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(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_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref *iref,
1665 u64 parent, u64 root_objectid,
1666 u64 owner, u64 offset, int refs_to_add,
1667 struct btrfs_delayed_extent_op *extent_op)
1669 struct extent_buffer *leaf;
1670 struct btrfs_extent_item *ei;
1673 unsigned long item_offset;
1678 leaf = path->nodes[0];
1679 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680 item_offset = (unsigned long)iref - (unsigned long)ei;
1682 type = extent_ref_type(parent, owner);
1683 size = btrfs_extent_inline_ref_size(type);
1685 btrfs_extend_item(root, path, size);
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 refs = btrfs_extent_refs(leaf, ei);
1689 refs += refs_to_add;
1690 btrfs_set_extent_refs(leaf, ei, refs);
1692 __run_delayed_extent_op(extent_op, leaf, ei);
1694 ptr = (unsigned long)ei + item_offset;
1695 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696 if (ptr < end - size)
1697 memmove_extent_buffer(leaf, ptr + size, ptr,
1700 iref = (struct btrfs_extent_inline_ref *)ptr;
1701 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703 struct btrfs_extent_data_ref *dref;
1704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 struct btrfs_shared_data_ref *sref;
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719 btrfs_mark_buffer_dirty(leaf);
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 struct btrfs_extent_inline_ref **ref_ret,
1726 u64 bytenr, u64 num_bytes, u64 parent,
1727 u64 root_objectid, u64 owner, u64 offset)
1731 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732 bytenr, num_bytes, parent,
1733 root_objectid, owner, offset, 0);
1737 btrfs_release_path(path);
1740 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1744 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745 root_objectid, owner, offset);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755 struct btrfs_path *path,
1756 struct btrfs_extent_inline_ref *iref,
1758 struct btrfs_delayed_extent_op *extent_op)
1760 struct extent_buffer *leaf;
1761 struct btrfs_extent_item *ei;
1762 struct btrfs_extent_data_ref *dref = NULL;
1763 struct btrfs_shared_data_ref *sref = NULL;
1771 leaf = path->nodes[0];
1772 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1773 refs = btrfs_extent_refs(leaf, ei);
1774 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1775 refs += refs_to_mod;
1776 btrfs_set_extent_refs(leaf, ei, refs);
1778 __run_delayed_extent_op(extent_op, leaf, ei);
1780 type = btrfs_extent_inline_ref_type(leaf, iref);
1782 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1783 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1784 refs = btrfs_extent_data_ref_count(leaf, dref);
1785 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1786 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1787 refs = btrfs_shared_data_ref_count(leaf, sref);
1790 BUG_ON(refs_to_mod != -1);
1793 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1794 refs += refs_to_mod;
1797 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1798 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1802 size = btrfs_extent_inline_ref_size(type);
1803 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1804 ptr = (unsigned long)iref;
1805 end = (unsigned long)ei + item_size;
1806 if (ptr + size < end)
1807 memmove_extent_buffer(leaf, ptr, ptr + size,
1810 btrfs_truncate_item(root, path, item_size, 1);
1812 btrfs_mark_buffer_dirty(leaf);
1815 static noinline_for_stack
1816 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1817 struct btrfs_root *root,
1818 struct btrfs_path *path,
1819 u64 bytenr, u64 num_bytes, u64 parent,
1820 u64 root_objectid, u64 owner,
1821 u64 offset, int refs_to_add,
1822 struct btrfs_delayed_extent_op *extent_op)
1824 struct btrfs_extent_inline_ref *iref;
1827 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1828 bytenr, num_bytes, parent,
1829 root_objectid, owner, offset, 1);
1831 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1832 update_inline_extent_backref(root, path, iref,
1833 refs_to_add, extent_op);
1834 } else if (ret == -ENOENT) {
1835 setup_inline_extent_backref(root, path, iref, parent,
1836 root_objectid, owner, offset,
1837 refs_to_add, extent_op);
1843 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1844 struct btrfs_root *root,
1845 struct btrfs_path *path,
1846 u64 bytenr, u64 parent, u64 root_objectid,
1847 u64 owner, u64 offset, int refs_to_add)
1850 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1851 BUG_ON(refs_to_add != 1);
1852 ret = insert_tree_block_ref(trans, root, path, bytenr,
1853 parent, root_objectid);
1855 ret = insert_extent_data_ref(trans, root, path, bytenr,
1856 parent, root_objectid,
1857 owner, offset, refs_to_add);
1862 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1863 struct btrfs_root *root,
1864 struct btrfs_path *path,
1865 struct btrfs_extent_inline_ref *iref,
1866 int refs_to_drop, int is_data)
1870 BUG_ON(!is_data && refs_to_drop != 1);
1872 update_inline_extent_backref(root, path, iref,
1873 -refs_to_drop, NULL);
1874 } else if (is_data) {
1875 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1877 ret = btrfs_del_item(trans, root, path);
1882 static int btrfs_issue_discard(struct block_device *bdev,
1885 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1888 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1889 u64 num_bytes, u64 *actual_bytes)
1892 u64 discarded_bytes = 0;
1893 struct btrfs_bio *bbio = NULL;
1896 /* Tell the block device(s) that the sectors can be discarded */
1897 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1898 bytenr, &num_bytes, &bbio, 0);
1899 /* Error condition is -ENOMEM */
1901 struct btrfs_bio_stripe *stripe = bbio->stripes;
1905 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1906 if (!stripe->dev->can_discard)
1909 ret = btrfs_issue_discard(stripe->dev->bdev,
1913 discarded_bytes += stripe->length;
1914 else if (ret != -EOPNOTSUPP)
1915 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1918 * Just in case we get back EOPNOTSUPP for some reason,
1919 * just ignore the return value so we don't screw up
1920 * people calling discard_extent.
1928 *actual_bytes = discarded_bytes;
1931 if (ret == -EOPNOTSUPP)
1936 /* Can return -ENOMEM */
1937 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1938 struct btrfs_root *root,
1939 u64 bytenr, u64 num_bytes, u64 parent,
1940 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1943 struct btrfs_fs_info *fs_info = root->fs_info;
1945 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1946 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1948 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1949 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1951 parent, root_objectid, (int)owner,
1952 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1954 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1956 parent, root_objectid, owner, offset,
1957 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1962 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1963 struct btrfs_root *root,
1964 u64 bytenr, u64 num_bytes,
1965 u64 parent, u64 root_objectid,
1966 u64 owner, u64 offset, int refs_to_add,
1967 struct btrfs_delayed_extent_op *extent_op)
1969 struct btrfs_path *path;
1970 struct extent_buffer *leaf;
1971 struct btrfs_extent_item *item;
1976 path = btrfs_alloc_path();
1981 path->leave_spinning = 1;
1982 /* this will setup the path even if it fails to insert the back ref */
1983 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1984 path, bytenr, num_bytes, parent,
1985 root_objectid, owner, offset,
1986 refs_to_add, extent_op);
1990 if (ret != -EAGAIN) {
1995 leaf = path->nodes[0];
1996 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1997 refs = btrfs_extent_refs(leaf, item);
1998 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2000 __run_delayed_extent_op(extent_op, leaf, item);
2002 btrfs_mark_buffer_dirty(leaf);
2003 btrfs_release_path(path);
2006 path->leave_spinning = 1;
2008 /* now insert the actual backref */
2009 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2010 path, bytenr, parent, root_objectid,
2011 owner, offset, refs_to_add);
2013 btrfs_abort_transaction(trans, root, ret);
2015 btrfs_free_path(path);
2019 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2020 struct btrfs_root *root,
2021 struct btrfs_delayed_ref_node *node,
2022 struct btrfs_delayed_extent_op *extent_op,
2023 int insert_reserved)
2026 struct btrfs_delayed_data_ref *ref;
2027 struct btrfs_key ins;
2032 ins.objectid = node->bytenr;
2033 ins.offset = node->num_bytes;
2034 ins.type = BTRFS_EXTENT_ITEM_KEY;
2036 ref = btrfs_delayed_node_to_data_ref(node);
2037 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2038 parent = ref->parent;
2040 ref_root = ref->root;
2042 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2044 flags |= extent_op->flags_to_set;
2045 ret = alloc_reserved_file_extent(trans, root,
2046 parent, ref_root, flags,
2047 ref->objectid, ref->offset,
2048 &ins, node->ref_mod);
2049 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2050 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2051 node->num_bytes, parent,
2052 ref_root, ref->objectid,
2053 ref->offset, node->ref_mod,
2055 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2056 ret = __btrfs_free_extent(trans, root, node->bytenr,
2057 node->num_bytes, parent,
2058 ref_root, ref->objectid,
2059 ref->offset, node->ref_mod,
2067 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2068 struct extent_buffer *leaf,
2069 struct btrfs_extent_item *ei)
2071 u64 flags = btrfs_extent_flags(leaf, ei);
2072 if (extent_op->update_flags) {
2073 flags |= extent_op->flags_to_set;
2074 btrfs_set_extent_flags(leaf, ei, flags);
2077 if (extent_op->update_key) {
2078 struct btrfs_tree_block_info *bi;
2079 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2080 bi = (struct btrfs_tree_block_info *)(ei + 1);
2081 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2085 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2086 struct btrfs_root *root,
2087 struct btrfs_delayed_ref_node *node,
2088 struct btrfs_delayed_extent_op *extent_op)
2090 struct btrfs_key key;
2091 struct btrfs_path *path;
2092 struct btrfs_extent_item *ei;
2093 struct extent_buffer *leaf;
2097 int metadata = (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2098 node->type == BTRFS_SHARED_BLOCK_REF_KEY);
2103 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2106 path = btrfs_alloc_path();
2110 key.objectid = node->bytenr;
2113 struct btrfs_delayed_tree_ref *tree_ref;
2115 tree_ref = btrfs_delayed_node_to_tree_ref(node);
2116 key.type = BTRFS_METADATA_ITEM_KEY;
2117 key.offset = tree_ref->level;
2119 key.type = BTRFS_EXTENT_ITEM_KEY;
2120 key.offset = node->num_bytes;
2125 path->leave_spinning = 1;
2126 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2134 btrfs_release_path(path);
2137 key.offset = node->num_bytes;
2138 key.type = BTRFS_EXTENT_ITEM_KEY;
2145 leaf = path->nodes[0];
2146 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2147 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2148 if (item_size < sizeof(*ei)) {
2149 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2155 leaf = path->nodes[0];
2156 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2159 BUG_ON(item_size < sizeof(*ei));
2160 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2161 __run_delayed_extent_op(extent_op, leaf, ei);
2163 btrfs_mark_buffer_dirty(leaf);
2165 btrfs_free_path(path);
2169 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2170 struct btrfs_root *root,
2171 struct btrfs_delayed_ref_node *node,
2172 struct btrfs_delayed_extent_op *extent_op,
2173 int insert_reserved)
2176 struct btrfs_delayed_tree_ref *ref;
2177 struct btrfs_key ins;
2180 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2183 ref = btrfs_delayed_node_to_tree_ref(node);
2184 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2185 parent = ref->parent;
2187 ref_root = ref->root;
2189 ins.objectid = node->bytenr;
2190 if (skinny_metadata) {
2191 ins.offset = ref->level;
2192 ins.type = BTRFS_METADATA_ITEM_KEY;
2194 ins.offset = node->num_bytes;
2195 ins.type = BTRFS_EXTENT_ITEM_KEY;
2198 BUG_ON(node->ref_mod != 1);
2199 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2200 BUG_ON(!extent_op || !extent_op->update_flags);
2201 ret = alloc_reserved_tree_block(trans, root,
2203 extent_op->flags_to_set,
2206 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2207 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2208 node->num_bytes, parent, ref_root,
2209 ref->level, 0, 1, extent_op);
2210 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2211 ret = __btrfs_free_extent(trans, root, node->bytenr,
2212 node->num_bytes, parent, ref_root,
2213 ref->level, 0, 1, extent_op);
2220 /* helper function to actually process a single delayed ref entry */
2221 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2222 struct btrfs_root *root,
2223 struct btrfs_delayed_ref_node *node,
2224 struct btrfs_delayed_extent_op *extent_op,
2225 int insert_reserved)
2232 if (btrfs_delayed_ref_is_head(node)) {
2233 struct btrfs_delayed_ref_head *head;
2235 * we've hit the end of the chain and we were supposed
2236 * to insert this extent into the tree. But, it got
2237 * deleted before we ever needed to insert it, so all
2238 * we have to do is clean up the accounting
2241 head = btrfs_delayed_node_to_head(node);
2242 if (insert_reserved) {
2243 btrfs_pin_extent(root, node->bytenr,
2244 node->num_bytes, 1);
2245 if (head->is_data) {
2246 ret = btrfs_del_csums(trans, root,
2254 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2255 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2256 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2258 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2259 node->type == BTRFS_SHARED_DATA_REF_KEY)
2260 ret = run_delayed_data_ref(trans, root, node, extent_op,
2267 static noinline struct btrfs_delayed_ref_node *
2268 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2270 struct rb_node *node;
2271 struct btrfs_delayed_ref_node *ref;
2272 int action = BTRFS_ADD_DELAYED_REF;
2275 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2276 * this prevents ref count from going down to zero when
2277 * there still are pending delayed ref.
2279 node = rb_prev(&head->node.rb_node);
2283 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2285 if (ref->bytenr != head->node.bytenr)
2287 if (ref->action == action)
2289 node = rb_prev(node);
2291 if (action == BTRFS_ADD_DELAYED_REF) {
2292 action = BTRFS_DROP_DELAYED_REF;
2299 * Returns 0 on success or if called with an already aborted transaction.
2300 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2302 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2303 struct btrfs_root *root,
2304 struct list_head *cluster)
2306 struct btrfs_delayed_ref_root *delayed_refs;
2307 struct btrfs_delayed_ref_node *ref;
2308 struct btrfs_delayed_ref_head *locked_ref = NULL;
2309 struct btrfs_delayed_extent_op *extent_op;
2310 struct btrfs_fs_info *fs_info = root->fs_info;
2313 int must_insert_reserved = 0;
2315 delayed_refs = &trans->transaction->delayed_refs;
2318 /* pick a new head ref from the cluster list */
2319 if (list_empty(cluster))
2322 locked_ref = list_entry(cluster->next,
2323 struct btrfs_delayed_ref_head, cluster);
2325 /* grab the lock that says we are going to process
2326 * all the refs for this head */
2327 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2330 * we may have dropped the spin lock to get the head
2331 * mutex lock, and that might have given someone else
2332 * time to free the head. If that's true, it has been
2333 * removed from our list and we can move on.
2335 if (ret == -EAGAIN) {
2343 * We need to try and merge add/drops of the same ref since we
2344 * can run into issues with relocate dropping the implicit ref
2345 * and then it being added back again before the drop can
2346 * finish. If we merged anything we need to re-loop so we can
2349 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2353 * locked_ref is the head node, so we have to go one
2354 * node back for any delayed ref updates
2356 ref = select_delayed_ref(locked_ref);
2358 if (ref && ref->seq &&
2359 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2361 * there are still refs with lower seq numbers in the
2362 * process of being added. Don't run this ref yet.
2364 list_del_init(&locked_ref->cluster);
2365 btrfs_delayed_ref_unlock(locked_ref);
2367 delayed_refs->num_heads_ready++;
2368 spin_unlock(&delayed_refs->lock);
2370 spin_lock(&delayed_refs->lock);
2375 * record the must insert reserved flag before we
2376 * drop the spin lock.
2378 must_insert_reserved = locked_ref->must_insert_reserved;
2379 locked_ref->must_insert_reserved = 0;
2381 extent_op = locked_ref->extent_op;
2382 locked_ref->extent_op = NULL;
2385 /* All delayed refs have been processed, Go ahead
2386 * and send the head node to run_one_delayed_ref,
2387 * so that any accounting fixes can happen
2389 ref = &locked_ref->node;
2391 if (extent_op && must_insert_reserved) {
2392 btrfs_free_delayed_extent_op(extent_op);
2397 spin_unlock(&delayed_refs->lock);
2399 ret = run_delayed_extent_op(trans, root,
2401 btrfs_free_delayed_extent_op(extent_op);
2404 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2405 spin_lock(&delayed_refs->lock);
2406 btrfs_delayed_ref_unlock(locked_ref);
2415 rb_erase(&ref->rb_node, &delayed_refs->root);
2416 delayed_refs->num_entries--;
2417 if (!btrfs_delayed_ref_is_head(ref)) {
2419 * when we play the delayed ref, also correct the
2422 switch (ref->action) {
2423 case BTRFS_ADD_DELAYED_REF:
2424 case BTRFS_ADD_DELAYED_EXTENT:
2425 locked_ref->node.ref_mod -= ref->ref_mod;
2427 case BTRFS_DROP_DELAYED_REF:
2428 locked_ref->node.ref_mod += ref->ref_mod;
2434 spin_unlock(&delayed_refs->lock);
2436 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2437 must_insert_reserved);
2439 btrfs_free_delayed_extent_op(extent_op);
2441 btrfs_delayed_ref_unlock(locked_ref);
2442 btrfs_put_delayed_ref(ref);
2443 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2444 spin_lock(&delayed_refs->lock);
2449 * If this node is a head, that means all the refs in this head
2450 * have been dealt with, and we will pick the next head to deal
2451 * with, so we must unlock the head and drop it from the cluster
2452 * list before we release it.
2454 if (btrfs_delayed_ref_is_head(ref)) {
2455 list_del_init(&locked_ref->cluster);
2456 btrfs_delayed_ref_unlock(locked_ref);
2459 btrfs_put_delayed_ref(ref);
2463 spin_lock(&delayed_refs->lock);
2468 #ifdef SCRAMBLE_DELAYED_REFS
2470 * Normally delayed refs get processed in ascending bytenr order. This
2471 * correlates in most cases to the order added. To expose dependencies on this
2472 * order, we start to process the tree in the middle instead of the beginning
2474 static u64 find_middle(struct rb_root *root)
2476 struct rb_node *n = root->rb_node;
2477 struct btrfs_delayed_ref_node *entry;
2480 u64 first = 0, last = 0;
2484 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2485 first = entry->bytenr;
2489 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2490 last = entry->bytenr;
2495 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2496 WARN_ON(!entry->in_tree);
2498 middle = entry->bytenr;
2511 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2512 struct btrfs_fs_info *fs_info)
2514 struct qgroup_update *qgroup_update;
2517 if (list_empty(&trans->qgroup_ref_list) !=
2518 !trans->delayed_ref_elem.seq) {
2519 /* list without seq or seq without list */
2521 "qgroup accounting update error, list is%s empty, seq is %llu",
2522 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2523 trans->delayed_ref_elem.seq);
2527 if (!trans->delayed_ref_elem.seq)
2530 while (!list_empty(&trans->qgroup_ref_list)) {
2531 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2532 struct qgroup_update, list);
2533 list_del(&qgroup_update->list);
2535 ret = btrfs_qgroup_account_ref(
2536 trans, fs_info, qgroup_update->node,
2537 qgroup_update->extent_op);
2538 kfree(qgroup_update);
2541 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2546 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2549 int val = atomic_read(&delayed_refs->ref_seq);
2551 if (val < seq || val >= seq + count)
2557 * this starts processing the delayed reference count updates and
2558 * extent insertions we have queued up so far. count can be
2559 * 0, which means to process everything in the tree at the start
2560 * of the run (but not newly added entries), or it can be some target
2561 * number you'd like to process.
2563 * Returns 0 on success or if called with an aborted transaction
2564 * Returns <0 on error and aborts the transaction
2566 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2567 struct btrfs_root *root, unsigned long count)
2569 struct rb_node *node;
2570 struct btrfs_delayed_ref_root *delayed_refs;
2571 struct btrfs_delayed_ref_node *ref;
2572 struct list_head cluster;
2575 int run_all = count == (unsigned long)-1;
2579 /* We'll clean this up in btrfs_cleanup_transaction */
2583 if (root == root->fs_info->extent_root)
2584 root = root->fs_info->tree_root;
2586 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2588 delayed_refs = &trans->transaction->delayed_refs;
2589 INIT_LIST_HEAD(&cluster);
2591 count = delayed_refs->num_entries * 2;
2595 if (!run_all && !run_most) {
2597 int seq = atomic_read(&delayed_refs->ref_seq);
2600 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2602 DEFINE_WAIT(__wait);
2603 if (delayed_refs->num_entries < 16348)
2606 prepare_to_wait(&delayed_refs->wait, &__wait,
2607 TASK_UNINTERRUPTIBLE);
2609 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2612 finish_wait(&delayed_refs->wait, &__wait);
2614 if (!refs_newer(delayed_refs, seq, 256))
2619 finish_wait(&delayed_refs->wait, &__wait);
2625 atomic_inc(&delayed_refs->procs_running_refs);
2630 spin_lock(&delayed_refs->lock);
2632 #ifdef SCRAMBLE_DELAYED_REFS
2633 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2637 if (!(run_all || run_most) &&
2638 delayed_refs->num_heads_ready < 64)
2642 * go find something we can process in the rbtree. We start at
2643 * the beginning of the tree, and then build a cluster
2644 * of refs to process starting at the first one we are able to
2647 delayed_start = delayed_refs->run_delayed_start;
2648 ret = btrfs_find_ref_cluster(trans, &cluster,
2649 delayed_refs->run_delayed_start);
2653 ret = run_clustered_refs(trans, root, &cluster);
2655 btrfs_release_ref_cluster(&cluster);
2656 spin_unlock(&delayed_refs->lock);
2657 btrfs_abort_transaction(trans, root, ret);
2658 atomic_dec(&delayed_refs->procs_running_refs);
2662 atomic_add(ret, &delayed_refs->ref_seq);
2664 count -= min_t(unsigned long, ret, count);
2669 if (delayed_start >= delayed_refs->run_delayed_start) {
2672 * btrfs_find_ref_cluster looped. let's do one
2673 * more cycle. if we don't run any delayed ref
2674 * during that cycle (because we can't because
2675 * all of them are blocked), bail out.
2680 * no runnable refs left, stop trying
2687 /* refs were run, let's reset staleness detection */
2693 if (!list_empty(&trans->new_bgs)) {
2694 spin_unlock(&delayed_refs->lock);
2695 btrfs_create_pending_block_groups(trans, root);
2696 spin_lock(&delayed_refs->lock);
2699 node = rb_first(&delayed_refs->root);
2702 count = (unsigned long)-1;
2705 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2707 if (btrfs_delayed_ref_is_head(ref)) {
2708 struct btrfs_delayed_ref_head *head;
2710 head = btrfs_delayed_node_to_head(ref);
2711 atomic_inc(&ref->refs);
2713 spin_unlock(&delayed_refs->lock);
2715 * Mutex was contended, block until it's
2716 * released and try again
2718 mutex_lock(&head->mutex);
2719 mutex_unlock(&head->mutex);
2721 btrfs_put_delayed_ref(ref);
2725 node = rb_next(node);
2727 spin_unlock(&delayed_refs->lock);
2728 schedule_timeout(1);
2732 atomic_dec(&delayed_refs->procs_running_refs);
2734 if (waitqueue_active(&delayed_refs->wait))
2735 wake_up(&delayed_refs->wait);
2737 spin_unlock(&delayed_refs->lock);
2738 assert_qgroups_uptodate(trans);
2742 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2743 struct btrfs_root *root,
2744 u64 bytenr, u64 num_bytes, u64 flags,
2747 struct btrfs_delayed_extent_op *extent_op;
2750 extent_op = btrfs_alloc_delayed_extent_op();
2754 extent_op->flags_to_set = flags;
2755 extent_op->update_flags = 1;
2756 extent_op->update_key = 0;
2757 extent_op->is_data = is_data ? 1 : 0;
2759 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2760 num_bytes, extent_op);
2762 btrfs_free_delayed_extent_op(extent_op);
2766 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2767 struct btrfs_root *root,
2768 struct btrfs_path *path,
2769 u64 objectid, u64 offset, u64 bytenr)
2771 struct btrfs_delayed_ref_head *head;
2772 struct btrfs_delayed_ref_node *ref;
2773 struct btrfs_delayed_data_ref *data_ref;
2774 struct btrfs_delayed_ref_root *delayed_refs;
2775 struct rb_node *node;
2779 delayed_refs = &trans->transaction->delayed_refs;
2780 spin_lock(&delayed_refs->lock);
2781 head = btrfs_find_delayed_ref_head(trans, bytenr);
2785 if (!mutex_trylock(&head->mutex)) {
2786 atomic_inc(&head->node.refs);
2787 spin_unlock(&delayed_refs->lock);
2789 btrfs_release_path(path);
2792 * Mutex was contended, block until it's released and let
2795 mutex_lock(&head->mutex);
2796 mutex_unlock(&head->mutex);
2797 btrfs_put_delayed_ref(&head->node);
2801 node = rb_prev(&head->node.rb_node);
2805 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2807 if (ref->bytenr != bytenr)
2811 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2814 data_ref = btrfs_delayed_node_to_data_ref(ref);
2816 node = rb_prev(node);
2820 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2821 if (ref->bytenr == bytenr && ref->seq == seq)
2825 if (data_ref->root != root->root_key.objectid ||
2826 data_ref->objectid != objectid || data_ref->offset != offset)
2831 mutex_unlock(&head->mutex);
2833 spin_unlock(&delayed_refs->lock);
2837 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2838 struct btrfs_root *root,
2839 struct btrfs_path *path,
2840 u64 objectid, u64 offset, u64 bytenr)
2842 struct btrfs_root *extent_root = root->fs_info->extent_root;
2843 struct extent_buffer *leaf;
2844 struct btrfs_extent_data_ref *ref;
2845 struct btrfs_extent_inline_ref *iref;
2846 struct btrfs_extent_item *ei;
2847 struct btrfs_key key;
2851 key.objectid = bytenr;
2852 key.offset = (u64)-1;
2853 key.type = BTRFS_EXTENT_ITEM_KEY;
2855 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2858 BUG_ON(ret == 0); /* Corruption */
2861 if (path->slots[0] == 0)
2865 leaf = path->nodes[0];
2866 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2868 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2872 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2874 if (item_size < sizeof(*ei)) {
2875 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2879 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2881 if (item_size != sizeof(*ei) +
2882 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2885 if (btrfs_extent_generation(leaf, ei) <=
2886 btrfs_root_last_snapshot(&root->root_item))
2889 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2890 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2891 BTRFS_EXTENT_DATA_REF_KEY)
2894 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2895 if (btrfs_extent_refs(leaf, ei) !=
2896 btrfs_extent_data_ref_count(leaf, ref) ||
2897 btrfs_extent_data_ref_root(leaf, ref) !=
2898 root->root_key.objectid ||
2899 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2900 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2908 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2909 struct btrfs_root *root,
2910 u64 objectid, u64 offset, u64 bytenr)
2912 struct btrfs_path *path;
2916 path = btrfs_alloc_path();
2921 ret = check_committed_ref(trans, root, path, objectid,
2923 if (ret && ret != -ENOENT)
2926 ret2 = check_delayed_ref(trans, root, path, objectid,
2928 } while (ret2 == -EAGAIN);
2930 if (ret2 && ret2 != -ENOENT) {
2935 if (ret != -ENOENT || ret2 != -ENOENT)
2938 btrfs_free_path(path);
2939 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2944 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2945 struct btrfs_root *root,
2946 struct extent_buffer *buf,
2947 int full_backref, int inc, int for_cow)
2954 struct btrfs_key key;
2955 struct btrfs_file_extent_item *fi;
2959 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2960 u64, u64, u64, u64, u64, u64, int);
2962 ref_root = btrfs_header_owner(buf);
2963 nritems = btrfs_header_nritems(buf);
2964 level = btrfs_header_level(buf);
2966 if (!root->ref_cows && level == 0)
2970 process_func = btrfs_inc_extent_ref;
2972 process_func = btrfs_free_extent;
2975 parent = buf->start;
2979 for (i = 0; i < nritems; i++) {
2981 btrfs_item_key_to_cpu(buf, &key, i);
2982 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2984 fi = btrfs_item_ptr(buf, i,
2985 struct btrfs_file_extent_item);
2986 if (btrfs_file_extent_type(buf, fi) ==
2987 BTRFS_FILE_EXTENT_INLINE)
2989 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2993 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2994 key.offset -= btrfs_file_extent_offset(buf, fi);
2995 ret = process_func(trans, root, bytenr, num_bytes,
2996 parent, ref_root, key.objectid,
2997 key.offset, for_cow);
3001 bytenr = btrfs_node_blockptr(buf, i);
3002 num_bytes = btrfs_level_size(root, level - 1);
3003 ret = process_func(trans, root, bytenr, num_bytes,
3004 parent, ref_root, level - 1, 0,
3015 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3016 struct extent_buffer *buf, int full_backref, int for_cow)
3018 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3021 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3022 struct extent_buffer *buf, int full_backref, int for_cow)
3024 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3027 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3028 struct btrfs_root *root,
3029 struct btrfs_path *path,
3030 struct btrfs_block_group_cache *cache)
3033 struct btrfs_root *extent_root = root->fs_info->extent_root;
3035 struct extent_buffer *leaf;
3037 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3040 BUG_ON(ret); /* Corruption */
3042 leaf = path->nodes[0];
3043 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3044 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3045 btrfs_mark_buffer_dirty(leaf);
3046 btrfs_release_path(path);
3049 btrfs_abort_transaction(trans, root, ret);
3056 static struct btrfs_block_group_cache *
3057 next_block_group(struct btrfs_root *root,
3058 struct btrfs_block_group_cache *cache)
3060 struct rb_node *node;
3061 spin_lock(&root->fs_info->block_group_cache_lock);
3062 node = rb_next(&cache->cache_node);
3063 btrfs_put_block_group(cache);
3065 cache = rb_entry(node, struct btrfs_block_group_cache,
3067 btrfs_get_block_group(cache);
3070 spin_unlock(&root->fs_info->block_group_cache_lock);
3074 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3075 struct btrfs_trans_handle *trans,
3076 struct btrfs_path *path)
3078 struct btrfs_root *root = block_group->fs_info->tree_root;
3079 struct inode *inode = NULL;
3081 int dcs = BTRFS_DC_ERROR;
3087 * If this block group is smaller than 100 megs don't bother caching the
3090 if (block_group->key.offset < (100 * 1024 * 1024)) {
3091 spin_lock(&block_group->lock);
3092 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3093 spin_unlock(&block_group->lock);
3098 inode = lookup_free_space_inode(root, block_group, path);
3099 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3100 ret = PTR_ERR(inode);
3101 btrfs_release_path(path);
3105 if (IS_ERR(inode)) {
3109 if (block_group->ro)
3112 ret = create_free_space_inode(root, trans, block_group, path);
3118 /* We've already setup this transaction, go ahead and exit */
3119 if (block_group->cache_generation == trans->transid &&
3120 i_size_read(inode)) {
3121 dcs = BTRFS_DC_SETUP;
3126 * We want to set the generation to 0, that way if anything goes wrong
3127 * from here on out we know not to trust this cache when we load up next
3130 BTRFS_I(inode)->generation = 0;
3131 ret = btrfs_update_inode(trans, root, inode);
3134 if (i_size_read(inode) > 0) {
3135 ret = btrfs_truncate_free_space_cache(root, trans, path,
3141 spin_lock(&block_group->lock);
3142 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3143 !btrfs_test_opt(root, SPACE_CACHE)) {
3145 * don't bother trying to write stuff out _if_
3146 * a) we're not cached,
3147 * b) we're with nospace_cache mount option.
3149 dcs = BTRFS_DC_WRITTEN;
3150 spin_unlock(&block_group->lock);
3153 spin_unlock(&block_group->lock);
3156 * Try to preallocate enough space based on how big the block group is.
3157 * Keep in mind this has to include any pinned space which could end up
3158 * taking up quite a bit since it's not folded into the other space
3161 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3166 num_pages *= PAGE_CACHE_SIZE;
3168 ret = btrfs_check_data_free_space(inode, num_pages);
3172 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3173 num_pages, num_pages,
3176 dcs = BTRFS_DC_SETUP;
3177 btrfs_free_reserved_data_space(inode, num_pages);
3182 btrfs_release_path(path);
3184 spin_lock(&block_group->lock);
3185 if (!ret && dcs == BTRFS_DC_SETUP)
3186 block_group->cache_generation = trans->transid;
3187 block_group->disk_cache_state = dcs;
3188 spin_unlock(&block_group->lock);
3193 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3194 struct btrfs_root *root)
3196 struct btrfs_block_group_cache *cache;
3198 struct btrfs_path *path;
3201 path = btrfs_alloc_path();
3207 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3209 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3211 cache = next_block_group(root, cache);
3219 err = cache_save_setup(cache, trans, path);
3220 last = cache->key.objectid + cache->key.offset;
3221 btrfs_put_block_group(cache);
3226 err = btrfs_run_delayed_refs(trans, root,
3228 if (err) /* File system offline */
3232 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3234 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3235 btrfs_put_block_group(cache);
3241 cache = next_block_group(root, cache);
3250 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3251 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3253 last = cache->key.objectid + cache->key.offset;
3255 err = write_one_cache_group(trans, root, path, cache);
3256 if (err) /* File system offline */
3259 btrfs_put_block_group(cache);
3264 * I don't think this is needed since we're just marking our
3265 * preallocated extent as written, but just in case it can't
3269 err = btrfs_run_delayed_refs(trans, root,
3271 if (err) /* File system offline */
3275 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3278 * Really this shouldn't happen, but it could if we
3279 * couldn't write the entire preallocated extent and
3280 * splitting the extent resulted in a new block.
3283 btrfs_put_block_group(cache);
3286 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3288 cache = next_block_group(root, cache);
3297 err = btrfs_write_out_cache(root, trans, cache, path);
3300 * If we didn't have an error then the cache state is still
3301 * NEED_WRITE, so we can set it to WRITTEN.
3303 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3304 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3305 last = cache->key.objectid + cache->key.offset;
3306 btrfs_put_block_group(cache);
3310 btrfs_free_path(path);
3314 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3316 struct btrfs_block_group_cache *block_group;
3319 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3320 if (!block_group || block_group->ro)
3323 btrfs_put_block_group(block_group);
3327 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3328 u64 total_bytes, u64 bytes_used,
3329 struct btrfs_space_info **space_info)
3331 struct btrfs_space_info *found;
3335 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3336 BTRFS_BLOCK_GROUP_RAID10))
3341 found = __find_space_info(info, flags);
3343 spin_lock(&found->lock);
3344 found->total_bytes += total_bytes;
3345 found->disk_total += total_bytes * factor;
3346 found->bytes_used += bytes_used;
3347 found->disk_used += bytes_used * factor;
3349 spin_unlock(&found->lock);
3350 *space_info = found;
3353 found = kzalloc(sizeof(*found), GFP_NOFS);
3357 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3358 INIT_LIST_HEAD(&found->block_groups[i]);
3359 init_rwsem(&found->groups_sem);
3360 spin_lock_init(&found->lock);
3361 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3362 found->total_bytes = total_bytes;
3363 found->disk_total = total_bytes * factor;
3364 found->bytes_used = bytes_used;
3365 found->disk_used = bytes_used * factor;
3366 found->bytes_pinned = 0;
3367 found->bytes_reserved = 0;
3368 found->bytes_readonly = 0;
3369 found->bytes_may_use = 0;
3371 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3372 found->chunk_alloc = 0;
3374 init_waitqueue_head(&found->wait);
3375 *space_info = found;
3376 list_add_rcu(&found->list, &info->space_info);
3377 if (flags & BTRFS_BLOCK_GROUP_DATA)
3378 info->data_sinfo = found;
3382 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3384 u64 extra_flags = chunk_to_extended(flags) &
3385 BTRFS_EXTENDED_PROFILE_MASK;
3387 write_seqlock(&fs_info->profiles_lock);
3388 if (flags & BTRFS_BLOCK_GROUP_DATA)
3389 fs_info->avail_data_alloc_bits |= extra_flags;
3390 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3391 fs_info->avail_metadata_alloc_bits |= extra_flags;
3392 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3393 fs_info->avail_system_alloc_bits |= extra_flags;
3394 write_sequnlock(&fs_info->profiles_lock);
3398 * returns target flags in extended format or 0 if restripe for this
3399 * chunk_type is not in progress
3401 * should be called with either volume_mutex or balance_lock held
3403 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3405 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3411 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3412 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3413 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3414 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3415 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3416 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3417 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3418 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3419 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3426 * @flags: available profiles in extended format (see ctree.h)
3428 * Returns reduced profile in chunk format. If profile changing is in
3429 * progress (either running or paused) picks the target profile (if it's
3430 * already available), otherwise falls back to plain reducing.
3432 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3435 * we add in the count of missing devices because we want
3436 * to make sure that any RAID levels on a degraded FS
3437 * continue to be honored.
3439 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3440 root->fs_info->fs_devices->missing_devices;
3445 * see if restripe for this chunk_type is in progress, if so
3446 * try to reduce to the target profile
3448 spin_lock(&root->fs_info->balance_lock);
3449 target = get_restripe_target(root->fs_info, flags);
3451 /* pick target profile only if it's already available */
3452 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3453 spin_unlock(&root->fs_info->balance_lock);
3454 return extended_to_chunk(target);
3457 spin_unlock(&root->fs_info->balance_lock);
3459 /* First, mask out the RAID levels which aren't possible */
3460 if (num_devices == 1)
3461 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3462 BTRFS_BLOCK_GROUP_RAID5);
3463 if (num_devices < 3)
3464 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3465 if (num_devices < 4)
3466 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3468 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3469 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3470 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3473 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3474 tmp = BTRFS_BLOCK_GROUP_RAID6;
3475 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3476 tmp = BTRFS_BLOCK_GROUP_RAID5;
3477 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3478 tmp = BTRFS_BLOCK_GROUP_RAID10;
3479 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3480 tmp = BTRFS_BLOCK_GROUP_RAID1;
3481 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3482 tmp = BTRFS_BLOCK_GROUP_RAID0;
3484 return extended_to_chunk(flags | tmp);
3487 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3492 seq = read_seqbegin(&root->fs_info->profiles_lock);
3494 if (flags & BTRFS_BLOCK_GROUP_DATA)
3495 flags |= root->fs_info->avail_data_alloc_bits;
3496 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3497 flags |= root->fs_info->avail_system_alloc_bits;
3498 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3499 flags |= root->fs_info->avail_metadata_alloc_bits;
3500 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3502 return btrfs_reduce_alloc_profile(root, flags);
3505 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3511 flags = BTRFS_BLOCK_GROUP_DATA;
3512 else if (root == root->fs_info->chunk_root)
3513 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3515 flags = BTRFS_BLOCK_GROUP_METADATA;
3517 ret = get_alloc_profile(root, flags);
3522 * This will check the space that the inode allocates from to make sure we have
3523 * enough space for bytes.
3525 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3527 struct btrfs_space_info *data_sinfo;
3528 struct btrfs_root *root = BTRFS_I(inode)->root;
3529 struct btrfs_fs_info *fs_info = root->fs_info;
3531 int ret = 0, committed = 0, alloc_chunk = 1;
3533 /* make sure bytes are sectorsize aligned */
3534 bytes = ALIGN(bytes, root->sectorsize);
3536 if (root == root->fs_info->tree_root ||
3537 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3542 data_sinfo = fs_info->data_sinfo;
3547 /* make sure we have enough space to handle the data first */
3548 spin_lock(&data_sinfo->lock);
3549 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3550 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3551 data_sinfo->bytes_may_use;
3553 if (used + bytes > data_sinfo->total_bytes) {
3554 struct btrfs_trans_handle *trans;
3557 * if we don't have enough free bytes in this space then we need
3558 * to alloc a new chunk.
3560 if (!data_sinfo->full && alloc_chunk) {
3563 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3564 spin_unlock(&data_sinfo->lock);
3566 alloc_target = btrfs_get_alloc_profile(root, 1);
3567 trans = btrfs_join_transaction(root);
3569 return PTR_ERR(trans);
3571 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3573 CHUNK_ALLOC_NO_FORCE);
3574 btrfs_end_transaction(trans, root);
3583 data_sinfo = fs_info->data_sinfo;
3589 * If we have less pinned bytes than we want to allocate then
3590 * don't bother committing the transaction, it won't help us.
3592 if (data_sinfo->bytes_pinned < bytes)
3594 spin_unlock(&data_sinfo->lock);
3596 /* commit the current transaction and try again */
3599 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3601 trans = btrfs_join_transaction(root);
3603 return PTR_ERR(trans);
3604 ret = btrfs_commit_transaction(trans, root);
3612 data_sinfo->bytes_may_use += bytes;
3613 trace_btrfs_space_reservation(root->fs_info, "space_info",
3614 data_sinfo->flags, bytes, 1);
3615 spin_unlock(&data_sinfo->lock);
3621 * Called if we need to clear a data reservation for this inode.
3623 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3625 struct btrfs_root *root = BTRFS_I(inode)->root;
3626 struct btrfs_space_info *data_sinfo;
3628 /* make sure bytes are sectorsize aligned */
3629 bytes = ALIGN(bytes, root->sectorsize);
3631 data_sinfo = root->fs_info->data_sinfo;
3632 spin_lock(&data_sinfo->lock);
3633 data_sinfo->bytes_may_use -= bytes;
3634 trace_btrfs_space_reservation(root->fs_info, "space_info",
3635 data_sinfo->flags, bytes, 0);
3636 spin_unlock(&data_sinfo->lock);
3639 static void force_metadata_allocation(struct btrfs_fs_info *info)
3641 struct list_head *head = &info->space_info;
3642 struct btrfs_space_info *found;
3645 list_for_each_entry_rcu(found, head, list) {
3646 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3647 found->force_alloc = CHUNK_ALLOC_FORCE;
3652 static int should_alloc_chunk(struct btrfs_root *root,
3653 struct btrfs_space_info *sinfo, int force)
3655 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3656 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3657 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3660 if (force == CHUNK_ALLOC_FORCE)
3664 * We need to take into account the global rsv because for all intents
3665 * and purposes it's used space. Don't worry about locking the
3666 * global_rsv, it doesn't change except when the transaction commits.
3668 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3669 num_allocated += global_rsv->size;
3672 * in limited mode, we want to have some free space up to
3673 * about 1% of the FS size.
3675 if (force == CHUNK_ALLOC_LIMITED) {
3676 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3677 thresh = max_t(u64, 64 * 1024 * 1024,
3678 div_factor_fine(thresh, 1));
3680 if (num_bytes - num_allocated < thresh)
3684 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3689 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3693 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3694 BTRFS_BLOCK_GROUP_RAID0 |
3695 BTRFS_BLOCK_GROUP_RAID5 |
3696 BTRFS_BLOCK_GROUP_RAID6))
3697 num_dev = root->fs_info->fs_devices->rw_devices;
3698 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3701 num_dev = 1; /* DUP or single */
3703 /* metadata for updaing devices and chunk tree */
3704 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3707 static void check_system_chunk(struct btrfs_trans_handle *trans,
3708 struct btrfs_root *root, u64 type)
3710 struct btrfs_space_info *info;
3714 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3715 spin_lock(&info->lock);
3716 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3717 info->bytes_reserved - info->bytes_readonly;
3718 spin_unlock(&info->lock);
3720 thresh = get_system_chunk_thresh(root, type);
3721 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3722 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3723 left, thresh, type);
3724 dump_space_info(info, 0, 0);
3727 if (left < thresh) {
3730 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3731 btrfs_alloc_chunk(trans, root, flags);
3735 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3736 struct btrfs_root *extent_root, u64 flags, int force)
3738 struct btrfs_space_info *space_info;
3739 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3740 int wait_for_alloc = 0;
3743 /* Don't re-enter if we're already allocating a chunk */
3744 if (trans->allocating_chunk)
3747 space_info = __find_space_info(extent_root->fs_info, flags);
3749 ret = update_space_info(extent_root->fs_info, flags,
3751 BUG_ON(ret); /* -ENOMEM */
3753 BUG_ON(!space_info); /* Logic error */
3756 spin_lock(&space_info->lock);
3757 if (force < space_info->force_alloc)
3758 force = space_info->force_alloc;
3759 if (space_info->full) {
3760 spin_unlock(&space_info->lock);
3764 if (!should_alloc_chunk(extent_root, space_info, force)) {
3765 spin_unlock(&space_info->lock);
3767 } else if (space_info->chunk_alloc) {
3770 space_info->chunk_alloc = 1;
3773 spin_unlock(&space_info->lock);
3775 mutex_lock(&fs_info->chunk_mutex);
3778 * The chunk_mutex is held throughout the entirety of a chunk
3779 * allocation, so once we've acquired the chunk_mutex we know that the
3780 * other guy is done and we need to recheck and see if we should
3783 if (wait_for_alloc) {
3784 mutex_unlock(&fs_info->chunk_mutex);
3789 trans->allocating_chunk = true;
3792 * If we have mixed data/metadata chunks we want to make sure we keep
3793 * allocating mixed chunks instead of individual chunks.
3795 if (btrfs_mixed_space_info(space_info))
3796 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3799 * if we're doing a data chunk, go ahead and make sure that
3800 * we keep a reasonable number of metadata chunks allocated in the
3803 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3804 fs_info->data_chunk_allocations++;
3805 if (!(fs_info->data_chunk_allocations %
3806 fs_info->metadata_ratio))
3807 force_metadata_allocation(fs_info);
3811 * Check if we have enough space in SYSTEM chunk because we may need
3812 * to update devices.
3814 check_system_chunk(trans, extent_root, flags);
3816 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3817 trans->allocating_chunk = false;
3819 spin_lock(&space_info->lock);
3820 if (ret < 0 && ret != -ENOSPC)
3823 space_info->full = 1;
3827 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3829 space_info->chunk_alloc = 0;
3830 spin_unlock(&space_info->lock);
3831 mutex_unlock(&fs_info->chunk_mutex);
3835 static int can_overcommit(struct btrfs_root *root,
3836 struct btrfs_space_info *space_info, u64 bytes,
3837 enum btrfs_reserve_flush_enum flush)
3839 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3840 u64 profile = btrfs_get_alloc_profile(root, 0);
3846 used = space_info->bytes_used + space_info->bytes_reserved +
3847 space_info->bytes_pinned + space_info->bytes_readonly;
3849 spin_lock(&global_rsv->lock);
3850 rsv_size = global_rsv->size;
3851 spin_unlock(&global_rsv->lock);
3854 * We only want to allow over committing if we have lots of actual space
3855 * free, but if we don't have enough space to handle the global reserve
3856 * space then we could end up having a real enospc problem when trying
3857 * to allocate a chunk or some other such important allocation.
3860 if (used + rsv_size >= space_info->total_bytes)
3863 used += space_info->bytes_may_use;
3865 spin_lock(&root->fs_info->free_chunk_lock);
3866 avail = root->fs_info->free_chunk_space;
3867 spin_unlock(&root->fs_info->free_chunk_lock);
3870 * If we have dup, raid1 or raid10 then only half of the free
3871 * space is actually useable. For raid56, the space info used
3872 * doesn't include the parity drive, so we don't have to
3875 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3876 BTRFS_BLOCK_GROUP_RAID1 |
3877 BTRFS_BLOCK_GROUP_RAID10))
3880 to_add = space_info->total_bytes;
3883 * If we aren't flushing all things, let us overcommit up to
3884 * 1/2th of the space. If we can flush, don't let us overcommit
3885 * too much, let it overcommit up to 1/8 of the space.
3887 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3893 * Limit the overcommit to the amount of free space we could possibly
3894 * allocate for chunks.
3896 to_add = min(avail, to_add);
3898 if (used + bytes < space_info->total_bytes + to_add)
3903 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3904 unsigned long nr_pages)
3906 struct super_block *sb = root->fs_info->sb;
3909 /* If we can not start writeback, just sync all the delalloc file. */
3910 started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
3911 WB_REASON_FS_FREE_SPACE);
3914 * We needn't worry the filesystem going from r/w to r/o though
3915 * we don't acquire ->s_umount mutex, because the filesystem
3916 * should guarantee the delalloc inodes list be empty after
3917 * the filesystem is readonly(all dirty pages are written to
3920 btrfs_start_delalloc_inodes(root, 0);
3921 if (!current->journal_info)
3922 btrfs_wait_ordered_extents(root, 0);
3927 * shrink metadata reservation for delalloc
3929 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3932 struct btrfs_block_rsv *block_rsv;
3933 struct btrfs_space_info *space_info;
3934 struct btrfs_trans_handle *trans;
3938 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3940 enum btrfs_reserve_flush_enum flush;
3942 trans = (struct btrfs_trans_handle *)current->journal_info;
3943 block_rsv = &root->fs_info->delalloc_block_rsv;
3944 space_info = block_rsv->space_info;
3947 delalloc_bytes = percpu_counter_sum_positive(
3948 &root->fs_info->delalloc_bytes);
3949 if (delalloc_bytes == 0) {
3952 btrfs_wait_ordered_extents(root, 0);
3956 while (delalloc_bytes && loops < 3) {
3957 max_reclaim = min(delalloc_bytes, to_reclaim);
3958 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3959 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3961 * We need to wait for the async pages to actually start before
3964 wait_event(root->fs_info->async_submit_wait,
3965 !atomic_read(&root->fs_info->async_delalloc_pages));
3968 flush = BTRFS_RESERVE_FLUSH_ALL;
3970 flush = BTRFS_RESERVE_NO_FLUSH;
3971 spin_lock(&space_info->lock);
3972 if (can_overcommit(root, space_info, orig, flush)) {
3973 spin_unlock(&space_info->lock);
3976 spin_unlock(&space_info->lock);
3979 if (wait_ordered && !trans) {
3980 btrfs_wait_ordered_extents(root, 0);
3982 time_left = schedule_timeout_killable(1);
3987 delalloc_bytes = percpu_counter_sum_positive(
3988 &root->fs_info->delalloc_bytes);
3993 * maybe_commit_transaction - possibly commit the transaction if its ok to
3994 * @root - the root we're allocating for
3995 * @bytes - the number of bytes we want to reserve
3996 * @force - force the commit
3998 * This will check to make sure that committing the transaction will actually
3999 * get us somewhere and then commit the transaction if it does. Otherwise it
4000 * will return -ENOSPC.
4002 static int may_commit_transaction(struct btrfs_root *root,
4003 struct btrfs_space_info *space_info,
4004 u64 bytes, int force)
4006 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4007 struct btrfs_trans_handle *trans;
4009 trans = (struct btrfs_trans_handle *)current->journal_info;
4016 /* See if there is enough pinned space to make this reservation */
4017 spin_lock(&space_info->lock);
4018 if (space_info->bytes_pinned >= bytes) {
4019 spin_unlock(&space_info->lock);
4022 spin_unlock(&space_info->lock);
4025 * See if there is some space in the delayed insertion reservation for
4028 if (space_info != delayed_rsv->space_info)
4031 spin_lock(&space_info->lock);
4032 spin_lock(&delayed_rsv->lock);
4033 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
4034 spin_unlock(&delayed_rsv->lock);
4035 spin_unlock(&space_info->lock);
4038 spin_unlock(&delayed_rsv->lock);
4039 spin_unlock(&space_info->lock);
4042 trans = btrfs_join_transaction(root);
4046 return btrfs_commit_transaction(trans, root);
4050 FLUSH_DELAYED_ITEMS_NR = 1,
4051 FLUSH_DELAYED_ITEMS = 2,
4053 FLUSH_DELALLOC_WAIT = 4,
4058 static int flush_space(struct btrfs_root *root,
4059 struct btrfs_space_info *space_info, u64 num_bytes,
4060 u64 orig_bytes, int state)
4062 struct btrfs_trans_handle *trans;
4067 case FLUSH_DELAYED_ITEMS_NR:
4068 case FLUSH_DELAYED_ITEMS:
4069 if (state == FLUSH_DELAYED_ITEMS_NR) {
4070 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4072 nr = (int)div64_u64(num_bytes, bytes);
4079 trans = btrfs_join_transaction(root);
4080 if (IS_ERR(trans)) {
4081 ret = PTR_ERR(trans);
4084 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4085 btrfs_end_transaction(trans, root);
4087 case FLUSH_DELALLOC:
4088 case FLUSH_DELALLOC_WAIT:
4089 shrink_delalloc(root, num_bytes, orig_bytes,
4090 state == FLUSH_DELALLOC_WAIT);
4093 trans = btrfs_join_transaction(root);
4094 if (IS_ERR(trans)) {
4095 ret = PTR_ERR(trans);
4098 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4099 btrfs_get_alloc_profile(root, 0),
4100 CHUNK_ALLOC_NO_FORCE);
4101 btrfs_end_transaction(trans, root);
4106 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4116 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4117 * @root - the root we're allocating for
4118 * @block_rsv - the block_rsv we're allocating for
4119 * @orig_bytes - the number of bytes we want
4120 * @flush - whether or not we can flush to make our reservation
4122 * This will reserve orgi_bytes number of bytes from the space info associated
4123 * with the block_rsv. If there is not enough space it will make an attempt to
4124 * flush out space to make room. It will do this by flushing delalloc if
4125 * possible or committing the transaction. If flush is 0 then no attempts to
4126 * regain reservations will be made and this will fail if there is not enough
4129 static int reserve_metadata_bytes(struct btrfs_root *root,
4130 struct btrfs_block_rsv *block_rsv,
4132 enum btrfs_reserve_flush_enum flush)
4134 struct btrfs_space_info *space_info = block_rsv->space_info;
4136 u64 num_bytes = orig_bytes;
4137 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4139 bool flushing = false;
4143 spin_lock(&space_info->lock);
4145 * We only want to wait if somebody other than us is flushing and we
4146 * are actually allowed to flush all things.
4148 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4149 space_info->flush) {
4150 spin_unlock(&space_info->lock);
4152 * If we have a trans handle we can't wait because the flusher
4153 * may have to commit the transaction, which would mean we would
4154 * deadlock since we are waiting for the flusher to finish, but
4155 * hold the current transaction open.
4157 if (current->journal_info)
4159 ret = wait_event_killable(space_info->wait, !space_info->flush);
4160 /* Must have been killed, return */
4164 spin_lock(&space_info->lock);
4168 used = space_info->bytes_used + space_info->bytes_reserved +
4169 space_info->bytes_pinned + space_info->bytes_readonly +
4170 space_info->bytes_may_use;
4173 * The idea here is that we've not already over-reserved the block group
4174 * then we can go ahead and save our reservation first and then start
4175 * flushing if we need to. Otherwise if we've already overcommitted
4176 * lets start flushing stuff first and then come back and try to make
4179 if (used <= space_info->total_bytes) {
4180 if (used + orig_bytes <= space_info->total_bytes) {
4181 space_info->bytes_may_use += orig_bytes;
4182 trace_btrfs_space_reservation(root->fs_info,
4183 "space_info", space_info->flags, orig_bytes, 1);
4187 * Ok set num_bytes to orig_bytes since we aren't
4188 * overocmmitted, this way we only try and reclaim what
4191 num_bytes = orig_bytes;
4195 * Ok we're over committed, set num_bytes to the overcommitted
4196 * amount plus the amount of bytes that we need for this
4199 num_bytes = used - space_info->total_bytes +
4203 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4204 space_info->bytes_may_use += orig_bytes;
4205 trace_btrfs_space_reservation(root->fs_info, "space_info",
4206 space_info->flags, orig_bytes,
4212 * Couldn't make our reservation, save our place so while we're trying
4213 * to reclaim space we can actually use it instead of somebody else
4214 * stealing it from us.
4216 * We make the other tasks wait for the flush only when we can flush
4219 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4221 space_info->flush = 1;
4224 spin_unlock(&space_info->lock);
4226 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4229 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4234 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4235 * would happen. So skip delalloc flush.
4237 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4238 (flush_state == FLUSH_DELALLOC ||
4239 flush_state == FLUSH_DELALLOC_WAIT))
4240 flush_state = ALLOC_CHUNK;
4244 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4245 flush_state < COMMIT_TRANS)
4247 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4248 flush_state <= COMMIT_TRANS)
4252 if (ret == -ENOSPC &&
4253 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4254 struct btrfs_block_rsv *global_rsv =
4255 &root->fs_info->global_block_rsv;
4257 if (block_rsv != global_rsv &&
4258 !block_rsv_use_bytes(global_rsv, orig_bytes))
4262 spin_lock(&space_info->lock);
4263 space_info->flush = 0;
4264 wake_up_all(&space_info->wait);
4265 spin_unlock(&space_info->lock);
4270 static struct btrfs_block_rsv *get_block_rsv(
4271 const struct btrfs_trans_handle *trans,
4272 const struct btrfs_root *root)
4274 struct btrfs_block_rsv *block_rsv = NULL;
4277 block_rsv = trans->block_rsv;
4279 if (root == root->fs_info->csum_root && trans->adding_csums)
4280 block_rsv = trans->block_rsv;
4283 block_rsv = root->block_rsv;
4286 block_rsv = &root->fs_info->empty_block_rsv;
4291 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4295 spin_lock(&block_rsv->lock);
4296 if (block_rsv->reserved >= num_bytes) {
4297 block_rsv->reserved -= num_bytes;
4298 if (block_rsv->reserved < block_rsv->size)
4299 block_rsv->full = 0;
4302 spin_unlock(&block_rsv->lock);
4306 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4307 u64 num_bytes, int update_size)
4309 spin_lock(&block_rsv->lock);
4310 block_rsv->reserved += num_bytes;
4312 block_rsv->size += num_bytes;
4313 else if (block_rsv->reserved >= block_rsv->size)
4314 block_rsv->full = 1;
4315 spin_unlock(&block_rsv->lock);
4318 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4319 struct btrfs_block_rsv *block_rsv,
4320 struct btrfs_block_rsv *dest, u64 num_bytes)
4322 struct btrfs_space_info *space_info = block_rsv->space_info;
4324 spin_lock(&block_rsv->lock);
4325 if (num_bytes == (u64)-1)
4326 num_bytes = block_rsv->size;
4327 block_rsv->size -= num_bytes;
4328 if (block_rsv->reserved >= block_rsv->size) {
4329 num_bytes = block_rsv->reserved - block_rsv->size;
4330 block_rsv->reserved = block_rsv->size;
4331 block_rsv->full = 1;
4335 spin_unlock(&block_rsv->lock);
4337 if (num_bytes > 0) {
4339 spin_lock(&dest->lock);
4343 bytes_to_add = dest->size - dest->reserved;
4344 bytes_to_add = min(num_bytes, bytes_to_add);
4345 dest->reserved += bytes_to_add;
4346 if (dest->reserved >= dest->size)
4348 num_bytes -= bytes_to_add;
4350 spin_unlock(&dest->lock);
4353 spin_lock(&space_info->lock);
4354 space_info->bytes_may_use -= num_bytes;
4355 trace_btrfs_space_reservation(fs_info, "space_info",
4356 space_info->flags, num_bytes, 0);
4357 space_info->reservation_progress++;
4358 spin_unlock(&space_info->lock);
4363 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4364 struct btrfs_block_rsv *dst, u64 num_bytes)
4368 ret = block_rsv_use_bytes(src, num_bytes);
4372 block_rsv_add_bytes(dst, num_bytes, 1);
4376 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4378 memset(rsv, 0, sizeof(*rsv));
4379 spin_lock_init(&rsv->lock);
4383 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4384 unsigned short type)
4386 struct btrfs_block_rsv *block_rsv;
4387 struct btrfs_fs_info *fs_info = root->fs_info;
4389 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4393 btrfs_init_block_rsv(block_rsv, type);
4394 block_rsv->space_info = __find_space_info(fs_info,
4395 BTRFS_BLOCK_GROUP_METADATA);
4399 void btrfs_free_block_rsv(struct btrfs_root *root,
4400 struct btrfs_block_rsv *rsv)
4404 btrfs_block_rsv_release(root, rsv, (u64)-1);
4408 int btrfs_block_rsv_add(struct btrfs_root *root,
4409 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4410 enum btrfs_reserve_flush_enum flush)
4417 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4419 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4426 int btrfs_block_rsv_check(struct btrfs_root *root,
4427 struct btrfs_block_rsv *block_rsv, int min_factor)
4435 spin_lock(&block_rsv->lock);
4436 num_bytes = div_factor(block_rsv->size, min_factor);
4437 if (block_rsv->reserved >= num_bytes)
4439 spin_unlock(&block_rsv->lock);
4444 int btrfs_block_rsv_refill(struct btrfs_root *root,
4445 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4446 enum btrfs_reserve_flush_enum flush)
4454 spin_lock(&block_rsv->lock);
4455 num_bytes = min_reserved;
4456 if (block_rsv->reserved >= num_bytes)
4459 num_bytes -= block_rsv->reserved;
4460 spin_unlock(&block_rsv->lock);
4465 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4467 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4474 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4475 struct btrfs_block_rsv *dst_rsv,
4478 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4481 void btrfs_block_rsv_release(struct btrfs_root *root,
4482 struct btrfs_block_rsv *block_rsv,
4485 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4486 if (global_rsv->full || global_rsv == block_rsv ||
4487 block_rsv->space_info != global_rsv->space_info)
4489 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4494 * helper to calculate size of global block reservation.
4495 * the desired value is sum of space used by extent tree,
4496 * checksum tree and root tree
4498 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4500 struct btrfs_space_info *sinfo;
4504 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4506 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4507 spin_lock(&sinfo->lock);
4508 data_used = sinfo->bytes_used;
4509 spin_unlock(&sinfo->lock);
4511 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4512 spin_lock(&sinfo->lock);
4513 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4515 meta_used = sinfo->bytes_used;
4516 spin_unlock(&sinfo->lock);
4518 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4520 num_bytes += div64_u64(data_used + meta_used, 50);
4522 if (num_bytes * 3 > meta_used)
4523 num_bytes = div64_u64(meta_used, 3);
4525 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4528 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4530 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4531 struct btrfs_space_info *sinfo = block_rsv->space_info;
4534 num_bytes = calc_global_metadata_size(fs_info);
4536 spin_lock(&sinfo->lock);
4537 spin_lock(&block_rsv->lock);
4539 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4541 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4542 sinfo->bytes_reserved + sinfo->bytes_readonly +
4543 sinfo->bytes_may_use;
4545 if (sinfo->total_bytes > num_bytes) {
4546 num_bytes = sinfo->total_bytes - num_bytes;
4547 block_rsv->reserved += num_bytes;
4548 sinfo->bytes_may_use += num_bytes;
4549 trace_btrfs_space_reservation(fs_info, "space_info",
4550 sinfo->flags, num_bytes, 1);
4553 if (block_rsv->reserved >= block_rsv->size) {
4554 num_bytes = block_rsv->reserved - block_rsv->size;
4555 sinfo->bytes_may_use -= num_bytes;
4556 trace_btrfs_space_reservation(fs_info, "space_info",
4557 sinfo->flags, num_bytes, 0);
4558 sinfo->reservation_progress++;
4559 block_rsv->reserved = block_rsv->size;
4560 block_rsv->full = 1;
4563 spin_unlock(&block_rsv->lock);
4564 spin_unlock(&sinfo->lock);
4567 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4569 struct btrfs_space_info *space_info;
4571 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4572 fs_info->chunk_block_rsv.space_info = space_info;
4574 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4575 fs_info->global_block_rsv.space_info = space_info;
4576 fs_info->delalloc_block_rsv.space_info = space_info;
4577 fs_info->trans_block_rsv.space_info = space_info;
4578 fs_info->empty_block_rsv.space_info = space_info;
4579 fs_info->delayed_block_rsv.space_info = space_info;
4581 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4582 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4583 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4584 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4585 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4587 update_global_block_rsv(fs_info);
4590 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4592 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4594 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4595 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4596 WARN_ON(fs_info->trans_block_rsv.size > 0);
4597 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4598 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4599 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4600 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4601 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4604 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4605 struct btrfs_root *root)
4607 if (!trans->block_rsv)
4610 if (!trans->bytes_reserved)
4613 trace_btrfs_space_reservation(root->fs_info, "transaction",
4614 trans->transid, trans->bytes_reserved, 0);
4615 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4616 trans->bytes_reserved = 0;
4619 /* Can only return 0 or -ENOSPC */
4620 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4621 struct inode *inode)
4623 struct btrfs_root *root = BTRFS_I(inode)->root;
4624 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4625 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4628 * We need to hold space in order to delete our orphan item once we've
4629 * added it, so this takes the reservation so we can release it later
4630 * when we are truly done with the orphan item.
4632 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4633 trace_btrfs_space_reservation(root->fs_info, "orphan",
4634 btrfs_ino(inode), num_bytes, 1);
4635 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4638 void btrfs_orphan_release_metadata(struct inode *inode)
4640 struct btrfs_root *root = BTRFS_I(inode)->root;
4641 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4642 trace_btrfs_space_reservation(root->fs_info, "orphan",
4643 btrfs_ino(inode), num_bytes, 0);
4644 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4648 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4649 * root: the root of the parent directory
4650 * rsv: block reservation
4651 * items: the number of items that we need do reservation
4652 * qgroup_reserved: used to return the reserved size in qgroup
4654 * This function is used to reserve the space for snapshot/subvolume
4655 * creation and deletion. Those operations are different with the
4656 * common file/directory operations, they change two fs/file trees
4657 * and root tree, the number of items that the qgroup reserves is
4658 * different with the free space reservation. So we can not use
4659 * the space reseravtion mechanism in start_transaction().
4661 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4662 struct btrfs_block_rsv *rsv,
4664 u64 *qgroup_reserved)
4669 if (root->fs_info->quota_enabled) {
4670 /* One for parent inode, two for dir entries */
4671 num_bytes = 3 * root->leafsize;
4672 ret = btrfs_qgroup_reserve(root, num_bytes);
4679 *qgroup_reserved = num_bytes;
4681 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4682 rsv->space_info = __find_space_info(root->fs_info,
4683 BTRFS_BLOCK_GROUP_METADATA);
4684 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4685 BTRFS_RESERVE_FLUSH_ALL);
4687 if (*qgroup_reserved)
4688 btrfs_qgroup_free(root, *qgroup_reserved);
4694 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4695 struct btrfs_block_rsv *rsv,
4696 u64 qgroup_reserved)
4698 btrfs_block_rsv_release(root, rsv, (u64)-1);
4699 if (qgroup_reserved)
4700 btrfs_qgroup_free(root, qgroup_reserved);
4704 * drop_outstanding_extent - drop an outstanding extent
4705 * @inode: the inode we're dropping the extent for
4707 * This is called when we are freeing up an outstanding extent, either called
4708 * after an error or after an extent is written. This will return the number of
4709 * reserved extents that need to be freed. This must be called with
4710 * BTRFS_I(inode)->lock held.
4712 static unsigned drop_outstanding_extent(struct inode *inode)
4714 unsigned drop_inode_space = 0;
4715 unsigned dropped_extents = 0;
4717 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4718 BTRFS_I(inode)->outstanding_extents--;
4720 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4721 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4722 &BTRFS_I(inode)->runtime_flags))
4723 drop_inode_space = 1;
4726 * If we have more or the same amount of outsanding extents than we have
4727 * reserved then we need to leave the reserved extents count alone.
4729 if (BTRFS_I(inode)->outstanding_extents >=
4730 BTRFS_I(inode)->reserved_extents)
4731 return drop_inode_space;
4733 dropped_extents = BTRFS_I(inode)->reserved_extents -
4734 BTRFS_I(inode)->outstanding_extents;
4735 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4736 return dropped_extents + drop_inode_space;
4740 * calc_csum_metadata_size - return the amount of metada space that must be
4741 * reserved/free'd for the given bytes.
4742 * @inode: the inode we're manipulating
4743 * @num_bytes: the number of bytes in question
4744 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4746 * This adjusts the number of csum_bytes in the inode and then returns the
4747 * correct amount of metadata that must either be reserved or freed. We
4748 * calculate how many checksums we can fit into one leaf and then divide the
4749 * number of bytes that will need to be checksumed by this value to figure out
4750 * how many checksums will be required. If we are adding bytes then the number
4751 * may go up and we will return the number of additional bytes that must be
4752 * reserved. If it is going down we will return the number of bytes that must
4755 * This must be called with BTRFS_I(inode)->lock held.
4757 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4760 struct btrfs_root *root = BTRFS_I(inode)->root;
4762 int num_csums_per_leaf;
4766 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4767 BTRFS_I(inode)->csum_bytes == 0)
4770 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4772 BTRFS_I(inode)->csum_bytes += num_bytes;
4774 BTRFS_I(inode)->csum_bytes -= num_bytes;
4775 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4776 num_csums_per_leaf = (int)div64_u64(csum_size,
4777 sizeof(struct btrfs_csum_item) +
4778 sizeof(struct btrfs_disk_key));
4779 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4780 num_csums = num_csums + num_csums_per_leaf - 1;
4781 num_csums = num_csums / num_csums_per_leaf;
4783 old_csums = old_csums + num_csums_per_leaf - 1;
4784 old_csums = old_csums / num_csums_per_leaf;
4786 /* No change, no need to reserve more */
4787 if (old_csums == num_csums)
4791 return btrfs_calc_trans_metadata_size(root,
4792 num_csums - old_csums);
4794 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4797 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4799 struct btrfs_root *root = BTRFS_I(inode)->root;
4800 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4803 unsigned nr_extents = 0;
4804 int extra_reserve = 0;
4805 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4807 bool delalloc_lock = true;
4811 /* If we are a free space inode we need to not flush since we will be in
4812 * the middle of a transaction commit. We also don't need the delalloc
4813 * mutex since we won't race with anybody. We need this mostly to make
4814 * lockdep shut its filthy mouth.
4816 if (btrfs_is_free_space_inode(inode)) {
4817 flush = BTRFS_RESERVE_NO_FLUSH;
4818 delalloc_lock = false;
4821 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4822 btrfs_transaction_in_commit(root->fs_info))
4823 schedule_timeout(1);
4826 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4828 num_bytes = ALIGN(num_bytes, root->sectorsize);
4830 spin_lock(&BTRFS_I(inode)->lock);
4831 BTRFS_I(inode)->outstanding_extents++;
4833 if (BTRFS_I(inode)->outstanding_extents >
4834 BTRFS_I(inode)->reserved_extents)
4835 nr_extents = BTRFS_I(inode)->outstanding_extents -
4836 BTRFS_I(inode)->reserved_extents;
4839 * Add an item to reserve for updating the inode when we complete the
4842 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4843 &BTRFS_I(inode)->runtime_flags)) {
4848 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4849 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4850 csum_bytes = BTRFS_I(inode)->csum_bytes;
4851 spin_unlock(&BTRFS_I(inode)->lock);
4853 if (root->fs_info->quota_enabled) {
4854 ret = btrfs_qgroup_reserve(root, num_bytes +
4855 nr_extents * root->leafsize);
4860 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4861 if (unlikely(ret)) {
4862 if (root->fs_info->quota_enabled)
4863 btrfs_qgroup_free(root, num_bytes +
4864 nr_extents * root->leafsize);
4868 spin_lock(&BTRFS_I(inode)->lock);
4869 if (extra_reserve) {
4870 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4871 &BTRFS_I(inode)->runtime_flags);
4874 BTRFS_I(inode)->reserved_extents += nr_extents;
4875 spin_unlock(&BTRFS_I(inode)->lock);
4878 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4881 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4882 btrfs_ino(inode), to_reserve, 1);
4883 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4888 spin_lock(&BTRFS_I(inode)->lock);
4889 dropped = drop_outstanding_extent(inode);
4891 * If the inodes csum_bytes is the same as the original
4892 * csum_bytes then we know we haven't raced with any free()ers
4893 * so we can just reduce our inodes csum bytes and carry on.
4895 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4896 calc_csum_metadata_size(inode, num_bytes, 0);
4898 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4902 * This is tricky, but first we need to figure out how much we
4903 * free'd from any free-ers that occured during this
4904 * reservation, so we reset ->csum_bytes to the csum_bytes
4905 * before we dropped our lock, and then call the free for the
4906 * number of bytes that were freed while we were trying our
4909 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4910 BTRFS_I(inode)->csum_bytes = csum_bytes;
4911 to_free = calc_csum_metadata_size(inode, bytes, 0);
4915 * Now we need to see how much we would have freed had we not
4916 * been making this reservation and our ->csum_bytes were not
4917 * artificially inflated.
4919 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4920 bytes = csum_bytes - orig_csum_bytes;
4921 bytes = calc_csum_metadata_size(inode, bytes, 0);
4924 * Now reset ->csum_bytes to what it should be. If bytes is
4925 * more than to_free then we would have free'd more space had we
4926 * not had an artificially high ->csum_bytes, so we need to free
4927 * the remainder. If bytes is the same or less then we don't
4928 * need to do anything, the other free-ers did the correct
4931 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4932 if (bytes > to_free)
4933 to_free = bytes - to_free;
4937 spin_unlock(&BTRFS_I(inode)->lock);
4939 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4942 btrfs_block_rsv_release(root, block_rsv, to_free);
4943 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4944 btrfs_ino(inode), to_free, 0);
4947 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4952 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4953 * @inode: the inode to release the reservation for
4954 * @num_bytes: the number of bytes we're releasing
4956 * This will release the metadata reservation for an inode. This can be called
4957 * once we complete IO for a given set of bytes to release their metadata
4960 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4962 struct btrfs_root *root = BTRFS_I(inode)->root;
4966 num_bytes = ALIGN(num_bytes, root->sectorsize);
4967 spin_lock(&BTRFS_I(inode)->lock);
4968 dropped = drop_outstanding_extent(inode);
4971 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4972 spin_unlock(&BTRFS_I(inode)->lock);
4974 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4976 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4977 btrfs_ino(inode), to_free, 0);
4978 if (root->fs_info->quota_enabled) {
4979 btrfs_qgroup_free(root, num_bytes +
4980 dropped * root->leafsize);
4983 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4988 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4989 * @inode: inode we're writing to
4990 * @num_bytes: the number of bytes we want to allocate
4992 * This will do the following things
4994 * o reserve space in the data space info for num_bytes
4995 * o reserve space in the metadata space info based on number of outstanding
4996 * extents and how much csums will be needed
4997 * o add to the inodes ->delalloc_bytes
4998 * o add it to the fs_info's delalloc inodes list.
5000 * This will return 0 for success and -ENOSPC if there is no space left.
5002 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5006 ret = btrfs_check_data_free_space(inode, num_bytes);
5010 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5012 btrfs_free_reserved_data_space(inode, num_bytes);
5020 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5021 * @inode: inode we're releasing space for
5022 * @num_bytes: the number of bytes we want to free up
5024 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5025 * called in the case that we don't need the metadata AND data reservations
5026 * anymore. So if there is an error or we insert an inline extent.
5028 * This function will release the metadata space that was not used and will
5029 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5030 * list if there are no delalloc bytes left.
5032 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5034 btrfs_delalloc_release_metadata(inode, num_bytes);
5035 btrfs_free_reserved_data_space(inode, num_bytes);
5038 static int update_block_group(struct btrfs_root *root,
5039 u64 bytenr, u64 num_bytes, int alloc)
5041 struct btrfs_block_group_cache *cache = NULL;
5042 struct btrfs_fs_info *info = root->fs_info;
5043 u64 total = num_bytes;
5048 /* block accounting for super block */
5049 spin_lock(&info->delalloc_lock);
5050 old_val = btrfs_super_bytes_used(info->super_copy);
5052 old_val += num_bytes;
5054 old_val -= num_bytes;
5055 btrfs_set_super_bytes_used(info->super_copy, old_val);
5056 spin_unlock(&info->delalloc_lock);
5059 cache = btrfs_lookup_block_group(info, bytenr);
5062 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5063 BTRFS_BLOCK_GROUP_RAID1 |
5064 BTRFS_BLOCK_GROUP_RAID10))
5069 * If this block group has free space cache written out, we
5070 * need to make sure to load it if we are removing space. This
5071 * is because we need the unpinning stage to actually add the
5072 * space back to the block group, otherwise we will leak space.
5074 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5075 cache_block_group(cache, 1);
5077 byte_in_group = bytenr - cache->key.objectid;
5078 WARN_ON(byte_in_group > cache->key.offset);
5080 spin_lock(&cache->space_info->lock);
5081 spin_lock(&cache->lock);
5083 if (btrfs_test_opt(root, SPACE_CACHE) &&
5084 cache->disk_cache_state < BTRFS_DC_CLEAR)
5085 cache->disk_cache_state = BTRFS_DC_CLEAR;
5088 old_val = btrfs_block_group_used(&cache->item);
5089 num_bytes = min(total, cache->key.offset - byte_in_group);
5091 old_val += num_bytes;
5092 btrfs_set_block_group_used(&cache->item, old_val);
5093 cache->reserved -= num_bytes;
5094 cache->space_info->bytes_reserved -= num_bytes;
5095 cache->space_info->bytes_used += num_bytes;
5096 cache->space_info->disk_used += num_bytes * factor;
5097 spin_unlock(&cache->lock);
5098 spin_unlock(&cache->space_info->lock);
5100 old_val -= num_bytes;
5101 btrfs_set_block_group_used(&cache->item, old_val);
5102 cache->pinned += num_bytes;
5103 cache->space_info->bytes_pinned += num_bytes;
5104 cache->space_info->bytes_used -= num_bytes;
5105 cache->space_info->disk_used -= num_bytes * factor;
5106 spin_unlock(&cache->lock);
5107 spin_unlock(&cache->space_info->lock);
5109 set_extent_dirty(info->pinned_extents,
5110 bytenr, bytenr + num_bytes - 1,
5111 GFP_NOFS | __GFP_NOFAIL);
5113 btrfs_put_block_group(cache);
5115 bytenr += num_bytes;
5120 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5122 struct btrfs_block_group_cache *cache;
5125 spin_lock(&root->fs_info->block_group_cache_lock);
5126 bytenr = root->fs_info->first_logical_byte;
5127 spin_unlock(&root->fs_info->block_group_cache_lock);
5129 if (bytenr < (u64)-1)
5132 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5136 bytenr = cache->key.objectid;
5137 btrfs_put_block_group(cache);
5142 static int pin_down_extent(struct btrfs_root *root,
5143 struct btrfs_block_group_cache *cache,
5144 u64 bytenr, u64 num_bytes, int reserved)
5146 spin_lock(&cache->space_info->lock);
5147 spin_lock(&cache->lock);
5148 cache->pinned += num_bytes;
5149 cache->space_info->bytes_pinned += num_bytes;
5151 cache->reserved -= num_bytes;
5152 cache->space_info->bytes_reserved -= num_bytes;
5154 spin_unlock(&cache->lock);
5155 spin_unlock(&cache->space_info->lock);
5157 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5158 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5163 * this function must be called within transaction
5165 int btrfs_pin_extent(struct btrfs_root *root,
5166 u64 bytenr, u64 num_bytes, int reserved)
5168 struct btrfs_block_group_cache *cache;
5170 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5171 BUG_ON(!cache); /* Logic error */
5173 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5175 btrfs_put_block_group(cache);
5180 * this function must be called within transaction
5182 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5183 u64 bytenr, u64 num_bytes)
5185 struct btrfs_block_group_cache *cache;
5187 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5188 BUG_ON(!cache); /* Logic error */
5191 * pull in the free space cache (if any) so that our pin
5192 * removes the free space from the cache. We have load_only set
5193 * to one because the slow code to read in the free extents does check
5194 * the pinned extents.
5196 cache_block_group(cache, 1);
5198 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5200 /* remove us from the free space cache (if we're there at all) */
5201 btrfs_remove_free_space(cache, bytenr, num_bytes);
5202 btrfs_put_block_group(cache);
5207 * btrfs_update_reserved_bytes - update the block_group and space info counters
5208 * @cache: The cache we are manipulating
5209 * @num_bytes: The number of bytes in question
5210 * @reserve: One of the reservation enums
5212 * This is called by the allocator when it reserves space, or by somebody who is
5213 * freeing space that was never actually used on disk. For example if you
5214 * reserve some space for a new leaf in transaction A and before transaction A
5215 * commits you free that leaf, you call this with reserve set to 0 in order to
5216 * clear the reservation.
5218 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5219 * ENOSPC accounting. For data we handle the reservation through clearing the
5220 * delalloc bits in the io_tree. We have to do this since we could end up
5221 * allocating less disk space for the amount of data we have reserved in the
5222 * case of compression.
5224 * If this is a reservation and the block group has become read only we cannot
5225 * make the reservation and return -EAGAIN, otherwise this function always
5228 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5229 u64 num_bytes, int reserve)
5231 struct btrfs_space_info *space_info = cache->space_info;
5234 spin_lock(&space_info->lock);
5235 spin_lock(&cache->lock);
5236 if (reserve != RESERVE_FREE) {
5240 cache->reserved += num_bytes;
5241 space_info->bytes_reserved += num_bytes;
5242 if (reserve == RESERVE_ALLOC) {
5243 trace_btrfs_space_reservation(cache->fs_info,
5244 "space_info", space_info->flags,
5246 space_info->bytes_may_use -= num_bytes;
5251 space_info->bytes_readonly += num_bytes;
5252 cache->reserved -= num_bytes;
5253 space_info->bytes_reserved -= num_bytes;
5254 space_info->reservation_progress++;
5256 spin_unlock(&cache->lock);
5257 spin_unlock(&space_info->lock);
5261 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5262 struct btrfs_root *root)
5264 struct btrfs_fs_info *fs_info = root->fs_info;
5265 struct btrfs_caching_control *next;
5266 struct btrfs_caching_control *caching_ctl;
5267 struct btrfs_block_group_cache *cache;
5269 down_write(&fs_info->extent_commit_sem);
5271 list_for_each_entry_safe(caching_ctl, next,
5272 &fs_info->caching_block_groups, list) {
5273 cache = caching_ctl->block_group;
5274 if (block_group_cache_done(cache)) {
5275 cache->last_byte_to_unpin = (u64)-1;
5276 list_del_init(&caching_ctl->list);
5277 put_caching_control(caching_ctl);
5279 cache->last_byte_to_unpin = caching_ctl->progress;
5283 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5284 fs_info->pinned_extents = &fs_info->freed_extents[1];
5286 fs_info->pinned_extents = &fs_info->freed_extents[0];
5288 up_write(&fs_info->extent_commit_sem);
5290 update_global_block_rsv(fs_info);
5293 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5295 struct btrfs_fs_info *fs_info = root->fs_info;
5296 struct btrfs_block_group_cache *cache = NULL;
5297 struct btrfs_space_info *space_info;
5298 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5302 while (start <= end) {
5305 start >= cache->key.objectid + cache->key.offset) {
5307 btrfs_put_block_group(cache);
5308 cache = btrfs_lookup_block_group(fs_info, start);
5309 BUG_ON(!cache); /* Logic error */
5312 len = cache->key.objectid + cache->key.offset - start;
5313 len = min(len, end + 1 - start);
5315 if (start < cache->last_byte_to_unpin) {
5316 len = min(len, cache->last_byte_to_unpin - start);
5317 btrfs_add_free_space(cache, start, len);
5321 space_info = cache->space_info;
5323 spin_lock(&space_info->lock);
5324 spin_lock(&cache->lock);
5325 cache->pinned -= len;
5326 space_info->bytes_pinned -= len;
5328 space_info->bytes_readonly += len;
5331 spin_unlock(&cache->lock);
5332 if (!readonly && global_rsv->space_info == space_info) {
5333 spin_lock(&global_rsv->lock);
5334 if (!global_rsv->full) {
5335 len = min(len, global_rsv->size -
5336 global_rsv->reserved);
5337 global_rsv->reserved += len;
5338 space_info->bytes_may_use += len;
5339 if (global_rsv->reserved >= global_rsv->size)
5340 global_rsv->full = 1;
5342 spin_unlock(&global_rsv->lock);
5344 spin_unlock(&space_info->lock);
5348 btrfs_put_block_group(cache);
5352 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5353 struct btrfs_root *root)
5355 struct btrfs_fs_info *fs_info = root->fs_info;
5356 struct extent_io_tree *unpin;
5364 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5365 unpin = &fs_info->freed_extents[1];
5367 unpin = &fs_info->freed_extents[0];
5370 ret = find_first_extent_bit(unpin, 0, &start, &end,
5371 EXTENT_DIRTY, NULL);
5375 if (btrfs_test_opt(root, DISCARD))
5376 ret = btrfs_discard_extent(root, start,
5377 end + 1 - start, NULL);
5379 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5380 unpin_extent_range(root, start, end);
5387 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5388 struct btrfs_root *root,
5389 u64 bytenr, u64 num_bytes, u64 parent,
5390 u64 root_objectid, u64 owner_objectid,
5391 u64 owner_offset, int refs_to_drop,
5392 struct btrfs_delayed_extent_op *extent_op)
5394 struct btrfs_key key;
5395 struct btrfs_path *path;
5396 struct btrfs_fs_info *info = root->fs_info;
5397 struct btrfs_root *extent_root = info->extent_root;
5398 struct extent_buffer *leaf;
5399 struct btrfs_extent_item *ei;
5400 struct btrfs_extent_inline_ref *iref;
5403 int extent_slot = 0;
5404 int found_extent = 0;
5408 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5411 path = btrfs_alloc_path();
5416 path->leave_spinning = 1;
5418 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5419 BUG_ON(!is_data && refs_to_drop != 1);
5422 skinny_metadata = 0;
5424 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5425 bytenr, num_bytes, parent,
5426 root_objectid, owner_objectid,
5429 extent_slot = path->slots[0];
5430 while (extent_slot >= 0) {
5431 btrfs_item_key_to_cpu(path->nodes[0], &key,
5433 if (key.objectid != bytenr)
5435 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5436 key.offset == num_bytes) {
5440 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5441 key.offset == owner_objectid) {
5445 if (path->slots[0] - extent_slot > 5)
5449 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5450 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5451 if (found_extent && item_size < sizeof(*ei))
5454 if (!found_extent) {
5456 ret = remove_extent_backref(trans, extent_root, path,
5460 btrfs_abort_transaction(trans, extent_root, ret);
5463 btrfs_release_path(path);
5464 path->leave_spinning = 1;
5466 key.objectid = bytenr;
5467 key.type = BTRFS_EXTENT_ITEM_KEY;
5468 key.offset = num_bytes;
5470 if (!is_data && skinny_metadata) {
5471 key.type = BTRFS_METADATA_ITEM_KEY;
5472 key.offset = owner_objectid;
5475 ret = btrfs_search_slot(trans, extent_root,
5477 if (ret > 0 && skinny_metadata && path->slots[0]) {
5479 * Couldn't find our skinny metadata item,
5480 * see if we have ye olde extent item.
5483 btrfs_item_key_to_cpu(path->nodes[0], &key,
5485 if (key.objectid == bytenr &&
5486 key.type == BTRFS_EXTENT_ITEM_KEY &&
5487 key.offset == num_bytes)
5491 if (ret > 0 && skinny_metadata) {
5492 skinny_metadata = false;
5493 key.type = BTRFS_EXTENT_ITEM_KEY;
5494 key.offset = num_bytes;
5495 btrfs_release_path(path);
5496 ret = btrfs_search_slot(trans, extent_root,
5501 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5502 ret, (unsigned long long)bytenr);
5504 btrfs_print_leaf(extent_root,
5508 btrfs_abort_transaction(trans, extent_root, ret);
5511 extent_slot = path->slots[0];
5513 } else if (ret == -ENOENT) {
5514 btrfs_print_leaf(extent_root, path->nodes[0]);
5517 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5518 (unsigned long long)bytenr,
5519 (unsigned long long)parent,
5520 (unsigned long long)root_objectid,
5521 (unsigned long long)owner_objectid,
5522 (unsigned long long)owner_offset);
5524 btrfs_abort_transaction(trans, extent_root, ret);
5528 leaf = path->nodes[0];
5529 item_size = btrfs_item_size_nr(leaf, extent_slot);
5530 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5531 if (item_size < sizeof(*ei)) {
5532 BUG_ON(found_extent || extent_slot != path->slots[0]);
5533 ret = convert_extent_item_v0(trans, extent_root, path,
5536 btrfs_abort_transaction(trans, extent_root, ret);
5540 btrfs_release_path(path);
5541 path->leave_spinning = 1;
5543 key.objectid = bytenr;
5544 key.type = BTRFS_EXTENT_ITEM_KEY;
5545 key.offset = num_bytes;
5547 ret = btrfs_search_slot(trans, extent_root, &key, path,
5550 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5551 ret, (unsigned long long)bytenr);
5552 btrfs_print_leaf(extent_root, path->nodes[0]);
5555 btrfs_abort_transaction(trans, extent_root, ret);
5559 extent_slot = path->slots[0];
5560 leaf = path->nodes[0];
5561 item_size = btrfs_item_size_nr(leaf, extent_slot);
5564 BUG_ON(item_size < sizeof(*ei));
5565 ei = btrfs_item_ptr(leaf, extent_slot,
5566 struct btrfs_extent_item);
5567 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5568 key.type == BTRFS_EXTENT_ITEM_KEY) {
5569 struct btrfs_tree_block_info *bi;
5570 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5571 bi = (struct btrfs_tree_block_info *)(ei + 1);
5572 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5575 refs = btrfs_extent_refs(leaf, ei);
5576 BUG_ON(refs < refs_to_drop);
5577 refs -= refs_to_drop;
5581 __run_delayed_extent_op(extent_op, leaf, ei);
5583 * In the case of inline back ref, reference count will
5584 * be updated by remove_extent_backref
5587 BUG_ON(!found_extent);
5589 btrfs_set_extent_refs(leaf, ei, refs);
5590 btrfs_mark_buffer_dirty(leaf);
5593 ret = remove_extent_backref(trans, extent_root, path,
5597 btrfs_abort_transaction(trans, extent_root, ret);
5603 BUG_ON(is_data && refs_to_drop !=
5604 extent_data_ref_count(root, path, iref));
5606 BUG_ON(path->slots[0] != extent_slot);
5608 BUG_ON(path->slots[0] != extent_slot + 1);
5609 path->slots[0] = extent_slot;
5614 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5617 btrfs_abort_transaction(trans, extent_root, ret);
5620 btrfs_release_path(path);
5623 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5625 btrfs_abort_transaction(trans, extent_root, ret);
5630 ret = update_block_group(root, bytenr, num_bytes, 0);
5632 btrfs_abort_transaction(trans, extent_root, ret);
5637 btrfs_free_path(path);
5642 * when we free an block, it is possible (and likely) that we free the last
5643 * delayed ref for that extent as well. This searches the delayed ref tree for
5644 * a given extent, and if there are no other delayed refs to be processed, it
5645 * removes it from the tree.
5647 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5648 struct btrfs_root *root, u64 bytenr)
5650 struct btrfs_delayed_ref_head *head;
5651 struct btrfs_delayed_ref_root *delayed_refs;
5652 struct btrfs_delayed_ref_node *ref;
5653 struct rb_node *node;
5656 delayed_refs = &trans->transaction->delayed_refs;
5657 spin_lock(&delayed_refs->lock);
5658 head = btrfs_find_delayed_ref_head(trans, bytenr);
5662 node = rb_prev(&head->node.rb_node);
5666 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5668 /* there are still entries for this ref, we can't drop it */
5669 if (ref->bytenr == bytenr)
5672 if (head->extent_op) {
5673 if (!head->must_insert_reserved)
5675 btrfs_free_delayed_extent_op(head->extent_op);
5676 head->extent_op = NULL;
5680 * waiting for the lock here would deadlock. If someone else has it
5681 * locked they are already in the process of dropping it anyway
5683 if (!mutex_trylock(&head->mutex))
5687 * at this point we have a head with no other entries. Go
5688 * ahead and process it.
5690 head->node.in_tree = 0;
5691 rb_erase(&head->node.rb_node, &delayed_refs->root);
5693 delayed_refs->num_entries--;
5696 * we don't take a ref on the node because we're removing it from the
5697 * tree, so we just steal the ref the tree was holding.
5699 delayed_refs->num_heads--;
5700 if (list_empty(&head->cluster))
5701 delayed_refs->num_heads_ready--;
5703 list_del_init(&head->cluster);
5704 spin_unlock(&delayed_refs->lock);
5706 BUG_ON(head->extent_op);
5707 if (head->must_insert_reserved)
5710 mutex_unlock(&head->mutex);
5711 btrfs_put_delayed_ref(&head->node);
5714 spin_unlock(&delayed_refs->lock);
5718 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5719 struct btrfs_root *root,
5720 struct extent_buffer *buf,
5721 u64 parent, int last_ref)
5723 struct btrfs_block_group_cache *cache = NULL;
5726 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5727 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5728 buf->start, buf->len,
5729 parent, root->root_key.objectid,
5730 btrfs_header_level(buf),
5731 BTRFS_DROP_DELAYED_REF, NULL, 0);
5732 BUG_ON(ret); /* -ENOMEM */
5738 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5740 if (btrfs_header_generation(buf) == trans->transid) {
5741 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5742 ret = check_ref_cleanup(trans, root, buf->start);
5747 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5748 pin_down_extent(root, cache, buf->start, buf->len, 1);
5752 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5754 btrfs_add_free_space(cache, buf->start, buf->len);
5755 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5759 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5762 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5763 btrfs_put_block_group(cache);
5766 /* Can return -ENOMEM */
5767 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5768 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5769 u64 owner, u64 offset, int for_cow)
5772 struct btrfs_fs_info *fs_info = root->fs_info;
5775 * tree log blocks never actually go into the extent allocation
5776 * tree, just update pinning info and exit early.
5778 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5779 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5780 /* unlocks the pinned mutex */
5781 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5783 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5784 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5786 parent, root_objectid, (int)owner,
5787 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5789 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5791 parent, root_objectid, owner,
5792 offset, BTRFS_DROP_DELAYED_REF,
5798 static u64 stripe_align(struct btrfs_root *root,
5799 struct btrfs_block_group_cache *cache,
5800 u64 val, u64 num_bytes)
5802 u64 ret = ALIGN(val, root->stripesize);
5807 * when we wait for progress in the block group caching, its because
5808 * our allocation attempt failed at least once. So, we must sleep
5809 * and let some progress happen before we try again.
5811 * This function will sleep at least once waiting for new free space to
5812 * show up, and then it will check the block group free space numbers
5813 * for our min num_bytes. Another option is to have it go ahead
5814 * and look in the rbtree for a free extent of a given size, but this
5818 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5821 struct btrfs_caching_control *caching_ctl;
5823 caching_ctl = get_caching_control(cache);
5827 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5828 (cache->free_space_ctl->free_space >= num_bytes));
5830 put_caching_control(caching_ctl);
5835 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5837 struct btrfs_caching_control *caching_ctl;
5839 caching_ctl = get_caching_control(cache);
5843 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5845 put_caching_control(caching_ctl);
5849 int __get_raid_index(u64 flags)
5851 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5852 return BTRFS_RAID_RAID10;
5853 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5854 return BTRFS_RAID_RAID1;
5855 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5856 return BTRFS_RAID_DUP;
5857 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5858 return BTRFS_RAID_RAID0;
5859 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5860 return BTRFS_RAID_RAID5;
5861 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5862 return BTRFS_RAID_RAID6;
5864 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
5867 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5869 return __get_raid_index(cache->flags);
5872 enum btrfs_loop_type {
5873 LOOP_CACHING_NOWAIT = 0,
5874 LOOP_CACHING_WAIT = 1,
5875 LOOP_ALLOC_CHUNK = 2,
5876 LOOP_NO_EMPTY_SIZE = 3,
5880 * walks the btree of allocated extents and find a hole of a given size.
5881 * The key ins is changed to record the hole:
5882 * ins->objectid == block start
5883 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5884 * ins->offset == number of blocks
5885 * Any available blocks before search_start are skipped.
5887 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5888 struct btrfs_root *orig_root,
5889 u64 num_bytes, u64 empty_size,
5890 u64 hint_byte, struct btrfs_key *ins,
5894 struct btrfs_root *root = orig_root->fs_info->extent_root;
5895 struct btrfs_free_cluster *last_ptr = NULL;
5896 struct btrfs_block_group_cache *block_group = NULL;
5897 struct btrfs_block_group_cache *used_block_group;
5898 u64 search_start = 0;
5899 int empty_cluster = 2 * 1024 * 1024;
5900 struct btrfs_space_info *space_info;
5902 int index = __get_raid_index(data);
5903 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5904 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5905 bool found_uncached_bg = false;
5906 bool failed_cluster_refill = false;
5907 bool failed_alloc = false;
5908 bool use_cluster = true;
5909 bool have_caching_bg = false;
5911 WARN_ON(num_bytes < root->sectorsize);
5912 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5916 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5918 space_info = __find_space_info(root->fs_info, data);
5920 btrfs_err(root->fs_info, "No space info for %llu", data);
5925 * If the space info is for both data and metadata it means we have a
5926 * small filesystem and we can't use the clustering stuff.
5928 if (btrfs_mixed_space_info(space_info))
5929 use_cluster = false;
5931 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5932 last_ptr = &root->fs_info->meta_alloc_cluster;
5933 if (!btrfs_test_opt(root, SSD))
5934 empty_cluster = 64 * 1024;
5937 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5938 btrfs_test_opt(root, SSD)) {
5939 last_ptr = &root->fs_info->data_alloc_cluster;
5943 spin_lock(&last_ptr->lock);
5944 if (last_ptr->block_group)
5945 hint_byte = last_ptr->window_start;
5946 spin_unlock(&last_ptr->lock);
5949 search_start = max(search_start, first_logical_byte(root, 0));
5950 search_start = max(search_start, hint_byte);
5955 if (search_start == hint_byte) {
5956 block_group = btrfs_lookup_block_group(root->fs_info,
5958 used_block_group = block_group;
5960 * we don't want to use the block group if it doesn't match our
5961 * allocation bits, or if its not cached.
5963 * However if we are re-searching with an ideal block group
5964 * picked out then we don't care that the block group is cached.
5966 if (block_group && block_group_bits(block_group, data) &&
5967 block_group->cached != BTRFS_CACHE_NO) {
5968 down_read(&space_info->groups_sem);
5969 if (list_empty(&block_group->list) ||
5972 * someone is removing this block group,
5973 * we can't jump into the have_block_group
5974 * target because our list pointers are not
5977 btrfs_put_block_group(block_group);
5978 up_read(&space_info->groups_sem);
5980 index = get_block_group_index(block_group);
5981 goto have_block_group;
5983 } else if (block_group) {
5984 btrfs_put_block_group(block_group);
5988 have_caching_bg = false;
5989 down_read(&space_info->groups_sem);
5990 list_for_each_entry(block_group, &space_info->block_groups[index],
5995 used_block_group = block_group;
5996 btrfs_get_block_group(block_group);
5997 search_start = block_group->key.objectid;
6000 * this can happen if we end up cycling through all the
6001 * raid types, but we want to make sure we only allocate
6002 * for the proper type.
6004 if (!block_group_bits(block_group, data)) {
6005 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6006 BTRFS_BLOCK_GROUP_RAID1 |
6007 BTRFS_BLOCK_GROUP_RAID5 |
6008 BTRFS_BLOCK_GROUP_RAID6 |
6009 BTRFS_BLOCK_GROUP_RAID10;
6012 * if they asked for extra copies and this block group
6013 * doesn't provide them, bail. This does allow us to
6014 * fill raid0 from raid1.
6016 if ((data & extra) && !(block_group->flags & extra))
6021 cached = block_group_cache_done(block_group);
6022 if (unlikely(!cached)) {
6023 found_uncached_bg = true;
6024 ret = cache_block_group(block_group, 0);
6029 if (unlikely(block_group->ro))
6033 * Ok we want to try and use the cluster allocator, so
6037 unsigned long aligned_cluster;
6039 * the refill lock keeps out other
6040 * people trying to start a new cluster
6042 spin_lock(&last_ptr->refill_lock);
6043 used_block_group = last_ptr->block_group;
6044 if (used_block_group != block_group &&
6045 (!used_block_group ||
6046 used_block_group->ro ||
6047 !block_group_bits(used_block_group, data))) {
6048 used_block_group = block_group;
6049 goto refill_cluster;
6052 if (used_block_group != block_group)
6053 btrfs_get_block_group(used_block_group);
6055 offset = btrfs_alloc_from_cluster(used_block_group,
6056 last_ptr, num_bytes, used_block_group->key.objectid);
6058 /* we have a block, we're done */
6059 spin_unlock(&last_ptr->refill_lock);
6060 trace_btrfs_reserve_extent_cluster(root,
6061 block_group, search_start, num_bytes);
6065 WARN_ON(last_ptr->block_group != used_block_group);
6066 if (used_block_group != block_group) {
6067 btrfs_put_block_group(used_block_group);
6068 used_block_group = block_group;
6071 BUG_ON(used_block_group != block_group);
6072 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6073 * set up a new clusters, so lets just skip it
6074 * and let the allocator find whatever block
6075 * it can find. If we reach this point, we
6076 * will have tried the cluster allocator
6077 * plenty of times and not have found
6078 * anything, so we are likely way too
6079 * fragmented for the clustering stuff to find
6082 * However, if the cluster is taken from the
6083 * current block group, release the cluster
6084 * first, so that we stand a better chance of
6085 * succeeding in the unclustered
6087 if (loop >= LOOP_NO_EMPTY_SIZE &&
6088 last_ptr->block_group != block_group) {
6089 spin_unlock(&last_ptr->refill_lock);
6090 goto unclustered_alloc;
6094 * this cluster didn't work out, free it and
6097 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6099 if (loop >= LOOP_NO_EMPTY_SIZE) {
6100 spin_unlock(&last_ptr->refill_lock);
6101 goto unclustered_alloc;
6104 aligned_cluster = max_t(unsigned long,
6105 empty_cluster + empty_size,
6106 block_group->full_stripe_len);
6108 /* allocate a cluster in this block group */
6109 ret = btrfs_find_space_cluster(trans, root,
6110 block_group, last_ptr,
6111 search_start, num_bytes,
6115 * now pull our allocation out of this
6118 offset = btrfs_alloc_from_cluster(block_group,
6119 last_ptr, num_bytes,
6122 /* we found one, proceed */
6123 spin_unlock(&last_ptr->refill_lock);
6124 trace_btrfs_reserve_extent_cluster(root,
6125 block_group, search_start,
6129 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6130 && !failed_cluster_refill) {
6131 spin_unlock(&last_ptr->refill_lock);
6133 failed_cluster_refill = true;
6134 wait_block_group_cache_progress(block_group,
6135 num_bytes + empty_cluster + empty_size);
6136 goto have_block_group;
6140 * at this point we either didn't find a cluster
6141 * or we weren't able to allocate a block from our
6142 * cluster. Free the cluster we've been trying
6143 * to use, and go to the next block group
6145 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6146 spin_unlock(&last_ptr->refill_lock);
6151 spin_lock(&block_group->free_space_ctl->tree_lock);
6153 block_group->free_space_ctl->free_space <
6154 num_bytes + empty_cluster + empty_size) {
6155 spin_unlock(&block_group->free_space_ctl->tree_lock);
6158 spin_unlock(&block_group->free_space_ctl->tree_lock);
6160 offset = btrfs_find_space_for_alloc(block_group, search_start,
6161 num_bytes, empty_size);
6163 * If we didn't find a chunk, and we haven't failed on this
6164 * block group before, and this block group is in the middle of
6165 * caching and we are ok with waiting, then go ahead and wait
6166 * for progress to be made, and set failed_alloc to true.
6168 * If failed_alloc is true then we've already waited on this
6169 * block group once and should move on to the next block group.
6171 if (!offset && !failed_alloc && !cached &&
6172 loop > LOOP_CACHING_NOWAIT) {
6173 wait_block_group_cache_progress(block_group,
6174 num_bytes + empty_size);
6175 failed_alloc = true;
6176 goto have_block_group;
6177 } else if (!offset) {
6179 have_caching_bg = true;
6183 search_start = stripe_align(root, used_block_group,
6186 /* move on to the next group */
6187 if (search_start + num_bytes >
6188 used_block_group->key.objectid + used_block_group->key.offset) {
6189 btrfs_add_free_space(used_block_group, offset, num_bytes);
6193 if (offset < search_start)
6194 btrfs_add_free_space(used_block_group, offset,
6195 search_start - offset);
6196 BUG_ON(offset > search_start);
6198 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6200 if (ret == -EAGAIN) {
6201 btrfs_add_free_space(used_block_group, offset, num_bytes);
6205 /* we are all good, lets return */
6206 ins->objectid = search_start;
6207 ins->offset = num_bytes;
6209 trace_btrfs_reserve_extent(orig_root, block_group,
6210 search_start, num_bytes);
6211 if (used_block_group != block_group)
6212 btrfs_put_block_group(used_block_group);
6213 btrfs_put_block_group(block_group);
6216 failed_cluster_refill = false;
6217 failed_alloc = false;
6218 BUG_ON(index != get_block_group_index(block_group));
6219 if (used_block_group != block_group)
6220 btrfs_put_block_group(used_block_group);
6221 btrfs_put_block_group(block_group);
6223 up_read(&space_info->groups_sem);
6225 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6228 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6232 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6233 * caching kthreads as we move along
6234 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6235 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6236 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6239 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6242 if (loop == LOOP_ALLOC_CHUNK) {
6243 ret = do_chunk_alloc(trans, root, data,
6246 * Do not bail out on ENOSPC since we
6247 * can do more things.
6249 if (ret < 0 && ret != -ENOSPC) {
6250 btrfs_abort_transaction(trans,
6256 if (loop == LOOP_NO_EMPTY_SIZE) {
6262 } else if (!ins->objectid) {
6264 } else if (ins->objectid) {
6272 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6273 int dump_block_groups)
6275 struct btrfs_block_group_cache *cache;
6278 spin_lock(&info->lock);
6279 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6280 (unsigned long long)info->flags,
6281 (unsigned long long)(info->total_bytes - info->bytes_used -
6282 info->bytes_pinned - info->bytes_reserved -
6283 info->bytes_readonly),
6284 (info->full) ? "" : "not ");
6285 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6286 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6287 (unsigned long long)info->total_bytes,
6288 (unsigned long long)info->bytes_used,
6289 (unsigned long long)info->bytes_pinned,
6290 (unsigned long long)info->bytes_reserved,
6291 (unsigned long long)info->bytes_may_use,
6292 (unsigned long long)info->bytes_readonly);
6293 spin_unlock(&info->lock);
6295 if (!dump_block_groups)
6298 down_read(&info->groups_sem);
6300 list_for_each_entry(cache, &info->block_groups[index], list) {
6301 spin_lock(&cache->lock);
6302 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6303 (unsigned long long)cache->key.objectid,
6304 (unsigned long long)cache->key.offset,
6305 (unsigned long long)btrfs_block_group_used(&cache->item),
6306 (unsigned long long)cache->pinned,
6307 (unsigned long long)cache->reserved,
6308 cache->ro ? "[readonly]" : "");
6309 btrfs_dump_free_space(cache, bytes);
6310 spin_unlock(&cache->lock);
6312 if (++index < BTRFS_NR_RAID_TYPES)
6314 up_read(&info->groups_sem);
6317 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6318 struct btrfs_root *root,
6319 u64 num_bytes, u64 min_alloc_size,
6320 u64 empty_size, u64 hint_byte,
6321 struct btrfs_key *ins, u64 data)
6323 bool final_tried = false;
6326 data = btrfs_get_alloc_profile(root, data);
6328 WARN_ON(num_bytes < root->sectorsize);
6329 ret = find_free_extent(trans, root, num_bytes, empty_size,
6330 hint_byte, ins, data);
6332 if (ret == -ENOSPC) {
6334 num_bytes = num_bytes >> 1;
6335 num_bytes = round_down(num_bytes, root->sectorsize);
6336 num_bytes = max(num_bytes, min_alloc_size);
6337 if (num_bytes == min_alloc_size)
6340 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6341 struct btrfs_space_info *sinfo;
6343 sinfo = __find_space_info(root->fs_info, data);
6344 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6345 (unsigned long long)data,
6346 (unsigned long long)num_bytes);
6348 dump_space_info(sinfo, num_bytes, 1);
6352 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6357 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6358 u64 start, u64 len, int pin)
6360 struct btrfs_block_group_cache *cache;
6363 cache = btrfs_lookup_block_group(root->fs_info, start);
6365 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6366 (unsigned long long)start);
6370 if (btrfs_test_opt(root, DISCARD))
6371 ret = btrfs_discard_extent(root, start, len, NULL);
6374 pin_down_extent(root, cache, start, len, 1);
6376 btrfs_add_free_space(cache, start, len);
6377 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6379 btrfs_put_block_group(cache);
6381 trace_btrfs_reserved_extent_free(root, start, len);
6386 int btrfs_free_reserved_extent(struct btrfs_root *root,
6389 return __btrfs_free_reserved_extent(root, start, len, 0);
6392 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6395 return __btrfs_free_reserved_extent(root, start, len, 1);
6398 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6399 struct btrfs_root *root,
6400 u64 parent, u64 root_objectid,
6401 u64 flags, u64 owner, u64 offset,
6402 struct btrfs_key *ins, int ref_mod)
6405 struct btrfs_fs_info *fs_info = root->fs_info;
6406 struct btrfs_extent_item *extent_item;
6407 struct btrfs_extent_inline_ref *iref;
6408 struct btrfs_path *path;
6409 struct extent_buffer *leaf;
6414 type = BTRFS_SHARED_DATA_REF_KEY;
6416 type = BTRFS_EXTENT_DATA_REF_KEY;
6418 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6420 path = btrfs_alloc_path();
6424 path->leave_spinning = 1;
6425 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6428 btrfs_free_path(path);
6432 leaf = path->nodes[0];
6433 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6434 struct btrfs_extent_item);
6435 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6436 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6437 btrfs_set_extent_flags(leaf, extent_item,
6438 flags | BTRFS_EXTENT_FLAG_DATA);
6440 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6441 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6443 struct btrfs_shared_data_ref *ref;
6444 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6445 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6446 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6448 struct btrfs_extent_data_ref *ref;
6449 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6450 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6451 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6452 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6453 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6456 btrfs_mark_buffer_dirty(path->nodes[0]);
6457 btrfs_free_path(path);
6459 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6460 if (ret) { /* -ENOENT, logic error */
6461 btrfs_err(fs_info, "update block group failed for %llu %llu",
6462 (unsigned long long)ins->objectid,
6463 (unsigned long long)ins->offset);
6469 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6470 struct btrfs_root *root,
6471 u64 parent, u64 root_objectid,
6472 u64 flags, struct btrfs_disk_key *key,
6473 int level, struct btrfs_key *ins)
6476 struct btrfs_fs_info *fs_info = root->fs_info;
6477 struct btrfs_extent_item *extent_item;
6478 struct btrfs_tree_block_info *block_info;
6479 struct btrfs_extent_inline_ref *iref;
6480 struct btrfs_path *path;
6481 struct extent_buffer *leaf;
6482 u32 size = sizeof(*extent_item) + sizeof(*iref);
6483 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6486 if (!skinny_metadata)
6487 size += sizeof(*block_info);
6489 path = btrfs_alloc_path();
6493 path->leave_spinning = 1;
6494 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6497 btrfs_free_path(path);
6501 leaf = path->nodes[0];
6502 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6503 struct btrfs_extent_item);
6504 btrfs_set_extent_refs(leaf, extent_item, 1);
6505 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6506 btrfs_set_extent_flags(leaf, extent_item,
6507 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6509 if (skinny_metadata) {
6510 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6512 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6513 btrfs_set_tree_block_key(leaf, block_info, key);
6514 btrfs_set_tree_block_level(leaf, block_info, level);
6515 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6519 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6520 btrfs_set_extent_inline_ref_type(leaf, iref,
6521 BTRFS_SHARED_BLOCK_REF_KEY);
6522 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6524 btrfs_set_extent_inline_ref_type(leaf, iref,
6525 BTRFS_TREE_BLOCK_REF_KEY);
6526 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6529 btrfs_mark_buffer_dirty(leaf);
6530 btrfs_free_path(path);
6532 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6533 if (ret) { /* -ENOENT, logic error */
6534 btrfs_err(fs_info, "update block group failed for %llu %llu",
6535 (unsigned long long)ins->objectid,
6536 (unsigned long long)ins->offset);
6542 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6543 struct btrfs_root *root,
6544 u64 root_objectid, u64 owner,
6545 u64 offset, struct btrfs_key *ins)
6549 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6551 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6553 root_objectid, owner, offset,
6554 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6559 * this is used by the tree logging recovery code. It records that
6560 * an extent has been allocated and makes sure to clear the free
6561 * space cache bits as well
6563 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6564 struct btrfs_root *root,
6565 u64 root_objectid, u64 owner, u64 offset,
6566 struct btrfs_key *ins)
6569 struct btrfs_block_group_cache *block_group;
6570 struct btrfs_caching_control *caching_ctl;
6571 u64 start = ins->objectid;
6572 u64 num_bytes = ins->offset;
6574 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6575 cache_block_group(block_group, 0);
6576 caching_ctl = get_caching_control(block_group);
6579 BUG_ON(!block_group_cache_done(block_group));
6580 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6581 BUG_ON(ret); /* -ENOMEM */
6583 mutex_lock(&caching_ctl->mutex);
6585 if (start >= caching_ctl->progress) {
6586 ret = add_excluded_extent(root, start, num_bytes);
6587 BUG_ON(ret); /* -ENOMEM */
6588 } else if (start + num_bytes <= caching_ctl->progress) {
6589 ret = btrfs_remove_free_space(block_group,
6591 BUG_ON(ret); /* -ENOMEM */
6593 num_bytes = caching_ctl->progress - start;
6594 ret = btrfs_remove_free_space(block_group,
6596 BUG_ON(ret); /* -ENOMEM */
6598 start = caching_ctl->progress;
6599 num_bytes = ins->objectid + ins->offset -
6600 caching_ctl->progress;
6601 ret = add_excluded_extent(root, start, num_bytes);
6602 BUG_ON(ret); /* -ENOMEM */
6605 mutex_unlock(&caching_ctl->mutex);
6606 put_caching_control(caching_ctl);
6609 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6610 RESERVE_ALLOC_NO_ACCOUNT);
6611 BUG_ON(ret); /* logic error */
6612 btrfs_put_block_group(block_group);
6613 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6614 0, owner, offset, ins, 1);
6618 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6619 struct btrfs_root *root,
6620 u64 bytenr, u32 blocksize,
6623 struct extent_buffer *buf;
6625 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6627 return ERR_PTR(-ENOMEM);
6628 btrfs_set_header_generation(buf, trans->transid);
6629 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6630 btrfs_tree_lock(buf);
6631 clean_tree_block(trans, root, buf);
6632 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6634 btrfs_set_lock_blocking(buf);
6635 btrfs_set_buffer_uptodate(buf);
6637 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6639 * we allow two log transactions at a time, use different
6640 * EXENT bit to differentiate dirty pages.
6642 if (root->log_transid % 2 == 0)
6643 set_extent_dirty(&root->dirty_log_pages, buf->start,
6644 buf->start + buf->len - 1, GFP_NOFS);
6646 set_extent_new(&root->dirty_log_pages, buf->start,
6647 buf->start + buf->len - 1, GFP_NOFS);
6649 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6650 buf->start + buf->len - 1, GFP_NOFS);
6652 trans->blocks_used++;
6653 /* this returns a buffer locked for blocking */
6657 static struct btrfs_block_rsv *
6658 use_block_rsv(struct btrfs_trans_handle *trans,
6659 struct btrfs_root *root, u32 blocksize)
6661 struct btrfs_block_rsv *block_rsv;
6662 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6665 block_rsv = get_block_rsv(trans, root);
6667 if (block_rsv->size == 0) {
6668 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6669 BTRFS_RESERVE_NO_FLUSH);
6671 * If we couldn't reserve metadata bytes try and use some from
6672 * the global reserve.
6674 if (ret && block_rsv != global_rsv) {
6675 ret = block_rsv_use_bytes(global_rsv, blocksize);
6678 return ERR_PTR(ret);
6680 return ERR_PTR(ret);
6685 ret = block_rsv_use_bytes(block_rsv, blocksize);
6688 if (ret && !block_rsv->failfast) {
6689 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6690 static DEFINE_RATELIMIT_STATE(_rs,
6691 DEFAULT_RATELIMIT_INTERVAL * 10,
6692 /*DEFAULT_RATELIMIT_BURST*/ 1);
6693 if (__ratelimit(&_rs))
6695 "btrfs: block rsv returned %d\n", ret);
6697 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6698 BTRFS_RESERVE_NO_FLUSH);
6701 } else if (ret && block_rsv != global_rsv) {
6702 ret = block_rsv_use_bytes(global_rsv, blocksize);
6708 return ERR_PTR(-ENOSPC);
6711 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6712 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6714 block_rsv_add_bytes(block_rsv, blocksize, 0);
6715 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6719 * finds a free extent and does all the dirty work required for allocation
6720 * returns the key for the extent through ins, and a tree buffer for
6721 * the first block of the extent through buf.
6723 * returns the tree buffer or NULL.
6725 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6726 struct btrfs_root *root, u32 blocksize,
6727 u64 parent, u64 root_objectid,
6728 struct btrfs_disk_key *key, int level,
6729 u64 hint, u64 empty_size)
6731 struct btrfs_key ins;
6732 struct btrfs_block_rsv *block_rsv;
6733 struct extent_buffer *buf;
6736 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6739 block_rsv = use_block_rsv(trans, root, blocksize);
6740 if (IS_ERR(block_rsv))
6741 return ERR_CAST(block_rsv);
6743 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6744 empty_size, hint, &ins, 0);
6746 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6747 return ERR_PTR(ret);
6750 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6752 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6754 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6756 parent = ins.objectid;
6757 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6761 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6762 struct btrfs_delayed_extent_op *extent_op;
6763 extent_op = btrfs_alloc_delayed_extent_op();
6764 BUG_ON(!extent_op); /* -ENOMEM */
6766 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6768 memset(&extent_op->key, 0, sizeof(extent_op->key));
6769 extent_op->flags_to_set = flags;
6770 if (skinny_metadata)
6771 extent_op->update_key = 0;
6773 extent_op->update_key = 1;
6774 extent_op->update_flags = 1;
6775 extent_op->is_data = 0;
6777 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6779 ins.offset, parent, root_objectid,
6780 level, BTRFS_ADD_DELAYED_EXTENT,
6782 BUG_ON(ret); /* -ENOMEM */
6787 struct walk_control {
6788 u64 refs[BTRFS_MAX_LEVEL];
6789 u64 flags[BTRFS_MAX_LEVEL];
6790 struct btrfs_key update_progress;
6801 #define DROP_REFERENCE 1
6802 #define UPDATE_BACKREF 2
6804 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6805 struct btrfs_root *root,
6806 struct walk_control *wc,
6807 struct btrfs_path *path)
6815 struct btrfs_key key;
6816 struct extent_buffer *eb;
6821 if (path->slots[wc->level] < wc->reada_slot) {
6822 wc->reada_count = wc->reada_count * 2 / 3;
6823 wc->reada_count = max(wc->reada_count, 2);
6825 wc->reada_count = wc->reada_count * 3 / 2;
6826 wc->reada_count = min_t(int, wc->reada_count,
6827 BTRFS_NODEPTRS_PER_BLOCK(root));
6830 eb = path->nodes[wc->level];
6831 nritems = btrfs_header_nritems(eb);
6832 blocksize = btrfs_level_size(root, wc->level - 1);
6834 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6835 if (nread >= wc->reada_count)
6839 bytenr = btrfs_node_blockptr(eb, slot);
6840 generation = btrfs_node_ptr_generation(eb, slot);
6842 if (slot == path->slots[wc->level])
6845 if (wc->stage == UPDATE_BACKREF &&
6846 generation <= root->root_key.offset)
6849 /* We don't lock the tree block, it's OK to be racy here */
6850 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6851 wc->level - 1, 1, &refs,
6853 /* We don't care about errors in readahead. */
6858 if (wc->stage == DROP_REFERENCE) {
6862 if (wc->level == 1 &&
6863 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6865 if (!wc->update_ref ||
6866 generation <= root->root_key.offset)
6868 btrfs_node_key_to_cpu(eb, &key, slot);
6869 ret = btrfs_comp_cpu_keys(&key,
6870 &wc->update_progress);
6874 if (wc->level == 1 &&
6875 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6879 ret = readahead_tree_block(root, bytenr, blocksize,
6885 wc->reada_slot = slot;
6889 * helper to process tree block while walking down the tree.
6891 * when wc->stage == UPDATE_BACKREF, this function updates
6892 * back refs for pointers in the block.
6894 * NOTE: return value 1 means we should stop walking down.
6896 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6897 struct btrfs_root *root,
6898 struct btrfs_path *path,
6899 struct walk_control *wc, int lookup_info)
6901 int level = wc->level;
6902 struct extent_buffer *eb = path->nodes[level];
6903 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6906 if (wc->stage == UPDATE_BACKREF &&
6907 btrfs_header_owner(eb) != root->root_key.objectid)
6911 * when reference count of tree block is 1, it won't increase
6912 * again. once full backref flag is set, we never clear it.
6915 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6916 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6917 BUG_ON(!path->locks[level]);
6918 ret = btrfs_lookup_extent_info(trans, root,
6919 eb->start, level, 1,
6922 BUG_ON(ret == -ENOMEM);
6925 BUG_ON(wc->refs[level] == 0);
6928 if (wc->stage == DROP_REFERENCE) {
6929 if (wc->refs[level] > 1)
6932 if (path->locks[level] && !wc->keep_locks) {
6933 btrfs_tree_unlock_rw(eb, path->locks[level]);
6934 path->locks[level] = 0;
6939 /* wc->stage == UPDATE_BACKREF */
6940 if (!(wc->flags[level] & flag)) {
6941 BUG_ON(!path->locks[level]);
6942 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6943 BUG_ON(ret); /* -ENOMEM */
6944 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6945 BUG_ON(ret); /* -ENOMEM */
6946 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6948 BUG_ON(ret); /* -ENOMEM */
6949 wc->flags[level] |= flag;
6953 * the block is shared by multiple trees, so it's not good to
6954 * keep the tree lock
6956 if (path->locks[level] && level > 0) {
6957 btrfs_tree_unlock_rw(eb, path->locks[level]);
6958 path->locks[level] = 0;
6964 * helper to process tree block pointer.
6966 * when wc->stage == DROP_REFERENCE, this function checks
6967 * reference count of the block pointed to. if the block
6968 * is shared and we need update back refs for the subtree
6969 * rooted at the block, this function changes wc->stage to
6970 * UPDATE_BACKREF. if the block is shared and there is no
6971 * need to update back, this function drops the reference
6974 * NOTE: return value 1 means we should stop walking down.
6976 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6977 struct btrfs_root *root,
6978 struct btrfs_path *path,
6979 struct walk_control *wc, int *lookup_info)
6985 struct btrfs_key key;
6986 struct extent_buffer *next;
6987 int level = wc->level;
6991 generation = btrfs_node_ptr_generation(path->nodes[level],
6992 path->slots[level]);
6994 * if the lower level block was created before the snapshot
6995 * was created, we know there is no need to update back refs
6998 if (wc->stage == UPDATE_BACKREF &&
6999 generation <= root->root_key.offset) {
7004 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7005 blocksize = btrfs_level_size(root, level - 1);
7007 next = btrfs_find_tree_block(root, bytenr, blocksize);
7009 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7014 btrfs_tree_lock(next);
7015 btrfs_set_lock_blocking(next);
7017 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7018 &wc->refs[level - 1],
7019 &wc->flags[level - 1]);
7021 btrfs_tree_unlock(next);
7025 if (unlikely(wc->refs[level - 1] == 0)) {
7026 btrfs_err(root->fs_info, "Missing references.");
7031 if (wc->stage == DROP_REFERENCE) {
7032 if (wc->refs[level - 1] > 1) {
7034 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7037 if (!wc->update_ref ||
7038 generation <= root->root_key.offset)
7041 btrfs_node_key_to_cpu(path->nodes[level], &key,
7042 path->slots[level]);
7043 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7047 wc->stage = UPDATE_BACKREF;
7048 wc->shared_level = level - 1;
7052 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7056 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7057 btrfs_tree_unlock(next);
7058 free_extent_buffer(next);
7064 if (reada && level == 1)
7065 reada_walk_down(trans, root, wc, path);
7066 next = read_tree_block(root, bytenr, blocksize, generation);
7069 btrfs_tree_lock(next);
7070 btrfs_set_lock_blocking(next);
7074 BUG_ON(level != btrfs_header_level(next));
7075 path->nodes[level] = next;
7076 path->slots[level] = 0;
7077 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7083 wc->refs[level - 1] = 0;
7084 wc->flags[level - 1] = 0;
7085 if (wc->stage == DROP_REFERENCE) {
7086 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7087 parent = path->nodes[level]->start;
7089 BUG_ON(root->root_key.objectid !=
7090 btrfs_header_owner(path->nodes[level]));
7094 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7095 root->root_key.objectid, level - 1, 0, 0);
7096 BUG_ON(ret); /* -ENOMEM */
7098 btrfs_tree_unlock(next);
7099 free_extent_buffer(next);
7105 * helper to process tree block while walking up the tree.
7107 * when wc->stage == DROP_REFERENCE, this function drops
7108 * reference count on the block.
7110 * when wc->stage == UPDATE_BACKREF, this function changes
7111 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7112 * to UPDATE_BACKREF previously while processing the block.
7114 * NOTE: return value 1 means we should stop walking up.
7116 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7117 struct btrfs_root *root,
7118 struct btrfs_path *path,
7119 struct walk_control *wc)
7122 int level = wc->level;
7123 struct extent_buffer *eb = path->nodes[level];
7126 if (wc->stage == UPDATE_BACKREF) {
7127 BUG_ON(wc->shared_level < level);
7128 if (level < wc->shared_level)
7131 ret = find_next_key(path, level + 1, &wc->update_progress);
7135 wc->stage = DROP_REFERENCE;
7136 wc->shared_level = -1;
7137 path->slots[level] = 0;
7140 * check reference count again if the block isn't locked.
7141 * we should start walking down the tree again if reference
7144 if (!path->locks[level]) {
7146 btrfs_tree_lock(eb);
7147 btrfs_set_lock_blocking(eb);
7148 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7150 ret = btrfs_lookup_extent_info(trans, root,
7151 eb->start, level, 1,
7155 btrfs_tree_unlock_rw(eb, path->locks[level]);
7156 path->locks[level] = 0;
7159 BUG_ON(wc->refs[level] == 0);
7160 if (wc->refs[level] == 1) {
7161 btrfs_tree_unlock_rw(eb, path->locks[level]);
7162 path->locks[level] = 0;
7168 /* wc->stage == DROP_REFERENCE */
7169 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7171 if (wc->refs[level] == 1) {
7173 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7174 ret = btrfs_dec_ref(trans, root, eb, 1,
7177 ret = btrfs_dec_ref(trans, root, eb, 0,
7179 BUG_ON(ret); /* -ENOMEM */
7181 /* make block locked assertion in clean_tree_block happy */
7182 if (!path->locks[level] &&
7183 btrfs_header_generation(eb) == trans->transid) {
7184 btrfs_tree_lock(eb);
7185 btrfs_set_lock_blocking(eb);
7186 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7188 clean_tree_block(trans, root, eb);
7191 if (eb == root->node) {
7192 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7195 BUG_ON(root->root_key.objectid !=
7196 btrfs_header_owner(eb));
7198 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7199 parent = path->nodes[level + 1]->start;
7201 BUG_ON(root->root_key.objectid !=
7202 btrfs_header_owner(path->nodes[level + 1]));
7205 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7207 wc->refs[level] = 0;
7208 wc->flags[level] = 0;
7212 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7213 struct btrfs_root *root,
7214 struct btrfs_path *path,
7215 struct walk_control *wc)
7217 int level = wc->level;
7218 int lookup_info = 1;
7221 while (level >= 0) {
7222 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7229 if (path->slots[level] >=
7230 btrfs_header_nritems(path->nodes[level]))
7233 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7235 path->slots[level]++;
7244 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7245 struct btrfs_root *root,
7246 struct btrfs_path *path,
7247 struct walk_control *wc, int max_level)
7249 int level = wc->level;
7252 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7253 while (level < max_level && path->nodes[level]) {
7255 if (path->slots[level] + 1 <
7256 btrfs_header_nritems(path->nodes[level])) {
7257 path->slots[level]++;
7260 ret = walk_up_proc(trans, root, path, wc);
7264 if (path->locks[level]) {
7265 btrfs_tree_unlock_rw(path->nodes[level],
7266 path->locks[level]);
7267 path->locks[level] = 0;
7269 free_extent_buffer(path->nodes[level]);
7270 path->nodes[level] = NULL;
7278 * drop a subvolume tree.
7280 * this function traverses the tree freeing any blocks that only
7281 * referenced by the tree.
7283 * when a shared tree block is found. this function decreases its
7284 * reference count by one. if update_ref is true, this function
7285 * also make sure backrefs for the shared block and all lower level
7286 * blocks are properly updated.
7288 * If called with for_reloc == 0, may exit early with -EAGAIN
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 if (!for_reloc && btrfs_fs_closing(root->fs_info)) {
7390 pr_debug("btrfs: drop snapshot early exit\n");
7395 ret = walk_down_tree(trans, root, path, wc);
7401 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7408 BUG_ON(wc->stage != DROP_REFERENCE);
7412 if (wc->stage == DROP_REFERENCE) {
7414 btrfs_node_key(path->nodes[level],
7415 &root_item->drop_progress,
7416 path->slots[level]);
7417 root_item->drop_level = level;
7420 BUG_ON(wc->level == 0);
7421 if (btrfs_should_end_transaction(trans, tree_root)) {
7422 ret = btrfs_update_root(trans, tree_root,
7426 btrfs_abort_transaction(trans, tree_root, ret);
7431 btrfs_end_transaction_throttle(trans, tree_root);
7432 trans = btrfs_start_transaction(tree_root, 0);
7433 if (IS_ERR(trans)) {
7434 err = PTR_ERR(trans);
7438 trans->block_rsv = block_rsv;
7441 btrfs_release_path(path);
7445 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7447 btrfs_abort_transaction(trans, tree_root, ret);
7451 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7452 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7455 btrfs_abort_transaction(trans, tree_root, ret);
7458 } else if (ret > 0) {
7459 /* if we fail to delete the orphan item this time
7460 * around, it'll get picked up the next time.
7462 * The most common failure here is just -ENOENT.
7464 btrfs_del_orphan_item(trans, tree_root,
7465 root->root_key.objectid);
7469 if (root->in_radix) {
7470 btrfs_free_fs_root(tree_root->fs_info, root);
7472 free_extent_buffer(root->node);
7473 free_extent_buffer(root->commit_root);
7477 btrfs_end_transaction_throttle(trans, tree_root);
7480 btrfs_free_path(path);
7483 btrfs_std_error(root->fs_info, err);
7488 * drop subtree rooted at tree block 'node'.
7490 * NOTE: this function will unlock and release tree block 'node'
7491 * only used by relocation code
7493 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7494 struct btrfs_root *root,
7495 struct extent_buffer *node,
7496 struct extent_buffer *parent)
7498 struct btrfs_path *path;
7499 struct walk_control *wc;
7505 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7507 path = btrfs_alloc_path();
7511 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7513 btrfs_free_path(path);
7517 btrfs_assert_tree_locked(parent);
7518 parent_level = btrfs_header_level(parent);
7519 extent_buffer_get(parent);
7520 path->nodes[parent_level] = parent;
7521 path->slots[parent_level] = btrfs_header_nritems(parent);
7523 btrfs_assert_tree_locked(node);
7524 level = btrfs_header_level(node);
7525 path->nodes[level] = node;
7526 path->slots[level] = 0;
7527 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7529 wc->refs[parent_level] = 1;
7530 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7532 wc->shared_level = -1;
7533 wc->stage = DROP_REFERENCE;
7537 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7540 wret = walk_down_tree(trans, root, path, wc);
7546 wret = walk_up_tree(trans, root, path, wc, parent_level);
7554 btrfs_free_path(path);
7558 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7564 * if restripe for this chunk_type is on pick target profile and
7565 * return, otherwise do the usual balance
7567 stripped = get_restripe_target(root->fs_info, flags);
7569 return extended_to_chunk(stripped);
7572 * we add in the count of missing devices because we want
7573 * to make sure that any RAID levels on a degraded FS
7574 * continue to be honored.
7576 num_devices = root->fs_info->fs_devices->rw_devices +
7577 root->fs_info->fs_devices->missing_devices;
7579 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7580 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7581 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7583 if (num_devices == 1) {
7584 stripped |= BTRFS_BLOCK_GROUP_DUP;
7585 stripped = flags & ~stripped;
7587 /* turn raid0 into single device chunks */
7588 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7591 /* turn mirroring into duplication */
7592 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7593 BTRFS_BLOCK_GROUP_RAID10))
7594 return stripped | BTRFS_BLOCK_GROUP_DUP;
7596 /* they already had raid on here, just return */
7597 if (flags & stripped)
7600 stripped |= BTRFS_BLOCK_GROUP_DUP;
7601 stripped = flags & ~stripped;
7603 /* switch duplicated blocks with raid1 */
7604 if (flags & BTRFS_BLOCK_GROUP_DUP)
7605 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7607 /* this is drive concat, leave it alone */
7613 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7615 struct btrfs_space_info *sinfo = cache->space_info;
7617 u64 min_allocable_bytes;
7622 * We need some metadata space and system metadata space for
7623 * allocating chunks in some corner cases until we force to set
7624 * it to be readonly.
7627 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7629 min_allocable_bytes = 1 * 1024 * 1024;
7631 min_allocable_bytes = 0;
7633 spin_lock(&sinfo->lock);
7634 spin_lock(&cache->lock);
7641 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7642 cache->bytes_super - btrfs_block_group_used(&cache->item);
7644 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7645 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7646 min_allocable_bytes <= sinfo->total_bytes) {
7647 sinfo->bytes_readonly += num_bytes;
7652 spin_unlock(&cache->lock);
7653 spin_unlock(&sinfo->lock);
7657 int btrfs_set_block_group_ro(struct btrfs_root *root,
7658 struct btrfs_block_group_cache *cache)
7661 struct btrfs_trans_handle *trans;
7667 trans = btrfs_join_transaction(root);
7669 return PTR_ERR(trans);
7671 alloc_flags = update_block_group_flags(root, cache->flags);
7672 if (alloc_flags != cache->flags) {
7673 ret = do_chunk_alloc(trans, root, alloc_flags,
7679 ret = set_block_group_ro(cache, 0);
7682 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7683 ret = do_chunk_alloc(trans, root, alloc_flags,
7687 ret = set_block_group_ro(cache, 0);
7689 btrfs_end_transaction(trans, root);
7693 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7694 struct btrfs_root *root, u64 type)
7696 u64 alloc_flags = get_alloc_profile(root, type);
7697 return do_chunk_alloc(trans, root, alloc_flags,
7702 * helper to account the unused space of all the readonly block group in the
7703 * list. takes mirrors into account.
7705 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7707 struct btrfs_block_group_cache *block_group;
7711 list_for_each_entry(block_group, groups_list, list) {
7712 spin_lock(&block_group->lock);
7714 if (!block_group->ro) {
7715 spin_unlock(&block_group->lock);
7719 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7720 BTRFS_BLOCK_GROUP_RAID10 |
7721 BTRFS_BLOCK_GROUP_DUP))
7726 free_bytes += (block_group->key.offset -
7727 btrfs_block_group_used(&block_group->item)) *
7730 spin_unlock(&block_group->lock);
7737 * helper to account the unused space of all the readonly block group in the
7738 * space_info. takes mirrors into account.
7740 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7745 spin_lock(&sinfo->lock);
7747 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7748 if (!list_empty(&sinfo->block_groups[i]))
7749 free_bytes += __btrfs_get_ro_block_group_free_space(
7750 &sinfo->block_groups[i]);
7752 spin_unlock(&sinfo->lock);
7757 void btrfs_set_block_group_rw(struct btrfs_root *root,
7758 struct btrfs_block_group_cache *cache)
7760 struct btrfs_space_info *sinfo = cache->space_info;
7765 spin_lock(&sinfo->lock);
7766 spin_lock(&cache->lock);
7767 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7768 cache->bytes_super - btrfs_block_group_used(&cache->item);
7769 sinfo->bytes_readonly -= num_bytes;
7771 spin_unlock(&cache->lock);
7772 spin_unlock(&sinfo->lock);
7776 * checks to see if its even possible to relocate this block group.
7778 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7779 * ok to go ahead and try.
7781 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7783 struct btrfs_block_group_cache *block_group;
7784 struct btrfs_space_info *space_info;
7785 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7786 struct btrfs_device *device;
7795 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7797 /* odd, couldn't find the block group, leave it alone */
7801 min_free = btrfs_block_group_used(&block_group->item);
7803 /* no bytes used, we're good */
7807 space_info = block_group->space_info;
7808 spin_lock(&space_info->lock);
7810 full = space_info->full;
7813 * if this is the last block group we have in this space, we can't
7814 * relocate it unless we're able to allocate a new chunk below.
7816 * Otherwise, we need to make sure we have room in the space to handle
7817 * all of the extents from this block group. If we can, we're good
7819 if ((space_info->total_bytes != block_group->key.offset) &&
7820 (space_info->bytes_used + space_info->bytes_reserved +
7821 space_info->bytes_pinned + space_info->bytes_readonly +
7822 min_free < space_info->total_bytes)) {
7823 spin_unlock(&space_info->lock);
7826 spin_unlock(&space_info->lock);
7829 * ok we don't have enough space, but maybe we have free space on our
7830 * devices to allocate new chunks for relocation, so loop through our
7831 * alloc devices and guess if we have enough space. if this block
7832 * group is going to be restriped, run checks against the target
7833 * profile instead of the current one.
7845 target = get_restripe_target(root->fs_info, block_group->flags);
7847 index = __get_raid_index(extended_to_chunk(target));
7850 * this is just a balance, so if we were marked as full
7851 * we know there is no space for a new chunk
7856 index = get_block_group_index(block_group);
7859 if (index == BTRFS_RAID_RAID10) {
7863 } else if (index == BTRFS_RAID_RAID1) {
7865 } else if (index == BTRFS_RAID_DUP) {
7868 } else if (index == BTRFS_RAID_RAID0) {
7869 dev_min = fs_devices->rw_devices;
7870 do_div(min_free, dev_min);
7873 mutex_lock(&root->fs_info->chunk_mutex);
7874 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7878 * check to make sure we can actually find a chunk with enough
7879 * space to fit our block group in.
7881 if (device->total_bytes > device->bytes_used + min_free &&
7882 !device->is_tgtdev_for_dev_replace) {
7883 ret = find_free_dev_extent(device, min_free,
7888 if (dev_nr >= dev_min)
7894 mutex_unlock(&root->fs_info->chunk_mutex);
7896 btrfs_put_block_group(block_group);
7900 static int find_first_block_group(struct btrfs_root *root,
7901 struct btrfs_path *path, struct btrfs_key *key)
7904 struct btrfs_key found_key;
7905 struct extent_buffer *leaf;
7908 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7913 slot = path->slots[0];
7914 leaf = path->nodes[0];
7915 if (slot >= btrfs_header_nritems(leaf)) {
7916 ret = btrfs_next_leaf(root, path);
7923 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7925 if (found_key.objectid >= key->objectid &&
7926 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7936 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7938 struct btrfs_block_group_cache *block_group;
7942 struct inode *inode;
7944 block_group = btrfs_lookup_first_block_group(info, last);
7945 while (block_group) {
7946 spin_lock(&block_group->lock);
7947 if (block_group->iref)
7949 spin_unlock(&block_group->lock);
7950 block_group = next_block_group(info->tree_root,
7960 inode = block_group->inode;
7961 block_group->iref = 0;
7962 block_group->inode = NULL;
7963 spin_unlock(&block_group->lock);
7965 last = block_group->key.objectid + block_group->key.offset;
7966 btrfs_put_block_group(block_group);
7970 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7972 struct btrfs_block_group_cache *block_group;
7973 struct btrfs_space_info *space_info;
7974 struct btrfs_caching_control *caching_ctl;
7977 down_write(&info->extent_commit_sem);
7978 while (!list_empty(&info->caching_block_groups)) {
7979 caching_ctl = list_entry(info->caching_block_groups.next,
7980 struct btrfs_caching_control, list);
7981 list_del(&caching_ctl->list);
7982 put_caching_control(caching_ctl);
7984 up_write(&info->extent_commit_sem);
7986 spin_lock(&info->block_group_cache_lock);
7987 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7988 block_group = rb_entry(n, struct btrfs_block_group_cache,
7990 rb_erase(&block_group->cache_node,
7991 &info->block_group_cache_tree);
7992 spin_unlock(&info->block_group_cache_lock);
7994 down_write(&block_group->space_info->groups_sem);
7995 list_del(&block_group->list);
7996 up_write(&block_group->space_info->groups_sem);
7998 if (block_group->cached == BTRFS_CACHE_STARTED)
7999 wait_block_group_cache_done(block_group);
8002 * We haven't cached this block group, which means we could
8003 * possibly have excluded extents on this block group.
8005 if (block_group->cached == BTRFS_CACHE_NO)
8006 free_excluded_extents(info->extent_root, block_group);
8008 btrfs_remove_free_space_cache(block_group);
8009 btrfs_put_block_group(block_group);
8011 spin_lock(&info->block_group_cache_lock);
8013 spin_unlock(&info->block_group_cache_lock);
8015 /* now that all the block groups are freed, go through and
8016 * free all the space_info structs. This is only called during
8017 * the final stages of unmount, and so we know nobody is
8018 * using them. We call synchronize_rcu() once before we start,
8019 * just to be on the safe side.
8023 release_global_block_rsv(info);
8025 while(!list_empty(&info->space_info)) {
8026 space_info = list_entry(info->space_info.next,
8027 struct btrfs_space_info,
8029 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8030 if (space_info->bytes_pinned > 0 ||
8031 space_info->bytes_reserved > 0 ||
8032 space_info->bytes_may_use > 0) {
8034 dump_space_info(space_info, 0, 0);
8037 list_del(&space_info->list);
8043 static void __link_block_group(struct btrfs_space_info *space_info,
8044 struct btrfs_block_group_cache *cache)
8046 int index = get_block_group_index(cache);
8048 down_write(&space_info->groups_sem);
8049 list_add_tail(&cache->list, &space_info->block_groups[index]);
8050 up_write(&space_info->groups_sem);
8053 int btrfs_read_block_groups(struct btrfs_root *root)
8055 struct btrfs_path *path;
8057 struct btrfs_block_group_cache *cache;
8058 struct btrfs_fs_info *info = root->fs_info;
8059 struct btrfs_space_info *space_info;
8060 struct btrfs_key key;
8061 struct btrfs_key found_key;
8062 struct extent_buffer *leaf;
8066 root = info->extent_root;
8069 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8070 path = btrfs_alloc_path();
8075 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8076 if (btrfs_test_opt(root, SPACE_CACHE) &&
8077 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8079 if (btrfs_test_opt(root, CLEAR_CACHE))
8083 ret = find_first_block_group(root, path, &key);
8088 leaf = path->nodes[0];
8089 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8090 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8095 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8097 if (!cache->free_space_ctl) {
8103 atomic_set(&cache->count, 1);
8104 spin_lock_init(&cache->lock);
8105 cache->fs_info = info;
8106 INIT_LIST_HEAD(&cache->list);
8107 INIT_LIST_HEAD(&cache->cluster_list);
8111 * When we mount with old space cache, we need to
8112 * set BTRFS_DC_CLEAR and set dirty flag.
8114 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8115 * truncate the old free space cache inode and
8117 * b) Setting 'dirty flag' makes sure that we flush
8118 * the new space cache info onto disk.
8120 cache->disk_cache_state = BTRFS_DC_CLEAR;
8121 if (btrfs_test_opt(root, SPACE_CACHE))
8125 read_extent_buffer(leaf, &cache->item,
8126 btrfs_item_ptr_offset(leaf, path->slots[0]),
8127 sizeof(cache->item));
8128 memcpy(&cache->key, &found_key, sizeof(found_key));
8130 key.objectid = found_key.objectid + found_key.offset;
8131 btrfs_release_path(path);
8132 cache->flags = btrfs_block_group_flags(&cache->item);
8133 cache->sectorsize = root->sectorsize;
8134 cache->full_stripe_len = btrfs_full_stripe_len(root,
8135 &root->fs_info->mapping_tree,
8136 found_key.objectid);
8137 btrfs_init_free_space_ctl(cache);
8140 * We need to exclude the super stripes now so that the space
8141 * info has super bytes accounted for, otherwise we'll think
8142 * we have more space than we actually do.
8144 ret = exclude_super_stripes(root, cache);
8147 * We may have excluded something, so call this just in
8150 free_excluded_extents(root, cache);
8151 kfree(cache->free_space_ctl);
8157 * check for two cases, either we are full, and therefore
8158 * don't need to bother with the caching work since we won't
8159 * find any space, or we are empty, and we can just add all
8160 * the space in and be done with it. This saves us _alot_ of
8161 * time, particularly in the full case.
8163 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8164 cache->last_byte_to_unpin = (u64)-1;
8165 cache->cached = BTRFS_CACHE_FINISHED;
8166 free_excluded_extents(root, cache);
8167 } else if (btrfs_block_group_used(&cache->item) == 0) {
8168 cache->last_byte_to_unpin = (u64)-1;
8169 cache->cached = BTRFS_CACHE_FINISHED;
8170 add_new_free_space(cache, root->fs_info,
8172 found_key.objectid +
8174 free_excluded_extents(root, cache);
8177 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8179 btrfs_remove_free_space_cache(cache);
8180 btrfs_put_block_group(cache);
8184 ret = update_space_info(info, cache->flags, found_key.offset,
8185 btrfs_block_group_used(&cache->item),
8188 btrfs_remove_free_space_cache(cache);
8189 spin_lock(&info->block_group_cache_lock);
8190 rb_erase(&cache->cache_node,
8191 &info->block_group_cache_tree);
8192 spin_unlock(&info->block_group_cache_lock);
8193 btrfs_put_block_group(cache);
8197 cache->space_info = space_info;
8198 spin_lock(&cache->space_info->lock);
8199 cache->space_info->bytes_readonly += cache->bytes_super;
8200 spin_unlock(&cache->space_info->lock);
8202 __link_block_group(space_info, cache);
8204 set_avail_alloc_bits(root->fs_info, cache->flags);
8205 if (btrfs_chunk_readonly(root, cache->key.objectid))
8206 set_block_group_ro(cache, 1);
8209 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8210 if (!(get_alloc_profile(root, space_info->flags) &
8211 (BTRFS_BLOCK_GROUP_RAID10 |
8212 BTRFS_BLOCK_GROUP_RAID1 |
8213 BTRFS_BLOCK_GROUP_RAID5 |
8214 BTRFS_BLOCK_GROUP_RAID6 |
8215 BTRFS_BLOCK_GROUP_DUP)))
8218 * avoid allocating from un-mirrored block group if there are
8219 * mirrored block groups.
8221 list_for_each_entry(cache, &space_info->block_groups[3], list)
8222 set_block_group_ro(cache, 1);
8223 list_for_each_entry(cache, &space_info->block_groups[4], list)
8224 set_block_group_ro(cache, 1);
8227 init_global_block_rsv(info);
8230 btrfs_free_path(path);
8234 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8235 struct btrfs_root *root)
8237 struct btrfs_block_group_cache *block_group, *tmp;
8238 struct btrfs_root *extent_root = root->fs_info->extent_root;
8239 struct btrfs_block_group_item item;
8240 struct btrfs_key key;
8243 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8245 list_del_init(&block_group->new_bg_list);
8250 spin_lock(&block_group->lock);
8251 memcpy(&item, &block_group->item, sizeof(item));
8252 memcpy(&key, &block_group->key, sizeof(key));
8253 spin_unlock(&block_group->lock);
8255 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8258 btrfs_abort_transaction(trans, extent_root, ret);
8262 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8263 struct btrfs_root *root, u64 bytes_used,
8264 u64 type, u64 chunk_objectid, u64 chunk_offset,
8268 struct btrfs_root *extent_root;
8269 struct btrfs_block_group_cache *cache;
8271 extent_root = root->fs_info->extent_root;
8273 root->fs_info->last_trans_log_full_commit = trans->transid;
8275 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8278 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8280 if (!cache->free_space_ctl) {
8285 cache->key.objectid = chunk_offset;
8286 cache->key.offset = size;
8287 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8288 cache->sectorsize = root->sectorsize;
8289 cache->fs_info = root->fs_info;
8290 cache->full_stripe_len = btrfs_full_stripe_len(root,
8291 &root->fs_info->mapping_tree,
8294 atomic_set(&cache->count, 1);
8295 spin_lock_init(&cache->lock);
8296 INIT_LIST_HEAD(&cache->list);
8297 INIT_LIST_HEAD(&cache->cluster_list);
8298 INIT_LIST_HEAD(&cache->new_bg_list);
8300 btrfs_init_free_space_ctl(cache);
8302 btrfs_set_block_group_used(&cache->item, bytes_used);
8303 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8304 cache->flags = type;
8305 btrfs_set_block_group_flags(&cache->item, type);
8307 cache->last_byte_to_unpin = (u64)-1;
8308 cache->cached = BTRFS_CACHE_FINISHED;
8309 ret = exclude_super_stripes(root, cache);
8312 * We may have excluded something, so call this just in
8315 free_excluded_extents(root, cache);
8316 kfree(cache->free_space_ctl);
8321 add_new_free_space(cache, root->fs_info, chunk_offset,
8322 chunk_offset + size);
8324 free_excluded_extents(root, cache);
8326 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8328 btrfs_remove_free_space_cache(cache);
8329 btrfs_put_block_group(cache);
8333 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8334 &cache->space_info);
8336 btrfs_remove_free_space_cache(cache);
8337 spin_lock(&root->fs_info->block_group_cache_lock);
8338 rb_erase(&cache->cache_node,
8339 &root->fs_info->block_group_cache_tree);
8340 spin_unlock(&root->fs_info->block_group_cache_lock);
8341 btrfs_put_block_group(cache);
8344 update_global_block_rsv(root->fs_info);
8346 spin_lock(&cache->space_info->lock);
8347 cache->space_info->bytes_readonly += cache->bytes_super;
8348 spin_unlock(&cache->space_info->lock);
8350 __link_block_group(cache->space_info, cache);
8352 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8354 set_avail_alloc_bits(extent_root->fs_info, type);
8359 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8361 u64 extra_flags = chunk_to_extended(flags) &
8362 BTRFS_EXTENDED_PROFILE_MASK;
8364 write_seqlock(&fs_info->profiles_lock);
8365 if (flags & BTRFS_BLOCK_GROUP_DATA)
8366 fs_info->avail_data_alloc_bits &= ~extra_flags;
8367 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8368 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8369 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8370 fs_info->avail_system_alloc_bits &= ~extra_flags;
8371 write_sequnlock(&fs_info->profiles_lock);
8374 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8375 struct btrfs_root *root, u64 group_start)
8377 struct btrfs_path *path;
8378 struct btrfs_block_group_cache *block_group;
8379 struct btrfs_free_cluster *cluster;
8380 struct btrfs_root *tree_root = root->fs_info->tree_root;
8381 struct btrfs_key key;
8382 struct inode *inode;
8387 root = root->fs_info->extent_root;
8389 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8390 BUG_ON(!block_group);
8391 BUG_ON(!block_group->ro);
8394 * Free the reserved super bytes from this block group before
8397 free_excluded_extents(root, block_group);
8399 memcpy(&key, &block_group->key, sizeof(key));
8400 index = get_block_group_index(block_group);
8401 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8402 BTRFS_BLOCK_GROUP_RAID1 |
8403 BTRFS_BLOCK_GROUP_RAID10))
8408 /* make sure this block group isn't part of an allocation cluster */
8409 cluster = &root->fs_info->data_alloc_cluster;
8410 spin_lock(&cluster->refill_lock);
8411 btrfs_return_cluster_to_free_space(block_group, cluster);
8412 spin_unlock(&cluster->refill_lock);
8415 * make sure this block group isn't part of a metadata
8416 * allocation cluster
8418 cluster = &root->fs_info->meta_alloc_cluster;
8419 spin_lock(&cluster->refill_lock);
8420 btrfs_return_cluster_to_free_space(block_group, cluster);
8421 spin_unlock(&cluster->refill_lock);
8423 path = btrfs_alloc_path();
8429 inode = lookup_free_space_inode(tree_root, block_group, path);
8430 if (!IS_ERR(inode)) {
8431 ret = btrfs_orphan_add(trans, inode);
8433 btrfs_add_delayed_iput(inode);
8437 /* One for the block groups ref */
8438 spin_lock(&block_group->lock);
8439 if (block_group->iref) {
8440 block_group->iref = 0;
8441 block_group->inode = NULL;
8442 spin_unlock(&block_group->lock);
8445 spin_unlock(&block_group->lock);
8447 /* One for our lookup ref */
8448 btrfs_add_delayed_iput(inode);
8451 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8452 key.offset = block_group->key.objectid;
8455 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8459 btrfs_release_path(path);
8461 ret = btrfs_del_item(trans, tree_root, path);
8464 btrfs_release_path(path);
8467 spin_lock(&root->fs_info->block_group_cache_lock);
8468 rb_erase(&block_group->cache_node,
8469 &root->fs_info->block_group_cache_tree);
8471 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8472 root->fs_info->first_logical_byte = (u64)-1;
8473 spin_unlock(&root->fs_info->block_group_cache_lock);
8475 down_write(&block_group->space_info->groups_sem);
8477 * we must use list_del_init so people can check to see if they
8478 * are still on the list after taking the semaphore
8480 list_del_init(&block_group->list);
8481 if (list_empty(&block_group->space_info->block_groups[index]))
8482 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8483 up_write(&block_group->space_info->groups_sem);
8485 if (block_group->cached == BTRFS_CACHE_STARTED)
8486 wait_block_group_cache_done(block_group);
8488 btrfs_remove_free_space_cache(block_group);
8490 spin_lock(&block_group->space_info->lock);
8491 block_group->space_info->total_bytes -= block_group->key.offset;
8492 block_group->space_info->bytes_readonly -= block_group->key.offset;
8493 block_group->space_info->disk_total -= block_group->key.offset * factor;
8494 spin_unlock(&block_group->space_info->lock);
8496 memcpy(&key, &block_group->key, sizeof(key));
8498 btrfs_clear_space_info_full(root->fs_info);
8500 btrfs_put_block_group(block_group);
8501 btrfs_put_block_group(block_group);
8503 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8509 ret = btrfs_del_item(trans, root, path);
8511 btrfs_free_path(path);
8515 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8517 struct btrfs_space_info *space_info;
8518 struct btrfs_super_block *disk_super;
8524 disk_super = fs_info->super_copy;
8525 if (!btrfs_super_root(disk_super))
8528 features = btrfs_super_incompat_flags(disk_super);
8529 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8532 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8533 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8538 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8539 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8541 flags = BTRFS_BLOCK_GROUP_METADATA;
8542 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8546 flags = BTRFS_BLOCK_GROUP_DATA;
8547 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8553 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8555 return unpin_extent_range(root, start, end);
8558 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8559 u64 num_bytes, u64 *actual_bytes)
8561 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8564 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8566 struct btrfs_fs_info *fs_info = root->fs_info;
8567 struct btrfs_block_group_cache *cache = NULL;
8572 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8576 * try to trim all FS space, our block group may start from non-zero.
8578 if (range->len == total_bytes)
8579 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8581 cache = btrfs_lookup_block_group(fs_info, range->start);
8584 if (cache->key.objectid >= (range->start + range->len)) {
8585 btrfs_put_block_group(cache);
8589 start = max(range->start, cache->key.objectid);
8590 end = min(range->start + range->len,
8591 cache->key.objectid + cache->key.offset);
8593 if (end - start >= range->minlen) {
8594 if (!block_group_cache_done(cache)) {
8595 ret = cache_block_group(cache, 0);
8597 wait_block_group_cache_done(cache);
8599 ret = btrfs_trim_block_group(cache,
8605 trimmed += group_trimmed;
8607 btrfs_put_block_group(cache);
8612 cache = next_block_group(fs_info->tree_root, cache);
8615 range->len = trimmed;