btrfs update block group failed for %llu "
6540 "%llu\n", (unsigned long long)ins->objectid,
6541 (unsigned long long)ins->offset);
6547 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6548 struct btrfs_root *root,
6549 u64 root_objectid, u64 owner,
6550 u64 offset, struct btrfs_key *ins)
6554 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6556 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6558 root_objectid, owner, offset,
6559 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6564 * this is used by the tree logging recovery code. It records that
6565 * an extent has been allocated and makes sure to clear the free
6566 * space cache bits as well
6568 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6569 struct btrfs_root *root,
6570 u64 root_objectid, u64 owner, u64 offset,
6571 struct btrfs_key *ins)
6574 struct btrfs_block_group_cache *block_group;
6575 struct btrfs_caching_control *caching_ctl;
6576 u64 start = ins->objectid;
6577 u64 num_bytes = ins->offset;
6579 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6580 cache_block_group(block_group, 0);
6581 caching_ctl = get_caching_control(block_group);
6584 BUG_ON(!block_group_cache_done(block_group));
6585 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6586 BUG_ON(ret); /* -ENOMEM */
6588 mutex_lock(&caching_ctl->mutex);
6590 if (start >= caching_ctl->progress) {
6591 ret = add_excluded_extent(root, start, num_bytes);
6592 BUG_ON(ret); /* -ENOMEM */
6593 } else if (start + num_bytes <= caching_ctl->progress) {
6594 ret = btrfs_remove_free_space(block_group,
6596 BUG_ON(ret); /* -ENOMEM */
6598 num_bytes = caching_ctl->progress - start;
6599 ret = btrfs_remove_free_space(block_group,
6601 BUG_ON(ret); /* -ENOMEM */
6603 start = caching_ctl->progress;
6604 num_bytes = ins->objectid + ins->offset -
6605 caching_ctl->progress;
6606 ret = add_excluded_extent(root, start, num_bytes);
6607 BUG_ON(ret); /* -ENOMEM */
6610 mutex_unlock(&caching_ctl->mutex);
6611 put_caching_control(caching_ctl);
6614 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6615 RESERVE_ALLOC_NO_ACCOUNT);
6616 BUG_ON(ret); /* logic error */
6617 btrfs_put_block_group(block_group);
6618 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6619 0, owner, offset, ins, 1);
6623 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6624 struct btrfs_root *root,
6625 u64 bytenr, u32 blocksize,
6628 struct extent_buffer *buf;
6630 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6632 return ERR_PTR(-ENOMEM);
6633 btrfs_set_header_generation(buf, trans->transid);
6634 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6635 btrfs_tree_lock(buf);
6636 clean_tree_block(trans, root, buf);
6637 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6639 btrfs_set_lock_blocking(buf);
6640 btrfs_set_buffer_uptodate(buf);
6642 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6644 * we allow two log transactions at a time, use different
6645 * EXENT bit to differentiate dirty pages.
6647 if (root->log_transid % 2 == 0)
6648 set_extent_dirty(&root->dirty_log_pages, buf->start,
6649 buf->start + buf->len - 1, GFP_NOFS);
6651 set_extent_new(&root->dirty_log_pages, buf->start,
6652 buf->start + buf->len - 1, GFP_NOFS);
6654 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6655 buf->start + buf->len - 1, GFP_NOFS);
6657 trans->blocks_used++;
6658 /* this returns a buffer locked for blocking */
6662 static struct btrfs_block_rsv *
6663 use_block_rsv(struct btrfs_trans_handle *trans,
6664 struct btrfs_root *root, u32 blocksize)
6666 struct btrfs_block_rsv *block_rsv;
6667 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6670 block_rsv = get_block_rsv(trans, root);
6672 if (block_rsv->size == 0) {
6673 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6674 BTRFS_RESERVE_NO_FLUSH);
6676 * If we couldn't reserve metadata bytes try and use some from
6677 * the global reserve.
6679 if (ret && block_rsv != global_rsv) {
6680 ret = block_rsv_use_bytes(global_rsv, blocksize);
6683 return ERR_PTR(ret);
6685 return ERR_PTR(ret);
6690 ret = block_rsv_use_bytes(block_rsv, blocksize);
6693 if (ret && !block_rsv->failfast) {
6694 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6695 static DEFINE_RATELIMIT_STATE(_rs,
6696 DEFAULT_RATELIMIT_INTERVAL * 10,
6697 /*DEFAULT_RATELIMIT_BURST*/ 1);
6698 if (__ratelimit(&_rs))
6700 "btrfs: block rsv returned %d\n", ret);
6702 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6703 BTRFS_RESERVE_NO_FLUSH);
6706 } else if (ret && block_rsv != global_rsv) {
6707 ret = block_rsv_use_bytes(global_rsv, blocksize);
6713 return ERR_PTR(-ENOSPC);
6716 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6717 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6719 block_rsv_add_bytes(block_rsv, blocksize, 0);
6720 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6724 * finds a free extent and does all the dirty work required for allocation
6725 * returns the key for the extent through ins, and a tree buffer for
6726 * the first block of the extent through buf.
6728 * returns the tree buffer or NULL.
6730 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6731 struct btrfs_root *root, u32 blocksize,
6732 u64 parent, u64 root_objectid,
6733 struct btrfs_disk_key *key, int level,
6734 u64 hint, u64 empty_size)
6736 struct btrfs_key ins;
6737 struct btrfs_block_rsv *block_rsv;
6738 struct extent_buffer *buf;
6741 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6744 block_rsv = use_block_rsv(trans, root, blocksize);
6745 if (IS_ERR(block_rsv))
6746 return ERR_CAST(block_rsv);
6748 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6749 empty_size, hint, &ins, 0);
6751 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6752 return ERR_PTR(ret);
6755 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6757 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6759 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6761 parent = ins.objectid;
6762 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6766 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6767 struct btrfs_delayed_extent_op *extent_op;
6768 extent_op = btrfs_alloc_delayed_extent_op();
6769 BUG_ON(!extent_op); /* -ENOMEM */
6771 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6773 memset(&extent_op->key, 0, sizeof(extent_op->key));
6774 extent_op->flags_to_set = flags;
6775 if (skinny_metadata)
6776 extent_op->update_key = 0;
6778 extent_op->update_key = 1;
6779 extent_op->update_flags = 1;
6780 extent_op->is_data = 0;
6782 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6784 ins.offset, parent, root_objectid,
6785 level, BTRFS_ADD_DELAYED_EXTENT,
6787 BUG_ON(ret); /* -ENOMEM */
6792 struct walk_control {
6793 u64 refs[BTRFS_MAX_LEVEL];
6794 u64 flags[BTRFS_MAX_LEVEL];
6795 struct btrfs_key update_progress;
6806 #define DROP_REFERENCE 1
6807 #define UPDATE_BACKREF 2
6809 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6810 struct btrfs_root *root,
6811 struct walk_control *wc,
6812 struct btrfs_path *path)
6820 struct btrfs_key key;
6821 struct extent_buffer *eb;
6826 if (path->slots[wc->level] < wc->reada_slot) {
6827 wc->reada_count = wc->reada_count * 2 / 3;
6828 wc->reada_count = max(wc->reada_count, 2);
6830 wc->reada_count = wc->reada_count * 3 / 2;
6831 wc->reada_count = min_t(int, wc->reada_count,
6832 BTRFS_NODEPTRS_PER_BLOCK(root));
6835 eb = path->nodes[wc->level];
6836 nritems = btrfs_header_nritems(eb);
6837 blocksize = btrfs_level_size(root, wc->level - 1);
6839 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6840 if (nread >= wc->reada_count)
6844 bytenr = btrfs_node_blockptr(eb, slot);
6845 generation = btrfs_node_ptr_generation(eb, slot);
6847 if (slot == path->slots[wc->level])
6850 if (wc->stage == UPDATE_BACKREF &&
6851 generation <= root->root_key.offset)
6854 /* We don't lock the tree block, it's OK to be racy here */
6855 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6856 wc->level - 1, 1, &refs,
6858 /* We don't care about errors in readahead. */
6863 if (wc->stage == DROP_REFERENCE) {
6867 if (wc->level == 1 &&
6868 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6870 if (!wc->update_ref ||
6871 generation <= root->root_key.offset)
6873 btrfs_node_key_to_cpu(eb, &key, slot);
6874 ret = btrfs_comp_cpu_keys(&key,
6875 &wc->update_progress);
6879 if (wc->level == 1 &&
6880 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6884 ret = readahead_tree_block(root, bytenr, blocksize,
6890 wc->reada_slot = slot;
6894 * helper to process tree block while walking down the tree.
6896 * when wc->stage == UPDATE_BACKREF, this function updates
6897 * back refs for pointers in the block.
6899 * NOTE: return value 1 means we should stop walking down.
6901 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6902 struct btrfs_root *root,
6903 struct btrfs_path *path,
6904 struct walk_control *wc, int lookup_info)
6906 int level = wc->level;
6907 struct extent_buffer *eb = path->nodes[level];
6908 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6911 if (wc->stage == UPDATE_BACKREF &&
6912 btrfs_header_owner(eb) != root->root_key.objectid)
6916 * when reference count of tree block is 1, it won't increase
6917 * again. once full backref flag is set, we never clear it.
6920 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6921 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6922 BUG_ON(!path->locks[level]);
6923 ret = btrfs_lookup_extent_info(trans, root,
6924 eb->start, level, 1,
6927 BUG_ON(ret == -ENOMEM);
6930 BUG_ON(wc->refs[level] == 0);
6933 if (wc->stage == DROP_REFERENCE) {
6934 if (wc->refs[level] > 1)
6937 if (path->locks[level] && !wc->keep_locks) {
6938 btrfs_tree_unlock_rw(eb, path->locks[level]);
6939 path->locks[level] = 0;
6944 /* wc->stage == UPDATE_BACKREF */
6945 if (!(wc->flags[level] & flag)) {
6946 BUG_ON(!path->locks[level]);
6947 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6948 BUG_ON(ret); /* -ENOMEM */
6949 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6950 BUG_ON(ret); /* -ENOMEM */
6951 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6953 BUG_ON(ret); /* -ENOMEM */
6954 wc->flags[level] |= flag;
6958 * the block is shared by multiple trees, so it's not good to
6959 * keep the tree lock
6961 if (path->locks[level] && level > 0) {
6962 btrfs_tree_unlock_rw(eb, path->locks[level]);
6963 path->locks[level] = 0;
6969 * helper to process tree block pointer.
6971 * when wc->stage == DROP_REFERENCE, this function checks
6972 * reference count of the block pointed to. if the block
6973 * is shared and we need update back refs for the subtree
6974 * rooted at the block, this function changes wc->stage to
6975 * UPDATE_BACKREF. if the block is shared and there is no
6976 * need to update back, this function drops the reference
6979 * NOTE: return value 1 means we should stop walking down.
6981 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6982 struct btrfs_root *root,
6983 struct btrfs_path *path,
6984 struct walk_control *wc, int *lookup_info)
6990 struct btrfs_key key;
6991 struct extent_buffer *next;
6992 int level = wc->level;
6996 generation = btrfs_node_ptr_generation(path->nodes[level],
6997 path->slots[level]);
6999 * if the lower level block was created before the snapshot
7000 * was created, we know there is no need to update back refs
7003 if (wc->stage == UPDATE_BACKREF &&
7004 generation <= root->root_key.offset) {
7009 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7010 blocksize = btrfs_level_size(root, level - 1);
7012 next = btrfs_find_tree_block(root, bytenr, blocksize);
7014 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7019 btrfs_tree_lock(next);
7020 btrfs_set_lock_blocking(next);
7022 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7023 &wc->refs[level - 1],
7024 &wc->flags[level - 1]);
7026 btrfs_tree_unlock(next);
7030 BUG_ON(wc->refs[level - 1] == 0);
7033 if (wc->stage == DROP_REFERENCE) {
7034 if (wc->refs[level - 1] > 1) {
7036 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7039 if (!wc->update_ref ||
7040 generation <= root->root_key.offset)
7043 btrfs_node_key_to_cpu(path->nodes[level], &key,
7044 path->slots[level]);
7045 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7049 wc->stage = UPDATE_BACKREF;
7050 wc->shared_level = level - 1;
7054 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7058 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7059 btrfs_tree_unlock(next);
7060 free_extent_buffer(next);
7066 if (reada && level == 1)
7067 reada_walk_down(trans, root, wc, path);
7068 next = read_tree_block(root, bytenr, blocksize, generation);
7071 btrfs_tree_lock(next);
7072 btrfs_set_lock_blocking(next);
7076 BUG_ON(level != btrfs_header_level(next));
7077 path->nodes[level] = next;
7078 path->slots[level] = 0;
7079 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7085 wc->refs[level - 1] = 0;
7086 wc->flags[level - 1] = 0;
7087 if (wc->stage == DROP_REFERENCE) {
7088 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7089 parent = path->nodes[level]->start;
7091 BUG_ON(root->root_key.objectid !=
7092 btrfs_header_owner(path->nodes[level]));
7096 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7097 root->root_key.objectid, level - 1, 0, 0);
7098 BUG_ON(ret); /* -ENOMEM */
7100 btrfs_tree_unlock(next);
7101 free_extent_buffer(next);
7107 * helper to process tree block while walking up the tree.
7109 * when wc->stage == DROP_REFERENCE, this function drops
7110 * reference count on the block.
7112 * when wc->stage == UPDATE_BACKREF, this function changes
7113 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7114 * to UPDATE_BACKREF previously while processing the block.
7116 * NOTE: return value 1 means we should stop walking up.
7118 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7119 struct btrfs_root *root,
7120 struct btrfs_path *path,
7121 struct walk_control *wc)
7124 int level = wc->level;
7125 struct extent_buffer *eb = path->nodes[level];
7128 if (wc->stage == UPDATE_BACKREF) {
7129 BUG_ON(wc->shared_level < level);
7130 if (level < wc->shared_level)
7133 ret = find_next_key(path, level + 1, &wc->update_progress);
7137 wc->stage = DROP_REFERENCE;
7138 wc->shared_level = -1;
7139 path->slots[level] = 0;
7142 * check reference count again if the block isn't locked.
7143 * we should start walking down the tree again if reference
7146 if (!path->locks[level]) {
7148 btrfs_tree_lock(eb);
7149 btrfs_set_lock_blocking(eb);
7150 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7152 ret = btrfs_lookup_extent_info(trans, root,
7153 eb->start, level, 1,
7157 btrfs_tree_unlock_rw(eb, path->locks[level]);
7158 path->locks[level] = 0;
7161 BUG_ON(wc->refs[level] == 0);
7162 if (wc->refs[level] == 1) {
7163 btrfs_tree_unlock_rw(eb, path->locks[level]);
7164 path->locks[level] = 0;
7170 /* wc->stage == DROP_REFERENCE */
7171 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7173 if (wc->refs[level] == 1) {
7175 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7176 ret = btrfs_dec_ref(trans, root, eb, 1,
7179 ret = btrfs_dec_ref(trans, root, eb, 0,
7181 BUG_ON(ret); /* -ENOMEM */
7183 /* make block locked assertion in clean_tree_block happy */
7184 if (!path->locks[level] &&
7185 btrfs_header_generation(eb) == trans->transid) {
7186 btrfs_tree_lock(eb);
7187 btrfs_set_lock_blocking(eb);
7188 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7190 clean_tree_block(trans, root, eb);
7193 if (eb == root->node) {
7194 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7197 BUG_ON(root->root_key.objectid !=
7198 btrfs_header_owner(eb));
7200 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7201 parent = path->nodes[level + 1]->start;
7203 BUG_ON(root->root_key.objectid !=
7204 btrfs_header_owner(path->nodes[level + 1]));
7207 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7209 wc->refs[level] = 0;
7210 wc->flags[level] = 0;
7214 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7215 struct btrfs_root *root,
7216 struct btrfs_path *path,
7217 struct walk_control *wc)
7219 int level = wc->level;
7220 int lookup_info = 1;
7223 while (level >= 0) {
7224 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7231 if (path->slots[level] >=
7232 btrfs_header_nritems(path->nodes[level]))
7235 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7237 path->slots[level]++;
7246 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7247 struct btrfs_root *root,
7248 struct btrfs_path *path,
7249 struct walk_control *wc, int max_level)
7251 int level = wc->level;
7254 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7255 while (level < max_level && path->nodes[level]) {
7257 if (path->slots[level] + 1 <
7258 btrfs_header_nritems(path->nodes[level])) {
7259 path->slots[level]++;
7262 ret = walk_up_proc(trans, root, path, wc);
7266 if (path->locks[level]) {
7267 btrfs_tree_unlock_rw(path->nodes[level],
7268 path->locks[level]);
7269 path->locks[level] = 0;
7271 free_extent_buffer(path->nodes[level]);
7272 path->nodes[level] = NULL;
7280 * drop a subvolume tree.
7282 * this function traverses the tree freeing any blocks that only
7283 * referenced by the tree.
7285 * when a shared tree block is found. this function decreases its
7286 * reference count by one. if update_ref is true, this function
7287 * also make sure backrefs for the shared block and all lower level
7288 * blocks are properly updated.
7290 int btrfs_drop_snapshot(struct btrfs_root *root,
7291 struct btrfs_block_rsv *block_rsv, int update_ref,
7294 struct btrfs_path *path;
7295 struct btrfs_trans_handle *trans;
7296 struct btrfs_root *tree_root = root->fs_info->tree_root;
7297 struct btrfs_root_item *root_item = &root->root_item;
7298 struct walk_control *wc;
7299 struct btrfs_key key;
7304 path = btrfs_alloc_path();
7310 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7312 btrfs_free_path(path);
7317 trans = btrfs_start_transaction(tree_root, 0);
7318 if (IS_ERR(trans)) {
7319 err = PTR_ERR(trans);
7324 trans->block_rsv = block_rsv;
7326 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7327 level = btrfs_header_level(root->node);
7328 path->nodes[level] = btrfs_lock_root_node(root);
7329 btrfs_set_lock_blocking(path->nodes[level]);
7330 path->slots[level] = 0;
7331 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7332 memset(&wc->update_progress, 0,
7333 sizeof(wc->update_progress));
7335 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7336 memcpy(&wc->update_progress, &key,
7337 sizeof(wc->update_progress));
7339 level = root_item->drop_level;
7341 path->lowest_level = level;
7342 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7343 path->lowest_level = 0;
7351 * unlock our path, this is safe because only this
7352 * function is allowed to delete this snapshot
7354 btrfs_unlock_up_safe(path, 0);
7356 level = btrfs_header_level(root->node);
7358 btrfs_tree_lock(path->nodes[level]);
7359 btrfs_set_lock_blocking(path->nodes[level]);
7361 ret = btrfs_lookup_extent_info(trans, root,
7362 path->nodes[level]->start,
7363 level, 1, &wc->refs[level],
7369 BUG_ON(wc->refs[level] == 0);
7371 if (level == root_item->drop_level)
7374 btrfs_tree_unlock(path->nodes[level]);
7375 WARN_ON(wc->refs[level] != 1);
7381 wc->shared_level = -1;
7382 wc->stage = DROP_REFERENCE;
7383 wc->update_ref = update_ref;
7385 wc->for_reloc = for_reloc;
7386 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7389 ret = walk_down_tree(trans, root, path, wc);
7395 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7402 BUG_ON(wc->stage != DROP_REFERENCE);
7406 if (wc->stage == DROP_REFERENCE) {
7408 btrfs_node_key(path->nodes[level],
7409 &root_item->drop_progress,
7410 path->slots[level]);
7411 root_item->drop_level = level;
7414 BUG_ON(wc->level == 0);
7415 if (btrfs_should_end_transaction(trans, tree_root)) {
7416 ret = btrfs_update_root(trans, tree_root,
7420 btrfs_abort_transaction(trans, tree_root, ret);
7425 btrfs_end_transaction_throttle(trans, tree_root);
7426 trans = btrfs_start_transaction(tree_root, 0);
7427 if (IS_ERR(trans)) {
7428 err = PTR_ERR(trans);
7432 trans->block_rsv = block_rsv;
7435 btrfs_release_path(path);
7439 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7441 btrfs_abort_transaction(trans, tree_root, ret);
7445 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7446 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7449 btrfs_abort_transaction(trans, tree_root, ret);
7452 } else if (ret > 0) {
7453 /* if we fail to delete the orphan item this time
7454 * around, it'll get picked up the next time.
7456 * The most common failure here is just -ENOENT.
7458 btrfs_del_orphan_item(trans, tree_root,
7459 root->root_key.objectid);
7463 if (root->in_radix) {
7464 btrfs_free_fs_root(tree_root->fs_info, root);
7466 free_extent_buffer(root->node);
7467 free_extent_buffer(root->commit_root);
7471 btrfs_end_transaction_throttle(trans, tree_root);
7474 btrfs_free_path(path);
7477 btrfs_std_error(root->fs_info, err);
7482 * drop subtree rooted at tree block 'node'.
7484 * NOTE: this function will unlock and release tree block 'node'
7485 * only used by relocation code
7487 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7488 struct btrfs_root *root,
7489 struct extent_buffer *node,
7490 struct extent_buffer *parent)
7492 struct btrfs_path *path;
7493 struct walk_control *wc;
7499 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7501 path = btrfs_alloc_path();
7505 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7507 btrfs_free_path(path);
7511 btrfs_assert_tree_locked(parent);
7512 parent_level = btrfs_header_level(parent);
7513 extent_buffer_get(parent);
7514 path->nodes[parent_level] = parent;
7515 path->slots[parent_level] = btrfs_header_nritems(parent);
7517 btrfs_assert_tree_locked(node);
7518 level = btrfs_header_level(node);
7519 path->nodes[level] = node;
7520 path->slots[level] = 0;
7521 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7523 wc->refs[parent_level] = 1;
7524 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7526 wc->shared_level = -1;
7527 wc->stage = DROP_REFERENCE;
7531 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7534 wret = walk_down_tree(trans, root, path, wc);
7540 wret = walk_up_tree(trans, root, path, wc, parent_level);
7548 btrfs_free_path(path);
7552 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7558 * if restripe for this chunk_type is on pick target profile and
7559 * return, otherwise do the usual balance
7561 stripped = get_restripe_target(root->fs_info, flags);
7563 return extended_to_chunk(stripped);
7566 * we add in the count of missing devices because we want
7567 * to make sure that any RAID levels on a degraded FS
7568 * continue to be honored.
7570 num_devices = root->fs_info->fs_devices->rw_devices +
7571 root->fs_info->fs_devices->missing_devices;
7573 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7574 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7575 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7577 if (num_devices == 1) {
7578 stripped |= BTRFS_BLOCK_GROUP_DUP;
7579 stripped = flags & ~stripped;
7581 /* turn raid0 into single device chunks */
7582 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7585 /* turn mirroring into duplication */
7586 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7587 BTRFS_BLOCK_GROUP_RAID10))
7588 return stripped | BTRFS_BLOCK_GROUP_DUP;
7590 /* they already had raid on here, just return */
7591 if (flags & stripped)
7594 stripped |= BTRFS_BLOCK_GROUP_DUP;
7595 stripped = flags & ~stripped;
7597 /* switch duplicated blocks with raid1 */
7598 if (flags & BTRFS_BLOCK_GROUP_DUP)
7599 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7601 /* this is drive concat, leave it alone */
7607 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7609 struct btrfs_space_info *sinfo = cache->space_info;
7611 u64 min_allocable_bytes;
7616 * We need some metadata space and system metadata space for
7617 * allocating chunks in some corner cases until we force to set
7618 * it to be readonly.
7621 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7623 min_allocable_bytes = 1 * 1024 * 1024;
7625 min_allocable_bytes = 0;
7627 spin_lock(&sinfo->lock);
7628 spin_lock(&cache->lock);
7635 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7636 cache->bytes_super - btrfs_block_group_used(&cache->item);
7638 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7639 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7640 min_allocable_bytes <= sinfo->total_bytes) {
7641 sinfo->bytes_readonly += num_bytes;
7646 spin_unlock(&cache->lock);
7647 spin_unlock(&sinfo->lock);
7651 int btrfs_set_block_group_ro(struct btrfs_root *root,
7652 struct btrfs_block_group_cache *cache)
7655 struct btrfs_trans_handle *trans;
7661 trans = btrfs_join_transaction(root);
7663 return PTR_ERR(trans);
7665 alloc_flags = update_block_group_flags(root, cache->flags);
7666 if (alloc_flags != cache->flags) {
7667 ret = do_chunk_alloc(trans, root, alloc_flags,
7673 ret = set_block_group_ro(cache, 0);
7676 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7677 ret = do_chunk_alloc(trans, root, alloc_flags,
7681 ret = set_block_group_ro(cache, 0);
7683 btrfs_end_transaction(trans, root);
7687 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7688 struct btrfs_root *root, u64 type)
7690 u64 alloc_flags = get_alloc_profile(root, type);
7691 return do_chunk_alloc(trans, root, alloc_flags,
7696 * helper to account the unused space of all the readonly block group in the
7697 * list. takes mirrors into account.
7699 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7701 struct btrfs_block_group_cache *block_group;
7705 list_for_each_entry(block_group, groups_list, list) {
7706 spin_lock(&block_group->lock);
7708 if (!block_group->ro) {
7709 spin_unlock(&block_group->lock);
7713 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7714 BTRFS_BLOCK_GROUP_RAID10 |
7715 BTRFS_BLOCK_GROUP_DUP))
7720 free_bytes += (block_group->key.offset -
7721 btrfs_block_group_used(&block_group->item)) *
7724 spin_unlock(&block_group->lock);
7731 * helper to account the unused space of all the readonly block group in the
7732 * space_info. takes mirrors into account.
7734 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7739 spin_lock(&sinfo->lock);
7741 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7742 if (!list_empty(&sinfo->block_groups[i]))
7743 free_bytes += __btrfs_get_ro_block_group_free_space(
7744 &sinfo->block_groups[i]);
7746 spin_unlock(&sinfo->lock);
7751 void btrfs_set_block_group_rw(struct btrfs_root *root,
7752 struct btrfs_block_group_cache *cache)
7754 struct btrfs_space_info *sinfo = cache->space_info;
7759 spin_lock(&sinfo->lock);
7760 spin_lock(&cache->lock);
7761 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7762 cache->bytes_super - btrfs_block_group_used(&cache->item);
7763 sinfo->bytes_readonly -= num_bytes;
7765 spin_unlock(&cache->lock);
7766 spin_unlock(&sinfo->lock);
7770 * checks to see if its even possible to relocate this block group.
7772 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7773 * ok to go ahead and try.
7775 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7777 struct btrfs_block_group_cache *block_group;
7778 struct btrfs_space_info *space_info;
7779 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7780 struct btrfs_device *device;
7789 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7791 /* odd, couldn't find the block group, leave it alone */
7795 min_free = btrfs_block_group_used(&block_group->item);
7797 /* no bytes used, we're good */
7801 space_info = block_group->space_info;
7802 spin_lock(&space_info->lock);
7804 full = space_info->full;
7807 * if this is the last block group we have in this space, we can't
7808 * relocate it unless we're able to allocate a new chunk below.
7810 * Otherwise, we need to make sure we have room in the space to handle
7811 * all of the extents from this block group. If we can, we're good
7813 if ((space_info->total_bytes != block_group->key.offset) &&
7814 (space_info->bytes_used + space_info->bytes_reserved +
7815 space_info->bytes_pinned + space_info->bytes_readonly +
7816 min_free < space_info->total_bytes)) {
7817 spin_unlock(&space_info->lock);
7820 spin_unlock(&space_info->lock);
7823 * ok we don't have enough space, but maybe we have free space on our
7824 * devices to allocate new chunks for relocation, so loop through our
7825 * alloc devices and guess if we have enough space. if this block
7826 * group is going to be restriped, run checks against the target
7827 * profile instead of the current one.
7839 target = get_restripe_target(root->fs_info, block_group->flags);
7841 index = __get_raid_index(extended_to_chunk(target));
7844 * this is just a balance, so if we were marked as full
7845 * we know there is no space for a new chunk
7850 index = get_block_group_index(block_group);
7853 if (index == BTRFS_RAID_RAID10) {
7857 } else if (index == BTRFS_RAID_RAID1) {
7859 } else if (index == BTRFS_RAID_DUP) {
7862 } else if (index == BTRFS_RAID_RAID0) {
7863 dev_min = fs_devices->rw_devices;
7864 do_div(min_free, dev_min);
7867 mutex_lock(&root->fs_info->chunk_mutex);
7868 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7872 * check to make sure we can actually find a chunk with enough
7873 * space to fit our block group in.
7875 if (device->total_bytes > device->bytes_used + min_free &&
7876 !device->is_tgtdev_for_dev_replace) {
7877 ret = find_free_dev_extent(device, min_free,
7882 if (dev_nr >= dev_min)
7888 mutex_unlock(&root->fs_info->chunk_mutex);
7890 btrfs_put_block_group(block_group);
7894 static int find_first_block_group(struct btrfs_root *root,
7895 struct btrfs_path *path, struct btrfs_key *key)
7898 struct btrfs_key found_key;
7899 struct extent_buffer *leaf;
7902 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7907 slot = path->slots[0];
7908 leaf = path->nodes[0];
7909 if (slot >= btrfs_header_nritems(leaf)) {
7910 ret = btrfs_next_leaf(root, path);
7917 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7919 if (found_key.objectid >= key->objectid &&
7920 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7930 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7932 struct btrfs_block_group_cache *block_group;
7936 struct inode *inode;
7938 block_group = btrfs_lookup_first_block_group(info, last);
7939 while (block_group) {
7940 spin_lock(&block_group->lock);
7941 if (block_group->iref)
7943 spin_unlock(&block_group->lock);
7944 block_group = next_block_group(info->tree_root,
7954 inode = block_group->inode;
7955 block_group->iref = 0;
7956 block_group->inode = NULL;
7957 spin_unlock(&block_group->lock);
7959 last = block_group->key.objectid + block_group->key.offset;
7960 btrfs_put_block_group(block_group);
7964 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7966 struct btrfs_block_group_cache *block_group;
7967 struct btrfs_space_info *space_info;
7968 struct btrfs_caching_control *caching_ctl;
7971 down_write(&info->extent_commit_sem);
7972 while (!list_empty(&info->caching_block_groups)) {
7973 caching_ctl = list_entry(info->caching_block_groups.next,
7974 struct btrfs_caching_control, list);
7975 list_del(&caching_ctl->list);
7976 put_caching_control(caching_ctl);
7978 up_write(&info->extent_commit_sem);
7980 spin_lock(&info->block_group_cache_lock);
7981 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7982 block_group = rb_entry(n, struct btrfs_block_group_cache,
7984 rb_erase(&block_group->cache_node,
7985 &info->block_group_cache_tree);
7986 spin_unlock(&info->block_group_cache_lock);
7988 down_write(&block_group->space_info->groups_sem);
7989 list_del(&block_group->list);
7990 up_write(&block_group->space_info->groups_sem);
7992 if (block_group->cached == BTRFS_CACHE_STARTED)
7993 wait_block_group_cache_done(block_group);
7996 * We haven't cached this block group, which means we could
7997 * possibly have excluded extents on this block group.
7999 if (block_group->cached == BTRFS_CACHE_NO)
8000 free_excluded_extents(info->extent_root, block_group);
8002 btrfs_remove_free_space_cache(block_group);
8003 btrfs_put_block_group(block_group);
8005 spin_lock(&info->block_group_cache_lock);
8007 spin_unlock(&info->block_group_cache_lock);
8009 /* now that all the block groups are freed, go through and
8010 * free all the space_info structs. This is only called during
8011 * the final stages of unmount, and so we know nobody is
8012 * using them. We call synchronize_rcu() once before we start,
8013 * just to be on the safe side.
8017 release_global_block_rsv(info);
8019 while(!list_empty(&info->space_info)) {
8020 space_info = list_entry(info->space_info.next,
8021 struct btrfs_space_info,
8023 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8024 if (space_info->bytes_pinned > 0 ||
8025 space_info->bytes_reserved > 0 ||
8026 space_info->bytes_may_use > 0) {
8028 dump_space_info(space_info, 0, 0);
8031 list_del(&space_info->list);
8037 static void __link_block_group(struct btrfs_space_info *space_info,
8038 struct btrfs_block_group_cache *cache)
8040 int index = get_block_group_index(cache);
8042 down_write(&space_info->groups_sem);
8043 list_add_tail(&cache->list, &space_info->block_groups[index]);
8044 up_write(&space_info->groups_sem);
8047 int btrfs_read_block_groups(struct btrfs_root *root)
8049 struct btrfs_path *path;
8051 struct btrfs_block_group_cache *cache;
8052 struct btrfs_fs_info *info = root->fs_info;
8053 struct btrfs_space_info *space_info;
8054 struct btrfs_key key;
8055 struct btrfs_key found_key;
8056 struct extent_buffer *leaf;
8060 root = info->extent_root;
8063 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8064 path = btrfs_alloc_path();
8069 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8070 if (btrfs_test_opt(root, SPACE_CACHE) &&
8071 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8073 if (btrfs_test_opt(root, CLEAR_CACHE))
8077 ret = find_first_block_group(root, path, &key);
8082 leaf = path->nodes[0];
8083 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8084 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8089 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8091 if (!cache->free_space_ctl) {
8097 atomic_set(&cache->count, 1);
8098 spin_lock_init(&cache->lock);
8099 cache->fs_info = info;
8100 INIT_LIST_HEAD(&cache->list);
8101 INIT_LIST_HEAD(&cache->cluster_list);
8105 * When we mount with old space cache, we need to
8106 * set BTRFS_DC_CLEAR and set dirty flag.
8108 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8109 * truncate the old free space cache inode and
8111 * b) Setting 'dirty flag' makes sure that we flush
8112 * the new space cache info onto disk.
8114 cache->disk_cache_state = BTRFS_DC_CLEAR;
8115 if (btrfs_test_opt(root, SPACE_CACHE))
8119 read_extent_buffer(leaf, &cache->item,
8120 btrfs_item_ptr_offset(leaf, path->slots[0]),
8121 sizeof(cache->item));
8122 memcpy(&cache->key, &found_key, sizeof(found_key));
8124 key.objectid = found_key.objectid + found_key.offset;
8125 btrfs_release_path(path);
8126 cache->flags = btrfs_block_group_flags(&cache->item);
8127 cache->sectorsize = root->sectorsize;
8128 cache->full_stripe_len = btrfs_full_stripe_len(root,
8129 &root->fs_info->mapping_tree,
8130 found_key.objectid);
8131 btrfs_init_free_space_ctl(cache);
8134 * We need to exclude the super stripes now so that the space
8135 * info has super bytes accounted for, otherwise we'll think
8136 * we have more space than we actually do.
8138 ret = exclude_super_stripes(root, cache);
8141 * We may have excluded something, so call this just in
8144 free_excluded_extents(root, cache);
8145 kfree(cache->free_space_ctl);
8151 * check for two cases, either we are full, and therefore
8152 * don't need to bother with the caching work since we won't
8153 * find any space, or we are empty, and we can just add all
8154 * the space in and be done with it. This saves us _alot_ of
8155 * time, particularly in the full case.
8157 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8158 cache->last_byte_to_unpin = (u64)-1;
8159 cache->cached = BTRFS_CACHE_FINISHED;
8160 free_excluded_extents(root, cache);
8161 } else if (btrfs_block_group_used(&cache->item) == 0) {
8162 cache->last_byte_to_unpin = (u64)-1;
8163 cache->cached = BTRFS_CACHE_FINISHED;
8164 add_new_free_space(cache, root->fs_info,
8166 found_key.objectid +
8168 free_excluded_extents(root, cache);
8171 ret = update_space_info(info, cache->flags, found_key.offset,
8172 btrfs_block_group_used(&cache->item),
8174 BUG_ON(ret); /* -ENOMEM */
8175 cache->space_info = space_info;
8176 spin_lock(&cache->space_info->lock);
8177 cache->space_info->bytes_readonly += cache->bytes_super;
8178 spin_unlock(&cache->space_info->lock);
8180 __link_block_group(space_info, cache);
8182 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8183 BUG_ON(ret); /* Logic error */
8185 set_avail_alloc_bits(root->fs_info, cache->flags);
8186 if (btrfs_chunk_readonly(root, cache->key.objectid))
8187 set_block_group_ro(cache, 1);
8190 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8191 if (!(get_alloc_profile(root, space_info->flags) &
8192 (BTRFS_BLOCK_GROUP_RAID10 |
8193 BTRFS_BLOCK_GROUP_RAID1 |
8194 BTRFS_BLOCK_GROUP_RAID5 |
8195 BTRFS_BLOCK_GROUP_RAID6 |
8196 BTRFS_BLOCK_GROUP_DUP)))
8199 * avoid allocating from un-mirrored block group if there are
8200 * mirrored block groups.
8202 list_for_each_entry(cache, &space_info->block_groups[3], list)
8203 set_block_group_ro(cache, 1);
8204 list_for_each_entry(cache, &space_info->block_groups[4], list)
8205 set_block_group_ro(cache, 1);
8208 init_global_block_rsv(info);
8211 btrfs_free_path(path);
8215 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8216 struct btrfs_root *root)
8218 struct btrfs_block_group_cache *block_group, *tmp;
8219 struct btrfs_root *extent_root = root->fs_info->extent_root;
8220 struct btrfs_block_group_item item;
8221 struct btrfs_key key;
8224 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8226 list_del_init(&block_group->new_bg_list);
8231 spin_lock(&block_group->lock);
8232 memcpy(&item, &block_group->item, sizeof(item));
8233 memcpy(&key, &block_group->key, sizeof(key));
8234 spin_unlock(&block_group->lock);
8236 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8239 btrfs_abort_transaction(trans, extent_root, ret);
8243 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8244 struct btrfs_root *root, u64 bytes_used,
8245 u64 type, u64 chunk_objectid, u64 chunk_offset,
8249 struct btrfs_root *extent_root;
8250 struct btrfs_block_group_cache *cache;
8252 extent_root = root->fs_info->extent_root;
8254 root->fs_info->last_trans_log_full_commit = trans->transid;
8256 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8259 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8261 if (!cache->free_space_ctl) {
8266 cache->key.objectid = chunk_offset;
8267 cache->key.offset = size;
8268 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8269 cache->sectorsize = root->sectorsize;
8270 cache->fs_info = root->fs_info;
8271 cache->full_stripe_len = btrfs_full_stripe_len(root,
8272 &root->fs_info->mapping_tree,
8275 atomic_set(&cache->count, 1);
8276 spin_lock_init(&cache->lock);
8277 INIT_LIST_HEAD(&cache->list);
8278 INIT_LIST_HEAD(&cache->cluster_list);
8279 INIT_LIST_HEAD(&cache->new_bg_list);
8281 btrfs_init_free_space_ctl(cache);
8283 btrfs_set_block_group_used(&cache->item, bytes_used);
8284 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8285 cache->flags = type;
8286 btrfs_set_block_group_flags(&cache->item, type);
8288 cache->last_byte_to_unpin = (u64)-1;
8289 cache->cached = BTRFS_CACHE_FINISHED;
8290 ret = exclude_super_stripes(root, cache);
8293 * We may have excluded something, so call this just in
8296 free_excluded_extents(root, cache);
8297 kfree(cache->free_space_ctl);
8302 add_new_free_space(cache, root->fs_info, chunk_offset,
8303 chunk_offset + size);
8305 free_excluded_extents(root, cache);
8307 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8308 &cache->space_info);
8309 BUG_ON(ret); /* -ENOMEM */
8310 update_global_block_rsv(root->fs_info);
8312 spin_lock(&cache->space_info->lock);
8313 cache->space_info->bytes_readonly += cache->bytes_super;
8314 spin_unlock(&cache->space_info->lock);
8316 __link_block_group(cache->space_info, cache);
8318 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8319 BUG_ON(ret); /* Logic error */
8321 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8323 set_avail_alloc_bits(extent_root->fs_info, type);
8328 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8330 u64 extra_flags = chunk_to_extended(flags) &
8331 BTRFS_EXTENDED_PROFILE_MASK;
8333 write_seqlock(&fs_info->profiles_lock);
8334 if (flags & BTRFS_BLOCK_GROUP_DATA)
8335 fs_info->avail_data_alloc_bits &= ~extra_flags;
8336 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8337 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8338 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8339 fs_info->avail_system_alloc_bits &= ~extra_flags;
8340 write_sequnlock(&fs_info->profiles_lock);
8343 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8344 struct btrfs_root *root, u64 group_start)
8346 struct btrfs_path *path;
8347 struct btrfs_block_group_cache *block_group;
8348 struct btrfs_free_cluster *cluster;
8349 struct btrfs_root *tree_root = root->fs_info->tree_root;
8350 struct btrfs_key key;
8351 struct inode *inode;
8356 root = root->fs_info->extent_root;
8358 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8359 BUG_ON(!block_group);
8360 BUG_ON(!block_group->ro);
8363 * Free the reserved super bytes from this block group before
8366 free_excluded_extents(root, block_group);
8368 memcpy(&key, &block_group->key, sizeof(key));
8369 index = get_block_group_index(block_group);
8370 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8371 BTRFS_BLOCK_GROUP_RAID1 |
8372 BTRFS_BLOCK_GROUP_RAID10))
8377 /* make sure this block group isn't part of an allocation cluster */
8378 cluster = &root->fs_info->data_alloc_cluster;
8379 spin_lock(&cluster->refill_lock);
8380 btrfs_return_cluster_to_free_space(block_group, cluster);
8381 spin_unlock(&cluster->refill_lock);
8384 * make sure this block group isn't part of a metadata
8385 * allocation cluster
8387 cluster = &root->fs_info->meta_alloc_cluster;
8388 spin_lock(&cluster->refill_lock);
8389 btrfs_return_cluster_to_free_space(block_group, cluster);
8390 spin_unlock(&cluster->refill_lock);
8392 path = btrfs_alloc_path();
8398 inode = lookup_free_space_inode(tree_root, block_group, path);
8399 if (!IS_ERR(inode)) {
8400 ret = btrfs_orphan_add(trans, inode);
8402 btrfs_add_delayed_iput(inode);
8406 /* One for the block groups ref */
8407 spin_lock(&block_group->lock);
8408 if (block_group->iref) {
8409 block_group->iref = 0;
8410 block_group->inode = NULL;
8411 spin_unlock(&block_group->lock);
8414 spin_unlock(&block_group->lock);
8416 /* One for our lookup ref */
8417 btrfs_add_delayed_iput(inode);
8420 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8421 key.offset = block_group->key.objectid;
8424 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8428 btrfs_release_path(path);
8430 ret = btrfs_del_item(trans, tree_root, path);
8433 btrfs_release_path(path);
8436 spin_lock(&root->fs_info->block_group_cache_lock);
8437 rb_erase(&block_group->cache_node,
8438 &root->fs_info->block_group_cache_tree);
8440 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8441 root->fs_info->first_logical_byte = (u64)-1;
8442 spin_unlock(&root->fs_info->block_group_cache_lock);
8444 down_write(&block_group->space_info->groups_sem);
8446 * we must use list_del_init so people can check to see if they
8447 * are still on the list after taking the semaphore
8449 list_del_init(&block_group->list);
8450 if (list_empty(&block_group->space_info->block_groups[index]))
8451 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8452 up_write(&block_group->space_info->groups_sem);
8454 if (block_group->cached == BTRFS_CACHE_STARTED)
8455 wait_block_group_cache_done(block_group);
8457 btrfs_remove_free_space_cache(block_group);
8459 spin_lock(&block_group->space_info->lock);
8460 block_group->space_info->total_bytes -= block_group->key.offset;
8461 block_group->space_info->bytes_readonly -= block_group->key.offset;
8462 block_group->space_info->disk_total -= block_group->key.offset * factor;
8463 spin_unlock(&block_group->space_info->lock);
8465 memcpy(&key, &block_group->key, sizeof(key));
8467 btrfs_clear_space_info_full(root->fs_info);
8469 btrfs_put_block_group(block_group);
8470 btrfs_put_block_group(block_group);
8472 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8478 ret = btrfs_del_item(trans, root, path);
8480 btrfs_free_path(path);
8484 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8486 struct btrfs_space_info *space_info;
8487 struct btrfs_super_block *disk_super;
8493 disk_super = fs_info->super_copy;
8494 if (!btrfs_super_root(disk_super))
8497 features = btrfs_super_incompat_flags(disk_super);
8498 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8501 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8502 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8507 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8508 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8510 flags = BTRFS_BLOCK_GROUP_METADATA;
8511 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8515 flags = BTRFS_BLOCK_GROUP_DATA;
8516 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8522 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8524 return unpin_extent_range(root, start, end);
8527 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8528 u64 num_bytes, u64 *actual_bytes)
8530 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8533 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8535 struct btrfs_fs_info *fs_info = root->fs_info;
8536 struct btrfs_block_group_cache *cache = NULL;
8541 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8545 * try to trim all FS space, our block group may start from non-zero.
8547 if (range->len == total_bytes)
8548 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8550 cache = btrfs_lookup_block_group(fs_info, range->start);
8553 if (cache->key.objectid >= (range->start + range->len)) {
8554 btrfs_put_block_group(cache);
8558 start = max(range->start, cache->key.objectid);
8559 end = min(range->start + range->len,
8560 cache->key.objectid + cache->key.offset);
8562 if (end - start >= range->minlen) {
8563 if (!block_group_cache_done(cache)) {
8564 ret = cache_block_group(cache, 0);
8566 wait_block_group_cache_done(cache);
8568 ret = btrfs_trim_block_group(cache,
8574 trimmed += group_trimmed;
8576 btrfs_put_block_group(cache);
8581 cache = next_block_group(fs_info->tree_root, cache);
8584 range->len = trimmed;