2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE = 0,
57 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 static int update_block_group(struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
110 block_group_cache_done(struct btrfs_block_group_cache *cache)
113 return cache->cached == BTRFS_CACHE_FINISHED;
116 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
118 return (cache->flags & bits) == bits;
121 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
123 atomic_inc(&cache->count);
126 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
128 if (atomic_dec_and_test(&cache->count)) {
129 WARN_ON(cache->pinned > 0);
130 WARN_ON(cache->reserved > 0);
131 kfree(cache->free_space_ctl);
137 * this adds the block group to the fs_info rb tree for the block group
140 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
141 struct btrfs_block_group_cache *block_group)
144 struct rb_node *parent = NULL;
145 struct btrfs_block_group_cache *cache;
147 spin_lock(&info->block_group_cache_lock);
148 p = &info->block_group_cache_tree.rb_node;
152 cache = rb_entry(parent, struct btrfs_block_group_cache,
154 if (block_group->key.objectid < cache->key.objectid) {
156 } else if (block_group->key.objectid > cache->key.objectid) {
159 spin_unlock(&info->block_group_cache_lock);
164 rb_link_node(&block_group->cache_node, parent, p);
165 rb_insert_color(&block_group->cache_node,
166 &info->block_group_cache_tree);
168 if (info->first_logical_byte > block_group->key.objectid)
169 info->first_logical_byte = block_group->key.objectid;
171 spin_unlock(&info->block_group_cache_lock);
177 * This will return the block group at or after bytenr if contains is 0, else
178 * it will return the block group that contains the bytenr
180 static struct btrfs_block_group_cache *
181 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
184 struct btrfs_block_group_cache *cache, *ret = NULL;
188 spin_lock(&info->block_group_cache_lock);
189 n = info->block_group_cache_tree.rb_node;
192 cache = rb_entry(n, struct btrfs_block_group_cache,
194 end = cache->key.objectid + cache->key.offset - 1;
195 start = cache->key.objectid;
197 if (bytenr < start) {
198 if (!contains && (!ret || start < ret->key.objectid))
201 } else if (bytenr > start) {
202 if (contains && bytenr <= end) {
213 btrfs_get_block_group(ret);
214 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
215 info->first_logical_byte = ret->key.objectid;
217 spin_unlock(&info->block_group_cache_lock);
222 static int add_excluded_extent(struct btrfs_root *root,
223 u64 start, u64 num_bytes)
225 u64 end = start + num_bytes - 1;
226 set_extent_bits(&root->fs_info->freed_extents[0],
227 start, end, EXTENT_UPTODATE, GFP_NOFS);
228 set_extent_bits(&root->fs_info->freed_extents[1],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 static void free_excluded_extents(struct btrfs_root *root,
234 struct btrfs_block_group_cache *cache)
238 start = cache->key.objectid;
239 end = start + cache->key.offset - 1;
241 clear_extent_bits(&root->fs_info->freed_extents[0],
242 start, end, EXTENT_UPTODATE, GFP_NOFS);
243 clear_extent_bits(&root->fs_info->freed_extents[1],
244 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 static int exclude_super_stripes(struct btrfs_root *root,
248 struct btrfs_block_group_cache *cache)
255 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
256 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
257 cache->bytes_super += stripe_len;
258 ret = add_excluded_extent(root, cache->key.objectid,
264 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
265 bytenr = btrfs_sb_offset(i);
266 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
267 cache->key.objectid, bytenr,
268 0, &logical, &nr, &stripe_len);
273 cache->bytes_super += stripe_len;
274 ret = add_excluded_extent(root, logical[nr],
287 static struct btrfs_caching_control *
288 get_caching_control(struct btrfs_block_group_cache *cache)
290 struct btrfs_caching_control *ctl;
292 spin_lock(&cache->lock);
293 if (cache->cached != BTRFS_CACHE_STARTED) {
294 spin_unlock(&cache->lock);
298 /* We're loading it the fast way, so we don't have a caching_ctl. */
299 if (!cache->caching_ctl) {
300 spin_unlock(&cache->lock);
304 ctl = cache->caching_ctl;
305 atomic_inc(&ctl->count);
306 spin_unlock(&cache->lock);
310 static void put_caching_control(struct btrfs_caching_control *ctl)
312 if (atomic_dec_and_test(&ctl->count))
317 * this is only called by cache_block_group, since we could have freed extents
318 * we need to check the pinned_extents for any extents that can't be used yet
319 * since their free space will be released as soon as the transaction commits.
321 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
322 struct btrfs_fs_info *info, u64 start, u64 end)
324 u64 extent_start, extent_end, size, total_added = 0;
327 while (start < end) {
328 ret = find_first_extent_bit(info->pinned_extents, start,
329 &extent_start, &extent_end,
330 EXTENT_DIRTY | EXTENT_UPTODATE,
335 if (extent_start <= start) {
336 start = extent_end + 1;
337 } else if (extent_start > start && extent_start < end) {
338 size = extent_start - start;
340 ret = btrfs_add_free_space(block_group, start,
342 BUG_ON(ret); /* -ENOMEM or logic error */
343 start = extent_end + 1;
352 ret = btrfs_add_free_space(block_group, start, size);
353 BUG_ON(ret); /* -ENOMEM or logic error */
359 static noinline void caching_thread(struct btrfs_work *work)
361 struct btrfs_block_group_cache *block_group;
362 struct btrfs_fs_info *fs_info;
363 struct btrfs_caching_control *caching_ctl;
364 struct btrfs_root *extent_root;
365 struct btrfs_path *path;
366 struct extent_buffer *leaf;
367 struct btrfs_key key;
373 caching_ctl = container_of(work, struct btrfs_caching_control, work);
374 block_group = caching_ctl->block_group;
375 fs_info = block_group->fs_info;
376 extent_root = fs_info->extent_root;
378 path = btrfs_alloc_path();
382 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
385 * We don't want to deadlock with somebody trying to allocate a new
386 * extent for the extent root while also trying to search the extent
387 * root to add free space. So we skip locking and search the commit
388 * root, since its read-only
390 path->skip_locking = 1;
391 path->search_commit_root = 1;
396 key.type = BTRFS_EXTENT_ITEM_KEY;
398 mutex_lock(&caching_ctl->mutex);
399 /* need to make sure the commit_root doesn't disappear */
400 down_read(&fs_info->extent_commit_sem);
402 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
406 leaf = path->nodes[0];
407 nritems = btrfs_header_nritems(leaf);
410 if (btrfs_fs_closing(fs_info) > 1) {
415 if (path->slots[0] < nritems) {
416 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
418 ret = find_next_key(path, 0, &key);
422 if (need_resched() ||
423 btrfs_next_leaf(extent_root, path)) {
424 caching_ctl->progress = last;
425 btrfs_release_path(path);
426 up_read(&fs_info->extent_commit_sem);
427 mutex_unlock(&caching_ctl->mutex);
431 leaf = path->nodes[0];
432 nritems = btrfs_header_nritems(leaf);
436 if (key.objectid < block_group->key.objectid) {
441 if (key.objectid >= block_group->key.objectid +
442 block_group->key.offset)
445 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
446 key.type == BTRFS_METADATA_ITEM_KEY) {
447 total_found += add_new_free_space(block_group,
450 if (key.type == BTRFS_METADATA_ITEM_KEY)
451 last = key.objectid +
452 fs_info->tree_root->leafsize;
454 last = key.objectid + key.offset;
456 if (total_found > (1024 * 1024 * 2)) {
458 wake_up(&caching_ctl->wait);
465 total_found += add_new_free_space(block_group, fs_info, last,
466 block_group->key.objectid +
467 block_group->key.offset);
468 caching_ctl->progress = (u64)-1;
470 spin_lock(&block_group->lock);
471 block_group->caching_ctl = NULL;
472 block_group->cached = BTRFS_CACHE_FINISHED;
473 spin_unlock(&block_group->lock);
476 btrfs_free_path(path);
477 up_read(&fs_info->extent_commit_sem);
479 free_excluded_extents(extent_root, block_group);
481 mutex_unlock(&caching_ctl->mutex);
483 wake_up(&caching_ctl->wait);
485 put_caching_control(caching_ctl);
486 btrfs_put_block_group(block_group);
489 static int cache_block_group(struct btrfs_block_group_cache *cache,
493 struct btrfs_fs_info *fs_info = cache->fs_info;
494 struct btrfs_caching_control *caching_ctl;
497 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
501 INIT_LIST_HEAD(&caching_ctl->list);
502 mutex_init(&caching_ctl->mutex);
503 init_waitqueue_head(&caching_ctl->wait);
504 caching_ctl->block_group = cache;
505 caching_ctl->progress = cache->key.objectid;
506 atomic_set(&caching_ctl->count, 1);
507 caching_ctl->work.func = caching_thread;
509 spin_lock(&cache->lock);
511 * This should be a rare occasion, but this could happen I think in the
512 * case where one thread starts to load the space cache info, and then
513 * some other thread starts a transaction commit which tries to do an
514 * allocation while the other thread is still loading the space cache
515 * info. The previous loop should have kept us from choosing this block
516 * group, but if we've moved to the state where we will wait on caching
517 * block groups we need to first check if we're doing a fast load here,
518 * so we can wait for it to finish, otherwise we could end up allocating
519 * from a block group who's cache gets evicted for one reason or
522 while (cache->cached == BTRFS_CACHE_FAST) {
523 struct btrfs_caching_control *ctl;
525 ctl = cache->caching_ctl;
526 atomic_inc(&ctl->count);
527 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
528 spin_unlock(&cache->lock);
532 finish_wait(&ctl->wait, &wait);
533 put_caching_control(ctl);
534 spin_lock(&cache->lock);
537 if (cache->cached != BTRFS_CACHE_NO) {
538 spin_unlock(&cache->lock);
542 WARN_ON(cache->caching_ctl);
543 cache->caching_ctl = caching_ctl;
544 cache->cached = BTRFS_CACHE_FAST;
545 spin_unlock(&cache->lock);
547 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
548 ret = load_free_space_cache(fs_info, cache);
550 spin_lock(&cache->lock);
552 cache->caching_ctl = NULL;
553 cache->cached = BTRFS_CACHE_FINISHED;
554 cache->last_byte_to_unpin = (u64)-1;
556 if (load_cache_only) {
557 cache->caching_ctl = NULL;
558 cache->cached = BTRFS_CACHE_NO;
560 cache->cached = BTRFS_CACHE_STARTED;
563 spin_unlock(&cache->lock);
564 wake_up(&caching_ctl->wait);
566 put_caching_control(caching_ctl);
567 free_excluded_extents(fs_info->extent_root, cache);
572 * We are not going to do the fast caching, set cached to the
573 * appropriate value and wakeup any waiters.
575 spin_lock(&cache->lock);
576 if (load_cache_only) {
577 cache->caching_ctl = NULL;
578 cache->cached = BTRFS_CACHE_NO;
580 cache->cached = BTRFS_CACHE_STARTED;
582 spin_unlock(&cache->lock);
583 wake_up(&caching_ctl->wait);
586 if (load_cache_only) {
587 put_caching_control(caching_ctl);
591 down_write(&fs_info->extent_commit_sem);
592 atomic_inc(&caching_ctl->count);
593 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
594 up_write(&fs_info->extent_commit_sem);
596 btrfs_get_block_group(cache);
598 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
604 * return the block group that starts at or after bytenr
606 static struct btrfs_block_group_cache *
607 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
609 struct btrfs_block_group_cache *cache;
611 cache = block_group_cache_tree_search(info, bytenr, 0);
617 * return the block group that contains the given bytenr
619 struct btrfs_block_group_cache *btrfs_lookup_block_group(
620 struct btrfs_fs_info *info,
623 struct btrfs_block_group_cache *cache;
625 cache = block_group_cache_tree_search(info, bytenr, 1);
630 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
633 struct list_head *head = &info->space_info;
634 struct btrfs_space_info *found;
636 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
639 list_for_each_entry_rcu(found, head, list) {
640 if (found->flags & flags) {
650 * after adding space to the filesystem, we need to clear the full flags
651 * on all the space infos.
653 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
655 struct list_head *head = &info->space_info;
656 struct btrfs_space_info *found;
659 list_for_each_entry_rcu(found, head, list)
664 u64 btrfs_find_block_group(struct btrfs_root *root,
665 u64 search_start, u64 search_hint, int owner)
667 struct btrfs_block_group_cache *cache;
669 u64 last = max(search_hint, search_start);
676 cache = btrfs_lookup_first_block_group(root->fs_info, last);
680 spin_lock(&cache->lock);
681 last = cache->key.objectid + cache->key.offset;
682 used = btrfs_block_group_used(&cache->item);
684 if ((full_search || !cache->ro) &&
685 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
686 if (used + cache->pinned + cache->reserved <
687 div_factor(cache->key.offset, factor)) {
688 group_start = cache->key.objectid;
689 spin_unlock(&cache->lock);
690 btrfs_put_block_group(cache);
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
703 if (!full_search && factor < 10) {
713 /* simple helper to search for an existing extent at a given offset */
714 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
730 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
731 if (key.objectid == start &&
732 key.type == BTRFS_METADATA_ITEM_KEY)
735 btrfs_free_path(path);
740 * helper function to lookup reference count and flags of a tree block.
742 * the head node for delayed ref is used to store the sum of all the
743 * reference count modifications queued up in the rbtree. the head
744 * node may also store the extent flags to set. This way you can check
745 * to see what the reference count and extent flags would be if all of
746 * the delayed refs are not processed.
748 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
749 struct btrfs_root *root, u64 bytenr,
750 u64 offset, int metadata, u64 *refs, u64 *flags)
752 struct btrfs_delayed_ref_head *head;
753 struct btrfs_delayed_ref_root *delayed_refs;
754 struct btrfs_path *path;
755 struct btrfs_extent_item *ei;
756 struct extent_buffer *leaf;
757 struct btrfs_key key;
764 * If we don't have skinny metadata, don't bother doing anything
767 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
768 offset = root->leafsize;
772 path = btrfs_alloc_path();
777 key.objectid = bytenr;
778 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.objectid = bytenr;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
787 path->skip_locking = 1;
788 path->search_commit_root = 1;
791 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
796 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
797 key.type = BTRFS_EXTENT_ITEM_KEY;
798 key.offset = root->leafsize;
799 btrfs_release_path(path);
804 leaf = path->nodes[0];
805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
806 if (item_size >= sizeof(*ei)) {
807 ei = btrfs_item_ptr(leaf, path->slots[0],
808 struct btrfs_extent_item);
809 num_refs = btrfs_extent_refs(leaf, ei);
810 extent_flags = btrfs_extent_flags(leaf, ei);
812 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
813 struct btrfs_extent_item_v0 *ei0;
814 BUG_ON(item_size != sizeof(*ei0));
815 ei0 = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_extent_item_v0);
817 num_refs = btrfs_extent_refs_v0(leaf, ei0);
818 /* FIXME: this isn't correct for data */
819 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
824 BUG_ON(num_refs == 0);
834 delayed_refs = &trans->transaction->delayed_refs;
835 spin_lock(&delayed_refs->lock);
836 head = btrfs_find_delayed_ref_head(trans, bytenr);
838 if (!mutex_trylock(&head->mutex)) {
839 atomic_inc(&head->node.refs);
840 spin_unlock(&delayed_refs->lock);
842 btrfs_release_path(path);
845 * Mutex was contended, block until it's released and try
848 mutex_lock(&head->mutex);
849 mutex_unlock(&head->mutex);
850 btrfs_put_delayed_ref(&head->node);
853 if (head->extent_op && head->extent_op->update_flags)
854 extent_flags |= head->extent_op->flags_to_set;
856 BUG_ON(num_refs == 0);
858 num_refs += head->node.ref_mod;
859 mutex_unlock(&head->mutex);
861 spin_unlock(&delayed_refs->lock);
863 WARN_ON(num_refs == 0);
867 *flags = extent_flags;
869 btrfs_free_path(path);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct btrfs_path *path,
983 u64 owner, u32 extra_size)
985 struct btrfs_extent_item *item;
986 struct btrfs_extent_item_v0 *ei0;
987 struct btrfs_extent_ref_v0 *ref0;
988 struct btrfs_tree_block_info *bi;
989 struct extent_buffer *leaf;
990 struct btrfs_key key;
991 struct btrfs_key found_key;
992 u32 new_size = sizeof(*item);
996 leaf = path->nodes[0];
997 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
999 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_extent_item_v0);
1002 refs = btrfs_extent_refs_v0(leaf, ei0);
1004 if (owner == (u64)-1) {
1006 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007 ret = btrfs_next_leaf(root, path);
1010 BUG_ON(ret > 0); /* Corruption */
1011 leaf = path->nodes[0];
1013 btrfs_item_key_to_cpu(leaf, &found_key,
1015 BUG_ON(key.objectid != found_key.objectid);
1016 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1020 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_extent_ref_v0);
1022 owner = btrfs_ref_objectid_v0(leaf, ref0);
1026 btrfs_release_path(path);
1028 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029 new_size += sizeof(*bi);
1031 new_size -= sizeof(*ei0);
1032 ret = btrfs_search_slot(trans, root, &key, path,
1033 new_size + extra_size, 1);
1036 BUG_ON(ret); /* Corruption */
1038 btrfs_extend_item(trans, root, path, new_size);
1040 leaf = path->nodes[0];
1041 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042 btrfs_set_extent_refs(leaf, item, refs);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf, item, 0);
1045 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046 btrfs_set_extent_flags(leaf, item,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049 bi = (struct btrfs_tree_block_info *)(item + 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1054 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1056 btrfs_mark_buffer_dirty(leaf);
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1063 u32 high_crc = ~(u32)0;
1064 u32 low_crc = ~(u32)0;
1067 lenum = cpu_to_le64(root_objectid);
1068 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1069 lenum = cpu_to_le64(owner);
1070 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1071 lenum = cpu_to_le64(offset);
1072 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1074 return ((u64)high_crc << 31) ^ (u64)low_crc;
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078 struct btrfs_extent_data_ref *ref)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081 btrfs_extent_data_ref_objectid(leaf, ref),
1082 btrfs_extent_data_ref_offset(leaf, ref));
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref,
1087 u64 root_objectid, u64 owner, u64 offset)
1089 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 u64 bytenr, u64 parent,
1101 u64 owner, u64 offset)
1103 struct btrfs_key key;
1104 struct btrfs_extent_data_ref *ref;
1105 struct extent_buffer *leaf;
1111 key.objectid = bytenr;
1113 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 key.offset = parent;
1116 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 key.offset = hash_extent_data_ref(root_objectid,
1122 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133 btrfs_release_path(path);
1134 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1145 leaf = path->nodes[0];
1146 nritems = btrfs_header_nritems(leaf);
1148 if (path->slots[0] >= nritems) {
1149 ret = btrfs_next_leaf(root, path);
1155 leaf = path->nodes[0];
1156 nritems = btrfs_header_nritems(leaf);
1160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161 if (key.objectid != bytenr ||
1162 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1165 ref = btrfs_item_ptr(leaf, path->slots[0],
1166 struct btrfs_extent_data_ref);
1168 if (match_extent_data_ref(leaf, ref, root_objectid,
1171 btrfs_release_path(path);
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 struct btrfs_path *path,
1186 u64 bytenr, u64 parent,
1187 u64 root_objectid, u64 owner,
1188 u64 offset, int refs_to_add)
1190 struct btrfs_key key;
1191 struct extent_buffer *leaf;
1196 key.objectid = bytenr;
1198 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199 key.offset = parent;
1200 size = sizeof(struct btrfs_shared_data_ref);
1202 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203 key.offset = hash_extent_data_ref(root_objectid,
1205 size = sizeof(struct btrfs_extent_data_ref);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209 if (ret && ret != -EEXIST)
1212 leaf = path->nodes[0];
1214 struct btrfs_shared_data_ref *ref;
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1218 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1220 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221 num_refs += refs_to_add;
1222 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1225 struct btrfs_extent_data_ref *ref;
1226 while (ret == -EEXIST) {
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1229 if (match_extent_data_ref(leaf, ref, root_objectid,
1232 btrfs_release_path(path);
1234 ret = btrfs_insert_empty_item(trans, root, path, &key,
1236 if (ret && ret != -EEXIST)
1239 leaf = path->nodes[0];
1241 ref = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_extent_data_ref);
1244 btrfs_set_extent_data_ref_root(leaf, ref,
1246 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1250 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251 num_refs += refs_to_add;
1252 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1255 btrfs_mark_buffer_dirty(leaf);
1258 btrfs_release_path(path);
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root,
1264 struct btrfs_path *path,
1267 struct btrfs_key key;
1268 struct btrfs_extent_data_ref *ref1 = NULL;
1269 struct btrfs_shared_data_ref *ref2 = NULL;
1270 struct extent_buffer *leaf;
1274 leaf = path->nodes[0];
1275 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_extent_data_ref);
1280 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283 struct btrfs_shared_data_ref);
1284 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287 struct btrfs_extent_ref_v0 *ref0;
1288 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_extent_ref_v0);
1290 num_refs = btrfs_ref_count_v0(leaf, ref0);
1296 BUG_ON(num_refs < refs_to_drop);
1297 num_refs -= refs_to_drop;
1299 if (num_refs == 0) {
1300 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1367 struct btrfs_key key;
1370 key.objectid = bytenr;
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1400 struct btrfs_key key;
1403 key.objectid = bytenr;
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1497 want = extent_ref_type(parent, owner);
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1536 key.type = BTRFS_EXTENT_ITEM_KEY;
1537 key.offset = num_bytes;
1538 btrfs_release_path(path);
1543 if (ret && !insert) {
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1570 BUG_ON(item_size < sizeof(*ei));
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1594 ptr += btrfs_extent_inline_ref_size(type);
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 if (parent == ref_offset) {
1617 if (ref_offset < parent)
1620 if (root_objectid == ref_offset) {
1624 if (ref_offset < root_objectid)
1628 ptr += btrfs_extent_inline_ref_size(type);
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_extent_inline_ref *iref,
1666 u64 parent, u64 root_objectid,
1667 u64 owner, u64 offset, int refs_to_add,
1668 struct btrfs_delayed_extent_op *extent_op)
1670 struct extent_buffer *leaf;
1671 struct btrfs_extent_item *ei;
1674 unsigned long item_offset;
1679 leaf = path->nodes[0];
1680 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681 item_offset = (unsigned long)iref - (unsigned long)ei;
1683 type = extent_ref_type(parent, owner);
1684 size = btrfs_extent_inline_ref_size(type);
1686 btrfs_extend_item(trans, root, path, size);
1688 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 refs = btrfs_extent_refs(leaf, ei);
1690 refs += refs_to_add;
1691 btrfs_set_extent_refs(leaf, ei, refs);
1693 __run_delayed_extent_op(extent_op, leaf, ei);
1695 ptr = (unsigned long)ei + item_offset;
1696 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697 if (ptr < end - size)
1698 memmove_extent_buffer(leaf, ptr + size, ptr,
1701 iref = (struct btrfs_extent_inline_ref *)ptr;
1702 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704 struct btrfs_extent_data_ref *dref;
1705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 struct btrfs_shared_data_ref *sref;
1712 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1720 btrfs_mark_buffer_dirty(leaf);
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref **ref_ret,
1727 u64 bytenr, u64 num_bytes, u64 parent,
1728 u64 root_objectid, u64 owner, u64 offset)
1732 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733 bytenr, num_bytes, parent,
1734 root_objectid, owner, offset, 0);
1738 btrfs_release_path(path);
1741 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1745 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746 root_objectid, owner, offset);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 struct btrfs_path *path,
1758 struct btrfs_extent_inline_ref *iref,
1760 struct btrfs_delayed_extent_op *extent_op)
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *ei;
1764 struct btrfs_extent_data_ref *dref = NULL;
1765 struct btrfs_shared_data_ref *sref = NULL;
1773 leaf = path->nodes[0];
1774 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775 refs = btrfs_extent_refs(leaf, ei);
1776 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777 refs += refs_to_mod;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1782 type = btrfs_extent_inline_ref_type(leaf, iref);
1784 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786 refs = btrfs_extent_data_ref_count(leaf, dref);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789 refs = btrfs_shared_data_ref_count(leaf, sref);
1792 BUG_ON(refs_to_mod != -1);
1795 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796 refs += refs_to_mod;
1799 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1802 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1812 btrfs_truncate_item(trans, root, path, item_size, 1);
1814 btrfs_mark_buffer_dirty(leaf);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct btrfs_extent_inline_ref *iref;
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(trans, root, path, iref,
1835 refs_to_add, extent_op);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(trans, root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data)
1872 BUG_ON(!is_data && refs_to_drop != 1);
1874 update_inline_extent_backref(trans, root, path, iref,
1875 -refs_to_drop, NULL);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1879 ret = btrfs_del_item(trans, root, path);
1884 static int btrfs_issue_discard(struct block_device *bdev,
1887 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1890 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1891 u64 num_bytes, u64 *actual_bytes)
1894 u64 discarded_bytes = 0;
1895 struct btrfs_bio *bbio = NULL;
1898 /* Tell the block device(s) that the sectors can be discarded */
1899 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1900 bytenr, &num_bytes, &bbio, 0);
1901 /* Error condition is -ENOMEM */
1903 struct btrfs_bio_stripe *stripe = bbio->stripes;
1907 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1908 if (!stripe->dev->can_discard)
1911 ret = btrfs_issue_discard(stripe->dev->bdev,
1915 discarded_bytes += stripe->length;
1916 else if (ret != -EOPNOTSUPP)
1917 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1920 * Just in case we get back EOPNOTSUPP for some reason,
1921 * just ignore the return value so we don't screw up
1922 * people calling discard_extent.
1930 *actual_bytes = discarded_bytes;
1933 if (ret == -EOPNOTSUPP)
1938 /* Can return -ENOMEM */
1939 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1940 struct btrfs_root *root,
1941 u64 bytenr, u64 num_bytes, u64 parent,
1942 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1945 struct btrfs_fs_info *fs_info = root->fs_info;
1947 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1948 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1950 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1951 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1953 parent, root_objectid, (int)owner,
1954 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1956 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1958 parent, root_objectid, owner, offset,
1959 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1964 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root,
1966 u64 bytenr, u64 num_bytes,
1967 u64 parent, u64 root_objectid,
1968 u64 owner, u64 offset, int refs_to_add,
1969 struct btrfs_delayed_extent_op *extent_op)
1971 struct btrfs_path *path;
1972 struct extent_buffer *leaf;
1973 struct btrfs_extent_item *item;
1978 path = btrfs_alloc_path();
1983 path->leave_spinning = 1;
1984 /* this will setup the path even if it fails to insert the back ref */
1985 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1986 path, bytenr, num_bytes, parent,
1987 root_objectid, owner, offset,
1988 refs_to_add, extent_op);
1992 if (ret != -EAGAIN) {
1997 leaf = path->nodes[0];
1998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 refs = btrfs_extent_refs(leaf, item);
2000 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2002 __run_delayed_extent_op(extent_op, leaf, item);
2004 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_release_path(path);
2008 path->leave_spinning = 1;
2010 /* now insert the actual backref */
2011 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012 path, bytenr, parent, root_objectid,
2013 owner, offset, refs_to_add);
2015 btrfs_abort_transaction(trans, root, ret);
2017 btrfs_free_path(path);
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op,
2025 int insert_reserved)
2028 struct btrfs_delayed_data_ref *ref;
2029 struct btrfs_key ins;
2034 ins.objectid = node->bytenr;
2035 ins.offset = node->num_bytes;
2036 ins.type = BTRFS_EXTENT_ITEM_KEY;
2038 ref = btrfs_delayed_node_to_data_ref(node);
2039 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2040 parent = ref->parent;
2042 ref_root = ref->root;
2044 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2046 flags |= extent_op->flags_to_set;
2047 ret = alloc_reserved_file_extent(trans, root,
2048 parent, ref_root, flags,
2049 ref->objectid, ref->offset,
2050 &ins, node->ref_mod);
2051 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2052 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2053 node->num_bytes, parent,
2054 ref_root, ref->objectid,
2055 ref->offset, node->ref_mod,
2057 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2058 ret = __btrfs_free_extent(trans, root, node->bytenr,
2059 node->num_bytes, parent,
2060 ref_root, ref->objectid,
2061 ref->offset, node->ref_mod,
2069 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2070 struct extent_buffer *leaf,
2071 struct btrfs_extent_item *ei)
2073 u64 flags = btrfs_extent_flags(leaf, ei);
2074 if (extent_op->update_flags) {
2075 flags |= extent_op->flags_to_set;
2076 btrfs_set_extent_flags(leaf, ei, flags);
2079 if (extent_op->update_key) {
2080 struct btrfs_tree_block_info *bi;
2081 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2082 bi = (struct btrfs_tree_block_info *)(ei + 1);
2083 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2087 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct btrfs_delayed_ref_node *node,
2090 struct btrfs_delayed_extent_op *extent_op)
2092 struct btrfs_key key;
2093 struct btrfs_path *path;
2094 struct btrfs_extent_item *ei;
2095 struct extent_buffer *leaf;
2099 int metadata = (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2100 node->type == BTRFS_SHARED_BLOCK_REF_KEY);
2105 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2108 path = btrfs_alloc_path();
2112 key.objectid = node->bytenr;
2115 struct btrfs_delayed_tree_ref *tree_ref;
2117 tree_ref = btrfs_delayed_node_to_tree_ref(node);
2118 key.type = BTRFS_METADATA_ITEM_KEY;
2119 key.offset = tree_ref->level;
2121 key.type = BTRFS_EXTENT_ITEM_KEY;
2122 key.offset = node->num_bytes;
2127 path->leave_spinning = 1;
2128 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2136 btrfs_release_path(path);
2139 key.offset = node->num_bytes;
2140 key.type = BTRFS_EXTENT_ITEM_KEY;
2147 leaf = path->nodes[0];
2148 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2149 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2150 if (item_size < sizeof(*ei)) {
2151 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2157 leaf = path->nodes[0];
2158 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2161 BUG_ON(item_size < sizeof(*ei));
2162 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2163 __run_delayed_extent_op(extent_op, leaf, ei);
2165 btrfs_mark_buffer_dirty(leaf);
2167 btrfs_free_path(path);
2171 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2172 struct btrfs_root *root,
2173 struct btrfs_delayed_ref_node *node,
2174 struct btrfs_delayed_extent_op *extent_op,
2175 int insert_reserved)
2178 struct btrfs_delayed_tree_ref *ref;
2179 struct btrfs_key ins;
2182 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2185 ref = btrfs_delayed_node_to_tree_ref(node);
2186 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2187 parent = ref->parent;
2189 ref_root = ref->root;
2191 ins.objectid = node->bytenr;
2192 if (skinny_metadata) {
2193 ins.offset = ref->level;
2194 ins.type = BTRFS_METADATA_ITEM_KEY;
2196 ins.offset = node->num_bytes;
2197 ins.type = BTRFS_EXTENT_ITEM_KEY;
2200 BUG_ON(node->ref_mod != 1);
2201 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2202 BUG_ON(!extent_op || !extent_op->update_flags);
2203 ret = alloc_reserved_tree_block(trans, root,
2205 extent_op->flags_to_set,
2208 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2209 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2210 node->num_bytes, parent, ref_root,
2211 ref->level, 0, 1, extent_op);
2212 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2213 ret = __btrfs_free_extent(trans, root, node->bytenr,
2214 node->num_bytes, parent, ref_root,
2215 ref->level, 0, 1, extent_op);
2222 /* helper function to actually process a single delayed ref entry */
2223 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2224 struct btrfs_root *root,
2225 struct btrfs_delayed_ref_node *node,
2226 struct btrfs_delayed_extent_op *extent_op,
2227 int insert_reserved)
2234 if (btrfs_delayed_ref_is_head(node)) {
2235 struct btrfs_delayed_ref_head *head;
2237 * we've hit the end of the chain and we were supposed
2238 * to insert this extent into the tree. But, it got
2239 * deleted before we ever needed to insert it, so all
2240 * we have to do is clean up the accounting
2243 head = btrfs_delayed_node_to_head(node);
2244 if (insert_reserved) {
2245 btrfs_pin_extent(root, node->bytenr,
2246 node->num_bytes, 1);
2247 if (head->is_data) {
2248 ret = btrfs_del_csums(trans, root,
2256 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2257 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2258 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2260 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2261 node->type == BTRFS_SHARED_DATA_REF_KEY)
2262 ret = run_delayed_data_ref(trans, root, node, extent_op,
2269 static noinline struct btrfs_delayed_ref_node *
2270 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2272 struct rb_node *node;
2273 struct btrfs_delayed_ref_node *ref;
2274 int action = BTRFS_ADD_DELAYED_REF;
2277 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2278 * this prevents ref count from going down to zero when
2279 * there still are pending delayed ref.
2281 node = rb_prev(&head->node.rb_node);
2285 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2287 if (ref->bytenr != head->node.bytenr)
2289 if (ref->action == action)
2291 node = rb_prev(node);
2293 if (action == BTRFS_ADD_DELAYED_REF) {
2294 action = BTRFS_DROP_DELAYED_REF;
2301 * Returns 0 on success or if called with an already aborted transaction.
2302 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2304 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2305 struct btrfs_root *root,
2306 struct list_head *cluster)
2308 struct btrfs_delayed_ref_root *delayed_refs;
2309 struct btrfs_delayed_ref_node *ref;
2310 struct btrfs_delayed_ref_head *locked_ref = NULL;
2311 struct btrfs_delayed_extent_op *extent_op;
2312 struct btrfs_fs_info *fs_info = root->fs_info;
2315 int must_insert_reserved = 0;
2317 delayed_refs = &trans->transaction->delayed_refs;
2320 /* pick a new head ref from the cluster list */
2321 if (list_empty(cluster))
2324 locked_ref = list_entry(cluster->next,
2325 struct btrfs_delayed_ref_head, cluster);
2327 /* grab the lock that says we are going to process
2328 * all the refs for this head */
2329 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2332 * we may have dropped the spin lock to get the head
2333 * mutex lock, and that might have given someone else
2334 * time to free the head. If that's true, it has been
2335 * removed from our list and we can move on.
2337 if (ret == -EAGAIN) {
2345 * We need to try and merge add/drops of the same ref since we
2346 * can run into issues with relocate dropping the implicit ref
2347 * and then it being added back again before the drop can
2348 * finish. If we merged anything we need to re-loop so we can
2351 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2355 * locked_ref is the head node, so we have to go one
2356 * node back for any delayed ref updates
2358 ref = select_delayed_ref(locked_ref);
2360 if (ref && ref->seq &&
2361 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2363 * there are still refs with lower seq numbers in the
2364 * process of being added. Don't run this ref yet.
2366 list_del_init(&locked_ref->cluster);
2367 btrfs_delayed_ref_unlock(locked_ref);
2369 delayed_refs->num_heads_ready++;
2370 spin_unlock(&delayed_refs->lock);
2372 spin_lock(&delayed_refs->lock);
2377 * record the must insert reserved flag before we
2378 * drop the spin lock.
2380 must_insert_reserved = locked_ref->must_insert_reserved;
2381 locked_ref->must_insert_reserved = 0;
2383 extent_op = locked_ref->extent_op;
2384 locked_ref->extent_op = NULL;
2387 /* All delayed refs have been processed, Go ahead
2388 * and send the head node to run_one_delayed_ref,
2389 * so that any accounting fixes can happen
2391 ref = &locked_ref->node;
2393 if (extent_op && must_insert_reserved) {
2394 btrfs_free_delayed_extent_op(extent_op);
2399 spin_unlock(&delayed_refs->lock);
2401 ret = run_delayed_extent_op(trans, root,
2403 btrfs_free_delayed_extent_op(extent_op);
2406 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2407 spin_lock(&delayed_refs->lock);
2408 btrfs_delayed_ref_unlock(locked_ref);
2417 rb_erase(&ref->rb_node, &delayed_refs->root);
2418 delayed_refs->num_entries--;
2419 if (!btrfs_delayed_ref_is_head(ref)) {
2421 * when we play the delayed ref, also correct the
2424 switch (ref->action) {
2425 case BTRFS_ADD_DELAYED_REF:
2426 case BTRFS_ADD_DELAYED_EXTENT:
2427 locked_ref->node.ref_mod -= ref->ref_mod;
2429 case BTRFS_DROP_DELAYED_REF:
2430 locked_ref->node.ref_mod += ref->ref_mod;
2436 spin_unlock(&delayed_refs->lock);
2438 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2439 must_insert_reserved);
2441 btrfs_free_delayed_extent_op(extent_op);
2443 btrfs_delayed_ref_unlock(locked_ref);
2444 btrfs_put_delayed_ref(ref);
2445 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2446 spin_lock(&delayed_refs->lock);
2451 * If this node is a head, that means all the refs in this head
2452 * have been dealt with, and we will pick the next head to deal
2453 * with, so we must unlock the head and drop it from the cluster
2454 * list before we release it.
2456 if (btrfs_delayed_ref_is_head(ref)) {
2457 list_del_init(&locked_ref->cluster);
2458 btrfs_delayed_ref_unlock(locked_ref);
2461 btrfs_put_delayed_ref(ref);
2465 spin_lock(&delayed_refs->lock);
2470 #ifdef SCRAMBLE_DELAYED_REFS
2472 * Normally delayed refs get processed in ascending bytenr order. This
2473 * correlates in most cases to the order added. To expose dependencies on this
2474 * order, we start to process the tree in the middle instead of the beginning
2476 static u64 find_middle(struct rb_root *root)
2478 struct rb_node *n = root->rb_node;
2479 struct btrfs_delayed_ref_node *entry;
2482 u64 first = 0, last = 0;
2486 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2487 first = entry->bytenr;
2491 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2492 last = entry->bytenr;
2497 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2498 WARN_ON(!entry->in_tree);
2500 middle = entry->bytenr;
2513 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2514 struct btrfs_fs_info *fs_info)
2516 struct qgroup_update *qgroup_update;
2519 if (list_empty(&trans->qgroup_ref_list) !=
2520 !trans->delayed_ref_elem.seq) {
2521 /* list without seq or seq without list */
2523 "qgroup accounting update error, list is%s empty, seq is %llu",
2524 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2525 trans->delayed_ref_elem.seq);
2529 if (!trans->delayed_ref_elem.seq)
2532 while (!list_empty(&trans->qgroup_ref_list)) {
2533 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2534 struct qgroup_update, list);
2535 list_del(&qgroup_update->list);
2537 ret = btrfs_qgroup_account_ref(
2538 trans, fs_info, qgroup_update->node,
2539 qgroup_update->extent_op);
2540 kfree(qgroup_update);
2543 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2548 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2551 int val = atomic_read(&delayed_refs->ref_seq);
2553 if (val < seq || val >= seq + count)
2559 * this starts processing the delayed reference count updates and
2560 * extent insertions we have queued up so far. count can be
2561 * 0, which means to process everything in the tree at the start
2562 * of the run (but not newly added entries), or it can be some target
2563 * number you'd like to process.
2565 * Returns 0 on success or if called with an aborted transaction
2566 * Returns <0 on error and aborts the transaction
2568 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2569 struct btrfs_root *root, unsigned long count)
2571 struct rb_node *node;
2572 struct btrfs_delayed_ref_root *delayed_refs;
2573 struct btrfs_delayed_ref_node *ref;
2574 struct list_head cluster;
2577 int run_all = count == (unsigned long)-1;
2581 /* We'll clean this up in btrfs_cleanup_transaction */
2585 if (root == root->fs_info->extent_root)
2586 root = root->fs_info->tree_root;
2588 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2590 delayed_refs = &trans->transaction->delayed_refs;
2591 INIT_LIST_HEAD(&cluster);
2593 count = delayed_refs->num_entries * 2;
2597 if (!run_all && !run_most) {
2599 int seq = atomic_read(&delayed_refs->ref_seq);
2602 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2604 DEFINE_WAIT(__wait);
2605 if (delayed_refs->num_entries < 16348)
2608 prepare_to_wait(&delayed_refs->wait, &__wait,
2609 TASK_UNINTERRUPTIBLE);
2611 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2614 finish_wait(&delayed_refs->wait, &__wait);
2616 if (!refs_newer(delayed_refs, seq, 256))
2621 finish_wait(&delayed_refs->wait, &__wait);
2627 atomic_inc(&delayed_refs->procs_running_refs);
2632 spin_lock(&delayed_refs->lock);
2634 #ifdef SCRAMBLE_DELAYED_REFS
2635 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2639 if (!(run_all || run_most) &&
2640 delayed_refs->num_heads_ready < 64)
2644 * go find something we can process in the rbtree. We start at
2645 * the beginning of the tree, and then build a cluster
2646 * of refs to process starting at the first one we are able to
2649 delayed_start = delayed_refs->run_delayed_start;
2650 ret = btrfs_find_ref_cluster(trans, &cluster,
2651 delayed_refs->run_delayed_start);
2655 ret = run_clustered_refs(trans, root, &cluster);
2657 btrfs_release_ref_cluster(&cluster);
2658 spin_unlock(&delayed_refs->lock);
2659 btrfs_abort_transaction(trans, root, ret);
2660 atomic_dec(&delayed_refs->procs_running_refs);
2664 atomic_add(ret, &delayed_refs->ref_seq);
2666 count -= min_t(unsigned long, ret, count);
2671 if (delayed_start >= delayed_refs->run_delayed_start) {
2674 * btrfs_find_ref_cluster looped. let's do one
2675 * more cycle. if we don't run any delayed ref
2676 * during that cycle (because we can't because
2677 * all of them are blocked), bail out.
2682 * no runnable refs left, stop trying
2689 /* refs were run, let's reset staleness detection */
2695 if (!list_empty(&trans->new_bgs)) {
2696 spin_unlock(&delayed_refs->lock);
2697 btrfs_create_pending_block_groups(trans, root);
2698 spin_lock(&delayed_refs->lock);
2701 node = rb_first(&delayed_refs->root);
2704 count = (unsigned long)-1;
2707 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2709 if (btrfs_delayed_ref_is_head(ref)) {
2710 struct btrfs_delayed_ref_head *head;
2712 head = btrfs_delayed_node_to_head(ref);
2713 atomic_inc(&ref->refs);
2715 spin_unlock(&delayed_refs->lock);
2717 * Mutex was contended, block until it's
2718 * released and try again
2720 mutex_lock(&head->mutex);
2721 mutex_unlock(&head->mutex);
2723 btrfs_put_delayed_ref(ref);
2727 node = rb_next(node);
2729 spin_unlock(&delayed_refs->lock);
2730 schedule_timeout(1);
2734 atomic_dec(&delayed_refs->procs_running_refs);
2736 if (waitqueue_active(&delayed_refs->wait))
2737 wake_up(&delayed_refs->wait);
2739 spin_unlock(&delayed_refs->lock);
2740 assert_qgroups_uptodate(trans);
2744 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2745 struct btrfs_root *root,
2746 u64 bytenr, u64 num_bytes, u64 flags,
2749 struct btrfs_delayed_extent_op *extent_op;
2752 extent_op = btrfs_alloc_delayed_extent_op();
2756 extent_op->flags_to_set = flags;
2757 extent_op->update_flags = 1;
2758 extent_op->update_key = 0;
2759 extent_op->is_data = is_data ? 1 : 0;
2761 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2762 num_bytes, extent_op);
2764 btrfs_free_delayed_extent_op(extent_op);
2768 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2769 struct btrfs_root *root,
2770 struct btrfs_path *path,
2771 u64 objectid, u64 offset, u64 bytenr)
2773 struct btrfs_delayed_ref_head *head;
2774 struct btrfs_delayed_ref_node *ref;
2775 struct btrfs_delayed_data_ref *data_ref;
2776 struct btrfs_delayed_ref_root *delayed_refs;
2777 struct rb_node *node;
2781 delayed_refs = &trans->transaction->delayed_refs;
2782 spin_lock(&delayed_refs->lock);
2783 head = btrfs_find_delayed_ref_head(trans, bytenr);
2787 if (!mutex_trylock(&head->mutex)) {
2788 atomic_inc(&head->node.refs);
2789 spin_unlock(&delayed_refs->lock);
2791 btrfs_release_path(path);
2794 * Mutex was contended, block until it's released and let
2797 mutex_lock(&head->mutex);
2798 mutex_unlock(&head->mutex);
2799 btrfs_put_delayed_ref(&head->node);
2803 node = rb_prev(&head->node.rb_node);
2807 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2809 if (ref->bytenr != bytenr)
2813 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2816 data_ref = btrfs_delayed_node_to_data_ref(ref);
2818 node = rb_prev(node);
2822 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2823 if (ref->bytenr == bytenr && ref->seq == seq)
2827 if (data_ref->root != root->root_key.objectid ||
2828 data_ref->objectid != objectid || data_ref->offset != offset)
2833 mutex_unlock(&head->mutex);
2835 spin_unlock(&delayed_refs->lock);
2839 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2840 struct btrfs_root *root,
2841 struct btrfs_path *path,
2842 u64 objectid, u64 offset, u64 bytenr)
2844 struct btrfs_root *extent_root = root->fs_info->extent_root;
2845 struct extent_buffer *leaf;
2846 struct btrfs_extent_data_ref *ref;
2847 struct btrfs_extent_inline_ref *iref;
2848 struct btrfs_extent_item *ei;
2849 struct btrfs_key key;
2853 key.objectid = bytenr;
2854 key.offset = (u64)-1;
2855 key.type = BTRFS_EXTENT_ITEM_KEY;
2857 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2860 BUG_ON(ret == 0); /* Corruption */
2863 if (path->slots[0] == 0)
2867 leaf = path->nodes[0];
2868 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2870 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2874 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2875 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2876 if (item_size < sizeof(*ei)) {
2877 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2881 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2883 if (item_size != sizeof(*ei) +
2884 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2887 if (btrfs_extent_generation(leaf, ei) <=
2888 btrfs_root_last_snapshot(&root->root_item))
2891 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2892 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2893 BTRFS_EXTENT_DATA_REF_KEY)
2896 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2897 if (btrfs_extent_refs(leaf, ei) !=
2898 btrfs_extent_data_ref_count(leaf, ref) ||
2899 btrfs_extent_data_ref_root(leaf, ref) !=
2900 root->root_key.objectid ||
2901 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2902 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2910 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2911 struct btrfs_root *root,
2912 u64 objectid, u64 offset, u64 bytenr)
2914 struct btrfs_path *path;
2918 path = btrfs_alloc_path();
2923 ret = check_committed_ref(trans, root, path, objectid,
2925 if (ret && ret != -ENOENT)
2928 ret2 = check_delayed_ref(trans, root, path, objectid,
2930 } while (ret2 == -EAGAIN);
2932 if (ret2 && ret2 != -ENOENT) {
2937 if (ret != -ENOENT || ret2 != -ENOENT)
2940 btrfs_free_path(path);
2941 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2946 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2947 struct btrfs_root *root,
2948 struct extent_buffer *buf,
2949 int full_backref, int inc, int for_cow)
2956 struct btrfs_key key;
2957 struct btrfs_file_extent_item *fi;
2961 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2962 u64, u64, u64, u64, u64, u64, int);
2964 ref_root = btrfs_header_owner(buf);
2965 nritems = btrfs_header_nritems(buf);
2966 level = btrfs_header_level(buf);
2968 if (!root->ref_cows && level == 0)
2972 process_func = btrfs_inc_extent_ref;
2974 process_func = btrfs_free_extent;
2977 parent = buf->start;
2981 for (i = 0; i < nritems; i++) {
2983 btrfs_item_key_to_cpu(buf, &key, i);
2984 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2986 fi = btrfs_item_ptr(buf, i,
2987 struct btrfs_file_extent_item);
2988 if (btrfs_file_extent_type(buf, fi) ==
2989 BTRFS_FILE_EXTENT_INLINE)
2991 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2995 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2996 key.offset -= btrfs_file_extent_offset(buf, fi);
2997 ret = process_func(trans, root, bytenr, num_bytes,
2998 parent, ref_root, key.objectid,
2999 key.offset, for_cow);
3003 bytenr = btrfs_node_blockptr(buf, i);
3004 num_bytes = btrfs_level_size(root, level - 1);
3005 ret = process_func(trans, root, bytenr, num_bytes,
3006 parent, ref_root, level - 1, 0,
3017 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3018 struct extent_buffer *buf, int full_backref, int for_cow)
3020 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3023 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3024 struct extent_buffer *buf, int full_backref, int for_cow)
3026 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3029 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3030 struct btrfs_root *root,
3031 struct btrfs_path *path,
3032 struct btrfs_block_group_cache *cache)
3035 struct btrfs_root *extent_root = root->fs_info->extent_root;
3037 struct extent_buffer *leaf;
3039 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3042 BUG_ON(ret); /* Corruption */
3044 leaf = path->nodes[0];
3045 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3046 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3047 btrfs_mark_buffer_dirty(leaf);
3048 btrfs_release_path(path);
3051 btrfs_abort_transaction(trans, root, ret);
3058 static struct btrfs_block_group_cache *
3059 next_block_group(struct btrfs_root *root,
3060 struct btrfs_block_group_cache *cache)
3062 struct rb_node *node;
3063 spin_lock(&root->fs_info->block_group_cache_lock);
3064 node = rb_next(&cache->cache_node);
3065 btrfs_put_block_group(cache);
3067 cache = rb_entry(node, struct btrfs_block_group_cache,
3069 btrfs_get_block_group(cache);
3072 spin_unlock(&root->fs_info->block_group_cache_lock);
3076 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3077 struct btrfs_trans_handle *trans,
3078 struct btrfs_path *path)
3080 struct btrfs_root *root = block_group->fs_info->tree_root;
3081 struct inode *inode = NULL;
3083 int dcs = BTRFS_DC_ERROR;
3089 * If this block group is smaller than 100 megs don't bother caching the
3092 if (block_group->key.offset < (100 * 1024 * 1024)) {
3093 spin_lock(&block_group->lock);
3094 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3095 spin_unlock(&block_group->lock);
3100 inode = lookup_free_space_inode(root, block_group, path);
3101 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3102 ret = PTR_ERR(inode);
3103 btrfs_release_path(path);
3107 if (IS_ERR(inode)) {
3111 if (block_group->ro)
3114 ret = create_free_space_inode(root, trans, block_group, path);
3120 /* We've already setup this transaction, go ahead and exit */
3121 if (block_group->cache_generation == trans->transid &&
3122 i_size_read(inode)) {
3123 dcs = BTRFS_DC_SETUP;
3128 * We want to set the generation to 0, that way if anything goes wrong
3129 * from here on out we know not to trust this cache when we load up next
3132 BTRFS_I(inode)->generation = 0;
3133 ret = btrfs_update_inode(trans, root, inode);
3136 if (i_size_read(inode) > 0) {
3137 ret = btrfs_truncate_free_space_cache(root, trans, path,
3143 spin_lock(&block_group->lock);
3144 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3145 !btrfs_test_opt(root, SPACE_CACHE)) {
3147 * don't bother trying to write stuff out _if_
3148 * a) we're not cached,
3149 * b) we're with nospace_cache mount option.
3151 dcs = BTRFS_DC_WRITTEN;
3152 spin_unlock(&block_group->lock);
3155 spin_unlock(&block_group->lock);
3158 * Try to preallocate enough space based on how big the block group is.
3159 * Keep in mind this has to include any pinned space which could end up
3160 * taking up quite a bit since it's not folded into the other space
3163 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3168 num_pages *= PAGE_CACHE_SIZE;
3170 ret = btrfs_check_data_free_space(inode, num_pages);
3174 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3175 num_pages, num_pages,
3178 dcs = BTRFS_DC_SETUP;
3179 btrfs_free_reserved_data_space(inode, num_pages);
3184 btrfs_release_path(path);
3186 spin_lock(&block_group->lock);
3187 if (!ret && dcs == BTRFS_DC_SETUP)
3188 block_group->cache_generation = trans->transid;
3189 block_group->disk_cache_state = dcs;
3190 spin_unlock(&block_group->lock);
3195 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3196 struct btrfs_root *root)
3198 struct btrfs_block_group_cache *cache;
3200 struct btrfs_path *path;
3203 path = btrfs_alloc_path();
3209 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3211 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3213 cache = next_block_group(root, cache);
3221 err = cache_save_setup(cache, trans, path);
3222 last = cache->key.objectid + cache->key.offset;
3223 btrfs_put_block_group(cache);
3228 err = btrfs_run_delayed_refs(trans, root,
3230 if (err) /* File system offline */
3234 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3236 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3237 btrfs_put_block_group(cache);
3243 cache = next_block_group(root, cache);
3252 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3253 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3255 last = cache->key.objectid + cache->key.offset;
3257 err = write_one_cache_group(trans, root, path, cache);
3258 if (err) /* File system offline */
3261 btrfs_put_block_group(cache);
3266 * I don't think this is needed since we're just marking our
3267 * preallocated extent as written, but just in case it can't
3271 err = btrfs_run_delayed_refs(trans, root,
3273 if (err) /* File system offline */
3277 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3280 * Really this shouldn't happen, but it could if we
3281 * couldn't write the entire preallocated extent and
3282 * splitting the extent resulted in a new block.
3285 btrfs_put_block_group(cache);
3288 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3290 cache = next_block_group(root, cache);
3299 err = btrfs_write_out_cache(root, trans, cache, path);
3302 * If we didn't have an error then the cache state is still
3303 * NEED_WRITE, so we can set it to WRITTEN.
3305 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3306 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3307 last = cache->key.objectid + cache->key.offset;
3308 btrfs_put_block_group(cache);
3312 btrfs_free_path(path);
3316 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3318 struct btrfs_block_group_cache *block_group;
3321 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3322 if (!block_group || block_group->ro)
3325 btrfs_put_block_group(block_group);
3329 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3330 u64 total_bytes, u64 bytes_used,
3331 struct btrfs_space_info **space_info)
3333 struct btrfs_space_info *found;
3337 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3338 BTRFS_BLOCK_GROUP_RAID10))
3343 found = __find_space_info(info, flags);
3345 spin_lock(&found->lock);
3346 found->total_bytes += total_bytes;
3347 found->disk_total += total_bytes * factor;
3348 found->bytes_used += bytes_used;
3349 found->disk_used += bytes_used * factor;
3351 spin_unlock(&found->lock);
3352 *space_info = found;
3355 found = kzalloc(sizeof(*found), GFP_NOFS);
3359 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3360 INIT_LIST_HEAD(&found->block_groups[i]);
3361 init_rwsem(&found->groups_sem);
3362 spin_lock_init(&found->lock);
3363 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3364 found->total_bytes = total_bytes;
3365 found->disk_total = total_bytes * factor;
3366 found->bytes_used = bytes_used;
3367 found->disk_used = bytes_used * factor;
3368 found->bytes_pinned = 0;
3369 found->bytes_reserved = 0;
3370 found->bytes_readonly = 0;
3371 found->bytes_may_use = 0;
3373 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3374 found->chunk_alloc = 0;
3376 init_waitqueue_head(&found->wait);
3377 *space_info = found;
3378 list_add_rcu(&found->list, &info->space_info);
3379 if (flags & BTRFS_BLOCK_GROUP_DATA)
3380 info->data_sinfo = found;
3384 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3386 u64 extra_flags = chunk_to_extended(flags) &
3387 BTRFS_EXTENDED_PROFILE_MASK;
3389 write_seqlock(&fs_info->profiles_lock);
3390 if (flags & BTRFS_BLOCK_GROUP_DATA)
3391 fs_info->avail_data_alloc_bits |= extra_flags;
3392 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3393 fs_info->avail_metadata_alloc_bits |= extra_flags;
3394 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3395 fs_info->avail_system_alloc_bits |= extra_flags;
3396 write_sequnlock(&fs_info->profiles_lock);
3400 * returns target flags in extended format or 0 if restripe for this
3401 * chunk_type is not in progress
3403 * should be called with either volume_mutex or balance_lock held
3405 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3407 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3413 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3414 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3415 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3416 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3417 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3418 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3419 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3420 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3421 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3428 * @flags: available profiles in extended format (see ctree.h)
3430 * Returns reduced profile in chunk format. If profile changing is in
3431 * progress (either running or paused) picks the target profile (if it's
3432 * already available), otherwise falls back to plain reducing.
3434 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3437 * we add in the count of missing devices because we want
3438 * to make sure that any RAID levels on a degraded FS
3439 * continue to be honored.
3441 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3442 root->fs_info->fs_devices->missing_devices;
3447 * see if restripe for this chunk_type is in progress, if so
3448 * try to reduce to the target profile
3450 spin_lock(&root->fs_info->balance_lock);
3451 target = get_restripe_target(root->fs_info, flags);
3453 /* pick target profile only if it's already available */
3454 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3455 spin_unlock(&root->fs_info->balance_lock);
3456 return extended_to_chunk(target);
3459 spin_unlock(&root->fs_info->balance_lock);
3461 /* First, mask out the RAID levels which aren't possible */
3462 if (num_devices == 1)
3463 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3464 BTRFS_BLOCK_GROUP_RAID5);
3465 if (num_devices < 3)
3466 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3467 if (num_devices < 4)
3468 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3470 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3471 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3472 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3475 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3476 tmp = BTRFS_BLOCK_GROUP_RAID6;
3477 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3478 tmp = BTRFS_BLOCK_GROUP_RAID5;
3479 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3480 tmp = BTRFS_BLOCK_GROUP_RAID10;
3481 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3482 tmp = BTRFS_BLOCK_GROUP_RAID1;
3483 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3484 tmp = BTRFS_BLOCK_GROUP_RAID0;
3486 return extended_to_chunk(flags | tmp);
3489 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3494 seq = read_seqbegin(&root->fs_info->profiles_lock);
3496 if (flags & BTRFS_BLOCK_GROUP_DATA)
3497 flags |= root->fs_info->avail_data_alloc_bits;
3498 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3499 flags |= root->fs_info->avail_system_alloc_bits;
3500 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3501 flags |= root->fs_info->avail_metadata_alloc_bits;
3502 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3504 return btrfs_reduce_alloc_profile(root, flags);
3507 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3513 flags = BTRFS_BLOCK_GROUP_DATA;
3514 else if (root == root->fs_info->chunk_root)
3515 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3517 flags = BTRFS_BLOCK_GROUP_METADATA;
3519 ret = get_alloc_profile(root, flags);
3524 * This will check the space that the inode allocates from to make sure we have
3525 * enough space for bytes.
3527 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3529 struct btrfs_space_info *data_sinfo;
3530 struct btrfs_root *root = BTRFS_I(inode)->root;
3531 struct btrfs_fs_info *fs_info = root->fs_info;
3533 int ret = 0, committed = 0, alloc_chunk = 1;
3535 /* make sure bytes are sectorsize aligned */
3536 bytes = ALIGN(bytes, root->sectorsize);
3538 if (root == root->fs_info->tree_root ||
3539 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3544 data_sinfo = fs_info->data_sinfo;
3549 /* make sure we have enough space to handle the data first */
3550 spin_lock(&data_sinfo->lock);
3551 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3552 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3553 data_sinfo->bytes_may_use;
3555 if (used + bytes > data_sinfo->total_bytes) {
3556 struct btrfs_trans_handle *trans;
3559 * if we don't have enough free bytes in this space then we need
3560 * to alloc a new chunk.
3562 if (!data_sinfo->full && alloc_chunk) {
3565 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3566 spin_unlock(&data_sinfo->lock);
3568 alloc_target = btrfs_get_alloc_profile(root, 1);
3569 trans = btrfs_join_transaction(root);
3571 return PTR_ERR(trans);
3573 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3575 CHUNK_ALLOC_NO_FORCE);
3576 btrfs_end_transaction(trans, root);
3585 data_sinfo = fs_info->data_sinfo;
3591 * If we have less pinned bytes than we want to allocate then
3592 * don't bother committing the transaction, it won't help us.
3594 if (data_sinfo->bytes_pinned < bytes)
3596 spin_unlock(&data_sinfo->lock);
3598 /* commit the current transaction and try again */
3601 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3603 trans = btrfs_join_transaction(root);
3605 return PTR_ERR(trans);
3606 ret = btrfs_commit_transaction(trans, root);
3614 data_sinfo->bytes_may_use += bytes;
3615 trace_btrfs_space_reservation(root->fs_info, "space_info",
3616 data_sinfo->flags, bytes, 1);
3617 spin_unlock(&data_sinfo->lock);
3623 * Called if we need to clear a data reservation for this inode.
3625 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3627 struct btrfs_root *root = BTRFS_I(inode)->root;
3628 struct btrfs_space_info *data_sinfo;
3630 /* make sure bytes are sectorsize aligned */
3631 bytes = ALIGN(bytes, root->sectorsize);
3633 data_sinfo = root->fs_info->data_sinfo;
3634 spin_lock(&data_sinfo->lock);
3635 data_sinfo->bytes_may_use -= bytes;
3636 trace_btrfs_space_reservation(root->fs_info, "space_info",
3637 data_sinfo->flags, bytes, 0);
3638 spin_unlock(&data_sinfo->lock);
3641 static void force_metadata_allocation(struct btrfs_fs_info *info)
3643 struct list_head *head = &info->space_info;
3644 struct btrfs_space_info *found;
3647 list_for_each_entry_rcu(found, head, list) {
3648 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3649 found->force_alloc = CHUNK_ALLOC_FORCE;
3654 static int should_alloc_chunk(struct btrfs_root *root,
3655 struct btrfs_space_info *sinfo, int force)
3657 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3658 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3659 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3662 if (force == CHUNK_ALLOC_FORCE)
3666 * We need to take into account the global rsv because for all intents
3667 * and purposes it's used space. Don't worry about locking the
3668 * global_rsv, it doesn't change except when the transaction commits.
3670 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3671 num_allocated += global_rsv->size;
3674 * in limited mode, we want to have some free space up to
3675 * about 1% of the FS size.
3677 if (force == CHUNK_ALLOC_LIMITED) {
3678 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3679 thresh = max_t(u64, 64 * 1024 * 1024,
3680 div_factor_fine(thresh, 1));
3682 if (num_bytes - num_allocated < thresh)
3686 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3691 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3695 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3696 BTRFS_BLOCK_GROUP_RAID0 |
3697 BTRFS_BLOCK_GROUP_RAID5 |
3698 BTRFS_BLOCK_GROUP_RAID6))
3699 num_dev = root->fs_info->fs_devices->rw_devices;
3700 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3703 num_dev = 1; /* DUP or single */
3705 /* metadata for updaing devices and chunk tree */
3706 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3709 static void check_system_chunk(struct btrfs_trans_handle *trans,
3710 struct btrfs_root *root, u64 type)
3712 struct btrfs_space_info *info;
3716 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3717 spin_lock(&info->lock);
3718 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3719 info->bytes_reserved - info->bytes_readonly;
3720 spin_unlock(&info->lock);
3722 thresh = get_system_chunk_thresh(root, type);
3723 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3724 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3725 left, thresh, type);
3726 dump_space_info(info, 0, 0);
3729 if (left < thresh) {
3732 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3733 btrfs_alloc_chunk(trans, root, flags);
3737 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3738 struct btrfs_root *extent_root, u64 flags, int force)
3740 struct btrfs_space_info *space_info;
3741 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3742 int wait_for_alloc = 0;
3745 /* Don't re-enter if we're already allocating a chunk */
3746 if (trans->allocating_chunk)
3749 space_info = __find_space_info(extent_root->fs_info, flags);
3751 ret = update_space_info(extent_root->fs_info, flags,
3753 BUG_ON(ret); /* -ENOMEM */
3755 BUG_ON(!space_info); /* Logic error */
3758 spin_lock(&space_info->lock);
3759 if (force < space_info->force_alloc)
3760 force = space_info->force_alloc;
3761 if (space_info->full) {
3762 spin_unlock(&space_info->lock);
3766 if (!should_alloc_chunk(extent_root, space_info, force)) {
3767 spin_unlock(&space_info->lock);
3769 } else if (space_info->chunk_alloc) {
3772 space_info->chunk_alloc = 1;
3775 spin_unlock(&space_info->lock);
3777 mutex_lock(&fs_info->chunk_mutex);
3780 * The chunk_mutex is held throughout the entirety of a chunk
3781 * allocation, so once we've acquired the chunk_mutex we know that the
3782 * other guy is done and we need to recheck and see if we should
3785 if (wait_for_alloc) {
3786 mutex_unlock(&fs_info->chunk_mutex);
3791 trans->allocating_chunk = true;
3794 * If we have mixed data/metadata chunks we want to make sure we keep
3795 * allocating mixed chunks instead of individual chunks.
3797 if (btrfs_mixed_space_info(space_info))
3798 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3801 * if we're doing a data chunk, go ahead and make sure that
3802 * we keep a reasonable number of metadata chunks allocated in the
3805 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3806 fs_info->data_chunk_allocations++;
3807 if (!(fs_info->data_chunk_allocations %
3808 fs_info->metadata_ratio))
3809 force_metadata_allocation(fs_info);
3813 * Check if we have enough space in SYSTEM chunk because we may need
3814 * to update devices.
3816 check_system_chunk(trans, extent_root, flags);
3818 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3819 trans->allocating_chunk = false;
3821 spin_lock(&space_info->lock);
3822 if (ret < 0 && ret != -ENOSPC)
3825 space_info->full = 1;
3829 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3831 space_info->chunk_alloc = 0;
3832 spin_unlock(&space_info->lock);
3833 mutex_unlock(&fs_info->chunk_mutex);
3837 static int can_overcommit(struct btrfs_root *root,
3838 struct btrfs_space_info *space_info, u64 bytes,
3839 enum btrfs_reserve_flush_enum flush)
3841 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3842 u64 profile = btrfs_get_alloc_profile(root, 0);
3848 used = space_info->bytes_used + space_info->bytes_reserved +
3849 space_info->bytes_pinned + space_info->bytes_readonly;
3851 spin_lock(&global_rsv->lock);
3852 rsv_size = global_rsv->size;
3853 spin_unlock(&global_rsv->lock);
3856 * We only want to allow over committing if we have lots of actual space
3857 * free, but if we don't have enough space to handle the global reserve
3858 * space then we could end up having a real enospc problem when trying
3859 * to allocate a chunk or some other such important allocation.
3862 if (used + rsv_size >= space_info->total_bytes)
3865 used += space_info->bytes_may_use;
3867 spin_lock(&root->fs_info->free_chunk_lock);
3868 avail = root->fs_info->free_chunk_space;
3869 spin_unlock(&root->fs_info->free_chunk_lock);
3872 * If we have dup, raid1 or raid10 then only half of the free
3873 * space is actually useable. For raid56, the space info used
3874 * doesn't include the parity drive, so we don't have to
3877 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3878 BTRFS_BLOCK_GROUP_RAID1 |
3879 BTRFS_BLOCK_GROUP_RAID10))
3882 to_add = space_info->total_bytes;
3885 * If we aren't flushing all things, let us overcommit up to
3886 * 1/2th of the space. If we can flush, don't let us overcommit
3887 * too much, let it overcommit up to 1/8 of the space.
3889 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3895 * Limit the overcommit to the amount of free space we could possibly
3896 * allocate for chunks.
3898 to_add = min(avail, to_add);
3900 if (used + bytes < space_info->total_bytes + to_add)
3905 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3906 unsigned long nr_pages)
3908 struct super_block *sb = root->fs_info->sb;
3911 /* If we can not start writeback, just sync all the delalloc file. */
3912 started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
3913 WB_REASON_FS_FREE_SPACE);
3916 * We needn't worry the filesystem going from r/w to r/o though
3917 * we don't acquire ->s_umount mutex, because the filesystem
3918 * should guarantee the delalloc inodes list be empty after
3919 * the filesystem is readonly(all dirty pages are written to
3922 btrfs_start_delalloc_inodes(root, 0);
3923 btrfs_wait_ordered_extents(root, 0);
3928 * shrink metadata reservation for delalloc
3930 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3933 struct btrfs_block_rsv *block_rsv;
3934 struct btrfs_space_info *space_info;
3935 struct btrfs_trans_handle *trans;
3939 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3941 enum btrfs_reserve_flush_enum flush;
3943 trans = (struct btrfs_trans_handle *)current->journal_info;
3944 block_rsv = &root->fs_info->delalloc_block_rsv;
3945 space_info = block_rsv->space_info;
3948 delalloc_bytes = percpu_counter_sum_positive(
3949 &root->fs_info->delalloc_bytes);
3950 if (delalloc_bytes == 0) {
3953 btrfs_wait_ordered_extents(root, 0);
3957 while (delalloc_bytes && loops < 3) {
3958 max_reclaim = min(delalloc_bytes, to_reclaim);
3959 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3960 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3962 * We need to wait for the async pages to actually start before
3965 wait_event(root->fs_info->async_submit_wait,
3966 !atomic_read(&root->fs_info->async_delalloc_pages));
3969 flush = BTRFS_RESERVE_FLUSH_ALL;
3971 flush = BTRFS_RESERVE_NO_FLUSH;
3972 spin_lock(&space_info->lock);
3973 if (can_overcommit(root, space_info, orig, flush)) {
3974 spin_unlock(&space_info->lock);
3977 spin_unlock(&space_info->lock);
3980 if (wait_ordered && !trans) {
3981 btrfs_wait_ordered_extents(root, 0);
3983 time_left = schedule_timeout_killable(1);
3988 delalloc_bytes = percpu_counter_sum_positive(
3989 &root->fs_info->delalloc_bytes);
3994 * maybe_commit_transaction - possibly commit the transaction if its ok to
3995 * @root - the root we're allocating for
3996 * @bytes - the number of bytes we want to reserve
3997 * @force - force the commit
3999 * This will check to make sure that committing the transaction will actually
4000 * get us somewhere and then commit the transaction if it does. Otherwise it
4001 * will return -ENOSPC.
4003 static int may_commit_transaction(struct btrfs_root *root,
4004 struct btrfs_space_info *space_info,
4005 u64 bytes, int force)
4007 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4008 struct btrfs_trans_handle *trans;
4010 trans = (struct btrfs_trans_handle *)current->journal_info;
4017 /* See if there is enough pinned space to make this reservation */
4018 spin_lock(&space_info->lock);
4019 if (space_info->bytes_pinned >= bytes) {
4020 spin_unlock(&space_info->lock);
4023 spin_unlock(&space_info->lock);
4026 * See if there is some space in the delayed insertion reservation for
4029 if (space_info != delayed_rsv->space_info)
4032 spin_lock(&space_info->lock);
4033 spin_lock(&delayed_rsv->lock);
4034 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
4035 spin_unlock(&delayed_rsv->lock);
4036 spin_unlock(&space_info->lock);
4039 spin_unlock(&delayed_rsv->lock);
4040 spin_unlock(&space_info->lock);
4043 trans = btrfs_join_transaction(root);
4047 return btrfs_commit_transaction(trans, root);
4051 FLUSH_DELAYED_ITEMS_NR = 1,
4052 FLUSH_DELAYED_ITEMS = 2,
4054 FLUSH_DELALLOC_WAIT = 4,
4059 static int flush_space(struct btrfs_root *root,
4060 struct btrfs_space_info *space_info, u64 num_bytes,
4061 u64 orig_bytes, int state)
4063 struct btrfs_trans_handle *trans;
4068 case FLUSH_DELAYED_ITEMS_NR:
4069 case FLUSH_DELAYED_ITEMS:
4070 if (state == FLUSH_DELAYED_ITEMS_NR) {
4071 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4073 nr = (int)div64_u64(num_bytes, bytes);
4080 trans = btrfs_join_transaction(root);
4081 if (IS_ERR(trans)) {
4082 ret = PTR_ERR(trans);
4085 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4086 btrfs_end_transaction(trans, root);
4088 case FLUSH_DELALLOC:
4089 case FLUSH_DELALLOC_WAIT:
4090 shrink_delalloc(root, num_bytes, orig_bytes,
4091 state == FLUSH_DELALLOC_WAIT);
4094 trans = btrfs_join_transaction(root);
4095 if (IS_ERR(trans)) {
4096 ret = PTR_ERR(trans);
4099 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4100 btrfs_get_alloc_profile(root, 0),
4101 CHUNK_ALLOC_NO_FORCE);
4102 btrfs_end_transaction(trans, root);
4107 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4117 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4118 * @root - the root we're allocating for
4119 * @block_rsv - the block_rsv we're allocating for
4120 * @orig_bytes - the number of bytes we want
4121 * @flush - whether or not we can flush to make our reservation
4123 * This will reserve orgi_bytes number of bytes from the space info associated
4124 * with the block_rsv. If there is not enough space it will make an attempt to
4125 * flush out space to make room. It will do this by flushing delalloc if
4126 * possible or committing the transaction. If flush is 0 then no attempts to
4127 * regain reservations will be made and this will fail if there is not enough
4130 static int reserve_metadata_bytes(struct btrfs_root *root,
4131 struct btrfs_block_rsv *block_rsv,
4133 enum btrfs_reserve_flush_enum flush)
4135 struct btrfs_space_info *space_info = block_rsv->space_info;
4137 u64 num_bytes = orig_bytes;
4138 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4140 bool flushing = false;
4144 spin_lock(&space_info->lock);
4146 * We only want to wait if somebody other than us is flushing and we
4147 * are actually allowed to flush all things.
4149 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4150 space_info->flush) {
4151 spin_unlock(&space_info->lock);
4153 * If we have a trans handle we can't wait because the flusher
4154 * may have to commit the transaction, which would mean we would
4155 * deadlock since we are waiting for the flusher to finish, but
4156 * hold the current transaction open.
4158 if (current->journal_info)
4160 ret = wait_event_killable(space_info->wait, !space_info->flush);
4161 /* Must have been killed, return */
4165 spin_lock(&space_info->lock);
4169 used = space_info->bytes_used + space_info->bytes_reserved +
4170 space_info->bytes_pinned + space_info->bytes_readonly +
4171 space_info->bytes_may_use;
4174 * The idea here is that we've not already over-reserved the block group
4175 * then we can go ahead and save our reservation first and then start
4176 * flushing if we need to. Otherwise if we've already overcommitted
4177 * lets start flushing stuff first and then come back and try to make
4180 if (used <= space_info->total_bytes) {
4181 if (used + orig_bytes <= space_info->total_bytes) {
4182 space_info->bytes_may_use += orig_bytes;
4183 trace_btrfs_space_reservation(root->fs_info,
4184 "space_info", space_info->flags, orig_bytes, 1);
4188 * Ok set num_bytes to orig_bytes since we aren't
4189 * overocmmitted, this way we only try and reclaim what
4192 num_bytes = orig_bytes;
4196 * Ok we're over committed, set num_bytes to the overcommitted
4197 * amount plus the amount of bytes that we need for this
4200 num_bytes = used - space_info->total_bytes +
4204 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4205 space_info->bytes_may_use += orig_bytes;
4206 trace_btrfs_space_reservation(root->fs_info, "space_info",
4207 space_info->flags, orig_bytes,
4213 * Couldn't make our reservation, save our place so while we're trying
4214 * to reclaim space we can actually use it instead of somebody else
4215 * stealing it from us.
4217 * We make the other tasks wait for the flush only when we can flush
4220 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4222 space_info->flush = 1;
4225 spin_unlock(&space_info->lock);
4227 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4230 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4235 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4236 * would happen. So skip delalloc flush.
4238 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4239 (flush_state == FLUSH_DELALLOC ||
4240 flush_state == FLUSH_DELALLOC_WAIT))
4241 flush_state = ALLOC_CHUNK;
4245 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4246 flush_state < COMMIT_TRANS)
4248 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4249 flush_state <= COMMIT_TRANS)
4253 if (ret == -ENOSPC &&
4254 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4255 struct btrfs_block_rsv *global_rsv =
4256 &root->fs_info->global_block_rsv;
4258 if (block_rsv != global_rsv &&
4259 !block_rsv_use_bytes(global_rsv, orig_bytes))
4263 spin_lock(&space_info->lock);
4264 space_info->flush = 0;
4265 wake_up_all(&space_info->wait);
4266 spin_unlock(&space_info->lock);
4271 static struct btrfs_block_rsv *get_block_rsv(
4272 const struct btrfs_trans_handle *trans,
4273 const struct btrfs_root *root)
4275 struct btrfs_block_rsv *block_rsv = NULL;
4278 block_rsv = trans->block_rsv;
4280 if (root == root->fs_info->csum_root && trans->adding_csums)
4281 block_rsv = trans->block_rsv;
4284 block_rsv = root->block_rsv;
4287 block_rsv = &root->fs_info->empty_block_rsv;
4292 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4296 spin_lock(&block_rsv->lock);
4297 if (block_rsv->reserved >= num_bytes) {
4298 block_rsv->reserved -= num_bytes;
4299 if (block_rsv->reserved < block_rsv->size)
4300 block_rsv->full = 0;
4303 spin_unlock(&block_rsv->lock);
4307 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4308 u64 num_bytes, int update_size)
4310 spin_lock(&block_rsv->lock);
4311 block_rsv->reserved += num_bytes;
4313 block_rsv->size += num_bytes;
4314 else if (block_rsv->reserved >= block_rsv->size)
4315 block_rsv->full = 1;
4316 spin_unlock(&block_rsv->lock);
4319 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4320 struct btrfs_block_rsv *block_rsv,
4321 struct btrfs_block_rsv *dest, u64 num_bytes)
4323 struct btrfs_space_info *space_info = block_rsv->space_info;
4325 spin_lock(&block_rsv->lock);
4326 if (num_bytes == (u64)-1)
4327 num_bytes = block_rsv->size;
4328 block_rsv->size -= num_bytes;
4329 if (block_rsv->reserved >= block_rsv->size) {
4330 num_bytes = block_rsv->reserved - block_rsv->size;
4331 block_rsv->reserved = block_rsv->size;
4332 block_rsv->full = 1;
4336 spin_unlock(&block_rsv->lock);
4338 if (num_bytes > 0) {
4340 spin_lock(&dest->lock);
4344 bytes_to_add = dest->size - dest->reserved;
4345 bytes_to_add = min(num_bytes, bytes_to_add);
4346 dest->reserved += bytes_to_add;
4347 if (dest->reserved >= dest->size)
4349 num_bytes -= bytes_to_add;
4351 spin_unlock(&dest->lock);
4354 spin_lock(&space_info->lock);
4355 space_info->bytes_may_use -= num_bytes;
4356 trace_btrfs_space_reservation(fs_info, "space_info",
4357 space_info->flags, num_bytes, 0);
4358 space_info->reservation_progress++;
4359 spin_unlock(&space_info->lock);
4364 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4365 struct btrfs_block_rsv *dst, u64 num_bytes)
4369 ret = block_rsv_use_bytes(src, num_bytes);
4373 block_rsv_add_bytes(dst, num_bytes, 1);
4377 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4379 memset(rsv, 0, sizeof(*rsv));
4380 spin_lock_init(&rsv->lock);
4384 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4385 unsigned short type)
4387 struct btrfs_block_rsv *block_rsv;
4388 struct btrfs_fs_info *fs_info = root->fs_info;
4390 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4394 btrfs_init_block_rsv(block_rsv, type);
4395 block_rsv->space_info = __find_space_info(fs_info,
4396 BTRFS_BLOCK_GROUP_METADATA);
4400 void btrfs_free_block_rsv(struct btrfs_root *root,
4401 struct btrfs_block_rsv *rsv)
4405 btrfs_block_rsv_release(root, rsv, (u64)-1);
4409 int btrfs_block_rsv_add(struct btrfs_root *root,
4410 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4411 enum btrfs_reserve_flush_enum flush)
4418 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4420 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4427 int btrfs_block_rsv_check(struct btrfs_root *root,
4428 struct btrfs_block_rsv *block_rsv, int min_factor)
4436 spin_lock(&block_rsv->lock);
4437 num_bytes = div_factor(block_rsv->size, min_factor);
4438 if (block_rsv->reserved >= num_bytes)
4440 spin_unlock(&block_rsv->lock);
4445 int btrfs_block_rsv_refill(struct btrfs_root *root,
4446 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4447 enum btrfs_reserve_flush_enum flush)
4455 spin_lock(&block_rsv->lock);
4456 num_bytes = min_reserved;
4457 if (block_rsv->reserved >= num_bytes)
4460 num_bytes -= block_rsv->reserved;
4461 spin_unlock(&block_rsv->lock);
4466 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4468 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4475 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4476 struct btrfs_block_rsv *dst_rsv,
4479 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4482 void btrfs_block_rsv_release(struct btrfs_root *root,
4483 struct btrfs_block_rsv *block_rsv,
4486 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4487 if (global_rsv->full || global_rsv == block_rsv ||
4488 block_rsv->space_info != global_rsv->space_info)
4490 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4495 * helper to calculate size of global block reservation.
4496 * the desired value is sum of space used by extent tree,
4497 * checksum tree and root tree
4499 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4501 struct btrfs_space_info *sinfo;
4505 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4507 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4508 spin_lock(&sinfo->lock);
4509 data_used = sinfo->bytes_used;
4510 spin_unlock(&sinfo->lock);
4512 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4513 spin_lock(&sinfo->lock);
4514 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4516 meta_used = sinfo->bytes_used;
4517 spin_unlock(&sinfo->lock);
4519 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4521 num_bytes += div64_u64(data_used + meta_used, 50);
4523 if (num_bytes * 3 > meta_used)
4524 num_bytes = div64_u64(meta_used, 3);
4526 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4529 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4531 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4532 struct btrfs_space_info *sinfo = block_rsv->space_info;
4535 num_bytes = calc_global_metadata_size(fs_info);
4537 spin_lock(&sinfo->lock);
4538 spin_lock(&block_rsv->lock);
4540 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4542 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4543 sinfo->bytes_reserved + sinfo->bytes_readonly +
4544 sinfo->bytes_may_use;
4546 if (sinfo->total_bytes > num_bytes) {
4547 num_bytes = sinfo->total_bytes - num_bytes;
4548 block_rsv->reserved += num_bytes;
4549 sinfo->bytes_may_use += num_bytes;
4550 trace_btrfs_space_reservation(fs_info, "space_info",
4551 sinfo->flags, num_bytes, 1);
4554 if (block_rsv->reserved >= block_rsv->size) {
4555 num_bytes = block_rsv->reserved - block_rsv->size;
4556 sinfo->bytes_may_use -= num_bytes;
4557 trace_btrfs_space_reservation(fs_info, "space_info",
4558 sinfo->flags, num_bytes, 0);
4559 sinfo->reservation_progress++;
4560 block_rsv->reserved = block_rsv->size;
4561 block_rsv->full = 1;
4564 spin_unlock(&block_rsv->lock);
4565 spin_unlock(&sinfo->lock);
4568 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4570 struct btrfs_space_info *space_info;
4572 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4573 fs_info->chunk_block_rsv.space_info = space_info;
4575 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4576 fs_info->global_block_rsv.space_info = space_info;
4577 fs_info->delalloc_block_rsv.space_info = space_info;
4578 fs_info->trans_block_rsv.space_info = space_info;
4579 fs_info->empty_block_rsv.space_info = space_info;
4580 fs_info->delayed_block_rsv.space_info = space_info;
4582 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4583 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4584 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4585 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4586 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4588 update_global_block_rsv(fs_info);
4591 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4593 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4595 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4596 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4597 WARN_ON(fs_info->trans_block_rsv.size > 0);
4598 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4599 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4600 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4601 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4602 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4605 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4606 struct btrfs_root *root)
4608 if (!trans->block_rsv)
4611 if (!trans->bytes_reserved)
4614 trace_btrfs_space_reservation(root->fs_info, "transaction",
4615 trans->transid, trans->bytes_reserved, 0);
4616 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4617 trans->bytes_reserved = 0;
4620 /* Can only return 0 or -ENOSPC */
4621 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4622 struct inode *inode)
4624 struct btrfs_root *root = BTRFS_I(inode)->root;
4625 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4626 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4629 * We need to hold space in order to delete our orphan item once we've
4630 * added it, so this takes the reservation so we can release it later
4631 * when we are truly done with the orphan item.
4633 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4634 trace_btrfs_space_reservation(root->fs_info, "orphan",
4635 btrfs_ino(inode), num_bytes, 1);
4636 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4639 void btrfs_orphan_release_metadata(struct inode *inode)
4641 struct btrfs_root *root = BTRFS_I(inode)->root;
4642 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4643 trace_btrfs_space_reservation(root->fs_info, "orphan",
4644 btrfs_ino(inode), num_bytes, 0);
4645 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4649 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4650 * root: the root of the parent directory
4651 * rsv: block reservation
4652 * items: the number of items that we need do reservation
4653 * qgroup_reserved: used to return the reserved size in qgroup
4655 * This function is used to reserve the space for snapshot/subvolume
4656 * creation and deletion. Those operations are different with the
4657 * common file/directory operations, they change two fs/file trees
4658 * and root tree, the number of items that the qgroup reserves is
4659 * different with the free space reservation. So we can not use
4660 * the space reseravtion mechanism in start_transaction().
4662 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4663 struct btrfs_block_rsv *rsv,
4665 u64 *qgroup_reserved)
4670 if (root->fs_info->quota_enabled) {
4671 /* One for parent inode, two for dir entries */
4672 num_bytes = 3 * root->leafsize;
4673 ret = btrfs_qgroup_reserve(root, num_bytes);
4680 *qgroup_reserved = num_bytes;
4682 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4683 rsv->space_info = __find_space_info(root->fs_info,
4684 BTRFS_BLOCK_GROUP_METADATA);
4685 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4686 BTRFS_RESERVE_FLUSH_ALL);
4688 if (*qgroup_reserved)
4689 btrfs_qgroup_free(root, *qgroup_reserved);
4695 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4696 struct btrfs_block_rsv *rsv,
4697 u64 qgroup_reserved)
4699 btrfs_block_rsv_release(root, rsv, (u64)-1);
4700 if (qgroup_reserved)
4701 btrfs_qgroup_free(root, qgroup_reserved);
4705 * drop_outstanding_extent - drop an outstanding extent
4706 * @inode: the inode we're dropping the extent for
4708 * This is called when we are freeing up an outstanding extent, either called
4709 * after an error or after an extent is written. This will return the number of
4710 * reserved extents that need to be freed. This must be called with
4711 * BTRFS_I(inode)->lock held.
4713 static unsigned drop_outstanding_extent(struct inode *inode)
4715 unsigned drop_inode_space = 0;
4716 unsigned dropped_extents = 0;
4718 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4719 BTRFS_I(inode)->outstanding_extents--;
4721 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4722 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4723 &BTRFS_I(inode)->runtime_flags))
4724 drop_inode_space = 1;
4727 * If we have more or the same amount of outsanding extents than we have
4728 * reserved then we need to leave the reserved extents count alone.
4730 if (BTRFS_I(inode)->outstanding_extents >=
4731 BTRFS_I(inode)->reserved_extents)
4732 return drop_inode_space;
4734 dropped_extents = BTRFS_I(inode)->reserved_extents -
4735 BTRFS_I(inode)->outstanding_extents;
4736 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4737 return dropped_extents + drop_inode_space;
4741 * calc_csum_metadata_size - return the amount of metada space that must be
4742 * reserved/free'd for the given bytes.
4743 * @inode: the inode we're manipulating
4744 * @num_bytes: the number of bytes in question
4745 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4747 * This adjusts the number of csum_bytes in the inode and then returns the
4748 * correct amount of metadata that must either be reserved or freed. We
4749 * calculate how many checksums we can fit into one leaf and then divide the
4750 * number of bytes that will need to be checksumed by this value to figure out
4751 * how many checksums will be required. If we are adding bytes then the number
4752 * may go up and we will return the number of additional bytes that must be
4753 * reserved. If it is going down we will return the number of bytes that must
4756 * This must be called with BTRFS_I(inode)->lock held.
4758 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4761 struct btrfs_root *root = BTRFS_I(inode)->root;
4763 int num_csums_per_leaf;
4767 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4768 BTRFS_I(inode)->csum_bytes == 0)
4771 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4773 BTRFS_I(inode)->csum_bytes += num_bytes;
4775 BTRFS_I(inode)->csum_bytes -= num_bytes;
4776 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4777 num_csums_per_leaf = (int)div64_u64(csum_size,
4778 sizeof(struct btrfs_csum_item) +
4779 sizeof(struct btrfs_disk_key));
4780 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4781 num_csums = num_csums + num_csums_per_leaf - 1;
4782 num_csums = num_csums / num_csums_per_leaf;
4784 old_csums = old_csums + num_csums_per_leaf - 1;
4785 old_csums = old_csums / num_csums_per_leaf;
4787 /* No change, no need to reserve more */
4788 if (old_csums == num_csums)
4792 return btrfs_calc_trans_metadata_size(root,
4793 num_csums - old_csums);
4795 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4798 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4800 struct btrfs_root *root = BTRFS_I(inode)->root;
4801 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4804 unsigned nr_extents = 0;
4805 int extra_reserve = 0;
4806 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4808 bool delalloc_lock = true;
4812 /* If we are a free space inode we need to not flush since we will be in
4813 * the middle of a transaction commit. We also don't need the delalloc
4814 * mutex since we won't race with anybody. We need this mostly to make
4815 * lockdep shut its filthy mouth.
4817 if (btrfs_is_free_space_inode(inode)) {
4818 flush = BTRFS_RESERVE_NO_FLUSH;
4819 delalloc_lock = false;
4822 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4823 btrfs_transaction_in_commit(root->fs_info))
4824 schedule_timeout(1);
4827 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4829 num_bytes = ALIGN(num_bytes, root->sectorsize);
4831 spin_lock(&BTRFS_I(inode)->lock);
4832 BTRFS_I(inode)->outstanding_extents++;
4834 if (BTRFS_I(inode)->outstanding_extents >
4835 BTRFS_I(inode)->reserved_extents)
4836 nr_extents = BTRFS_I(inode)->outstanding_extents -
4837 BTRFS_I(inode)->reserved_extents;
4840 * Add an item to reserve for updating the inode when we complete the
4843 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4844 &BTRFS_I(inode)->runtime_flags)) {
4849 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4850 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4851 csum_bytes = BTRFS_I(inode)->csum_bytes;
4852 spin_unlock(&BTRFS_I(inode)->lock);
4854 if (root->fs_info->quota_enabled) {
4855 ret = btrfs_qgroup_reserve(root, num_bytes +
4856 nr_extents * root->leafsize);
4861 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4862 if (unlikely(ret)) {
4863 if (root->fs_info->quota_enabled)
4864 btrfs_qgroup_free(root, num_bytes +
4865 nr_extents * root->leafsize);
4869 spin_lock(&BTRFS_I(inode)->lock);
4870 if (extra_reserve) {
4871 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4872 &BTRFS_I(inode)->runtime_flags);
4875 BTRFS_I(inode)->reserved_extents += nr_extents;
4876 spin_unlock(&BTRFS_I(inode)->lock);
4879 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4882 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4883 btrfs_ino(inode), to_reserve, 1);
4884 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4889 spin_lock(&BTRFS_I(inode)->lock);
4890 dropped = drop_outstanding_extent(inode);
4892 * If the inodes csum_bytes is the same as the original
4893 * csum_bytes then we know we haven't raced with any free()ers
4894 * so we can just reduce our inodes csum bytes and carry on.
4896 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4897 calc_csum_metadata_size(inode, num_bytes, 0);
4899 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4903 * This is tricky, but first we need to figure out how much we
4904 * free'd from any free-ers that occured during this
4905 * reservation, so we reset ->csum_bytes to the csum_bytes
4906 * before we dropped our lock, and then call the free for the
4907 * number of bytes that were freed while we were trying our
4910 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4911 BTRFS_I(inode)->csum_bytes = csum_bytes;
4912 to_free = calc_csum_metadata_size(inode, bytes, 0);
4916 * Now we need to see how much we would have freed had we not
4917 * been making this reservation and our ->csum_bytes were not
4918 * artificially inflated.
4920 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4921 bytes = csum_bytes - orig_csum_bytes;
4922 bytes = calc_csum_metadata_size(inode, bytes, 0);
4925 * Now reset ->csum_bytes to what it should be. If bytes is
4926 * more than to_free then we would have free'd more space had we
4927 * not had an artificially high ->csum_bytes, so we need to free
4928 * the remainder. If bytes is the same or less then we don't
4929 * need to do anything, the other free-ers did the correct
4932 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4933 if (bytes > to_free)
4934 to_free = bytes - to_free;
4938 spin_unlock(&BTRFS_I(inode)->lock);
4940 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4943 btrfs_block_rsv_release(root, block_rsv, to_free);
4944 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4945 btrfs_ino(inode), to_free, 0);
4948 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4953 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4954 * @inode: the inode to release the reservation for
4955 * @num_bytes: the number of bytes we're releasing
4957 * This will release the metadata reservation for an inode. This can be called
4958 * once we complete IO for a given set of bytes to release their metadata
4961 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4963 struct btrfs_root *root = BTRFS_I(inode)->root;
4967 num_bytes = ALIGN(num_bytes, root->sectorsize);
4968 spin_lock(&BTRFS_I(inode)->lock);
4969 dropped = drop_outstanding_extent(inode);
4972 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4973 spin_unlock(&BTRFS_I(inode)->lock);
4975 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4977 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4978 btrfs_ino(inode), to_free, 0);
4979 if (root->fs_info->quota_enabled) {
4980 btrfs_qgroup_free(root, num_bytes +
4981 dropped * root->leafsize);
4984 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4989 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4990 * @inode: inode we're writing to
4991 * @num_bytes: the number of bytes we want to allocate
4993 * This will do the following things
4995 * o reserve space in the data space info for num_bytes
4996 * o reserve space in the metadata space info based on number of outstanding
4997 * extents and how much csums will be needed
4998 * o add to the inodes ->delalloc_bytes
4999 * o add it to the fs_info's delalloc inodes list.
5001 * This will return 0 for success and -ENOSPC if there is no space left.
5003 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5007 ret = btrfs_check_data_free_space(inode, num_bytes);
5011 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5013 btrfs_free_reserved_data_space(inode, num_bytes);
5021 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5022 * @inode: inode we're releasing space for
5023 * @num_bytes: the number of bytes we want to free up
5025 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5026 * called in the case that we don't need the metadata AND data reservations
5027 * anymore. So if there is an error or we insert an inline extent.
5029 * This function will release the metadata space that was not used and will
5030 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5031 * list if there are no delalloc bytes left.
5033 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5035 btrfs_delalloc_release_metadata(inode, num_bytes);
5036 btrfs_free_reserved_data_space(inode, num_bytes);
5039 static int update_block_group(struct btrfs_root *root,
5040 u64 bytenr, u64 num_bytes, int alloc)
5042 struct btrfs_block_group_cache *cache = NULL;
5043 struct btrfs_fs_info *info = root->fs_info;
5044 u64 total = num_bytes;
5049 /* block accounting for super block */
5050 spin_lock(&info->delalloc_lock);
5051 old_val = btrfs_super_bytes_used(info->super_copy);
5053 old_val += num_bytes;
5055 old_val -= num_bytes;
5056 btrfs_set_super_bytes_used(info->super_copy, old_val);
5057 spin_unlock(&info->delalloc_lock);
5060 cache = btrfs_lookup_block_group(info, bytenr);
5063 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5064 BTRFS_BLOCK_GROUP_RAID1 |
5065 BTRFS_BLOCK_GROUP_RAID10))
5070 * If this block group has free space cache written out, we
5071 * need to make sure to load it if we are removing space. This
5072 * is because we need the unpinning stage to actually add the
5073 * space back to the block group, otherwise we will leak space.
5075 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5076 cache_block_group(cache, 1);
5078 byte_in_group = bytenr - cache->key.objectid;
5079 WARN_ON(byte_in_group > cache->key.offset);
5081 spin_lock(&cache->space_info->lock);
5082 spin_lock(&cache->lock);
5084 if (btrfs_test_opt(root, SPACE_CACHE) &&
5085 cache->disk_cache_state < BTRFS_DC_CLEAR)
5086 cache->disk_cache_state = BTRFS_DC_CLEAR;
5089 old_val = btrfs_block_group_used(&cache->item);
5090 num_bytes = min(total, cache->key.offset - byte_in_group);
5092 old_val += num_bytes;
5093 btrfs_set_block_group_used(&cache->item, old_val);
5094 cache->reserved -= num_bytes;
5095 cache->space_info->bytes_reserved -= num_bytes;
5096 cache->space_info->bytes_used += num_bytes;
5097 cache->space_info->disk_used += num_bytes * factor;
5098 spin_unlock(&cache->lock);
5099 spin_unlock(&cache->space_info->lock);
5101 old_val -= num_bytes;
5102 btrfs_set_block_group_used(&cache->item, old_val);
5103 cache->pinned += num_bytes;
5104 cache->space_info->bytes_pinned += num_bytes;
5105 cache->space_info->bytes_used -= num_bytes;
5106 cache->space_info->disk_used -= num_bytes * factor;
5107 spin_unlock(&cache->lock);
5108 spin_unlock(&cache->space_info->lock);
5110 set_extent_dirty(info->pinned_extents,
5111 bytenr, bytenr + num_bytes - 1,
5112 GFP_NOFS | __GFP_NOFAIL);
5114 btrfs_put_block_group(cache);
5116 bytenr += num_bytes;
5121 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5123 struct btrfs_block_group_cache *cache;
5126 spin_lock(&root->fs_info->block_group_cache_lock);
5127 bytenr = root->fs_info->first_logical_byte;
5128 spin_unlock(&root->fs_info->block_group_cache_lock);
5130 if (bytenr < (u64)-1)
5133 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5137 bytenr = cache->key.objectid;
5138 btrfs_put_block_group(cache);
5143 static int pin_down_extent(struct btrfs_root *root,
5144 struct btrfs_block_group_cache *cache,
5145 u64 bytenr, u64 num_bytes, int reserved)
5147 spin_lock(&cache->space_info->lock);
5148 spin_lock(&cache->lock);
5149 cache->pinned += num_bytes;
5150 cache->space_info->bytes_pinned += num_bytes;
5152 cache->reserved -= num_bytes;
5153 cache->space_info->bytes_reserved -= num_bytes;
5155 spin_unlock(&cache->lock);
5156 spin_unlock(&cache->space_info->lock);
5158 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5159 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5164 * this function must be called within transaction
5166 int btrfs_pin_extent(struct btrfs_root *root,
5167 u64 bytenr, u64 num_bytes, int reserved)
5169 struct btrfs_block_group_cache *cache;
5171 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5172 BUG_ON(!cache); /* Logic error */
5174 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5176 btrfs_put_block_group(cache);
5181 * this function must be called within transaction
5183 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5184 u64 bytenr, u64 num_bytes)
5186 struct btrfs_block_group_cache *cache;
5188 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5189 BUG_ON(!cache); /* Logic error */
5192 * pull in the free space cache (if any) so that our pin
5193 * removes the free space from the cache. We have load_only set
5194 * to one because the slow code to read in the free extents does check
5195 * the pinned extents.
5197 cache_block_group(cache, 1);
5199 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5201 /* remove us from the free space cache (if we're there at all) */
5202 btrfs_remove_free_space(cache, bytenr, num_bytes);
5203 btrfs_put_block_group(cache);
5208 * btrfs_update_reserved_bytes - update the block_group and space info counters
5209 * @cache: The cache we are manipulating
5210 * @num_bytes: The number of bytes in question
5211 * @reserve: One of the reservation enums
5213 * This is called by the allocator when it reserves space, or by somebody who is
5214 * freeing space that was never actually used on disk. For example if you
5215 * reserve some space for a new leaf in transaction A and before transaction A
5216 * commits you free that leaf, you call this with reserve set to 0 in order to
5217 * clear the reservation.
5219 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5220 * ENOSPC accounting. For data we handle the reservation through clearing the
5221 * delalloc bits in the io_tree. We have to do this since we could end up
5222 * allocating less disk space for the amount of data we have reserved in the
5223 * case of compression.
5225 * If this is a reservation and the block group has become read only we cannot
5226 * make the reservation and return -EAGAIN, otherwise this function always
5229 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5230 u64 num_bytes, int reserve)
5232 struct btrfs_space_info *space_info = cache->space_info;
5235 spin_lock(&space_info->lock);
5236 spin_lock(&cache->lock);
5237 if (reserve != RESERVE_FREE) {
5241 cache->reserved += num_bytes;
5242 space_info->bytes_reserved += num_bytes;
5243 if (reserve == RESERVE_ALLOC) {
5244 trace_btrfs_space_reservation(cache->fs_info,
5245 "space_info", space_info->flags,
5247 space_info->bytes_may_use -= num_bytes;
5252 space_info->bytes_readonly += num_bytes;
5253 cache->reserved -= num_bytes;
5254 space_info->bytes_reserved -= num_bytes;
5255 space_info->reservation_progress++;
5257 spin_unlock(&cache->lock);
5258 spin_unlock(&space_info->lock);
5262 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5263 struct btrfs_root *root)
5265 struct btrfs_fs_info *fs_info = root->fs_info;
5266 struct btrfs_caching_control *next;
5267 struct btrfs_caching_control *caching_ctl;
5268 struct btrfs_block_group_cache *cache;
5270 down_write(&fs_info->extent_commit_sem);
5272 list_for_each_entry_safe(caching_ctl, next,
5273 &fs_info->caching_block_groups, list) {
5274 cache = caching_ctl->block_group;
5275 if (block_group_cache_done(cache)) {
5276 cache->last_byte_to_unpin = (u64)-1;
5277 list_del_init(&caching_ctl->list);
5278 put_caching_control(caching_ctl);
5280 cache->last_byte_to_unpin = caching_ctl->progress;
5284 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5285 fs_info->pinned_extents = &fs_info->freed_extents[1];
5287 fs_info->pinned_extents = &fs_info->freed_extents[0];
5289 up_write(&fs_info->extent_commit_sem);
5291 update_global_block_rsv(fs_info);
5294 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5296 struct btrfs_fs_info *fs_info = root->fs_info;
5297 struct btrfs_block_group_cache *cache = NULL;
5298 struct btrfs_space_info *space_info;
5299 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5303 while (start <= end) {
5306 start >= cache->key.objectid + cache->key.offset) {
5308 btrfs_put_block_group(cache);
5309 cache = btrfs_lookup_block_group(fs_info, start);
5310 BUG_ON(!cache); /* Logic error */
5313 len = cache->key.objectid + cache->key.offset - start;
5314 len = min(len, end + 1 - start);
5316 if (start < cache->last_byte_to_unpin) {
5317 len = min(len, cache->last_byte_to_unpin - start);
5318 btrfs_add_free_space(cache, start, len);
5322 space_info = cache->space_info;
5324 spin_lock(&space_info->lock);
5325 spin_lock(&cache->lock);
5326 cache->pinned -= len;
5327 space_info->bytes_pinned -= len;
5329 space_info->bytes_readonly += len;
5332 spin_unlock(&cache->lock);
5333 if (!readonly && global_rsv->space_info == space_info) {
5334 spin_lock(&global_rsv->lock);
5335 if (!global_rsv->full) {
5336 len = min(len, global_rsv->size -
5337 global_rsv->reserved);
5338 global_rsv->reserved += len;
5339 space_info->bytes_may_use += len;
5340 if (global_rsv->reserved >= global_rsv->size)
5341 global_rsv->full = 1;
5343 spin_unlock(&global_rsv->lock);
5345 spin_unlock(&space_info->lock);
5349 btrfs_put_block_group(cache);
5353 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5354 struct btrfs_root *root)
5356 struct btrfs_fs_info *fs_info = root->fs_info;
5357 struct extent_io_tree *unpin;
5365 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5366 unpin = &fs_info->freed_extents[1];
5368 unpin = &fs_info->freed_extents[0];
5371 ret = find_first_extent_bit(unpin, 0, &start, &end,
5372 EXTENT_DIRTY, NULL);
5376 if (btrfs_test_opt(root, DISCARD))
5377 ret = btrfs_discard_extent(root, start,
5378 end + 1 - start, NULL);
5380 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5381 unpin_extent_range(root, start, end);
5388 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5389 struct btrfs_root *root,
5390 u64 bytenr, u64 num_bytes, u64 parent,
5391 u64 root_objectid, u64 owner_objectid,
5392 u64 owner_offset, int refs_to_drop,
5393 struct btrfs_delayed_extent_op *extent_op)
5395 struct btrfs_key key;
5396 struct btrfs_path *path;
5397 struct btrfs_fs_info *info = root->fs_info;
5398 struct btrfs_root *extent_root = info->extent_root;
5399 struct extent_buffer *leaf;
5400 struct btrfs_extent_item *ei;
5401 struct btrfs_extent_inline_ref *iref;
5404 int extent_slot = 0;
5405 int found_extent = 0;
5409 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5412 path = btrfs_alloc_path();
5417 path->leave_spinning = 1;
5419 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5420 BUG_ON(!is_data && refs_to_drop != 1);
5423 skinny_metadata = 0;
5425 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5426 bytenr, num_bytes, parent,
5427 root_objectid, owner_objectid,
5430 extent_slot = path->slots[0];
5431 while (extent_slot >= 0) {
5432 btrfs_item_key_to_cpu(path->nodes[0], &key,
5434 if (key.objectid != bytenr)
5436 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5437 key.offset == num_bytes) {
5441 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5442 key.offset == owner_objectid) {
5446 if (path->slots[0] - extent_slot > 5)
5450 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5451 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5452 if (found_extent && item_size < sizeof(*ei))
5455 if (!found_extent) {
5457 ret = remove_extent_backref(trans, extent_root, path,
5461 btrfs_abort_transaction(trans, extent_root, ret);
5464 btrfs_release_path(path);
5465 path->leave_spinning = 1;
5467 key.objectid = bytenr;
5468 key.type = BTRFS_EXTENT_ITEM_KEY;
5469 key.offset = num_bytes;
5471 if (!is_data && skinny_metadata) {
5472 key.type = BTRFS_METADATA_ITEM_KEY;
5473 key.offset = owner_objectid;
5476 ret = btrfs_search_slot(trans, extent_root,
5478 if (ret > 0 && skinny_metadata && path->slots[0]) {
5480 * Couldn't find our skinny metadata item,
5481 * see if we have ye olde extent item.
5484 btrfs_item_key_to_cpu(path->nodes[0], &key,
5486 if (key.objectid == bytenr &&
5487 key.type == BTRFS_EXTENT_ITEM_KEY &&
5488 key.offset == num_bytes)
5492 if (ret > 0 && skinny_metadata) {
5493 skinny_metadata = false;
5494 key.type = BTRFS_EXTENT_ITEM_KEY;
5495 key.offset = num_bytes;
5496 btrfs_release_path(path);
5497 ret = btrfs_search_slot(trans, extent_root,
5502 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5503 ret, (unsigned long long)bytenr);
5505 btrfs_print_leaf(extent_root,
5509 btrfs_abort_transaction(trans, extent_root, ret);
5512 extent_slot = path->slots[0];
5514 } else if (ret == -ENOENT) {
5515 btrfs_print_leaf(extent_root, path->nodes[0]);
5518 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5519 (unsigned long long)bytenr,
5520 (unsigned long long)parent,
5521 (unsigned long long)root_objectid,
5522 (unsigned long long)owner_objectid,
5523 (unsigned long long)owner_offset);
5525 btrfs_abort_transaction(trans, extent_root, ret);
5529 leaf = path->nodes[0];
5530 item_size = btrfs_item_size_nr(leaf, extent_slot);
5531 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5532 if (item_size < sizeof(*ei)) {
5533 BUG_ON(found_extent || extent_slot != path->slots[0]);
5534 ret = convert_extent_item_v0(trans, extent_root, path,
5537 btrfs_abort_transaction(trans, extent_root, ret);
5541 btrfs_release_path(path);
5542 path->leave_spinning = 1;
5544 key.objectid = bytenr;
5545 key.type = BTRFS_EXTENT_ITEM_KEY;
5546 key.offset = num_bytes;
5548 ret = btrfs_search_slot(trans, extent_root, &key, path,
5551 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5552 ret, (unsigned long long)bytenr);
5553 btrfs_print_leaf(extent_root, path->nodes[0]);
5556 btrfs_abort_transaction(trans, extent_root, ret);
5560 extent_slot = path->slots[0];
5561 leaf = path->nodes[0];
5562 item_size = btrfs_item_size_nr(leaf, extent_slot);
5565 BUG_ON(item_size < sizeof(*ei));
5566 ei = btrfs_item_ptr(leaf, extent_slot,
5567 struct btrfs_extent_item);
5568 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5569 key.type == BTRFS_EXTENT_ITEM_KEY) {
5570 struct btrfs_tree_block_info *bi;
5571 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5572 bi = (struct btrfs_tree_block_info *)(ei + 1);
5573 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5576 refs = btrfs_extent_refs(leaf, ei);
5577 BUG_ON(refs < refs_to_drop);
5578 refs -= refs_to_drop;
5582 __run_delayed_extent_op(extent_op, leaf, ei);
5584 * In the case of inline back ref, reference count will
5585 * be updated by remove_extent_backref
5588 BUG_ON(!found_extent);
5590 btrfs_set_extent_refs(leaf, ei, refs);
5591 btrfs_mark_buffer_dirty(leaf);
5594 ret = remove_extent_backref(trans, extent_root, path,
5598 btrfs_abort_transaction(trans, extent_root, ret);
5604 BUG_ON(is_data && refs_to_drop !=
5605 extent_data_ref_count(root, path, iref));
5607 BUG_ON(path->slots[0] != extent_slot);
5609 BUG_ON(path->slots[0] != extent_slot + 1);
5610 path->slots[0] = extent_slot;
5615 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5618 btrfs_abort_transaction(trans, extent_root, ret);
5621 btrfs_release_path(path);
5624 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5626 btrfs_abort_transaction(trans, extent_root, ret);
5631 ret = update_block_group(root, bytenr, num_bytes, 0);
5633 btrfs_abort_transaction(trans, extent_root, ret);
5638 btrfs_free_path(path);
5643 * when we free an block, it is possible (and likely) that we free the last
5644 * delayed ref for that extent as well. This searches the delayed ref tree for
5645 * a given extent, and if there are no other delayed refs to be processed, it
5646 * removes it from the tree.
5648 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5649 struct btrfs_root *root, u64 bytenr)
5651 struct btrfs_delayed_ref_head *head;
5652 struct btrfs_delayed_ref_root *delayed_refs;
5653 struct btrfs_delayed_ref_node *ref;
5654 struct rb_node *node;
5657 delayed_refs = &trans->transaction->delayed_refs;
5658 spin_lock(&delayed_refs->lock);
5659 head = btrfs_find_delayed_ref_head(trans, bytenr);
5663 node = rb_prev(&head->node.rb_node);
5667 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5669 /* there are still entries for this ref, we can't drop it */
5670 if (ref->bytenr == bytenr)
5673 if (head->extent_op) {
5674 if (!head->must_insert_reserved)
5676 btrfs_free_delayed_extent_op(head->extent_op);
5677 head->extent_op = NULL;
5681 * waiting for the lock here would deadlock. If someone else has it
5682 * locked they are already in the process of dropping it anyway
5684 if (!mutex_trylock(&head->mutex))
5688 * at this point we have a head with no other entries. Go
5689 * ahead and process it.
5691 head->node.in_tree = 0;
5692 rb_erase(&head->node.rb_node, &delayed_refs->root);
5694 delayed_refs->num_entries--;
5697 * we don't take a ref on the node because we're removing it from the
5698 * tree, so we just steal the ref the tree was holding.
5700 delayed_refs->num_heads--;
5701 if (list_empty(&head->cluster))
5702 delayed_refs->num_heads_ready--;
5704 list_del_init(&head->cluster);
5705 spin_unlock(&delayed_refs->lock);
5707 BUG_ON(head->extent_op);
5708 if (head->must_insert_reserved)
5711 mutex_unlock(&head->mutex);
5712 btrfs_put_delayed_ref(&head->node);
5715 spin_unlock(&delayed_refs->lock);
5719 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5720 struct btrfs_root *root,
5721 struct extent_buffer *buf,
5722 u64 parent, int last_ref)
5724 struct btrfs_block_group_cache *cache = NULL;
5727 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5728 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5729 buf->start, buf->len,
5730 parent, root->root_key.objectid,
5731 btrfs_header_level(buf),
5732 BTRFS_DROP_DELAYED_REF, NULL, 0);
5733 BUG_ON(ret); /* -ENOMEM */
5739 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5741 if (btrfs_header_generation(buf) == trans->transid) {
5742 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5743 ret = check_ref_cleanup(trans, root, buf->start);
5748 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5749 pin_down_extent(root, cache, buf->start, buf->len, 1);
5753 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5755 btrfs_add_free_space(cache, buf->start, buf->len);
5756 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5760 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5763 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5764 btrfs_put_block_group(cache);
5767 /* Can return -ENOMEM */
5768 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5769 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5770 u64 owner, u64 offset, int for_cow)
5773 struct btrfs_fs_info *fs_info = root->fs_info;
5776 * tree log blocks never actually go into the extent allocation
5777 * tree, just update pinning info and exit early.
5779 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5780 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5781 /* unlocks the pinned mutex */
5782 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5784 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5785 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5787 parent, root_objectid, (int)owner,
5788 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5790 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5792 parent, root_objectid, owner,
5793 offset, BTRFS_DROP_DELAYED_REF,
5799 static u64 stripe_align(struct btrfs_root *root,
5800 struct btrfs_block_group_cache *cache,
5801 u64 val, u64 num_bytes)
5803 u64 ret = ALIGN(val, root->stripesize);
5808 * when we wait for progress in the block group caching, its because
5809 * our allocation attempt failed at least once. So, we must sleep
5810 * and let some progress happen before we try again.
5812 * This function will sleep at least once waiting for new free space to
5813 * show up, and then it will check the block group free space numbers
5814 * for our min num_bytes. Another option is to have it go ahead
5815 * and look in the rbtree for a free extent of a given size, but this
5819 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5822 struct btrfs_caching_control *caching_ctl;
5824 caching_ctl = get_caching_control(cache);
5828 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5829 (cache->free_space_ctl->free_space >= num_bytes));
5831 put_caching_control(caching_ctl);
5836 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5838 struct btrfs_caching_control *caching_ctl;
5840 caching_ctl = get_caching_control(cache);
5844 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5846 put_caching_control(caching_ctl);
5850 int __get_raid_index(u64 flags)
5852 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5853 return BTRFS_RAID_RAID10;
5854 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5855 return BTRFS_RAID_RAID1;
5856 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5857 return BTRFS_RAID_DUP;
5858 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5859 return BTRFS_RAID_RAID0;
5860 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5861 return BTRFS_RAID_RAID5;
5862 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5863 return BTRFS_RAID_RAID6;
5865 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
5868 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5870 return __get_raid_index(cache->flags);
5873 enum btrfs_loop_type {
5874 LOOP_CACHING_NOWAIT = 0,
5875 LOOP_CACHING_WAIT = 1,
5876 LOOP_ALLOC_CHUNK = 2,
5877 LOOP_NO_EMPTY_SIZE = 3,
5881 * walks the btree of allocated extents and find a hole of a given size.
5882 * The key ins is changed to record the hole:
5883 * ins->objectid == block start
5884 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5885 * ins->offset == number of blocks
5886 * Any available blocks before search_start are skipped.
5888 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5889 struct btrfs_root *orig_root,
5890 u64 num_bytes, u64 empty_size,
5891 u64 hint_byte, struct btrfs_key *ins,
5895 struct btrfs_root *root = orig_root->fs_info->extent_root;
5896 struct btrfs_free_cluster *last_ptr = NULL;
5897 struct btrfs_block_group_cache *block_group = NULL;
5898 struct btrfs_block_group_cache *used_block_group;
5899 u64 search_start = 0;
5900 int empty_cluster = 2 * 1024 * 1024;
5901 struct btrfs_space_info *space_info;
5903 int index = __get_raid_index(data);
5904 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5905 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5906 bool found_uncached_bg = false;
5907 bool failed_cluster_refill = false;
5908 bool failed_alloc = false;
5909 bool use_cluster = true;
5910 bool have_caching_bg = false;
5912 WARN_ON(num_bytes < root->sectorsize);
5913 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5917 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5919 space_info = __find_space_info(root->fs_info, data);
5921 btrfs_err(root->fs_info, "No space info for %llu", data);
5926 * If the space info is for both data and metadata it means we have a
5927 * small filesystem and we can't use the clustering stuff.
5929 if (btrfs_mixed_space_info(space_info))
5930 use_cluster = false;
5932 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5933 last_ptr = &root->fs_info->meta_alloc_cluster;
5934 if (!btrfs_test_opt(root, SSD))
5935 empty_cluster = 64 * 1024;
5938 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5939 btrfs_test_opt(root, SSD)) {
5940 last_ptr = &root->fs_info->data_alloc_cluster;
5944 spin_lock(&last_ptr->lock);
5945 if (last_ptr->block_group)
5946 hint_byte = last_ptr->window_start;
5947 spin_unlock(&last_ptr->lock);
5950 search_start = max(search_start, first_logical_byte(root, 0));
5951 search_start = max(search_start, hint_byte);
5956 if (search_start == hint_byte) {
5957 block_group = btrfs_lookup_block_group(root->fs_info,
5959 used_block_group = block_group;
5961 * we don't want to use the block group if it doesn't match our
5962 * allocation bits, or if its not cached.
5964 * However if we are re-searching with an ideal block group
5965 * picked out then we don't care that the block group is cached.
5967 if (block_group && block_group_bits(block_group, data) &&
5968 block_group->cached != BTRFS_CACHE_NO) {
5969 down_read(&space_info->groups_sem);
5970 if (list_empty(&block_group->list) ||
5973 * someone is removing this block group,
5974 * we can't jump into the have_block_group
5975 * target because our list pointers are not
5978 btrfs_put_block_group(block_group);
5979 up_read(&space_info->groups_sem);
5981 index = get_block_group_index(block_group);
5982 goto have_block_group;
5984 } else if (block_group) {
5985 btrfs_put_block_group(block_group);
5989 have_caching_bg = false;
5990 down_read(&space_info->groups_sem);
5991 list_for_each_entry(block_group, &space_info->block_groups[index],
5996 used_block_group = block_group;
5997 btrfs_get_block_group(block_group);
5998 search_start = block_group->key.objectid;
6001 * this can happen if we end up cycling through all the
6002 * raid types, but we want to make sure we only allocate
6003 * for the proper type.
6005 if (!block_group_bits(block_group, data)) {
6006 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6007 BTRFS_BLOCK_GROUP_RAID1 |
6008 BTRFS_BLOCK_GROUP_RAID5 |
6009 BTRFS_BLOCK_GROUP_RAID6 |
6010 BTRFS_BLOCK_GROUP_RAID10;
6013 * if they asked for extra copies and this block group
6014 * doesn't provide them, bail. This does allow us to
6015 * fill raid0 from raid1.
6017 if ((data & extra) && !(block_group->flags & extra))
6022 cached = block_group_cache_done(block_group);
6023 if (unlikely(!cached)) {
6024 found_uncached_bg = true;
6025 ret = cache_block_group(block_group, 0);
6030 if (unlikely(block_group->ro))
6034 * Ok we want to try and use the cluster allocator, so
6038 unsigned long aligned_cluster;
6040 * the refill lock keeps out other
6041 * people trying to start a new cluster
6043 spin_lock(&last_ptr->refill_lock);
6044 used_block_group = last_ptr->block_group;
6045 if (used_block_group != block_group &&
6046 (!used_block_group ||
6047 used_block_group->ro ||
6048 !block_group_bits(used_block_group, data))) {
6049 used_block_group = block_group;
6050 goto refill_cluster;
6053 if (used_block_group != block_group)
6054 btrfs_get_block_group(used_block_group);
6056 offset = btrfs_alloc_from_cluster(used_block_group,
6057 last_ptr, num_bytes, used_block_group->key.objectid);
6059 /* we have a block, we're done */
6060 spin_unlock(&last_ptr->refill_lock);
6061 trace_btrfs_reserve_extent_cluster(root,
6062 block_group, search_start, num_bytes);
6066 WARN_ON(last_ptr->block_group != used_block_group);
6067 if (used_block_group != block_group) {
6068 btrfs_put_block_group(used_block_group);
6069 used_block_group = block_group;
6072 BUG_ON(used_block_group != block_group);
6073 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6074 * set up a new clusters, so lets just skip it
6075 * and let the allocator find whatever block
6076 * it can find. If we reach this point, we
6077 * will have tried the cluster allocator
6078 * plenty of times and not have found
6079 * anything, so we are likely way too
6080 * fragmented for the clustering stuff to find
6083 * However, if the cluster is taken from the
6084 * current block group, release the cluster
6085 * first, so that we stand a better chance of
6086 * succeeding in the unclustered
6088 if (loop >= LOOP_NO_EMPTY_SIZE &&
6089 last_ptr->block_group != block_group) {
6090 spin_unlock(&last_ptr->refill_lock);
6091 goto unclustered_alloc;
6095 * this cluster didn't work out, free it and
6098 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6100 if (loop >= LOOP_NO_EMPTY_SIZE) {
6101 spin_unlock(&last_ptr->refill_lock);
6102 goto unclustered_alloc;
6105 aligned_cluster = max_t(unsigned long,
6106 empty_cluster + empty_size,
6107 block_group->full_stripe_len);
6109 /* allocate a cluster in this block group */
6110 ret = btrfs_find_space_cluster(trans, root,
6111 block_group, last_ptr,
6112 search_start, num_bytes,
6116 * now pull our allocation out of this
6119 offset = btrfs_alloc_from_cluster(block_group,
6120 last_ptr, num_bytes,
6123 /* we found one, proceed */
6124 spin_unlock(&last_ptr->refill_lock);
6125 trace_btrfs_reserve_extent_cluster(root,
6126 block_group, search_start,
6130 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6131 && !failed_cluster_refill) {
6132 spin_unlock(&last_ptr->refill_lock);
6134 failed_cluster_refill = true;
6135 wait_block_group_cache_progress(block_group,
6136 num_bytes + empty_cluster + empty_size);
6137 goto have_block_group;
6141 * at this point we either didn't find a cluster
6142 * or we weren't able to allocate a block from our
6143 * cluster. Free the cluster we've been trying
6144 * to use, and go to the next block group
6146 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6147 spin_unlock(&last_ptr->refill_lock);
6152 spin_lock(&block_group->free_space_ctl->tree_lock);
6154 block_group->free_space_ctl->free_space <
6155 num_bytes + empty_cluster + empty_size) {
6156 spin_unlock(&block_group->free_space_ctl->tree_lock);
6159 spin_unlock(&block_group->free_space_ctl->tree_lock);
6161 offset = btrfs_find_space_for_alloc(block_group, search_start,
6162 num_bytes, empty_size);
6164 * If we didn't find a chunk, and we haven't failed on this
6165 * block group before, and this block group is in the middle of
6166 * caching and we are ok with waiting, then go ahead and wait
6167 * for progress to be made, and set failed_alloc to true.
6169 * If failed_alloc is true then we've already waited on this
6170 * block group once and should move on to the next block group.
6172 if (!offset && !failed_alloc && !cached &&
6173 loop > LOOP_CACHING_NOWAIT) {
6174 wait_block_group_cache_progress(block_group,
6175 num_bytes + empty_size);
6176 failed_alloc = true;
6177 goto have_block_group;
6178 } else if (!offset) {
6180 have_caching_bg = true;
6184 search_start = stripe_align(root, used_block_group,
6187 /* move on to the next group */
6188 if (search_start + num_bytes >
6189 used_block_group->key.objectid + used_block_group->key.offset) {
6190 btrfs_add_free_space(used_block_group, offset, num_bytes);
6194 if (offset < search_start)
6195 btrfs_add_free_space(used_block_group, offset,
6196 search_start - offset);
6197 BUG_ON(offset > search_start);
6199 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6201 if (ret == -EAGAIN) {
6202 btrfs_add_free_space(used_block_group, offset, num_bytes);
6206 /* we are all good, lets return */
6207 ins->objectid = search_start;
6208 ins->offset = num_bytes;
6210 trace_btrfs_reserve_extent(orig_root, block_group,
6211 search_start, num_bytes);
6212 if (used_block_group != block_group)
6213 btrfs_put_block_group(used_block_group);
6214 btrfs_put_block_group(block_group);
6217 failed_cluster_refill = false;
6218 failed_alloc = false;
6219 BUG_ON(index != get_block_group_index(block_group));
6220 if (used_block_group != block_group)
6221 btrfs_put_block_group(used_block_group);
6222 btrfs_put_block_group(block_group);
6224 up_read(&space_info->groups_sem);
6226 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6229 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6233 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6234 * caching kthreads as we move along
6235 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6236 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6237 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6240 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6243 if (loop == LOOP_ALLOC_CHUNK) {
6244 ret = do_chunk_alloc(trans, root, data,
6247 * Do not bail out on ENOSPC since we
6248 * can do more things.
6250 if (ret < 0 && ret != -ENOSPC) {
6251 btrfs_abort_transaction(trans,
6257 if (loop == LOOP_NO_EMPTY_SIZE) {
6263 } else if (!ins->objectid) {
6265 } else if (ins->objectid) {
6273 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6274 int dump_block_groups)
6276 struct btrfs_block_group_cache *cache;
6279 spin_lock(&info->lock);
6280 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6281 (unsigned long long)info->flags,
6282 (unsigned long long)(info->total_bytes - info->bytes_used -
6283 info->bytes_pinned - info->bytes_reserved -
6284 info->bytes_readonly),
6285 (info->full) ? "" : "not ");
6286 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6287 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6288 (unsigned long long)info->total_bytes,
6289 (unsigned long long)info->bytes_used,
6290 (unsigned long long)info->bytes_pinned,
6291 (unsigned long long)info->bytes_reserved,
6292 (unsigned long long)info->bytes_may_use,
6293 (unsigned long long)info->bytes_readonly);
6294 spin_unlock(&info->lock);
6296 if (!dump_block_groups)
6299 down_read(&info->groups_sem);
6301 list_for_each_entry(cache, &info->block_groups[index], list) {
6302 spin_lock(&cache->lock);
6303 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6304 (unsigned long long)cache->key.objectid,
6305 (unsigned long long)cache->key.offset,
6306 (unsigned long long)btrfs_block_group_used(&cache->item),
6307 (unsigned long long)cache->pinned,
6308 (unsigned long long)cache->reserved,
6309 cache->ro ? "[readonly]" : "");
6310 btrfs_dump_free_space(cache, bytes);
6311 spin_unlock(&cache->lock);
6313 if (++index < BTRFS_NR_RAID_TYPES)
6315 up_read(&info->groups_sem);
6318 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6319 struct btrfs_root *root,
6320 u64 num_bytes, u64 min_alloc_size,
6321 u64 empty_size, u64 hint_byte,
6322 struct btrfs_key *ins, u64 data)
6324 bool final_tried = false;
6327 data = btrfs_get_alloc_profile(root, data);
6329 WARN_ON(num_bytes < root->sectorsize);
6330 ret = find_free_extent(trans, root, num_bytes, empty_size,
6331 hint_byte, ins, data);
6333 if (ret == -ENOSPC) {
6335 num_bytes = num_bytes >> 1;
6336 num_bytes = round_down(num_bytes, root->sectorsize);
6337 num_bytes = max(num_bytes, min_alloc_size);
6338 if (num_bytes == min_alloc_size)
6341 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6342 struct btrfs_space_info *sinfo;
6344 sinfo = __find_space_info(root->fs_info, data);
6345 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6346 (unsigned long long)data,
6347 (unsigned long long)num_bytes);
6349 dump_space_info(sinfo, num_bytes, 1);
6353 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6358 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6359 u64 start, u64 len, int pin)
6361 struct btrfs_block_group_cache *cache;
6364 cache = btrfs_lookup_block_group(root->fs_info, start);
6366 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6367 (unsigned long long)start);
6371 if (btrfs_test_opt(root, DISCARD))
6372 ret = btrfs_discard_extent(root, start, len, NULL);
6375 pin_down_extent(root, cache, start, len, 1);
6377 btrfs_add_free_space(cache, start, len);
6378 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6380 btrfs_put_block_group(cache);
6382 trace_btrfs_reserved_extent_free(root, start, len);
6387 int btrfs_free_reserved_extent(struct btrfs_root *root,
6390 return __btrfs_free_reserved_extent(root, start, len, 0);
6393 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6396 return __btrfs_free_reserved_extent(root, start, len, 1);
6399 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6400 struct btrfs_root *root,
6401 u64 parent, u64 root_objectid,
6402 u64 flags, u64 owner, u64 offset,
6403 struct btrfs_key *ins, int ref_mod)
6406 struct btrfs_fs_info *fs_info = root->fs_info;
6407 struct btrfs_extent_item *extent_item;
6408 struct btrfs_extent_inline_ref *iref;
6409 struct btrfs_path *path;
6410 struct extent_buffer *leaf;
6415 type = BTRFS_SHARED_DATA_REF_KEY;
6417 type = BTRFS_EXTENT_DATA_REF_KEY;
6419 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6421 path = btrfs_alloc_path();
6425 path->leave_spinning = 1;
6426 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6429 btrfs_free_path(path);
6433 leaf = path->nodes[0];
6434 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6435 struct btrfs_extent_item);
6436 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6437 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6438 btrfs_set_extent_flags(leaf, extent_item,
6439 flags | BTRFS_EXTENT_FLAG_DATA);
6441 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6442 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6444 struct btrfs_shared_data_ref *ref;
6445 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6446 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6447 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6449 struct btrfs_extent_data_ref *ref;
6450 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6451 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6452 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6453 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6454 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6457 btrfs_mark_buffer_dirty(path->nodes[0]);
6458 btrfs_free_path(path);
6460 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6461 if (ret) { /* -ENOENT, logic error */
6462 btrfs_err(fs_info, "update block group failed for %llu %llu",
6463 (unsigned long long)ins->objectid,
6464 (unsigned long long)ins->offset);
6470 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6471 struct btrfs_root *root,
6472 u64 parent, u64 root_objectid,
6473 u64 flags, struct btrfs_disk_key *key,
6474 int level, struct btrfs_key *ins)
6477 struct btrfs_fs_info *fs_info = root->fs_info;
6478 struct btrfs_extent_item *extent_item;
6479 struct btrfs_tree_block_info *block_info;
6480 struct btrfs_extent_inline_ref *iref;
6481 struct btrfs_path *path;
6482 struct extent_buffer *leaf;
6483 u32 size = sizeof(*extent_item) + sizeof(*iref);
6484 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6487 if (!skinny_metadata)
6488 size += sizeof(*block_info);
6490 path = btrfs_alloc_path();
6494 path->leave_spinning = 1;
6495 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6498 btrfs_free_path(path);
6502 leaf = path->nodes[0];
6503 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6504 struct btrfs_extent_item);
6505 btrfs_set_extent_refs(leaf, extent_item, 1);
6506 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6507 btrfs_set_extent_flags(leaf, extent_item,
6508 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6510 if (skinny_metadata) {
6511 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6513 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6514 btrfs_set_tree_block_key(leaf, block_info, key);
6515 btrfs_set_tree_block_level(leaf, block_info, level);
6516 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6520 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6521 btrfs_set_extent_inline_ref_type(leaf, iref,
6522 BTRFS_SHARED_BLOCK_REF_KEY);
6523 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6525 btrfs_set_extent_inline_ref_type(leaf, iref,
6526 BTRFS_TREE_BLOCK_REF_KEY);
6527 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6530 btrfs_mark_buffer_dirty(leaf);
6531 btrfs_free_path(path);
6533 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6534 if (ret) { /* -ENOENT, logic error */
6535 btrfs_err(fs_info, "update block group failed for %llu %llu",
6536 (unsigned long long)ins->objectid,
6537 (unsigned long long)ins->offset);
6543 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6544 struct btrfs_root *root,
6545 u64 root_objectid, u64 owner,
6546 u64 offset, struct btrfs_key *ins)
6550 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6552 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6554 root_objectid, owner, offset,
6555 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6560 * this is used by the tree logging recovery code. It records that
6561 * an extent has been allocated and makes sure to clear the free
6562 * space cache bits as well
6564 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6565 struct btrfs_root *root,
6566 u64 root_objectid, u64 owner, u64 offset,
6567 struct btrfs_key *ins)
6570 struct btrfs_block_group_cache *block_group;
6571 struct btrfs_caching_control *caching_ctl;
6572 u64 start = ins->objectid;
6573 u64 num_bytes = ins->offset;
6575 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6576 cache_block_group(block_group, 0);
6577 caching_ctl = get_caching_control(block_group);
6580 BUG_ON(!block_group_cache_done(block_group));
6581 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6582 BUG_ON(ret); /* -ENOMEM */
6584 mutex_lock(&caching_ctl->mutex);
6586 if (start >= caching_ctl->progress) {
6587 ret = add_excluded_extent(root, start, num_bytes);
6588 BUG_ON(ret); /* -ENOMEM */
6589 } else if (start + num_bytes <= caching_ctl->progress) {
6590 ret = btrfs_remove_free_space(block_group,
6592 BUG_ON(ret); /* -ENOMEM */
6594 num_bytes = caching_ctl->progress - start;
6595 ret = btrfs_remove_free_space(block_group,
6597 BUG_ON(ret); /* -ENOMEM */
6599 start = caching_ctl->progress;
6600 num_bytes = ins->objectid + ins->offset -
6601 caching_ctl->progress;
6602 ret = add_excluded_extent(root, start, num_bytes);
6603 BUG_ON(ret); /* -ENOMEM */
6606 mutex_unlock(&caching_ctl->mutex);
6607 put_caching_control(caching_ctl);
6610 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6611 RESERVE_ALLOC_NO_ACCOUNT);
6612 BUG_ON(ret); /* logic error */
6613 btrfs_put_block_group(block_group);
6614 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6615 0, owner, offset, ins, 1);
6619 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6620 struct btrfs_root *root,
6621 u64 bytenr, u32 blocksize,
6624 struct extent_buffer *buf;
6626 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6628 return ERR_PTR(-ENOMEM);
6629 btrfs_set_header_generation(buf, trans->transid);
6630 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6631 btrfs_tree_lock(buf);
6632 clean_tree_block(trans, root, buf);
6633 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6635 btrfs_set_lock_blocking(buf);
6636 btrfs_set_buffer_uptodate(buf);
6638 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6640 * we allow two log transactions at a time, use different
6641 * EXENT bit to differentiate dirty pages.
6643 if (root->log_transid % 2 == 0)
6644 set_extent_dirty(&root->dirty_log_pages, buf->start,
6645 buf->start + buf->len - 1, GFP_NOFS);
6647 set_extent_new(&root->dirty_log_pages, buf->start,
6648 buf->start + buf->len - 1, GFP_NOFS);
6650 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6651 buf->start + buf->len - 1, GFP_NOFS);
6653 trans->blocks_used++;
6654 /* this returns a buffer locked for blocking */
6658 static struct btrfs_block_rsv *
6659 use_block_rsv(struct btrfs_trans_handle *trans,
6660 struct btrfs_root *root, u32 blocksize)
6662 struct btrfs_block_rsv *block_rsv;
6663 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6666 block_rsv = get_block_rsv(trans, root);
6668 if (block_rsv->size == 0) {
6669 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6670 BTRFS_RESERVE_NO_FLUSH);
6672 * If we couldn't reserve metadata bytes try and use some from
6673 * the global reserve.
6675 if (ret && block_rsv != global_rsv) {
6676 ret = block_rsv_use_bytes(global_rsv, blocksize);
6679 return ERR_PTR(ret);
6681 return ERR_PTR(ret);
6686 ret = block_rsv_use_bytes(block_rsv, blocksize);
6689 if (ret && !block_rsv->failfast) {
6690 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6691 static DEFINE_RATELIMIT_STATE(_rs,
6692 DEFAULT_RATELIMIT_INTERVAL * 10,
6693 /*DEFAULT_RATELIMIT_BURST*/ 1);
6694 if (__ratelimit(&_rs))
6696 "btrfs: block rsv returned %d\n", ret);
6698 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6699 BTRFS_RESERVE_NO_FLUSH);
6702 } else if (ret && block_rsv != global_rsv) {
6703 ret = block_rsv_use_bytes(global_rsv, blocksize);
6709 return ERR_PTR(-ENOSPC);
6712 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6713 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6715 block_rsv_add_bytes(block_rsv, blocksize, 0);
6716 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6720 * finds a free extent and does all the dirty work required for allocation
6721 * returns the key for the extent through ins, and a tree buffer for
6722 * the first block of the extent through buf.
6724 * returns the tree buffer or NULL.
6726 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6727 struct btrfs_root *root, u32 blocksize,
6728 u64 parent, u64 root_objectid,
6729 struct btrfs_disk_key *key, int level,
6730 u64 hint, u64 empty_size)
6732 struct btrfs_key ins;
6733 struct btrfs_block_rsv *block_rsv;
6734 struct extent_buffer *buf;
6737 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6740 block_rsv = use_block_rsv(trans, root, blocksize);
6741 if (IS_ERR(block_rsv))
6742 return ERR_CAST(block_rsv);
6744 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6745 empty_size, hint, &ins, 0);
6747 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6748 return ERR_PTR(ret);
6751 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6753 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6755 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6757 parent = ins.objectid;
6758 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6762 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6763 struct btrfs_delayed_extent_op *extent_op;
6764 extent_op = btrfs_alloc_delayed_extent_op();
6765 BUG_ON(!extent_op); /* -ENOMEM */
6767 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6769 memset(&extent_op->key, 0, sizeof(extent_op->key));
6770 extent_op->flags_to_set = flags;
6771 if (skinny_metadata)
6772 extent_op->update_key = 0;
6774 extent_op->update_key = 1;
6775 extent_op->update_flags = 1;
6776 extent_op->is_data = 0;
6778 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6780 ins.offset, parent, root_objectid,
6781 level, BTRFS_ADD_DELAYED_EXTENT,
6783 BUG_ON(ret); /* -ENOMEM */
6788 struct walk_control {
6789 u64 refs[BTRFS_MAX_LEVEL];
6790 u64 flags[BTRFS_MAX_LEVEL];
6791 struct btrfs_key update_progress;
6802 #define DROP_REFERENCE 1
6803 #define UPDATE_BACKREF 2
6805 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6806 struct btrfs_root *root,
6807 struct walk_control *wc,
6808 struct btrfs_path *path)
6816 struct btrfs_key key;
6817 struct extent_buffer *eb;
6822 if (path->slots[wc->level] < wc->reada_slot) {
6823 wc->reada_count = wc->reada_count * 2 / 3;
6824 wc->reada_count = max(wc->reada_count, 2);
6826 wc->reada_count = wc->reada_count * 3 / 2;
6827 wc->reada_count = min_t(int, wc->reada_count,
6828 BTRFS_NODEPTRS_PER_BLOCK(root));
6831 eb = path->nodes[wc->level];
6832 nritems = btrfs_header_nritems(eb);
6833 blocksize = btrfs_level_size(root, wc->level - 1);
6835 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6836 if (nread >= wc->reada_count)
6840 bytenr = btrfs_node_blockptr(eb, slot);
6841 generation = btrfs_node_ptr_generation(eb, slot);
6843 if (slot == path->slots[wc->level])
6846 if (wc->stage == UPDATE_BACKREF &&
6847 generation <= root->root_key.offset)
6850 /* We don't lock the tree block, it's OK to be racy here */
6851 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6852 wc->level - 1, 1, &refs,
6854 /* We don't care about errors in readahead. */
6859 if (wc->stage == DROP_REFERENCE) {
6863 if (wc->level == 1 &&
6864 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6866 if (!wc->update_ref ||
6867 generation <= root->root_key.offset)
6869 btrfs_node_key_to_cpu(eb, &key, slot);
6870 ret = btrfs_comp_cpu_keys(&key,
6871 &wc->update_progress);
6875 if (wc->level == 1 &&
6876 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6880 ret = readahead_tree_block(root, bytenr, blocksize,
6886 wc->reada_slot = slot;
6890 * helper to process tree block while walking down the tree.
6892 * when wc->stage == UPDATE_BACKREF, this function updates
6893 * back refs for pointers in the block.
6895 * NOTE: return value 1 means we should stop walking down.
6897 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6898 struct btrfs_root *root,
6899 struct btrfs_path *path,
6900 struct walk_control *wc, int lookup_info)
6902 int level = wc->level;
6903 struct extent_buffer *eb = path->nodes[level];
6904 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6907 if (wc->stage == UPDATE_BACKREF &&
6908 btrfs_header_owner(eb) != root->root_key.objectid)
6912 * when reference count of tree block is 1, it won't increase
6913 * again. once full backref flag is set, we never clear it.
6916 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6917 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6918 BUG_ON(!path->locks[level]);
6919 ret = btrfs_lookup_extent_info(trans, root,
6920 eb->start, level, 1,
6923 BUG_ON(ret == -ENOMEM);
6926 BUG_ON(wc->refs[level] == 0);
6929 if (wc->stage == DROP_REFERENCE) {
6930 if (wc->refs[level] > 1)
6933 if (path->locks[level] && !wc->keep_locks) {
6934 btrfs_tree_unlock_rw(eb, path->locks[level]);
6935 path->locks[level] = 0;
6940 /* wc->stage == UPDATE_BACKREF */
6941 if (!(wc->flags[level] & flag)) {
6942 BUG_ON(!path->locks[level]);
6943 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6944 BUG_ON(ret); /* -ENOMEM */
6945 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6946 BUG_ON(ret); /* -ENOMEM */
6947 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6949 BUG_ON(ret); /* -ENOMEM */
6950 wc->flags[level] |= flag;
6954 * the block is shared by multiple trees, so it's not good to
6955 * keep the tree lock
6957 if (path->locks[level] && level > 0) {
6958 btrfs_tree_unlock_rw(eb, path->locks[level]);
6959 path->locks[level] = 0;
6965 * helper to process tree block pointer.
6967 * when wc->stage == DROP_REFERENCE, this function checks
6968 * reference count of the block pointed to. if the block
6969 * is shared and we need update back refs for the subtree
6970 * rooted at the block, this function changes wc->stage to
6971 * UPDATE_BACKREF. if the block is shared and there is no
6972 * need to update back, this function drops the reference
6975 * NOTE: return value 1 means we should stop walking down.
6977 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6978 struct btrfs_root *root,
6979 struct btrfs_path *path,
6980 struct walk_control *wc, int *lookup_info)
6986 struct btrfs_key key;
6987 struct extent_buffer *next;
6988 int level = wc->level;
6992 generation = btrfs_node_ptr_generation(path->nodes[level],
6993 path->slots[level]);
6995 * if the lower level block was created before the snapshot
6996 * was created, we know there is no need to update back refs
6999 if (wc->stage == UPDATE_BACKREF &&
7000 generation <= root->root_key.offset) {
7005 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7006 blocksize = btrfs_level_size(root, level - 1);
7008 next = btrfs_find_tree_block(root, bytenr, blocksize);
7010 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7015 btrfs_tree_lock(next);
7016 btrfs_set_lock_blocking(next);
7018 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7019 &wc->refs[level - 1],
7020 &wc->flags[level - 1]);
7022 btrfs_tree_unlock(next);
7026 if (unlikely(wc->refs[level - 1] == 0)) {
7027 btrfs_err(root->fs_info, "Missing references.");
7032 if (wc->stage == DROP_REFERENCE) {
7033 if (wc->refs[level - 1] > 1) {
7035 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7038 if (!wc->update_ref ||
7039 generation <= root->root_key.offset)
7042 btrfs_node_key_to_cpu(path->nodes[level], &key,
7043 path->slots[level]);
7044 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7048 wc->stage = UPDATE_BACKREF;
7049 wc->shared_level = level - 1;
7053 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7057 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7058 btrfs_tree_unlock(next);
7059 free_extent_buffer(next);
7065 if (reada && level == 1)
7066 reada_walk_down(trans, root, wc, path);
7067 next = read_tree_block(root, bytenr, blocksize, generation);
7070 btrfs_tree_lock(next);
7071 btrfs_set_lock_blocking(next);
7075 BUG_ON(level != btrfs_header_level(next));
7076 path->nodes[level] = next;
7077 path->slots[level] = 0;
7078 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7084 wc->refs[level - 1] = 0;
7085 wc->flags[level - 1] = 0;
7086 if (wc->stage == DROP_REFERENCE) {
7087 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7088 parent = path->nodes[level]->start;
7090 BUG_ON(root->root_key.objectid !=
7091 btrfs_header_owner(path->nodes[level]));
7095 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7096 root->root_key.objectid, level - 1, 0, 0);
7097 BUG_ON(ret); /* -ENOMEM */
7099 btrfs_tree_unlock(next);
7100 free_extent_buffer(next);
7106 * helper to process tree block while walking up the tree.
7108 * when wc->stage == DROP_REFERENCE, this function drops
7109 * reference count on the block.
7111 * when wc->stage == UPDATE_BACKREF, this function changes
7112 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7113 * to UPDATE_BACKREF previously while processing the block.
7115 * NOTE: return value 1 means we should stop walking up.
7117 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7118 struct btrfs_root *root,
7119 struct btrfs_path *path,
7120 struct walk_control *wc)
7123 int level = wc->level;
7124 struct extent_buffer *eb = path->nodes[level];
7127 if (wc->stage == UPDATE_BACKREF) {
7128 BUG_ON(wc->shared_level < level);
7129 if (level < wc->shared_level)
7132 ret = find_next_key(path, level + 1, &wc->update_progress);
7136 wc->stage = DROP_REFERENCE;
7137 wc->shared_level = -1;
7138 path->slots[level] = 0;
7141 * check reference count again if the block isn't locked.
7142 * we should start walking down the tree again if reference
7145 if (!path->locks[level]) {
7147 btrfs_tree_lock(eb);
7148 btrfs_set_lock_blocking(eb);
7149 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7151 ret = btrfs_lookup_extent_info(trans, root,
7152 eb->start, level, 1,
7156 btrfs_tree_unlock_rw(eb, path->locks[level]);
7157 path->locks[level] = 0;
7160 BUG_ON(wc->refs[level] == 0);
7161 if (wc->refs[level] == 1) {
7162 btrfs_tree_unlock_rw(eb, path->locks[level]);
7163 path->locks[level] = 0;
7169 /* wc->stage == DROP_REFERENCE */
7170 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7172 if (wc->refs[level] == 1) {
7174 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7175 ret = btrfs_dec_ref(trans, root, eb, 1,
7178 ret = btrfs_dec_ref(trans, root, eb, 0,
7180 BUG_ON(ret); /* -ENOMEM */
7182 /* make block locked assertion in clean_tree_block happy */
7183 if (!path->locks[level] &&
7184 btrfs_header_generation(eb) == trans->transid) {
7185 btrfs_tree_lock(eb);
7186 btrfs_set_lock_blocking(eb);
7187 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7189 clean_tree_block(trans, root, eb);
7192 if (eb == root->node) {
7193 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7196 BUG_ON(root->root_key.objectid !=
7197 btrfs_header_owner(eb));
7199 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7200 parent = path->nodes[level + 1]->start;
7202 BUG_ON(root->root_key.objectid !=
7203 btrfs_header_owner(path->nodes[level + 1]));
7206 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7208 wc->refs[level] = 0;
7209 wc->flags[level] = 0;
7213 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7214 struct btrfs_root *root,
7215 struct btrfs_path *path,
7216 struct walk_control *wc)
7218 int level = wc->level;
7219 int lookup_info = 1;
7222 while (level >= 0) {
7223 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7230 if (path->slots[level] >=
7231 btrfs_header_nritems(path->nodes[level]))
7234 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7236 path->slots[level]++;
7245 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7246 struct btrfs_root *root,
7247 struct btrfs_path *path,
7248 struct walk_control *wc, int max_level)
7250 int level = wc->level;
7253 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7254 while (level < max_level && path->nodes[level]) {
7256 if (path->slots[level] + 1 <
7257 btrfs_header_nritems(path->nodes[level])) {
7258 path->slots[level]++;
7261 ret = walk_up_proc(trans, root, path, wc);
7265 if (path->locks[level]) {
7266 btrfs_tree_unlock_rw(path->nodes[level],
7267 path->locks[level]);
7268 path->locks[level] = 0;
7270 free_extent_buffer(path->nodes[level]);
7271 path->nodes[level] = NULL;
7279 * drop a subvolume tree.
7281 * this function traverses the tree freeing any blocks that only
7282 * referenced by the tree.
7284 * when a shared tree block is found. this function decreases its
7285 * reference count by one. if update_ref is true, this function
7286 * also make sure backrefs for the shared block and all lower level
7287 * blocks are properly updated.
7289 * If called with for_reloc == 0, may exit early with -EAGAIN
7291 int btrfs_drop_snapshot(struct btrfs_root *root,
7292 struct btrfs_block_rsv *block_rsv, int update_ref,
7295 struct btrfs_path *path;
7296 struct btrfs_trans_handle *trans;
7297 struct btrfs_root *tree_root = root->fs_info->tree_root;
7298 struct btrfs_root_item *root_item = &root->root_item;
7299 struct walk_control *wc;
7300 struct btrfs_key key;
7305 path = btrfs_alloc_path();
7311 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7313 btrfs_free_path(path);
7318 trans = btrfs_start_transaction(tree_root, 0);
7319 if (IS_ERR(trans)) {
7320 err = PTR_ERR(trans);
7325 trans->block_rsv = block_rsv;
7327 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7328 level = btrfs_header_level(root->node);
7329 path->nodes[level] = btrfs_lock_root_node(root);
7330 btrfs_set_lock_blocking(path->nodes[level]);
7331 path->slots[level] = 0;
7332 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7333 memset(&wc->update_progress, 0,
7334 sizeof(wc->update_progress));
7336 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7337 memcpy(&wc->update_progress, &key,
7338 sizeof(wc->update_progress));
7340 level = root_item->drop_level;
7342 path->lowest_level = level;
7343 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7344 path->lowest_level = 0;
7352 * unlock our path, this is safe because only this
7353 * function is allowed to delete this snapshot
7355 btrfs_unlock_up_safe(path, 0);
7357 level = btrfs_header_level(root->node);
7359 btrfs_tree_lock(path->nodes[level]);
7360 btrfs_set_lock_blocking(path->nodes[level]);
7362 ret = btrfs_lookup_extent_info(trans, root,
7363 path->nodes[level]->start,
7364 level, 1, &wc->refs[level],
7370 BUG_ON(wc->refs[level] == 0);
7372 if (level == root_item->drop_level)
7375 btrfs_tree_unlock(path->nodes[level]);
7376 WARN_ON(wc->refs[level] != 1);
7382 wc->shared_level = -1;
7383 wc->stage = DROP_REFERENCE;
7384 wc->update_ref = update_ref;
7386 wc->for_reloc = for_reloc;
7387 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7390 if (!for_reloc && btrfs_fs_closing(root->fs_info)) {
7391 pr_debug("btrfs: drop snapshot early exit\n");
7396 ret = walk_down_tree(trans, root, path, wc);
7402 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7409 BUG_ON(wc->stage != DROP_REFERENCE);
7413 if (wc->stage == DROP_REFERENCE) {
7415 btrfs_node_key(path->nodes[level],
7416 &root_item->drop_progress,
7417 path->slots[level]);
7418 root_item->drop_level = level;
7421 BUG_ON(wc->level == 0);
7422 if (btrfs_should_end_transaction(trans, tree_root)) {
7423 ret = btrfs_update_root(trans, tree_root,
7427 btrfs_abort_transaction(trans, tree_root, ret);
7432 btrfs_end_transaction_throttle(trans, tree_root);
7433 trans = btrfs_start_transaction(tree_root, 0);
7434 if (IS_ERR(trans)) {
7435 err = PTR_ERR(trans);
7439 trans->block_rsv = block_rsv;
7442 btrfs_release_path(path);
7446 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7448 btrfs_abort_transaction(trans, tree_root, ret);
7452 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7453 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7456 btrfs_abort_transaction(trans, tree_root, ret);
7459 } else if (ret > 0) {
7460 /* if we fail to delete the orphan item this time
7461 * around, it'll get picked up the next time.
7463 * The most common failure here is just -ENOENT.
7465 btrfs_del_orphan_item(trans, tree_root,
7466 root->root_key.objectid);
7470 if (root->in_radix) {
7471 btrfs_free_fs_root(tree_root->fs_info, root);
7473 free_extent_buffer(root->node);
7474 free_extent_buffer(root->commit_root);
7478 btrfs_end_transaction_throttle(trans, tree_root);
7481 btrfs_free_path(path);
7484 btrfs_std_error(root->fs_info, err);
7489 * drop subtree rooted at tree block 'node'.
7491 * NOTE: this function will unlock and release tree block 'node'
7492 * only used by relocation code
7494 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7495 struct btrfs_root *root,
7496 struct extent_buffer *node,
7497 struct extent_buffer *parent)
7499 struct btrfs_path *path;
7500 struct walk_control *wc;
7506 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7508 path = btrfs_alloc_path();
7512 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7514 btrfs_free_path(path);
7518 btrfs_assert_tree_locked(parent);
7519 parent_level = btrfs_header_level(parent);
7520 extent_buffer_get(parent);
7521 path->nodes[parent_level] = parent;
7522 path->slots[parent_level] = btrfs_header_nritems(parent);
7524 btrfs_assert_tree_locked(node);
7525 level = btrfs_header_level(node);
7526 path->nodes[level] = node;
7527 path->slots[level] = 0;
7528 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7530 wc->refs[parent_level] = 1;
7531 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7533 wc->shared_level = -1;
7534 wc->stage = DROP_REFERENCE;
7538 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7541 wret = walk_down_tree(trans, root, path, wc);
7547 wret = walk_up_tree(trans, root, path, wc, parent_level);
7555 btrfs_free_path(path);
7559 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7565 * if restripe for this chunk_type is on pick target profile and
7566 * return, otherwise do the usual balance
7568 stripped = get_restripe_target(root->fs_info, flags);
7570 return extended_to_chunk(stripped);
7573 * we add in the count of missing devices because we want
7574 * to make sure that any RAID levels on a degraded FS
7575 * continue to be honored.
7577 num_devices = root->fs_info->fs_devices->rw_devices +
7578 root->fs_info->fs_devices->missing_devices;
7580 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7581 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7582 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7584 if (num_devices == 1) {
7585 stripped |= BTRFS_BLOCK_GROUP_DUP;
7586 stripped = flags & ~stripped;
7588 /* turn raid0 into single device chunks */
7589 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7592 /* turn mirroring into duplication */
7593 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7594 BTRFS_BLOCK_GROUP_RAID10))
7595 return stripped | BTRFS_BLOCK_GROUP_DUP;
7597 /* they already had raid on here, just return */
7598 if (flags & stripped)
7601 stripped |= BTRFS_BLOCK_GROUP_DUP;
7602 stripped = flags & ~stripped;
7604 /* switch duplicated blocks with raid1 */
7605 if (flags & BTRFS_BLOCK_GROUP_DUP)
7606 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7608 /* this is drive concat, leave it alone */
7614 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7616 struct btrfs_space_info *sinfo = cache->space_info;
7618 u64 min_allocable_bytes;
7623 * We need some metadata space and system metadata space for
7624 * allocating chunks in some corner cases until we force to set
7625 * it to be readonly.
7628 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7630 min_allocable_bytes = 1 * 1024 * 1024;
7632 min_allocable_bytes = 0;
7634 spin_lock(&sinfo->lock);
7635 spin_lock(&cache->lock);
7642 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7643 cache->bytes_super - btrfs_block_group_used(&cache->item);
7645 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7646 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7647 min_allocable_bytes <= sinfo->total_bytes) {
7648 sinfo->bytes_readonly += num_bytes;
7653 spin_unlock(&cache->lock);
7654 spin_unlock(&sinfo->lock);
7658 int btrfs_set_block_group_ro(struct btrfs_root *root,
7659 struct btrfs_block_group_cache *cache)
7662 struct btrfs_trans_handle *trans;
7668 trans = btrfs_join_transaction(root);
7670 return PTR_ERR(trans);
7672 alloc_flags = update_block_group_flags(root, cache->flags);
7673 if (alloc_flags != cache->flags) {
7674 ret = do_chunk_alloc(trans, root, alloc_flags,
7680 ret = set_block_group_ro(cache, 0);
7683 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7684 ret = do_chunk_alloc(trans, root, alloc_flags,
7688 ret = set_block_group_ro(cache, 0);
7690 btrfs_end_transaction(trans, root);
7694 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7695 struct btrfs_root *root, u64 type)
7697 u64 alloc_flags = get_alloc_profile(root, type);
7698 return do_chunk_alloc(trans, root, alloc_flags,
7703 * helper to account the unused space of all the readonly block group in the
7704 * list. takes mirrors into account.
7706 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7708 struct btrfs_block_group_cache *block_group;
7712 list_for_each_entry(block_group, groups_list, list) {
7713 spin_lock(&block_group->lock);
7715 if (!block_group->ro) {
7716 spin_unlock(&block_group->lock);
7720 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7721 BTRFS_BLOCK_GROUP_RAID10 |
7722 BTRFS_BLOCK_GROUP_DUP))
7727 free_bytes += (block_group->key.offset -
7728 btrfs_block_group_used(&block_group->item)) *
7731 spin_unlock(&block_group->lock);
7738 * helper to account the unused space of all the readonly block group in the
7739 * space_info. takes mirrors into account.
7741 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7746 spin_lock(&sinfo->lock);
7748 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7749 if (!list_empty(&sinfo->block_groups[i]))
7750 free_bytes += __btrfs_get_ro_block_group_free_space(
7751 &sinfo->block_groups[i]);
7753 spin_unlock(&sinfo->lock);
7758 void btrfs_set_block_group_rw(struct btrfs_root *root,
7759 struct btrfs_block_group_cache *cache)
7761 struct btrfs_space_info *sinfo = cache->space_info;
7766 spin_lock(&sinfo->lock);
7767 spin_lock(&cache->lock);
7768 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7769 cache->bytes_super - btrfs_block_group_used(&cache->item);
7770 sinfo->bytes_readonly -= num_bytes;
7772 spin_unlock(&cache->lock);
7773 spin_unlock(&sinfo->lock);
7777 * checks to see if its even possible to relocate this block group.
7779 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7780 * ok to go ahead and try.
7782 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7784 struct btrfs_block_group_cache *block_group;
7785 struct btrfs_space_info *space_info;
7786 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7787 struct btrfs_device *device;
7796 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7798 /* odd, couldn't find the block group, leave it alone */
7802 min_free = btrfs_block_group_used(&block_group->item);
7804 /* no bytes used, we're good */
7808 space_info = block_group->space_info;
7809 spin_lock(&space_info->lock);
7811 full = space_info->full;
7814 * if this is the last block group we have in this space, we can't
7815 * relocate it unless we're able to allocate a new chunk below.
7817 * Otherwise, we need to make sure we have room in the space to handle
7818 * all of the extents from this block group. If we can, we're good
7820 if ((space_info->total_bytes != block_group->key.offset) &&
7821 (space_info->bytes_used + space_info->bytes_reserved +
7822 space_info->bytes_pinned + space_info->bytes_readonly +
7823 min_free < space_info->total_bytes)) {
7824 spin_unlock(&space_info->lock);
7827 spin_unlock(&space_info->lock);
7830 * ok we don't have enough space, but maybe we have free space on our
7831 * devices to allocate new chunks for relocation, so loop through our
7832 * alloc devices and guess if we have enough space. if this block
7833 * group is going to be restriped, run checks against the target
7834 * profile instead of the current one.
7846 target = get_restripe_target(root->fs_info, block_group->flags);
7848 index = __get_raid_index(extended_to_chunk(target));
7851 * this is just a balance, so if we were marked as full
7852 * we know there is no space for a new chunk
7857 index = get_block_group_index(block_group);
7860 if (index == BTRFS_RAID_RAID10) {
7864 } else if (index == BTRFS_RAID_RAID1) {
7866 } else if (index == BTRFS_RAID_DUP) {
7869 } else if (index == BTRFS_RAID_RAID0) {
7870 dev_min = fs_devices->rw_devices;
7871 do_div(min_free, dev_min);
7874 mutex_lock(&root->fs_info->chunk_mutex);
7875 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7879 * check to make sure we can actually find a chunk with enough
7880 * space to fit our block group in.
7882 if (device->total_bytes > device->bytes_used + min_free &&
7883 !device->is_tgtdev_for_dev_replace) {
7884 ret = find_free_dev_extent(device, min_free,
7889 if (dev_nr >= dev_min)
7895 mutex_unlock(&root->fs_info->chunk_mutex);
7897 btrfs_put_block_group(block_group);
7901 static int find_first_block_group(struct btrfs_root *root,
7902 struct btrfs_path *path, struct btrfs_key *key)
7905 struct btrfs_key found_key;
7906 struct extent_buffer *leaf;
7909 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7914 slot = path->slots[0];
7915 leaf = path->nodes[0];
7916 if (slot >= btrfs_header_nritems(leaf)) {
7917 ret = btrfs_next_leaf(root, path);
7924 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7926 if (found_key.objectid >= key->objectid &&
7927 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7937 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7939 struct btrfs_block_group_cache *block_group;
7943 struct inode *inode;
7945 block_group = btrfs_lookup_first_block_group(info, last);
7946 while (block_group) {
7947 spin_lock(&block_group->lock);
7948 if (block_group->iref)
7950 spin_unlock(&block_group->lock);
7951 block_group = next_block_group(info->tree_root,
7961 inode = block_group->inode;
7962 block_group->iref = 0;
7963 block_group->inode = NULL;
7964 spin_unlock(&block_group->lock);
7966 last = block_group->key.objectid + block_group->key.offset;
7967 btrfs_put_block_group(block_group);
7971 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7973 struct btrfs_block_group_cache *block_group;
7974 struct btrfs_space_info *space_info;
7975 struct btrfs_caching_control *caching_ctl;
7978 down_write(&info->extent_commit_sem);
7979 while (!list_empty(&info->caching_block_groups)) {
7980 caching_ctl = list_entry(info->caching_block_groups.next,
7981 struct btrfs_caching_control, list);
7982 list_del(&caching_ctl->list);
7983 put_caching_control(caching_ctl);
7985 up_write(&info->extent_commit_sem);
7987 spin_lock(&info->block_group_cache_lock);
7988 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7989 block_group = rb_entry(n, struct btrfs_block_group_cache,
7991 rb_erase(&block_group->cache_node,
7992 &info->block_group_cache_tree);
7993 spin_unlock(&info->block_group_cache_lock);
7995 down_write(&block_group->space_info->groups_sem);
7996 list_del(&block_group->list);
7997 up_write(&block_group->space_info->groups_sem);
7999 if (block_group->cached == BTRFS_CACHE_STARTED)
8000 wait_block_group_cache_done(block_group);
8003 * We haven't cached this block group, which means we could
8004 * possibly have excluded extents on this block group.
8006 if (block_group->cached == BTRFS_CACHE_NO)
8007 free_excluded_extents(info->extent_root, block_group);
8009 btrfs_remove_free_space_cache(block_group);
8010 btrfs_put_block_group(block_group);
8012 spin_lock(&info->block_group_cache_lock);
8014 spin_unlock(&info->block_group_cache_lock);
8016 /* now that all the block groups are freed, go through and
8017 * free all the space_info structs. This is only called during
8018 * the final stages of unmount, and so we know nobody is
8019 * using them. We call synchronize_rcu() once before we start,
8020 * just to be on the safe side.
8024 release_global_block_rsv(info);
8026 while(!list_empty(&info->space_info)) {
8027 space_info = list_entry(info->space_info.next,
8028 struct btrfs_space_info,
8030 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8031 if (space_info->bytes_pinned > 0 ||
8032 space_info->bytes_reserved > 0 ||
8033 space_info->bytes_may_use > 0) {
8035 dump_space_info(space_info, 0, 0);
8038 list_del(&space_info->list);
8044 static void __link_block_group(struct btrfs_space_info *space_info,
8045 struct btrfs_block_group_cache *cache)
8047 int index = get_block_group_index(cache);
8049 down_write(&space_info->groups_sem);
8050 list_add_tail(&cache->list, &space_info->block_groups[index]);
8051 up_write(&space_info->groups_sem);
8054 int btrfs_read_block_groups(struct btrfs_root *root)
8056 struct btrfs_path *path;
8058 struct btrfs_block_group_cache *cache;
8059 struct btrfs_fs_info *info = root->fs_info;
8060 struct btrfs_space_info *space_info;
8061 struct btrfs_key key;
8062 struct btrfs_key found_key;
8063 struct extent_buffer *leaf;
8067 root = info->extent_root;
8070 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8071 path = btrfs_alloc_path();
8076 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8077 if (btrfs_test_opt(root, SPACE_CACHE) &&
8078 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8080 if (btrfs_test_opt(root, CLEAR_CACHE))
8084 ret = find_first_block_group(root, path, &key);
8089 leaf = path->nodes[0];
8090 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8091 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8096 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8098 if (!cache->free_space_ctl) {
8104 atomic_set(&cache->count, 1);
8105 spin_lock_init(&cache->lock);
8106 cache->fs_info = info;
8107 INIT_LIST_HEAD(&cache->list);
8108 INIT_LIST_HEAD(&cache->cluster_list);
8112 * When we mount with old space cache, we need to
8113 * set BTRFS_DC_CLEAR and set dirty flag.
8115 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8116 * truncate the old free space cache inode and
8118 * b) Setting 'dirty flag' makes sure that we flush
8119 * the new space cache info onto disk.
8121 cache->disk_cache_state = BTRFS_DC_CLEAR;
8122 if (btrfs_test_opt(root, SPACE_CACHE))
8126 read_extent_buffer(leaf, &cache->item,
8127 btrfs_item_ptr_offset(leaf, path->slots[0]),
8128 sizeof(cache->item));
8129 memcpy(&cache->key, &found_key, sizeof(found_key));
8131 key.objectid = found_key.objectid + found_key.offset;
8132 btrfs_release_path(path);
8133 cache->flags = btrfs_block_group_flags(&cache->item);
8134 cache->sectorsize = root->sectorsize;
8135 cache->full_stripe_len = btrfs_full_stripe_len(root,
8136 &root->fs_info->mapping_tree,
8137 found_key.objectid);
8138 btrfs_init_free_space_ctl(cache);
8141 * We need to exclude the super stripes now so that the space
8142 * info has super bytes accounted for, otherwise we'll think
8143 * we have more space than we actually do.
8145 ret = exclude_super_stripes(root, cache);
8148 * We may have excluded something, so call this just in
8151 free_excluded_extents(root, cache);
8152 kfree(cache->free_space_ctl);
8158 * check for two cases, either we are full, and therefore
8159 * don't need to bother with the caching work since we won't
8160 * find any space, or we are empty, and we can just add all
8161 * the space in and be done with it. This saves us _alot_ of
8162 * time, particularly in the full case.
8164 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8165 cache->last_byte_to_unpin = (u64)-1;
8166 cache->cached = BTRFS_CACHE_FINISHED;
8167 free_excluded_extents(root, cache);
8168 } else if (btrfs_block_group_used(&cache->item) == 0) {
8169 cache->last_byte_to_unpin = (u64)-1;
8170 cache->cached = BTRFS_CACHE_FINISHED;
8171 add_new_free_space(cache, root->fs_info,
8173 found_key.objectid +
8175 free_excluded_extents(root, cache);
8178 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8180 btrfs_remove_free_space_cache(cache);
8181 btrfs_put_block_group(cache);
8185 ret = update_space_info(info, cache->flags, found_key.offset,
8186 btrfs_block_group_used(&cache->item),
8189 btrfs_remove_free_space_cache(cache);
8190 spin_lock(&info->block_group_cache_lock);
8191 rb_erase(&cache->cache_node,
8192 &info->block_group_cache_tree);
8193 spin_unlock(&info->block_group_cache_lock);
8194 btrfs_put_block_group(cache);
8198 cache->space_info = space_info;
8199 spin_lock(&cache->space_info->lock);
8200 cache->space_info->bytes_readonly += cache->bytes_super;
8201 spin_unlock(&cache->space_info->lock);
8203 __link_block_group(space_info, cache);
8205 set_avail_alloc_bits(root->fs_info, cache->flags);
8206 if (btrfs_chunk_readonly(root, cache->key.objectid))
8207 set_block_group_ro(cache, 1);
8210 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8211 if (!(get_alloc_profile(root, space_info->flags) &
8212 (BTRFS_BLOCK_GROUP_RAID10 |
8213 BTRFS_BLOCK_GROUP_RAID1 |
8214 BTRFS_BLOCK_GROUP_RAID5 |
8215 BTRFS_BLOCK_GROUP_RAID6 |
8216 BTRFS_BLOCK_GROUP_DUP)))
8219 * avoid allocating from un-mirrored block group if there are
8220 * mirrored block groups.
8222 list_for_each_entry(cache, &space_info->block_groups[3], list)
8223 set_block_group_ro(cache, 1);
8224 list_for_each_entry(cache, &space_info->block_groups[4], list)
8225 set_block_group_ro(cache, 1);
8228 init_global_block_rsv(info);
8231 btrfs_free_path(path);
8235 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8236 struct btrfs_root *root)
8238 struct btrfs_block_group_cache *block_group, *tmp;
8239 struct btrfs_root *extent_root = root->fs_info->extent_root;
8240 struct btrfs_block_group_item item;
8241 struct btrfs_key key;
8244 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8246 list_del_init(&block_group->new_bg_list);
8251 spin_lock(&block_group->lock);
8252 memcpy(&item, &block_group->item, sizeof(item));
8253 memcpy(&key, &block_group->key, sizeof(key));
8254 spin_unlock(&block_group->lock);
8256 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8259 btrfs_abort_transaction(trans, extent_root, ret);
8263 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8264 struct btrfs_root *root, u64 bytes_used,
8265 u64 type, u64 chunk_objectid, u64 chunk_offset,
8269 struct btrfs_root *extent_root;
8270 struct btrfs_block_group_cache *cache;
8272 extent_root = root->fs_info->extent_root;
8274 root->fs_info->last_trans_log_full_commit = trans->transid;
8276 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8279 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8281 if (!cache->free_space_ctl) {
8286 cache->key.objectid = chunk_offset;
8287 cache->key.offset = size;
8288 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8289 cache->sectorsize = root->sectorsize;
8290 cache->fs_info = root->fs_info;
8291 cache->full_stripe_len = btrfs_full_stripe_len(root,
8292 &root->fs_info->mapping_tree,
8295 atomic_set(&cache->count, 1);
8296 spin_lock_init(&cache->lock);
8297 INIT_LIST_HEAD(&cache->list);
8298 INIT_LIST_HEAD(&cache->cluster_list);
8299 INIT_LIST_HEAD(&cache->new_bg_list);
8301 btrfs_init_free_space_ctl(cache);
8303 btrfs_set_block_group_used(&cache->item, bytes_used);
8304 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8305 cache->flags = type;
8306 btrfs_set_block_group_flags(&cache->item, type);
8308 cache->last_byte_to_unpin = (u64)-1;
8309 cache->cached = BTRFS_CACHE_FINISHED;
8310 ret = exclude_super_stripes(root, cache);
8313 * We may have excluded something, so call this just in
8316 free_excluded_extents(root, cache);
8317 kfree(cache->free_space_ctl);
8322 add_new_free_space(cache, root->fs_info, chunk_offset,
8323 chunk_offset + size);
8325 free_excluded_extents(root, cache);
8327 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8329 btrfs_remove_free_space_cache(cache);
8330 btrfs_put_block_group(cache);
8334 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8335 &cache->space_info);
8337 btrfs_remove_free_space_cache(cache);
8338 spin_lock(&root->fs_info->block_group_cache_lock);
8339 rb_erase(&cache->cache_node,
8340 &root->fs_info->block_group_cache_tree);
8341 spin_unlock(&root->fs_info->block_group_cache_lock);
8342 btrfs_put_block_group(cache);
8345 update_global_block_rsv(root->fs_info);
8347 spin_lock(&cache->space_info->lock);
8348 cache->space_info->bytes_readonly += cache->bytes_super;
8349 spin_unlock(&cache->space_info->lock);
8351 __link_block_group(cache->space_info, cache);
8353 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8355 set_avail_alloc_bits(extent_root->fs_info, type);
8360 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8362 u64 extra_flags = chunk_to_extended(flags) &
8363 BTRFS_EXTENDED_PROFILE_MASK;
8365 write_seqlock(&fs_info->profiles_lock);
8366 if (flags & BTRFS_BLOCK_GROUP_DATA)
8367 fs_info->avail_data_alloc_bits &= ~extra_flags;
8368 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8369 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8370 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8371 fs_info->avail_system_alloc_bits &= ~extra_flags;
8372 write_sequnlock(&fs_info->profiles_lock);
8375 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8376 struct btrfs_root *root, u64 group_start)
8378 struct btrfs_path *path;
8379 struct btrfs_block_group_cache *block_group;
8380 struct btrfs_free_cluster *cluster;
8381 struct btrfs_root *tree_root = root->fs_info->tree_root;
8382 struct btrfs_key key;
8383 struct inode *inode;
8388 root = root->fs_info->extent_root;
8390 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8391 BUG_ON(!block_group);
8392 BUG_ON(!block_group->ro);
8395 * Free the reserved super bytes from this block group before
8398 free_excluded_extents(root, block_group);
8400 memcpy(&key, &block_group->key, sizeof(key));
8401 index = get_block_group_index(block_group);
8402 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8403 BTRFS_BLOCK_GROUP_RAID1 |
8404 BTRFS_BLOCK_GROUP_RAID10))
8409 /* make sure this block group isn't part of an allocation cluster */
8410 cluster = &root->fs_info->data_alloc_cluster;
8411 spin_lock(&cluster->refill_lock);
8412 btrfs_return_cluster_to_free_space(block_group, cluster);
8413 spin_unlock(&cluster->refill_lock);
8416 * make sure this block group isn't part of a metadata
8417 * allocation cluster
8419 cluster = &root->fs_info->meta_alloc_cluster;
8420 spin_lock(&cluster->refill_lock);
8421 btrfs_return_cluster_to_free_space(block_group, cluster);
8422 spin_unlock(&cluster->refill_lock);
8424 path = btrfs_alloc_path();
8430 inode = lookup_free_space_inode(tree_root, block_group, path);
8431 if (!IS_ERR(inode)) {
8432 ret = btrfs_orphan_add(trans, inode);
8434 btrfs_add_delayed_iput(inode);
8438 /* One for the block groups ref */
8439 spin_lock(&block_group->lock);
8440 if (block_group->iref) {
8441 block_group->iref = 0;
8442 block_group->inode = NULL;
8443 spin_unlock(&block_group->lock);
8446 spin_unlock(&block_group->lock);
8448 /* One for our lookup ref */
8449 btrfs_add_delayed_iput(inode);
8452 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8453 key.offset = block_group->key.objectid;
8456 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8460 btrfs_release_path(path);
8462 ret = btrfs_del_item(trans, tree_root, path);
8465 btrfs_release_path(path);
8468 spin_lock(&root->fs_info->block_group_cache_lock);
8469 rb_erase(&block_group->cache_node,
8470 &root->fs_info->block_group_cache_tree);
8472 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8473 root->fs_info->first_logical_byte = (u64)-1;
8474 spin_unlock(&root->fs_info->block_group_cache_lock);
8476 down_write(&block_group->space_info->groups_sem);
8478 * we must use list_del_init so people can check to see if they
8479 * are still on the list after taking the semaphore
8481 list_del_init(&block_group->list);
8482 if (list_empty(&block_group->space_info->block_groups[index]))
8483 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8484 up_write(&block_group->space_info->groups_sem);
8486 if (block_group->cached == BTRFS_CACHE_STARTED)
8487 wait_block_group_cache_done(block_group);
8489 btrfs_remove_free_space_cache(block_group);
8491 spin_lock(&block_group->space_info->lock);
8492 block_group->space_info->total_bytes -= block_group->key.offset;
8493 block_group->space_info->bytes_readonly -= block_group->key.offset;
8494 block_group->space_info->disk_total -= block_group->key.offset * factor;
8495 spin_unlock(&block_group->space_info->lock);
8497 memcpy(&key, &block_group->key, sizeof(key));
8499 btrfs_clear_space_info_full(root->fs_info);
8501 btrfs_put_block_group(block_group);
8502 btrfs_put_block_group(block_group);
8504 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8510 ret = btrfs_del_item(trans, root, path);
8512 btrfs_free_path(path);
8516 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8518 struct btrfs_space_info *space_info;
8519 struct btrfs_super_block *disk_super;
8525 disk_super = fs_info->super_copy;
8526 if (!btrfs_super_root(disk_super))
8529 features = btrfs_super_incompat_flags(disk_super);
8530 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8533 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8534 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8539 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8540 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8542 flags = BTRFS_BLOCK_GROUP_METADATA;
8543 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8547 flags = BTRFS_BLOCK_GROUP_DATA;
8548 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8554 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8556 return unpin_extent_range(root, start, end);
8559 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8560 u64 num_bytes, u64 *actual_bytes)
8562 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8565 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8567 struct btrfs_fs_info *fs_info = root->fs_info;
8568 struct btrfs_block_group_cache *cache = NULL;
8573 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8577 * try to trim all FS space, our block group may start from non-zero.
8579 if (range->len == total_bytes)
8580 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8582 cache = btrfs_lookup_block_group(fs_info, range->start);
8585 if (cache->key.objectid >= (range->start + range->len)) {
8586 btrfs_put_block_group(cache);
8590 start = max(range->start, cache->key.objectid);
8591 end = min(range->start + range->len,
8592 cache->key.objectid + cache->key.offset);
8594 if (end - start >= range->minlen) {
8595 if (!block_group_cache_done(cache)) {
8596 ret = cache_block_group(cache, 0);
8598 wait_block_group_cache_done(cache);
8600 ret = btrfs_trim_block_group(cache,
8606 trimmed += group_trimmed;
8608 btrfs_put_block_group(cache);
8613 cache = next_block_group(fs_info->tree_root, cache);
8616 range->len = trimmed;