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>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
341 struct btrfs_fs_info *info, u64 start, u64 end)
343 u64 extent_start, extent_end, size, total_added = 0;
346 while (start < end) {
347 ret = find_first_extent_bit(info->pinned_extents, start,
348 &extent_start, &extent_end,
349 EXTENT_DIRTY | EXTENT_UPTODATE,
354 if (extent_start <= start) {
355 start = extent_end + 1;
356 } else if (extent_start > start && extent_start < end) {
357 size = extent_start - start;
359 ret = btrfs_add_free_space(block_group, start,
361 BUG_ON(ret); /* -ENOMEM or logic error */
362 start = extent_end + 1;
371 ret = btrfs_add_free_space(block_group, start, size);
372 BUG_ON(ret); /* -ENOMEM or logic error */
378 static noinline void caching_thread(struct btrfs_work *work)
380 struct btrfs_block_group_cache *block_group;
381 struct btrfs_fs_info *fs_info;
382 struct btrfs_caching_control *caching_ctl;
383 struct btrfs_root *extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
392 caching_ctl = container_of(work, struct btrfs_caching_control, work);
393 block_group = caching_ctl->block_group;
394 fs_info = block_group->fs_info;
395 extent_root = fs_info->extent_root;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path->skip_locking = 1;
410 path->search_commit_root = 1;
415 key.type = BTRFS_EXTENT_ITEM_KEY;
417 mutex_lock(&caching_ctl->mutex);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info->commit_root_sem);
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
430 if (btrfs_fs_closing(fs_info) > 1) {
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 ret = find_next_key(path, 0, &key);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info->commit_root_sem)) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->commit_root_sem);
447 mutex_unlock(&caching_ctl->mutex);
452 ret = btrfs_next_leaf(extent_root, path);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
462 if (key.objectid < last) {
465 key.type = BTRFS_EXTENT_ITEM_KEY;
467 caching_ctl->progress = last;
468 btrfs_release_path(path);
472 if (key.objectid < block_group->key.objectid) {
477 if (key.objectid >= block_group->key.objectid +
478 block_group->key.offset)
481 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
482 key.type == BTRFS_METADATA_ITEM_KEY) {
483 total_found += add_new_free_space(block_group,
486 if (key.type == BTRFS_METADATA_ITEM_KEY)
487 last = key.objectid +
488 fs_info->tree_root->nodesize;
490 last = key.objectid + key.offset;
492 if (total_found > (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl->wait);
501 total_found += add_new_free_space(block_group, fs_info, last,
502 block_group->key.objectid +
503 block_group->key.offset);
504 caching_ctl->progress = (u64)-1;
506 spin_lock(&block_group->lock);
507 block_group->caching_ctl = NULL;
508 block_group->cached = BTRFS_CACHE_FINISHED;
509 spin_unlock(&block_group->lock);
512 btrfs_free_path(path);
513 up_read(&fs_info->commit_root_sem);
515 free_excluded_extents(extent_root, block_group);
517 mutex_unlock(&caching_ctl->mutex);
520 spin_lock(&block_group->lock);
521 block_group->caching_ctl = NULL;
522 block_group->cached = BTRFS_CACHE_ERROR;
523 spin_unlock(&block_group->lock);
525 wake_up(&caching_ctl->wait);
527 put_caching_control(caching_ctl);
528 btrfs_put_block_group(block_group);
531 static int cache_block_group(struct btrfs_block_group_cache *cache,
535 struct btrfs_fs_info *fs_info = cache->fs_info;
536 struct btrfs_caching_control *caching_ctl;
539 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
543 INIT_LIST_HEAD(&caching_ctl->list);
544 mutex_init(&caching_ctl->mutex);
545 init_waitqueue_head(&caching_ctl->wait);
546 caching_ctl->block_group = cache;
547 caching_ctl->progress = cache->key.objectid;
548 atomic_set(&caching_ctl->count, 1);
549 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
550 caching_thread, NULL, NULL);
552 spin_lock(&cache->lock);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 mutex_lock(&caching_ctl->mutex);
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 caching_ctl->progress = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
606 cache->has_caching_ctl = 1;
609 spin_unlock(&cache->lock);
610 mutex_unlock(&caching_ctl->mutex);
612 wake_up(&caching_ctl->wait);
614 put_caching_control(caching_ctl);
615 free_excluded_extents(fs_info->extent_root, cache);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache->lock);
624 if (load_cache_only) {
625 cache->caching_ctl = NULL;
626 cache->cached = BTRFS_CACHE_NO;
628 cache->cached = BTRFS_CACHE_STARTED;
629 cache->has_caching_ctl = 1;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_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,
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
821 BUG_ON(num_refs == 0);
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
839 btrfs_release_path(path);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
854 BUG_ON(num_refs == 0);
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
860 spin_unlock(&delayed_refs->lock);
862 WARN_ON(num_refs == 0);
866 *flags = extent_flags;
868 btrfs_free_path(path);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1003 if (owner == (u64)-1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1025 btrfs_release_path(path);
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1035 BUG_ON(ret); /* Corruption */
1037 btrfs_extend_item(root, path, new_size);
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 btrfs_mark_buffer_dirty(leaf);
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1100 u64 owner, u64 offset)
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1170 btrfs_release_path(path);
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1195 key.objectid = bytenr;
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1204 size = sizeof(struct btrfs_extent_data_ref);
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1211 leaf = path->nodes[0];
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1231 btrfs_release_path(path);
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 if (ret && ret != -EEXIST)
1238 leaf = path->nodes[0];
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1254 btrfs_mark_buffer_dirty(leaf);
1257 btrfs_release_path(path);
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1298 if (num_refs == 0) {
1299 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_path *path,
1320 struct btrfs_extent_inline_ref *iref)
1322 struct btrfs_key key;
1323 struct extent_buffer *leaf;
1324 struct btrfs_extent_data_ref *ref1;
1325 struct btrfs_shared_data_ref *ref2;
1328 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1332 BTRFS_EXTENT_DATA_REF_KEY) {
1333 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1334 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1337 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1341 struct btrfs_extent_data_ref);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1344 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_shared_data_ref);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1349 struct btrfs_extent_ref_v0 *ref0;
1350 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_ref_v0);
1352 num_refs = btrfs_ref_count_v0(leaf, ref0);
1360 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root,
1362 struct btrfs_path *path,
1363 u64 bytenr, u64 parent,
1366 struct btrfs_key key;
1369 key.objectid = bytenr;
1371 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1372 key.offset = parent;
1374 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1375 key.offset = root_objectid;
1378 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1382 if (ret == -ENOENT && parent) {
1383 btrfs_release_path(path);
1384 key.type = BTRFS_EXTENT_REF_V0_KEY;
1385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1393 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root,
1395 struct btrfs_path *path,
1396 u64 bytenr, u64 parent,
1399 struct btrfs_key key;
1402 key.objectid = bytenr;
1404 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1405 key.offset = parent;
1407 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1408 key.offset = root_objectid;
1411 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1412 btrfs_release_path(path);
1416 static inline int extent_ref_type(u64 parent, u64 owner)
1419 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421 type = BTRFS_SHARED_BLOCK_REF_KEY;
1423 type = BTRFS_TREE_BLOCK_REF_KEY;
1426 type = BTRFS_SHARED_DATA_REF_KEY;
1428 type = BTRFS_EXTENT_DATA_REF_KEY;
1433 static int find_next_key(struct btrfs_path *path, int level,
1434 struct btrfs_key *key)
1437 for (; level < BTRFS_MAX_LEVEL; level++) {
1438 if (!path->nodes[level])
1440 if (path->slots[level] + 1 >=
1441 btrfs_header_nritems(path->nodes[level]))
1444 btrfs_item_key_to_cpu(path->nodes[level], key,
1445 path->slots[level] + 1);
1447 btrfs_node_key_to_cpu(path->nodes[level], key,
1448 path->slots[level] + 1);
1455 * look for inline back ref. if back ref is found, *ref_ret is set
1456 * to the address of inline back ref, and 0 is returned.
1458 * if back ref isn't found, *ref_ret is set to the address where it
1459 * should be inserted, and -ENOENT is returned.
1461 * if insert is true and there are too many inline back refs, the path
1462 * points to the extent item, and -EAGAIN is returned.
1464 * NOTE: inline back refs are ordered in the same way that back ref
1465 * items in the tree are ordered.
1467 static noinline_for_stack
1468 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1469 struct btrfs_root *root,
1470 struct btrfs_path *path,
1471 struct btrfs_extent_inline_ref **ref_ret,
1472 u64 bytenr, u64 num_bytes,
1473 u64 parent, u64 root_objectid,
1474 u64 owner, u64 offset, int insert)
1476 struct btrfs_key key;
1477 struct extent_buffer *leaf;
1478 struct btrfs_extent_item *ei;
1479 struct btrfs_extent_inline_ref *iref;
1489 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1492 key.objectid = bytenr;
1493 key.type = BTRFS_EXTENT_ITEM_KEY;
1494 key.offset = num_bytes;
1496 want = extent_ref_type(parent, owner);
1498 extra_size = btrfs_extent_inline_ref_size(want);
1499 path->keep_locks = 1;
1504 * Owner is our parent level, so we can just add one to get the level
1505 * for the block we are interested in.
1507 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1508 key.type = BTRFS_METADATA_ITEM_KEY;
1513 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1520 * We may be a newly converted file system which still has the old fat
1521 * extent entries for metadata, so try and see if we have one of those.
1523 if (ret > 0 && skinny_metadata) {
1524 skinny_metadata = false;
1525 if (path->slots[0]) {
1527 btrfs_item_key_to_cpu(path->nodes[0], &key,
1529 if (key.objectid == bytenr &&
1530 key.type == BTRFS_EXTENT_ITEM_KEY &&
1531 key.offset == num_bytes)
1535 key.objectid = bytenr;
1536 key.type = BTRFS_EXTENT_ITEM_KEY;
1537 key.offset = num_bytes;
1538 btrfs_release_path(path);
1543 if (ret && !insert) {
1546 } else if (WARN_ON(ret)) {
1551 leaf = path->nodes[0];
1552 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1553 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1554 if (item_size < sizeof(*ei)) {
1559 ret = convert_extent_item_v0(trans, root, path, owner,
1565 leaf = path->nodes[0];
1566 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1569 BUG_ON(item_size < sizeof(*ei));
1571 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1572 flags = btrfs_extent_flags(leaf, ei);
1574 ptr = (unsigned long)(ei + 1);
1575 end = (unsigned long)ei + item_size;
1577 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1578 ptr += sizeof(struct btrfs_tree_block_info);
1588 iref = (struct btrfs_extent_inline_ref *)ptr;
1589 type = btrfs_extent_inline_ref_type(leaf, iref);
1593 ptr += btrfs_extent_inline_ref_size(type);
1597 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1598 struct btrfs_extent_data_ref *dref;
1599 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1600 if (match_extent_data_ref(leaf, dref, root_objectid,
1605 if (hash_extent_data_ref_item(leaf, dref) <
1606 hash_extent_data_ref(root_objectid, owner, offset))
1610 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612 if (parent == ref_offset) {
1616 if (ref_offset < parent)
1619 if (root_objectid == ref_offset) {
1623 if (ref_offset < root_objectid)
1627 ptr += btrfs_extent_inline_ref_size(type);
1629 if (err == -ENOENT && insert) {
1630 if (item_size + extra_size >=
1631 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1636 * To add new inline back ref, we have to make sure
1637 * there is no corresponding back ref item.
1638 * For simplicity, we just do not add new inline back
1639 * ref if there is any kind of item for this block
1641 if (find_next_key(path, 0, &key) == 0 &&
1642 key.objectid == bytenr &&
1643 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1648 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1651 path->keep_locks = 0;
1652 btrfs_unlock_up_safe(path, 1);
1658 * helper to add new inline back ref
1660 static noinline_for_stack
1661 void setup_inline_extent_backref(struct btrfs_root *root,
1662 struct btrfs_path *path,
1663 struct btrfs_extent_inline_ref *iref,
1664 u64 parent, u64 root_objectid,
1665 u64 owner, u64 offset, int refs_to_add,
1666 struct btrfs_delayed_extent_op *extent_op)
1668 struct extent_buffer *leaf;
1669 struct btrfs_extent_item *ei;
1672 unsigned long item_offset;
1677 leaf = path->nodes[0];
1678 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1679 item_offset = (unsigned long)iref - (unsigned long)ei;
1681 type = extent_ref_type(parent, owner);
1682 size = btrfs_extent_inline_ref_size(type);
1684 btrfs_extend_item(root, path, size);
1686 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1687 refs = btrfs_extent_refs(leaf, ei);
1688 refs += refs_to_add;
1689 btrfs_set_extent_refs(leaf, ei, refs);
1691 __run_delayed_extent_op(extent_op, leaf, ei);
1693 ptr = (unsigned long)ei + item_offset;
1694 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1695 if (ptr < end - size)
1696 memmove_extent_buffer(leaf, ptr + size, ptr,
1699 iref = (struct btrfs_extent_inline_ref *)ptr;
1700 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1701 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1702 struct btrfs_extent_data_ref *dref;
1703 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1704 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1705 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1706 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1707 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1708 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1709 struct btrfs_shared_data_ref *sref;
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1712 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1713 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1718 btrfs_mark_buffer_dirty(leaf);
1721 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1722 struct btrfs_root *root,
1723 struct btrfs_path *path,
1724 struct btrfs_extent_inline_ref **ref_ret,
1725 u64 bytenr, u64 num_bytes, u64 parent,
1726 u64 root_objectid, u64 owner, u64 offset)
1730 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1731 bytenr, num_bytes, parent,
1732 root_objectid, owner, offset, 0);
1736 btrfs_release_path(path);
1739 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1740 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1744 root_objectid, owner, offset);
1750 * helper to update/remove inline back ref
1752 static noinline_for_stack
1753 void update_inline_extent_backref(struct btrfs_root *root,
1754 struct btrfs_path *path,
1755 struct btrfs_extent_inline_ref *iref,
1757 struct btrfs_delayed_extent_op *extent_op,
1760 struct extent_buffer *leaf;
1761 struct btrfs_extent_item *ei;
1762 struct btrfs_extent_data_ref *dref = NULL;
1763 struct btrfs_shared_data_ref *sref = NULL;
1771 leaf = path->nodes[0];
1772 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1773 refs = btrfs_extent_refs(leaf, ei);
1774 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1775 refs += refs_to_mod;
1776 btrfs_set_extent_refs(leaf, ei, refs);
1778 __run_delayed_extent_op(extent_op, leaf, ei);
1780 type = btrfs_extent_inline_ref_type(leaf, iref);
1782 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1783 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1784 refs = btrfs_extent_data_ref_count(leaf, dref);
1785 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1786 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1787 refs = btrfs_shared_data_ref_count(leaf, sref);
1790 BUG_ON(refs_to_mod != -1);
1793 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1794 refs += refs_to_mod;
1797 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1798 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803 size = btrfs_extent_inline_ref_size(type);
1804 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1805 ptr = (unsigned long)iref;
1806 end = (unsigned long)ei + item_size;
1807 if (ptr + size < end)
1808 memmove_extent_buffer(leaf, ptr, ptr + size,
1811 btrfs_truncate_item(root, path, item_size, 1);
1813 btrfs_mark_buffer_dirty(leaf);
1816 static noinline_for_stack
1817 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1818 struct btrfs_root *root,
1819 struct btrfs_path *path,
1820 u64 bytenr, u64 num_bytes, u64 parent,
1821 u64 root_objectid, u64 owner,
1822 u64 offset, int refs_to_add,
1823 struct btrfs_delayed_extent_op *extent_op)
1825 struct btrfs_extent_inline_ref *iref;
1828 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1829 bytenr, num_bytes, parent,
1830 root_objectid, owner, offset, 1);
1832 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1833 update_inline_extent_backref(root, path, iref,
1834 refs_to_add, extent_op, NULL);
1835 } else if (ret == -ENOENT) {
1836 setup_inline_extent_backref(root, path, iref, parent,
1837 root_objectid, owner, offset,
1838 refs_to_add, extent_op);
1844 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1845 struct btrfs_root *root,
1846 struct btrfs_path *path,
1847 u64 bytenr, u64 parent, u64 root_objectid,
1848 u64 owner, u64 offset, int refs_to_add)
1851 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1852 BUG_ON(refs_to_add != 1);
1853 ret = insert_tree_block_ref(trans, root, path, bytenr,
1854 parent, root_objectid);
1856 ret = insert_extent_data_ref(trans, root, path, bytenr,
1857 parent, root_objectid,
1858 owner, offset, refs_to_add);
1863 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1864 struct btrfs_root *root,
1865 struct btrfs_path *path,
1866 struct btrfs_extent_inline_ref *iref,
1867 int refs_to_drop, int is_data, int *last_ref)
1871 BUG_ON(!is_data && refs_to_drop != 1);
1873 update_inline_extent_backref(root, path, iref,
1874 -refs_to_drop, NULL, last_ref);
1875 } else if (is_data) {
1876 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1880 ret = btrfs_del_item(trans, root, path);
1885 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1886 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1887 u64 *discarded_bytes)
1890 u64 bytes_left, end;
1891 u64 aligned_start = ALIGN(start, 1 << 9);
1893 if (WARN_ON(start != aligned_start)) {
1894 len -= aligned_start - start;
1895 len = round_down(len, 1 << 9);
1896 start = aligned_start;
1899 *discarded_bytes = 0;
1907 /* Skip any superblocks on this device. */
1908 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1909 u64 sb_start = btrfs_sb_offset(j);
1910 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1911 u64 size = sb_start - start;
1913 if (!in_range(sb_start, start, bytes_left) &&
1914 !in_range(sb_end, start, bytes_left) &&
1915 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1919 * Superblock spans beginning of range. Adjust start and
1922 if (sb_start <= start) {
1923 start += sb_end - start;
1928 bytes_left = end - start;
1933 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1936 *discarded_bytes += size;
1937 else if (ret != -EOPNOTSUPP)
1946 bytes_left = end - start;
1950 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1953 *discarded_bytes += bytes_left;
1958 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1959 u64 num_bytes, u64 *actual_bytes)
1962 u64 discarded_bytes = 0;
1963 struct btrfs_bio *bbio = NULL;
1966 /* Tell the block device(s) that the sectors can be discarded */
1967 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1968 bytenr, &num_bytes, &bbio, 0);
1969 /* Error condition is -ENOMEM */
1971 struct btrfs_bio_stripe *stripe = bbio->stripes;
1975 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1977 if (!stripe->dev->can_discard)
1980 ret = btrfs_issue_discard(stripe->dev->bdev,
1985 discarded_bytes += bytes;
1986 else if (ret != -EOPNOTSUPP)
1987 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1990 * Just in case we get back EOPNOTSUPP for some reason,
1991 * just ignore the return value so we don't screw up
1992 * people calling discard_extent.
1996 btrfs_put_bbio(bbio);
2000 *actual_bytes = discarded_bytes;
2003 if (ret == -EOPNOTSUPP)
2008 /* Can return -ENOMEM */
2009 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 u64 bytenr, u64 num_bytes, u64 parent,
2012 u64 root_objectid, u64 owner, u64 offset,
2016 struct btrfs_fs_info *fs_info = root->fs_info;
2018 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2019 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2022 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2024 parent, root_objectid, (int)owner,
2025 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2027 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2029 parent, root_objectid, owner, offset,
2030 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2035 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2036 struct btrfs_root *root,
2037 struct btrfs_delayed_ref_node *node,
2038 u64 parent, u64 root_objectid,
2039 u64 owner, u64 offset, int refs_to_add,
2040 struct btrfs_delayed_extent_op *extent_op)
2042 struct btrfs_fs_info *fs_info = root->fs_info;
2043 struct btrfs_path *path;
2044 struct extent_buffer *leaf;
2045 struct btrfs_extent_item *item;
2046 struct btrfs_key key;
2047 u64 bytenr = node->bytenr;
2048 u64 num_bytes = node->num_bytes;
2051 int no_quota = node->no_quota;
2053 path = btrfs_alloc_path();
2057 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2061 path->leave_spinning = 1;
2062 /* this will setup the path even if it fails to insert the back ref */
2063 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2064 bytenr, num_bytes, parent,
2065 root_objectid, owner, offset,
2066 refs_to_add, extent_op);
2067 if ((ret < 0 && ret != -EAGAIN) || !ret)
2071 * Ok we had -EAGAIN which means we didn't have space to insert and
2072 * inline extent ref, so just update the reference count and add a
2075 leaf = path->nodes[0];
2076 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2077 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 refs = btrfs_extent_refs(leaf, item);
2079 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2081 __run_delayed_extent_op(extent_op, leaf, item);
2083 btrfs_mark_buffer_dirty(leaf);
2084 btrfs_release_path(path);
2087 path->leave_spinning = 1;
2088 /* now insert the actual backref */
2089 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2090 path, bytenr, parent, root_objectid,
2091 owner, offset, refs_to_add);
2093 btrfs_abort_transaction(trans, root, ret);
2095 btrfs_free_path(path);
2099 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 struct btrfs_delayed_extent_op *extent_op,
2103 int insert_reserved)
2106 struct btrfs_delayed_data_ref *ref;
2107 struct btrfs_key ins;
2112 ins.objectid = node->bytenr;
2113 ins.offset = node->num_bytes;
2114 ins.type = BTRFS_EXTENT_ITEM_KEY;
2116 ref = btrfs_delayed_node_to_data_ref(node);
2117 trace_run_delayed_data_ref(node, ref, node->action);
2119 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2120 parent = ref->parent;
2121 ref_root = ref->root;
2123 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2125 flags |= extent_op->flags_to_set;
2126 ret = alloc_reserved_file_extent(trans, root,
2127 parent, ref_root, flags,
2128 ref->objectid, ref->offset,
2129 &ins, node->ref_mod);
2130 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2131 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2132 ref_root, ref->objectid,
2133 ref->offset, node->ref_mod,
2135 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2136 ret = __btrfs_free_extent(trans, root, node, parent,
2137 ref_root, ref->objectid,
2138 ref->offset, node->ref_mod,
2146 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2147 struct extent_buffer *leaf,
2148 struct btrfs_extent_item *ei)
2150 u64 flags = btrfs_extent_flags(leaf, ei);
2151 if (extent_op->update_flags) {
2152 flags |= extent_op->flags_to_set;
2153 btrfs_set_extent_flags(leaf, ei, flags);
2156 if (extent_op->update_key) {
2157 struct btrfs_tree_block_info *bi;
2158 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2159 bi = (struct btrfs_tree_block_info *)(ei + 1);
2160 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2164 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2165 struct btrfs_root *root,
2166 struct btrfs_delayed_ref_node *node,
2167 struct btrfs_delayed_extent_op *extent_op)
2169 struct btrfs_key key;
2170 struct btrfs_path *path;
2171 struct btrfs_extent_item *ei;
2172 struct extent_buffer *leaf;
2176 int metadata = !extent_op->is_data;
2181 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2184 path = btrfs_alloc_path();
2188 key.objectid = node->bytenr;
2191 key.type = BTRFS_METADATA_ITEM_KEY;
2192 key.offset = extent_op->level;
2194 key.type = BTRFS_EXTENT_ITEM_KEY;
2195 key.offset = node->num_bytes;
2200 path->leave_spinning = 1;
2201 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2209 if (path->slots[0] > 0) {
2211 btrfs_item_key_to_cpu(path->nodes[0], &key,
2213 if (key.objectid == node->bytenr &&
2214 key.type == BTRFS_EXTENT_ITEM_KEY &&
2215 key.offset == node->num_bytes)
2219 btrfs_release_path(path);
2222 key.objectid = node->bytenr;
2223 key.offset = node->num_bytes;
2224 key.type = BTRFS_EXTENT_ITEM_KEY;
2233 leaf = path->nodes[0];
2234 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2235 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2236 if (item_size < sizeof(*ei)) {
2237 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2243 leaf = path->nodes[0];
2244 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2247 BUG_ON(item_size < sizeof(*ei));
2248 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2249 __run_delayed_extent_op(extent_op, leaf, ei);
2251 btrfs_mark_buffer_dirty(leaf);
2253 btrfs_free_path(path);
2257 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2258 struct btrfs_root *root,
2259 struct btrfs_delayed_ref_node *node,
2260 struct btrfs_delayed_extent_op *extent_op,
2261 int insert_reserved)
2264 struct btrfs_delayed_tree_ref *ref;
2265 struct btrfs_key ins;
2268 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2271 ref = btrfs_delayed_node_to_tree_ref(node);
2272 trace_run_delayed_tree_ref(node, ref, node->action);
2274 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275 parent = ref->parent;
2276 ref_root = ref->root;
2278 ins.objectid = node->bytenr;
2279 if (skinny_metadata) {
2280 ins.offset = ref->level;
2281 ins.type = BTRFS_METADATA_ITEM_KEY;
2283 ins.offset = node->num_bytes;
2284 ins.type = BTRFS_EXTENT_ITEM_KEY;
2287 BUG_ON(node->ref_mod != 1);
2288 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2289 BUG_ON(!extent_op || !extent_op->update_flags);
2290 ret = alloc_reserved_tree_block(trans, root,
2292 extent_op->flags_to_set,
2296 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2297 ret = __btrfs_inc_extent_ref(trans, root, node,
2301 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2302 ret = __btrfs_free_extent(trans, root, node,
2304 ref->level, 0, 1, extent_op);
2311 /* helper function to actually process a single delayed ref entry */
2312 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct btrfs_delayed_ref_node *node,
2315 struct btrfs_delayed_extent_op *extent_op,
2316 int insert_reserved)
2320 if (trans->aborted) {
2321 if (insert_reserved)
2322 btrfs_pin_extent(root, node->bytenr,
2323 node->num_bytes, 1);
2327 if (btrfs_delayed_ref_is_head(node)) {
2328 struct btrfs_delayed_ref_head *head;
2330 * we've hit the end of the chain and we were supposed
2331 * to insert this extent into the tree. But, it got
2332 * deleted before we ever needed to insert it, so all
2333 * we have to do is clean up the accounting
2336 head = btrfs_delayed_node_to_head(node);
2337 trace_run_delayed_ref_head(node, head, node->action);
2339 if (insert_reserved) {
2340 btrfs_pin_extent(root, node->bytenr,
2341 node->num_bytes, 1);
2342 if (head->is_data) {
2343 ret = btrfs_del_csums(trans, root,
2349 /* Also free its reserved qgroup space */
2350 btrfs_qgroup_free_delayed_ref(root->fs_info,
2351 head->qgroup_ref_root,
2352 head->qgroup_reserved);
2356 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2357 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2358 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2360 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2361 node->type == BTRFS_SHARED_DATA_REF_KEY)
2362 ret = run_delayed_data_ref(trans, root, node, extent_op,
2369 static inline struct btrfs_delayed_ref_node *
2370 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2372 struct btrfs_delayed_ref_node *ref;
2374 if (list_empty(&head->ref_list))
2378 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2379 * This is to prevent a ref count from going down to zero, which deletes
2380 * the extent item from the extent tree, when there still are references
2381 * to add, which would fail because they would not find the extent item.
2383 list_for_each_entry(ref, &head->ref_list, list) {
2384 if (ref->action == BTRFS_ADD_DELAYED_REF)
2388 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2393 * Returns 0 on success or if called with an already aborted transaction.
2394 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2396 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2397 struct btrfs_root *root,
2400 struct btrfs_delayed_ref_root *delayed_refs;
2401 struct btrfs_delayed_ref_node *ref;
2402 struct btrfs_delayed_ref_head *locked_ref = NULL;
2403 struct btrfs_delayed_extent_op *extent_op;
2404 struct btrfs_fs_info *fs_info = root->fs_info;
2405 ktime_t start = ktime_get();
2407 unsigned long count = 0;
2408 unsigned long actual_count = 0;
2409 int must_insert_reserved = 0;
2411 delayed_refs = &trans->transaction->delayed_refs;
2417 spin_lock(&delayed_refs->lock);
2418 locked_ref = btrfs_select_ref_head(trans);
2420 spin_unlock(&delayed_refs->lock);
2424 /* grab the lock that says we are going to process
2425 * all the refs for this head */
2426 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2427 spin_unlock(&delayed_refs->lock);
2429 * we may have dropped the spin lock to get the head
2430 * mutex lock, and that might have given someone else
2431 * time to free the head. If that's true, it has been
2432 * removed from our list and we can move on.
2434 if (ret == -EAGAIN) {
2441 spin_lock(&locked_ref->lock);
2444 * locked_ref is the head node, so we have to go one
2445 * node back for any delayed ref updates
2447 ref = select_delayed_ref(locked_ref);
2449 if (ref && ref->seq &&
2450 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2451 spin_unlock(&locked_ref->lock);
2452 btrfs_delayed_ref_unlock(locked_ref);
2453 spin_lock(&delayed_refs->lock);
2454 locked_ref->processing = 0;
2455 delayed_refs->num_heads_ready++;
2456 spin_unlock(&delayed_refs->lock);
2464 * record the must insert reserved flag before we
2465 * drop the spin lock.
2467 must_insert_reserved = locked_ref->must_insert_reserved;
2468 locked_ref->must_insert_reserved = 0;
2470 extent_op = locked_ref->extent_op;
2471 locked_ref->extent_op = NULL;
2476 /* All delayed refs have been processed, Go ahead
2477 * and send the head node to run_one_delayed_ref,
2478 * so that any accounting fixes can happen
2480 ref = &locked_ref->node;
2482 if (extent_op && must_insert_reserved) {
2483 btrfs_free_delayed_extent_op(extent_op);
2488 spin_unlock(&locked_ref->lock);
2489 ret = run_delayed_extent_op(trans, root,
2491 btrfs_free_delayed_extent_op(extent_op);
2495 * Need to reset must_insert_reserved if
2496 * there was an error so the abort stuff
2497 * can cleanup the reserved space
2500 if (must_insert_reserved)
2501 locked_ref->must_insert_reserved = 1;
2502 locked_ref->processing = 0;
2503 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2504 btrfs_delayed_ref_unlock(locked_ref);
2511 * Need to drop our head ref lock and re-aqcuire the
2512 * delayed ref lock and then re-check to make sure
2515 spin_unlock(&locked_ref->lock);
2516 spin_lock(&delayed_refs->lock);
2517 spin_lock(&locked_ref->lock);
2518 if (!list_empty(&locked_ref->ref_list) ||
2519 locked_ref->extent_op) {
2520 spin_unlock(&locked_ref->lock);
2521 spin_unlock(&delayed_refs->lock);
2525 delayed_refs->num_heads--;
2526 rb_erase(&locked_ref->href_node,
2527 &delayed_refs->href_root);
2528 spin_unlock(&delayed_refs->lock);
2532 list_del(&ref->list);
2534 atomic_dec(&delayed_refs->num_entries);
2536 if (!btrfs_delayed_ref_is_head(ref)) {
2538 * when we play the delayed ref, also correct the
2541 switch (ref->action) {
2542 case BTRFS_ADD_DELAYED_REF:
2543 case BTRFS_ADD_DELAYED_EXTENT:
2544 locked_ref->node.ref_mod -= ref->ref_mod;
2546 case BTRFS_DROP_DELAYED_REF:
2547 locked_ref->node.ref_mod += ref->ref_mod;
2553 spin_unlock(&locked_ref->lock);
2555 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2556 must_insert_reserved);
2558 btrfs_free_delayed_extent_op(extent_op);
2560 locked_ref->processing = 0;
2561 btrfs_delayed_ref_unlock(locked_ref);
2562 btrfs_put_delayed_ref(ref);
2563 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2568 * If this node is a head, that means all the refs in this head
2569 * have been dealt with, and we will pick the next head to deal
2570 * with, so we must unlock the head and drop it from the cluster
2571 * list before we release it.
2573 if (btrfs_delayed_ref_is_head(ref)) {
2574 if (locked_ref->is_data &&
2575 locked_ref->total_ref_mod < 0) {
2576 spin_lock(&delayed_refs->lock);
2577 delayed_refs->pending_csums -= ref->num_bytes;
2578 spin_unlock(&delayed_refs->lock);
2580 btrfs_delayed_ref_unlock(locked_ref);
2583 btrfs_put_delayed_ref(ref);
2589 * We don't want to include ref heads since we can have empty ref heads
2590 * and those will drastically skew our runtime down since we just do
2591 * accounting, no actual extent tree updates.
2593 if (actual_count > 0) {
2594 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2598 * We weigh the current average higher than our current runtime
2599 * to avoid large swings in the average.
2601 spin_lock(&delayed_refs->lock);
2602 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2603 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2604 spin_unlock(&delayed_refs->lock);
2609 #ifdef SCRAMBLE_DELAYED_REFS
2611 * Normally delayed refs get processed in ascending bytenr order. This
2612 * correlates in most cases to the order added. To expose dependencies on this
2613 * order, we start to process the tree in the middle instead of the beginning
2615 static u64 find_middle(struct rb_root *root)
2617 struct rb_node *n = root->rb_node;
2618 struct btrfs_delayed_ref_node *entry;
2621 u64 first = 0, last = 0;
2625 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2626 first = entry->bytenr;
2630 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2631 last = entry->bytenr;
2636 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2637 WARN_ON(!entry->in_tree);
2639 middle = entry->bytenr;
2652 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2656 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2657 sizeof(struct btrfs_extent_inline_ref));
2658 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2659 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2662 * We don't ever fill up leaves all the way so multiply by 2 just to be
2663 * closer to what we're really going to want to ouse.
2665 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2669 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2670 * would require to store the csums for that many bytes.
2672 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2675 u64 num_csums_per_leaf;
2678 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2679 num_csums_per_leaf = div64_u64(csum_size,
2680 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2681 num_csums = div64_u64(csum_bytes, root->sectorsize);
2682 num_csums += num_csums_per_leaf - 1;
2683 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2687 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2688 struct btrfs_root *root)
2690 struct btrfs_block_rsv *global_rsv;
2691 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2692 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2693 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2694 u64 num_bytes, num_dirty_bgs_bytes;
2697 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2698 num_heads = heads_to_leaves(root, num_heads);
2700 num_bytes += (num_heads - 1) * root->nodesize;
2702 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2703 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2705 global_rsv = &root->fs_info->global_block_rsv;
2708 * If we can't allocate any more chunks lets make sure we have _lots_ of
2709 * wiggle room since running delayed refs can create more delayed refs.
2711 if (global_rsv->space_info->full) {
2712 num_dirty_bgs_bytes <<= 1;
2716 spin_lock(&global_rsv->lock);
2717 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2719 spin_unlock(&global_rsv->lock);
2723 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root)
2726 struct btrfs_fs_info *fs_info = root->fs_info;
2728 atomic_read(&trans->transaction->delayed_refs.num_entries);
2733 avg_runtime = fs_info->avg_delayed_ref_runtime;
2734 val = num_entries * avg_runtime;
2735 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2737 if (val >= NSEC_PER_SEC / 2)
2740 return btrfs_check_space_for_delayed_refs(trans, root);
2743 struct async_delayed_refs {
2744 struct btrfs_root *root;
2748 struct completion wait;
2749 struct btrfs_work work;
2752 static void delayed_ref_async_start(struct btrfs_work *work)
2754 struct async_delayed_refs *async;
2755 struct btrfs_trans_handle *trans;
2758 async = container_of(work, struct async_delayed_refs, work);
2760 trans = btrfs_join_transaction(async->root);
2761 if (IS_ERR(trans)) {
2762 async->error = PTR_ERR(trans);
2767 * trans->sync means that when we call end_transaciton, we won't
2768 * wait on delayed refs
2771 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2775 ret = btrfs_end_transaction(trans, async->root);
2776 if (ret && !async->error)
2780 complete(&async->wait);
2785 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2786 unsigned long count, int wait)
2788 struct async_delayed_refs *async;
2791 async = kmalloc(sizeof(*async), GFP_NOFS);
2795 async->root = root->fs_info->tree_root;
2796 async->count = count;
2802 init_completion(&async->wait);
2804 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2805 delayed_ref_async_start, NULL, NULL);
2807 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2810 wait_for_completion(&async->wait);
2819 * this starts processing the delayed reference count updates and
2820 * extent insertions we have queued up so far. count can be
2821 * 0, which means to process everything in the tree at the start
2822 * of the run (but not newly added entries), or it can be some target
2823 * number you'd like to process.
2825 * Returns 0 on success or if called with an aborted transaction
2826 * Returns <0 on error and aborts the transaction
2828 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2829 struct btrfs_root *root, unsigned long count)
2831 struct rb_node *node;
2832 struct btrfs_delayed_ref_root *delayed_refs;
2833 struct btrfs_delayed_ref_head *head;
2835 int run_all = count == (unsigned long)-1;
2836 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2838 /* We'll clean this up in btrfs_cleanup_transaction */
2842 if (root == root->fs_info->extent_root)
2843 root = root->fs_info->tree_root;
2845 delayed_refs = &trans->transaction->delayed_refs;
2847 count = atomic_read(&delayed_refs->num_entries) * 2;
2850 #ifdef SCRAMBLE_DELAYED_REFS
2851 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2853 trans->can_flush_pending_bgs = false;
2854 ret = __btrfs_run_delayed_refs(trans, root, count);
2856 btrfs_abort_transaction(trans, root, ret);
2861 if (!list_empty(&trans->new_bgs))
2862 btrfs_create_pending_block_groups(trans, root);
2864 spin_lock(&delayed_refs->lock);
2865 node = rb_first(&delayed_refs->href_root);
2867 spin_unlock(&delayed_refs->lock);
2870 count = (unsigned long)-1;
2873 head = rb_entry(node, struct btrfs_delayed_ref_head,
2875 if (btrfs_delayed_ref_is_head(&head->node)) {
2876 struct btrfs_delayed_ref_node *ref;
2879 atomic_inc(&ref->refs);
2881 spin_unlock(&delayed_refs->lock);
2883 * Mutex was contended, block until it's
2884 * released and try again
2886 mutex_lock(&head->mutex);
2887 mutex_unlock(&head->mutex);
2889 btrfs_put_delayed_ref(ref);
2895 node = rb_next(node);
2897 spin_unlock(&delayed_refs->lock);
2902 assert_qgroups_uptodate(trans);
2903 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2907 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2908 struct btrfs_root *root,
2909 u64 bytenr, u64 num_bytes, u64 flags,
2910 int level, int is_data)
2912 struct btrfs_delayed_extent_op *extent_op;
2915 extent_op = btrfs_alloc_delayed_extent_op();
2919 extent_op->flags_to_set = flags;
2920 extent_op->update_flags = 1;
2921 extent_op->update_key = 0;
2922 extent_op->is_data = is_data ? 1 : 0;
2923 extent_op->level = level;
2925 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2926 num_bytes, extent_op);
2928 btrfs_free_delayed_extent_op(extent_op);
2932 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2933 struct btrfs_root *root,
2934 struct btrfs_path *path,
2935 u64 objectid, u64 offset, u64 bytenr)
2937 struct btrfs_delayed_ref_head *head;
2938 struct btrfs_delayed_ref_node *ref;
2939 struct btrfs_delayed_data_ref *data_ref;
2940 struct btrfs_delayed_ref_root *delayed_refs;
2943 delayed_refs = &trans->transaction->delayed_refs;
2944 spin_lock(&delayed_refs->lock);
2945 head = btrfs_find_delayed_ref_head(trans, bytenr);
2947 spin_unlock(&delayed_refs->lock);
2951 if (!mutex_trylock(&head->mutex)) {
2952 atomic_inc(&head->node.refs);
2953 spin_unlock(&delayed_refs->lock);
2955 btrfs_release_path(path);
2958 * Mutex was contended, block until it's released and let
2961 mutex_lock(&head->mutex);
2962 mutex_unlock(&head->mutex);
2963 btrfs_put_delayed_ref(&head->node);
2966 spin_unlock(&delayed_refs->lock);
2968 spin_lock(&head->lock);
2969 list_for_each_entry(ref, &head->ref_list, list) {
2970 /* If it's a shared ref we know a cross reference exists */
2971 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2976 data_ref = btrfs_delayed_node_to_data_ref(ref);
2979 * If our ref doesn't match the one we're currently looking at
2980 * then we have a cross reference.
2982 if (data_ref->root != root->root_key.objectid ||
2983 data_ref->objectid != objectid ||
2984 data_ref->offset != offset) {
2989 spin_unlock(&head->lock);
2990 mutex_unlock(&head->mutex);
2994 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2995 struct btrfs_root *root,
2996 struct btrfs_path *path,
2997 u64 objectid, u64 offset, u64 bytenr)
2999 struct btrfs_root *extent_root = root->fs_info->extent_root;
3000 struct extent_buffer *leaf;
3001 struct btrfs_extent_data_ref *ref;
3002 struct btrfs_extent_inline_ref *iref;
3003 struct btrfs_extent_item *ei;
3004 struct btrfs_key key;
3008 key.objectid = bytenr;
3009 key.offset = (u64)-1;
3010 key.type = BTRFS_EXTENT_ITEM_KEY;
3012 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3015 BUG_ON(ret == 0); /* Corruption */
3018 if (path->slots[0] == 0)
3022 leaf = path->nodes[0];
3023 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3025 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3029 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3030 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3031 if (item_size < sizeof(*ei)) {
3032 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3036 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3038 if (item_size != sizeof(*ei) +
3039 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3042 if (btrfs_extent_generation(leaf, ei) <=
3043 btrfs_root_last_snapshot(&root->root_item))
3046 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3047 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3048 BTRFS_EXTENT_DATA_REF_KEY)
3051 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3052 if (btrfs_extent_refs(leaf, ei) !=
3053 btrfs_extent_data_ref_count(leaf, ref) ||
3054 btrfs_extent_data_ref_root(leaf, ref) !=
3055 root->root_key.objectid ||
3056 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3057 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3065 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3066 struct btrfs_root *root,
3067 u64 objectid, u64 offset, u64 bytenr)
3069 struct btrfs_path *path;
3073 path = btrfs_alloc_path();
3078 ret = check_committed_ref(trans, root, path, objectid,
3080 if (ret && ret != -ENOENT)
3083 ret2 = check_delayed_ref(trans, root, path, objectid,
3085 } while (ret2 == -EAGAIN);
3087 if (ret2 && ret2 != -ENOENT) {
3092 if (ret != -ENOENT || ret2 != -ENOENT)
3095 btrfs_free_path(path);
3096 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3101 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3102 struct btrfs_root *root,
3103 struct extent_buffer *buf,
3104 int full_backref, int inc)
3111 struct btrfs_key key;
3112 struct btrfs_file_extent_item *fi;
3116 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3117 u64, u64, u64, u64, u64, u64, int);
3120 if (btrfs_test_is_dummy_root(root))
3123 ref_root = btrfs_header_owner(buf);
3124 nritems = btrfs_header_nritems(buf);
3125 level = btrfs_header_level(buf);
3127 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3131 process_func = btrfs_inc_extent_ref;
3133 process_func = btrfs_free_extent;
3136 parent = buf->start;
3140 for (i = 0; i < nritems; i++) {
3142 btrfs_item_key_to_cpu(buf, &key, i);
3143 if (key.type != BTRFS_EXTENT_DATA_KEY)
3145 fi = btrfs_item_ptr(buf, i,
3146 struct btrfs_file_extent_item);
3147 if (btrfs_file_extent_type(buf, fi) ==
3148 BTRFS_FILE_EXTENT_INLINE)
3150 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3154 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3155 key.offset -= btrfs_file_extent_offset(buf, fi);
3156 ret = process_func(trans, root, bytenr, num_bytes,
3157 parent, ref_root, key.objectid,
3162 bytenr = btrfs_node_blockptr(buf, i);
3163 num_bytes = root->nodesize;
3164 ret = process_func(trans, root, bytenr, num_bytes,
3165 parent, ref_root, level - 1, 0,
3176 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3177 struct extent_buffer *buf, int full_backref)
3179 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3182 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3183 struct extent_buffer *buf, int full_backref)
3185 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3188 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3189 struct btrfs_root *root,
3190 struct btrfs_path *path,
3191 struct btrfs_block_group_cache *cache)
3194 struct btrfs_root *extent_root = root->fs_info->extent_root;
3196 struct extent_buffer *leaf;
3198 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3205 leaf = path->nodes[0];
3206 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3207 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3208 btrfs_mark_buffer_dirty(leaf);
3210 btrfs_release_path(path);
3215 static struct btrfs_block_group_cache *
3216 next_block_group(struct btrfs_root *root,
3217 struct btrfs_block_group_cache *cache)
3219 struct rb_node *node;
3221 spin_lock(&root->fs_info->block_group_cache_lock);
3223 /* If our block group was removed, we need a full search. */
3224 if (RB_EMPTY_NODE(&cache->cache_node)) {
3225 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3227 spin_unlock(&root->fs_info->block_group_cache_lock);
3228 btrfs_put_block_group(cache);
3229 cache = btrfs_lookup_first_block_group(root->fs_info,
3233 node = rb_next(&cache->cache_node);
3234 btrfs_put_block_group(cache);
3236 cache = rb_entry(node, struct btrfs_block_group_cache,
3238 btrfs_get_block_group(cache);
3241 spin_unlock(&root->fs_info->block_group_cache_lock);
3245 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3246 struct btrfs_trans_handle *trans,
3247 struct btrfs_path *path)
3249 struct btrfs_root *root = block_group->fs_info->tree_root;
3250 struct inode *inode = NULL;
3252 int dcs = BTRFS_DC_ERROR;
3258 * If this block group is smaller than 100 megs don't bother caching the
3261 if (block_group->key.offset < (100 * 1024 * 1024)) {
3262 spin_lock(&block_group->lock);
3263 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3264 spin_unlock(&block_group->lock);
3271 inode = lookup_free_space_inode(root, block_group, path);
3272 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3273 ret = PTR_ERR(inode);
3274 btrfs_release_path(path);
3278 if (IS_ERR(inode)) {
3282 if (block_group->ro)
3285 ret = create_free_space_inode(root, trans, block_group, path);
3291 /* We've already setup this transaction, go ahead and exit */
3292 if (block_group->cache_generation == trans->transid &&
3293 i_size_read(inode)) {
3294 dcs = BTRFS_DC_SETUP;
3299 * We want to set the generation to 0, that way if anything goes wrong
3300 * from here on out we know not to trust this cache when we load up next
3303 BTRFS_I(inode)->generation = 0;
3304 ret = btrfs_update_inode(trans, root, inode);
3307 * So theoretically we could recover from this, simply set the
3308 * super cache generation to 0 so we know to invalidate the
3309 * cache, but then we'd have to keep track of the block groups
3310 * that fail this way so we know we _have_ to reset this cache
3311 * before the next commit or risk reading stale cache. So to
3312 * limit our exposure to horrible edge cases lets just abort the
3313 * transaction, this only happens in really bad situations
3316 btrfs_abort_transaction(trans, root, ret);
3321 if (i_size_read(inode) > 0) {
3322 ret = btrfs_check_trunc_cache_free_space(root,
3323 &root->fs_info->global_block_rsv);
3327 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3332 spin_lock(&block_group->lock);
3333 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3334 !btrfs_test_opt(root, SPACE_CACHE)) {
3336 * don't bother trying to write stuff out _if_
3337 * a) we're not cached,
3338 * b) we're with nospace_cache mount option.
3340 dcs = BTRFS_DC_WRITTEN;
3341 spin_unlock(&block_group->lock);
3344 spin_unlock(&block_group->lock);
3347 * Try to preallocate enough space based on how big the block group is.
3348 * Keep in mind this has to include any pinned space which could end up
3349 * taking up quite a bit since it's not folded into the other space
3352 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3357 num_pages *= PAGE_CACHE_SIZE;
3359 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3363 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3364 num_pages, num_pages,
3367 dcs = BTRFS_DC_SETUP;
3368 btrfs_free_reserved_data_space(inode, 0, num_pages);
3373 btrfs_release_path(path);
3375 spin_lock(&block_group->lock);
3376 if (!ret && dcs == BTRFS_DC_SETUP)
3377 block_group->cache_generation = trans->transid;
3378 block_group->disk_cache_state = dcs;
3379 spin_unlock(&block_group->lock);
3384 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3385 struct btrfs_root *root)
3387 struct btrfs_block_group_cache *cache, *tmp;
3388 struct btrfs_transaction *cur_trans = trans->transaction;
3389 struct btrfs_path *path;
3391 if (list_empty(&cur_trans->dirty_bgs) ||
3392 !btrfs_test_opt(root, SPACE_CACHE))
3395 path = btrfs_alloc_path();
3399 /* Could add new block groups, use _safe just in case */
3400 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3402 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3403 cache_save_setup(cache, trans, path);
3406 btrfs_free_path(path);
3411 * transaction commit does final block group cache writeback during a
3412 * critical section where nothing is allowed to change the FS. This is
3413 * required in order for the cache to actually match the block group,
3414 * but can introduce a lot of latency into the commit.
3416 * So, btrfs_start_dirty_block_groups is here to kick off block group
3417 * cache IO. There's a chance we'll have to redo some of it if the
3418 * block group changes again during the commit, but it greatly reduces
3419 * the commit latency by getting rid of the easy block groups while
3420 * we're still allowing others to join the commit.
3422 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3423 struct btrfs_root *root)
3425 struct btrfs_block_group_cache *cache;
3426 struct btrfs_transaction *cur_trans = trans->transaction;
3429 struct btrfs_path *path = NULL;
3431 struct list_head *io = &cur_trans->io_bgs;
3432 int num_started = 0;
3435 spin_lock(&cur_trans->dirty_bgs_lock);
3436 if (list_empty(&cur_trans->dirty_bgs)) {
3437 spin_unlock(&cur_trans->dirty_bgs_lock);
3440 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3441 spin_unlock(&cur_trans->dirty_bgs_lock);
3445 * make sure all the block groups on our dirty list actually
3448 btrfs_create_pending_block_groups(trans, root);
3451 path = btrfs_alloc_path();
3457 * cache_write_mutex is here only to save us from balance or automatic
3458 * removal of empty block groups deleting this block group while we are
3459 * writing out the cache
3461 mutex_lock(&trans->transaction->cache_write_mutex);
3462 while (!list_empty(&dirty)) {
3463 cache = list_first_entry(&dirty,
3464 struct btrfs_block_group_cache,
3467 * this can happen if something re-dirties a block
3468 * group that is already under IO. Just wait for it to
3469 * finish and then do it all again
3471 if (!list_empty(&cache->io_list)) {
3472 list_del_init(&cache->io_list);
3473 btrfs_wait_cache_io(root, trans, cache,
3474 &cache->io_ctl, path,
3475 cache->key.objectid);
3476 btrfs_put_block_group(cache);
3481 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3482 * if it should update the cache_state. Don't delete
3483 * until after we wait.
3485 * Since we're not running in the commit critical section
3486 * we need the dirty_bgs_lock to protect from update_block_group
3488 spin_lock(&cur_trans->dirty_bgs_lock);
3489 list_del_init(&cache->dirty_list);
3490 spin_unlock(&cur_trans->dirty_bgs_lock);
3494 cache_save_setup(cache, trans, path);
3496 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3497 cache->io_ctl.inode = NULL;
3498 ret = btrfs_write_out_cache(root, trans, cache, path);
3499 if (ret == 0 && cache->io_ctl.inode) {
3504 * the cache_write_mutex is protecting
3507 list_add_tail(&cache->io_list, io);
3510 * if we failed to write the cache, the
3511 * generation will be bad and life goes on
3517 ret = write_one_cache_group(trans, root, path, cache);
3519 * Our block group might still be attached to the list
3520 * of new block groups in the transaction handle of some
3521 * other task (struct btrfs_trans_handle->new_bgs). This
3522 * means its block group item isn't yet in the extent
3523 * tree. If this happens ignore the error, as we will
3524 * try again later in the critical section of the
3525 * transaction commit.
3527 if (ret == -ENOENT) {
3529 spin_lock(&cur_trans->dirty_bgs_lock);
3530 if (list_empty(&cache->dirty_list)) {
3531 list_add_tail(&cache->dirty_list,
3532 &cur_trans->dirty_bgs);
3533 btrfs_get_block_group(cache);
3535 spin_unlock(&cur_trans->dirty_bgs_lock);
3537 btrfs_abort_transaction(trans, root, ret);
3541 /* if its not on the io list, we need to put the block group */
3543 btrfs_put_block_group(cache);
3549 * Avoid blocking other tasks for too long. It might even save
3550 * us from writing caches for block groups that are going to be
3553 mutex_unlock(&trans->transaction->cache_write_mutex);
3554 mutex_lock(&trans->transaction->cache_write_mutex);
3556 mutex_unlock(&trans->transaction->cache_write_mutex);
3559 * go through delayed refs for all the stuff we've just kicked off
3560 * and then loop back (just once)
3562 ret = btrfs_run_delayed_refs(trans, root, 0);
3563 if (!ret && loops == 0) {
3565 spin_lock(&cur_trans->dirty_bgs_lock);
3566 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3568 * dirty_bgs_lock protects us from concurrent block group
3569 * deletes too (not just cache_write_mutex).
3571 if (!list_empty(&dirty)) {
3572 spin_unlock(&cur_trans->dirty_bgs_lock);
3575 spin_unlock(&cur_trans->dirty_bgs_lock);
3578 btrfs_free_path(path);
3582 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3583 struct btrfs_root *root)
3585 struct btrfs_block_group_cache *cache;
3586 struct btrfs_transaction *cur_trans = trans->transaction;
3589 struct btrfs_path *path;
3590 struct list_head *io = &cur_trans->io_bgs;
3591 int num_started = 0;
3593 path = btrfs_alloc_path();
3598 * We don't need the lock here since we are protected by the transaction
3599 * commit. We want to do the cache_save_setup first and then run the
3600 * delayed refs to make sure we have the best chance at doing this all
3603 while (!list_empty(&cur_trans->dirty_bgs)) {
3604 cache = list_first_entry(&cur_trans->dirty_bgs,
3605 struct btrfs_block_group_cache,
3609 * this can happen if cache_save_setup re-dirties a block
3610 * group that is already under IO. Just wait for it to
3611 * finish and then do it all again
3613 if (!list_empty(&cache->io_list)) {
3614 list_del_init(&cache->io_list);
3615 btrfs_wait_cache_io(root, trans, cache,
3616 &cache->io_ctl, path,
3617 cache->key.objectid);
3618 btrfs_put_block_group(cache);
3622 * don't remove from the dirty list until after we've waited
3625 list_del_init(&cache->dirty_list);
3628 cache_save_setup(cache, trans, path);
3631 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3633 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3634 cache->io_ctl.inode = NULL;
3635 ret = btrfs_write_out_cache(root, trans, cache, path);
3636 if (ret == 0 && cache->io_ctl.inode) {
3639 list_add_tail(&cache->io_list, io);
3642 * if we failed to write the cache, the
3643 * generation will be bad and life goes on
3649 ret = write_one_cache_group(trans, root, path, cache);
3651 btrfs_abort_transaction(trans, root, ret);
3654 /* if its not on the io list, we need to put the block group */
3656 btrfs_put_block_group(cache);
3659 while (!list_empty(io)) {
3660 cache = list_first_entry(io, struct btrfs_block_group_cache,
3662 list_del_init(&cache->io_list);
3663 btrfs_wait_cache_io(root, trans, cache,
3664 &cache->io_ctl, path, cache->key.objectid);
3665 btrfs_put_block_group(cache);
3668 btrfs_free_path(path);
3672 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3674 struct btrfs_block_group_cache *block_group;
3677 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3678 if (!block_group || block_group->ro)
3681 btrfs_put_block_group(block_group);
3685 static const char *alloc_name(u64 flags)
3688 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3690 case BTRFS_BLOCK_GROUP_METADATA:
3692 case BTRFS_BLOCK_GROUP_DATA:
3694 case BTRFS_BLOCK_GROUP_SYSTEM:
3698 return "invalid-combination";
3702 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3703 u64 total_bytes, u64 bytes_used,
3704 struct btrfs_space_info **space_info)
3706 struct btrfs_space_info *found;
3711 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3712 BTRFS_BLOCK_GROUP_RAID10))
3717 found = __find_space_info(info, flags);
3719 spin_lock(&found->lock);
3720 found->total_bytes += total_bytes;
3721 found->disk_total += total_bytes * factor;
3722 found->bytes_used += bytes_used;
3723 found->disk_used += bytes_used * factor;
3724 if (total_bytes > 0)
3726 spin_unlock(&found->lock);
3727 *space_info = found;
3730 found = kzalloc(sizeof(*found), GFP_NOFS);
3734 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3740 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3741 INIT_LIST_HEAD(&found->block_groups[i]);
3742 init_rwsem(&found->groups_sem);
3743 spin_lock_init(&found->lock);
3744 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3745 found->total_bytes = total_bytes;
3746 found->disk_total = total_bytes * factor;
3747 found->bytes_used = bytes_used;
3748 found->disk_used = bytes_used * factor;
3749 found->bytes_pinned = 0;
3750 found->bytes_reserved = 0;
3751 found->bytes_readonly = 0;
3752 found->bytes_may_use = 0;
3754 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3755 found->chunk_alloc = 0;
3757 init_waitqueue_head(&found->wait);
3758 INIT_LIST_HEAD(&found->ro_bgs);
3760 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3761 info->space_info_kobj, "%s",
3762 alloc_name(found->flags));
3768 *space_info = found;
3769 list_add_rcu(&found->list, &info->space_info);
3770 if (flags & BTRFS_BLOCK_GROUP_DATA)
3771 info->data_sinfo = found;
3776 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3778 u64 extra_flags = chunk_to_extended(flags) &
3779 BTRFS_EXTENDED_PROFILE_MASK;
3781 write_seqlock(&fs_info->profiles_lock);
3782 if (flags & BTRFS_BLOCK_GROUP_DATA)
3783 fs_info->avail_data_alloc_bits |= extra_flags;
3784 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3785 fs_info->avail_metadata_alloc_bits |= extra_flags;
3786 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3787 fs_info->avail_system_alloc_bits |= extra_flags;
3788 write_sequnlock(&fs_info->profiles_lock);
3792 * returns target flags in extended format or 0 if restripe for this
3793 * chunk_type is not in progress
3795 * should be called with either volume_mutex or balance_lock held
3797 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3799 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3805 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3806 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3807 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3808 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3809 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3810 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3811 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3812 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3813 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3820 * @flags: available profiles in extended format (see ctree.h)
3822 * Returns reduced profile in chunk format. If profile changing is in
3823 * progress (either running or paused) picks the target profile (if it's
3824 * already available), otherwise falls back to plain reducing.
3826 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3828 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3834 * see if restripe for this chunk_type is in progress, if so
3835 * try to reduce to the target profile
3837 spin_lock(&root->fs_info->balance_lock);
3838 target = get_restripe_target(root->fs_info, flags);
3840 /* pick target profile only if it's already available */
3841 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3842 spin_unlock(&root->fs_info->balance_lock);
3843 return extended_to_chunk(target);
3846 spin_unlock(&root->fs_info->balance_lock);
3848 /* First, mask out the RAID levels which aren't possible */
3849 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3850 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3851 allowed |= btrfs_raid_group[raid_type];
3855 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3856 allowed = BTRFS_BLOCK_GROUP_RAID6;
3857 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3858 allowed = BTRFS_BLOCK_GROUP_RAID5;
3859 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3860 allowed = BTRFS_BLOCK_GROUP_RAID10;
3861 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3862 allowed = BTRFS_BLOCK_GROUP_RAID1;
3863 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3864 allowed = BTRFS_BLOCK_GROUP_RAID0;
3866 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3868 return extended_to_chunk(flags | allowed);
3871 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3878 seq = read_seqbegin(&root->fs_info->profiles_lock);
3880 if (flags & BTRFS_BLOCK_GROUP_DATA)
3881 flags |= root->fs_info->avail_data_alloc_bits;
3882 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3883 flags |= root->fs_info->avail_system_alloc_bits;
3884 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3885 flags |= root->fs_info->avail_metadata_alloc_bits;
3886 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3888 return btrfs_reduce_alloc_profile(root, flags);
3891 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3897 flags = BTRFS_BLOCK_GROUP_DATA;
3898 else if (root == root->fs_info->chunk_root)
3899 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3901 flags = BTRFS_BLOCK_GROUP_METADATA;
3903 ret = get_alloc_profile(root, flags);
3907 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
3909 struct btrfs_space_info *data_sinfo;
3910 struct btrfs_root *root = BTRFS_I(inode)->root;
3911 struct btrfs_fs_info *fs_info = root->fs_info;
3914 int need_commit = 2;
3915 int have_pinned_space;
3917 /* make sure bytes are sectorsize aligned */
3918 bytes = ALIGN(bytes, root->sectorsize);
3920 if (btrfs_is_free_space_inode(inode)) {
3922 ASSERT(current->journal_info);
3925 data_sinfo = fs_info->data_sinfo;
3930 /* make sure we have enough space to handle the data first */
3931 spin_lock(&data_sinfo->lock);
3932 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3933 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3934 data_sinfo->bytes_may_use;
3936 if (used + bytes > data_sinfo->total_bytes) {
3937 struct btrfs_trans_handle *trans;
3940 * if we don't have enough free bytes in this space then we need
3941 * to alloc a new chunk.
3943 if (!data_sinfo->full) {
3946 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3947 spin_unlock(&data_sinfo->lock);
3949 alloc_target = btrfs_get_alloc_profile(root, 1);
3951 * It is ugly that we don't call nolock join
3952 * transaction for the free space inode case here.
3953 * But it is safe because we only do the data space
3954 * reservation for the free space cache in the
3955 * transaction context, the common join transaction
3956 * just increase the counter of the current transaction
3957 * handler, doesn't try to acquire the trans_lock of
3960 trans = btrfs_join_transaction(root);
3962 return PTR_ERR(trans);
3964 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3966 CHUNK_ALLOC_NO_FORCE);
3967 btrfs_end_transaction(trans, root);
3972 have_pinned_space = 1;
3978 data_sinfo = fs_info->data_sinfo;
3984 * If we don't have enough pinned space to deal with this
3985 * allocation, and no removed chunk in current transaction,
3986 * don't bother committing the transaction.
3988 have_pinned_space = percpu_counter_compare(
3989 &data_sinfo->total_bytes_pinned,
3990 used + bytes - data_sinfo->total_bytes);
3991 spin_unlock(&data_sinfo->lock);
3993 /* commit the current transaction and try again */
3996 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3999 if (need_commit > 0)
4000 btrfs_wait_ordered_roots(fs_info, -1);
4002 trans = btrfs_join_transaction(root);
4004 return PTR_ERR(trans);
4005 if (have_pinned_space >= 0 ||
4006 trans->transaction->have_free_bgs ||
4008 ret = btrfs_commit_transaction(trans, root);
4012 * make sure that all running delayed iput are
4015 down_write(&root->fs_info->delayed_iput_sem);
4016 up_write(&root->fs_info->delayed_iput_sem);
4019 btrfs_end_transaction(trans, root);
4023 trace_btrfs_space_reservation(root->fs_info,
4024 "space_info:enospc",
4025 data_sinfo->flags, bytes, 1);
4028 data_sinfo->bytes_may_use += bytes;
4029 trace_btrfs_space_reservation(root->fs_info, "space_info",
4030 data_sinfo->flags, bytes, 1);
4031 spin_unlock(&data_sinfo->lock);
4037 * New check_data_free_space() with ability for precious data reservation
4038 * Will replace old btrfs_check_data_free_space(), but for patch split,
4039 * add a new function first and then replace it.
4041 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4043 struct btrfs_root *root = BTRFS_I(inode)->root;
4046 /* align the range */
4047 len = round_up(start + len, root->sectorsize) -
4048 round_down(start, root->sectorsize);
4049 start = round_down(start, root->sectorsize);
4051 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4056 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4058 * TODO: Find a good method to avoid reserve data space for NOCOW
4059 * range, but don't impact performance on quota disable case.
4061 ret = btrfs_qgroup_reserve_data(inode, start, len);
4066 * Called if we need to clear a data reservation for this inode
4067 * Normally in a error case.
4069 * This one will *NOT* use accurate qgroup reserved space API, just for case
4070 * which we can't sleep and is sure it won't affect qgroup reserved space.
4071 * Like clear_bit_hook().
4073 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4076 struct btrfs_root *root = BTRFS_I(inode)->root;
4077 struct btrfs_space_info *data_sinfo;
4079 /* Make sure the range is aligned to sectorsize */
4080 len = round_up(start + len, root->sectorsize) -
4081 round_down(start, root->sectorsize);
4082 start = round_down(start, root->sectorsize);
4084 data_sinfo = root->fs_info->data_sinfo;
4085 spin_lock(&data_sinfo->lock);
4086 if (WARN_ON(data_sinfo->bytes_may_use < len))
4087 data_sinfo->bytes_may_use = 0;
4089 data_sinfo->bytes_may_use -= len;
4090 trace_btrfs_space_reservation(root->fs_info, "space_info",
4091 data_sinfo->flags, len, 0);
4092 spin_unlock(&data_sinfo->lock);
4096 * Called if we need to clear a data reservation for this inode
4097 * Normally in a error case.
4099 * This one will handle the per-indoe data rsv map for accurate reserved
4102 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4104 btrfs_free_reserved_data_space_noquota(inode, start, len);
4105 btrfs_qgroup_free_data(inode, start, len);
4108 static void force_metadata_allocation(struct btrfs_fs_info *info)
4110 struct list_head *head = &info->space_info;
4111 struct btrfs_space_info *found;
4114 list_for_each_entry_rcu(found, head, list) {
4115 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4116 found->force_alloc = CHUNK_ALLOC_FORCE;
4121 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4123 return (global->size << 1);
4126 static int should_alloc_chunk(struct btrfs_root *root,
4127 struct btrfs_space_info *sinfo, int force)
4129 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4130 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4131 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4134 if (force == CHUNK_ALLOC_FORCE)
4138 * We need to take into account the global rsv because for all intents
4139 * and purposes it's used space. Don't worry about locking the
4140 * global_rsv, it doesn't change except when the transaction commits.
4142 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4143 num_allocated += calc_global_rsv_need_space(global_rsv);
4146 * in limited mode, we want to have some free space up to
4147 * about 1% of the FS size.
4149 if (force == CHUNK_ALLOC_LIMITED) {
4150 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4151 thresh = max_t(u64, 64 * 1024 * 1024,
4152 div_factor_fine(thresh, 1));
4154 if (num_bytes - num_allocated < thresh)
4158 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4163 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4167 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4168 BTRFS_BLOCK_GROUP_RAID0 |
4169 BTRFS_BLOCK_GROUP_RAID5 |
4170 BTRFS_BLOCK_GROUP_RAID6))
4171 num_dev = root->fs_info->fs_devices->rw_devices;
4172 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4175 num_dev = 1; /* DUP or single */
4181 * If @is_allocation is true, reserve space in the system space info necessary
4182 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4185 void check_system_chunk(struct btrfs_trans_handle *trans,
4186 struct btrfs_root *root,
4189 struct btrfs_space_info *info;
4196 * Needed because we can end up allocating a system chunk and for an
4197 * atomic and race free space reservation in the chunk block reserve.
4199 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4201 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4202 spin_lock(&info->lock);
4203 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4204 info->bytes_reserved - info->bytes_readonly -
4205 info->bytes_may_use;
4206 spin_unlock(&info->lock);
4208 num_devs = get_profile_num_devs(root, type);
4210 /* num_devs device items to update and 1 chunk item to add or remove */
4211 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4212 btrfs_calc_trans_metadata_size(root, 1);
4214 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4215 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4216 left, thresh, type);
4217 dump_space_info(info, 0, 0);
4220 if (left < thresh) {
4223 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4225 * Ignore failure to create system chunk. We might end up not
4226 * needing it, as we might not need to COW all nodes/leafs from
4227 * the paths we visit in the chunk tree (they were already COWed
4228 * or created in the current transaction for example).
4230 ret = btrfs_alloc_chunk(trans, root, flags);
4234 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4235 &root->fs_info->chunk_block_rsv,
4236 thresh, BTRFS_RESERVE_NO_FLUSH);
4238 trans->chunk_bytes_reserved += thresh;
4242 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4243 struct btrfs_root *extent_root, u64 flags, int force)
4245 struct btrfs_space_info *space_info;
4246 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4247 int wait_for_alloc = 0;
4250 /* Don't re-enter if we're already allocating a chunk */
4251 if (trans->allocating_chunk)
4254 space_info = __find_space_info(extent_root->fs_info, flags);
4256 ret = update_space_info(extent_root->fs_info, flags,
4258 BUG_ON(ret); /* -ENOMEM */
4260 BUG_ON(!space_info); /* Logic error */
4263 spin_lock(&space_info->lock);
4264 if (force < space_info->force_alloc)
4265 force = space_info->force_alloc;
4266 if (space_info->full) {
4267 if (should_alloc_chunk(extent_root, space_info, force))
4271 spin_unlock(&space_info->lock);
4275 if (!should_alloc_chunk(extent_root, space_info, force)) {
4276 spin_unlock(&space_info->lock);
4278 } else if (space_info->chunk_alloc) {
4281 space_info->chunk_alloc = 1;
4284 spin_unlock(&space_info->lock);
4286 mutex_lock(&fs_info->chunk_mutex);
4289 * The chunk_mutex is held throughout the entirety of a chunk
4290 * allocation, so once we've acquired the chunk_mutex we know that the
4291 * other guy is done and we need to recheck and see if we should
4294 if (wait_for_alloc) {
4295 mutex_unlock(&fs_info->chunk_mutex);
4300 trans->allocating_chunk = true;
4303 * If we have mixed data/metadata chunks we want to make sure we keep
4304 * allocating mixed chunks instead of individual chunks.
4306 if (btrfs_mixed_space_info(space_info))
4307 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4310 * if we're doing a data chunk, go ahead and make sure that
4311 * we keep a reasonable number of metadata chunks allocated in the
4314 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4315 fs_info->data_chunk_allocations++;
4316 if (!(fs_info->data_chunk_allocations %
4317 fs_info->metadata_ratio))
4318 force_metadata_allocation(fs_info);
4322 * Check if we have enough space in SYSTEM chunk because we may need
4323 * to update devices.
4325 check_system_chunk(trans, extent_root, flags);
4327 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4328 trans->allocating_chunk = false;
4330 spin_lock(&space_info->lock);
4331 if (ret < 0 && ret != -ENOSPC)
4334 space_info->full = 1;
4338 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4340 space_info->chunk_alloc = 0;
4341 spin_unlock(&space_info->lock);
4342 mutex_unlock(&fs_info->chunk_mutex);
4344 * When we allocate a new chunk we reserve space in the chunk block
4345 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4346 * add new nodes/leafs to it if we end up needing to do it when
4347 * inserting the chunk item and updating device items as part of the
4348 * second phase of chunk allocation, performed by
4349 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4350 * large number of new block groups to create in our transaction
4351 * handle's new_bgs list to avoid exhausting the chunk block reserve
4352 * in extreme cases - like having a single transaction create many new
4353 * block groups when starting to write out the free space caches of all
4354 * the block groups that were made dirty during the lifetime of the
4357 if (trans->can_flush_pending_bgs &&
4358 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4359 btrfs_create_pending_block_groups(trans, trans->root);
4360 btrfs_trans_release_chunk_metadata(trans);
4365 static int can_overcommit(struct btrfs_root *root,
4366 struct btrfs_space_info *space_info, u64 bytes,
4367 enum btrfs_reserve_flush_enum flush)
4369 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4370 u64 profile = btrfs_get_alloc_profile(root, 0);
4375 used = space_info->bytes_used + space_info->bytes_reserved +
4376 space_info->bytes_pinned + space_info->bytes_readonly;
4379 * We only want to allow over committing if we have lots of actual space
4380 * free, but if we don't have enough space to handle the global reserve
4381 * space then we could end up having a real enospc problem when trying
4382 * to allocate a chunk or some other such important allocation.
4384 spin_lock(&global_rsv->lock);
4385 space_size = calc_global_rsv_need_space(global_rsv);
4386 spin_unlock(&global_rsv->lock);
4387 if (used + space_size >= space_info->total_bytes)
4390 used += space_info->bytes_may_use;
4392 spin_lock(&root->fs_info->free_chunk_lock);
4393 avail = root->fs_info->free_chunk_space;
4394 spin_unlock(&root->fs_info->free_chunk_lock);
4397 * If we have dup, raid1 or raid10 then only half of the free
4398 * space is actually useable. For raid56, the space info used
4399 * doesn't include the parity drive, so we don't have to
4402 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4403 BTRFS_BLOCK_GROUP_RAID1 |
4404 BTRFS_BLOCK_GROUP_RAID10))
4408 * If we aren't flushing all things, let us overcommit up to
4409 * 1/2th of the space. If we can flush, don't let us overcommit
4410 * too much, let it overcommit up to 1/8 of the space.
4412 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4417 if (used + bytes < space_info->total_bytes + avail)
4422 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4423 unsigned long nr_pages, int nr_items)
4425 struct super_block *sb = root->fs_info->sb;
4427 if (down_read_trylock(&sb->s_umount)) {
4428 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4429 up_read(&sb->s_umount);
4432 * We needn't worry the filesystem going from r/w to r/o though
4433 * we don't acquire ->s_umount mutex, because the filesystem
4434 * should guarantee the delalloc inodes list be empty after
4435 * the filesystem is readonly(all dirty pages are written to
4438 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4439 if (!current->journal_info)
4440 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4444 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4449 bytes = btrfs_calc_trans_metadata_size(root, 1);
4450 nr = (int)div64_u64(to_reclaim, bytes);
4456 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4459 * shrink metadata reservation for delalloc
4461 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4464 struct btrfs_block_rsv *block_rsv;
4465 struct btrfs_space_info *space_info;
4466 struct btrfs_trans_handle *trans;
4470 unsigned long nr_pages;
4473 enum btrfs_reserve_flush_enum flush;
4475 /* Calc the number of the pages we need flush for space reservation */
4476 items = calc_reclaim_items_nr(root, to_reclaim);
4477 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4479 trans = (struct btrfs_trans_handle *)current->journal_info;
4480 block_rsv = &root->fs_info->delalloc_block_rsv;
4481 space_info = block_rsv->space_info;
4483 delalloc_bytes = percpu_counter_sum_positive(
4484 &root->fs_info->delalloc_bytes);
4485 if (delalloc_bytes == 0) {
4489 btrfs_wait_ordered_roots(root->fs_info, items);
4494 while (delalloc_bytes && loops < 3) {
4495 max_reclaim = min(delalloc_bytes, to_reclaim);
4496 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4497 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4499 * We need to wait for the async pages to actually start before
4502 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4506 if (max_reclaim <= nr_pages)
4509 max_reclaim -= nr_pages;
4511 wait_event(root->fs_info->async_submit_wait,
4512 atomic_read(&root->fs_info->async_delalloc_pages) <=
4516 flush = BTRFS_RESERVE_FLUSH_ALL;
4518 flush = BTRFS_RESERVE_NO_FLUSH;
4519 spin_lock(&space_info->lock);
4520 if (can_overcommit(root, space_info, orig, flush)) {
4521 spin_unlock(&space_info->lock);
4524 spin_unlock(&space_info->lock);
4527 if (wait_ordered && !trans) {
4528 btrfs_wait_ordered_roots(root->fs_info, items);
4530 time_left = schedule_timeout_killable(1);
4534 delalloc_bytes = percpu_counter_sum_positive(
4535 &root->fs_info->delalloc_bytes);
4540 * maybe_commit_transaction - possibly commit the transaction if its ok to
4541 * @root - the root we're allocating for
4542 * @bytes - the number of bytes we want to reserve
4543 * @force - force the commit
4545 * This will check to make sure that committing the transaction will actually
4546 * get us somewhere and then commit the transaction if it does. Otherwise it
4547 * will return -ENOSPC.
4549 static int may_commit_transaction(struct btrfs_root *root,
4550 struct btrfs_space_info *space_info,
4551 u64 bytes, int force)
4553 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4554 struct btrfs_trans_handle *trans;
4556 trans = (struct btrfs_trans_handle *)current->journal_info;
4563 /* See if there is enough pinned space to make this reservation */
4564 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4569 * See if there is some space in the delayed insertion reservation for
4572 if (space_info != delayed_rsv->space_info)
4575 spin_lock(&delayed_rsv->lock);
4576 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4577 bytes - delayed_rsv->size) >= 0) {
4578 spin_unlock(&delayed_rsv->lock);
4581 spin_unlock(&delayed_rsv->lock);
4584 trans = btrfs_join_transaction(root);
4588 return btrfs_commit_transaction(trans, root);
4592 FLUSH_DELAYED_ITEMS_NR = 1,
4593 FLUSH_DELAYED_ITEMS = 2,
4595 FLUSH_DELALLOC_WAIT = 4,
4600 static int flush_space(struct btrfs_root *root,
4601 struct btrfs_space_info *space_info, u64 num_bytes,
4602 u64 orig_bytes, int state)
4604 struct btrfs_trans_handle *trans;
4609 case FLUSH_DELAYED_ITEMS_NR:
4610 case FLUSH_DELAYED_ITEMS:
4611 if (state == FLUSH_DELAYED_ITEMS_NR)
4612 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4616 trans = btrfs_join_transaction(root);
4617 if (IS_ERR(trans)) {
4618 ret = PTR_ERR(trans);
4621 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4622 btrfs_end_transaction(trans, root);
4624 case FLUSH_DELALLOC:
4625 case FLUSH_DELALLOC_WAIT:
4626 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4627 state == FLUSH_DELALLOC_WAIT);
4630 trans = btrfs_join_transaction(root);
4631 if (IS_ERR(trans)) {
4632 ret = PTR_ERR(trans);
4635 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4636 btrfs_get_alloc_profile(root, 0),
4637 CHUNK_ALLOC_NO_FORCE);
4638 btrfs_end_transaction(trans, root);
4643 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4654 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4655 struct btrfs_space_info *space_info)
4661 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4663 spin_lock(&space_info->lock);
4664 if (can_overcommit(root, space_info, to_reclaim,
4665 BTRFS_RESERVE_FLUSH_ALL)) {
4670 used = space_info->bytes_used + space_info->bytes_reserved +
4671 space_info->bytes_pinned + space_info->bytes_readonly +
4672 space_info->bytes_may_use;
4673 if (can_overcommit(root, space_info, 1024 * 1024,
4674 BTRFS_RESERVE_FLUSH_ALL))
4675 expected = div_factor_fine(space_info->total_bytes, 95);
4677 expected = div_factor_fine(space_info->total_bytes, 90);
4679 if (used > expected)
4680 to_reclaim = used - expected;
4683 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4684 space_info->bytes_reserved);
4686 spin_unlock(&space_info->lock);
4691 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4692 struct btrfs_fs_info *fs_info, u64 used)
4694 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4696 /* If we're just plain full then async reclaim just slows us down. */
4697 if (space_info->bytes_used >= thresh)
4700 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4701 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4704 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4705 struct btrfs_fs_info *fs_info,
4710 spin_lock(&space_info->lock);
4712 * We run out of space and have not got any free space via flush_space,
4713 * so don't bother doing async reclaim.
4715 if (flush_state > COMMIT_TRANS && space_info->full) {
4716 spin_unlock(&space_info->lock);
4720 used = space_info->bytes_used + space_info->bytes_reserved +
4721 space_info->bytes_pinned + space_info->bytes_readonly +
4722 space_info->bytes_may_use;
4723 if (need_do_async_reclaim(space_info, fs_info, used)) {
4724 spin_unlock(&space_info->lock);
4727 spin_unlock(&space_info->lock);
4732 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4734 struct btrfs_fs_info *fs_info;
4735 struct btrfs_space_info *space_info;
4739 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4740 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4742 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4747 flush_state = FLUSH_DELAYED_ITEMS_NR;
4749 flush_space(fs_info->fs_root, space_info, to_reclaim,
4750 to_reclaim, flush_state);
4752 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4755 } while (flush_state < COMMIT_TRANS);
4758 void btrfs_init_async_reclaim_work(struct work_struct *work)
4760 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4764 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4765 * @root - the root we're allocating for
4766 * @block_rsv - the block_rsv we're allocating for
4767 * @orig_bytes - the number of bytes we want
4768 * @flush - whether or not we can flush to make our reservation
4770 * This will reserve orgi_bytes number of bytes from the space info associated
4771 * with the block_rsv. If there is not enough space it will make an attempt to
4772 * flush out space to make room. It will do this by flushing delalloc if
4773 * possible or committing the transaction. If flush is 0 then no attempts to
4774 * regain reservations will be made and this will fail if there is not enough
4777 static int reserve_metadata_bytes(struct btrfs_root *root,
4778 struct btrfs_block_rsv *block_rsv,
4780 enum btrfs_reserve_flush_enum flush)
4782 struct btrfs_space_info *space_info = block_rsv->space_info;
4784 u64 num_bytes = orig_bytes;
4785 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4787 bool flushing = false;
4791 spin_lock(&space_info->lock);
4793 * We only want to wait if somebody other than us is flushing and we
4794 * are actually allowed to flush all things.
4796 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4797 space_info->flush) {
4798 spin_unlock(&space_info->lock);
4800 * If we have a trans handle we can't wait because the flusher
4801 * may have to commit the transaction, which would mean we would
4802 * deadlock since we are waiting for the flusher to finish, but
4803 * hold the current transaction open.
4805 if (current->journal_info)
4807 ret = wait_event_killable(space_info->wait, !space_info->flush);
4808 /* Must have been killed, return */
4812 spin_lock(&space_info->lock);
4816 used = space_info->bytes_used + space_info->bytes_reserved +
4817 space_info->bytes_pinned + space_info->bytes_readonly +
4818 space_info->bytes_may_use;
4821 * The idea here is that we've not already over-reserved the block group
4822 * then we can go ahead and save our reservation first and then start
4823 * flushing if we need to. Otherwise if we've already overcommitted
4824 * lets start flushing stuff first and then come back and try to make
4827 if (used <= space_info->total_bytes) {
4828 if (used + orig_bytes <= space_info->total_bytes) {
4829 space_info->bytes_may_use += orig_bytes;
4830 trace_btrfs_space_reservation(root->fs_info,
4831 "space_info", space_info->flags, orig_bytes, 1);
4835 * Ok set num_bytes to orig_bytes since we aren't
4836 * overocmmitted, this way we only try and reclaim what
4839 num_bytes = orig_bytes;
4843 * Ok we're over committed, set num_bytes to the overcommitted
4844 * amount plus the amount of bytes that we need for this
4847 num_bytes = used - space_info->total_bytes +
4851 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4852 space_info->bytes_may_use += orig_bytes;
4853 trace_btrfs_space_reservation(root->fs_info, "space_info",
4854 space_info->flags, orig_bytes,
4860 * Couldn't make our reservation, save our place so while we're trying
4861 * to reclaim space we can actually use it instead of somebody else
4862 * stealing it from us.
4864 * We make the other tasks wait for the flush only when we can flush
4867 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4869 space_info->flush = 1;
4870 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4873 * We will do the space reservation dance during log replay,
4874 * which means we won't have fs_info->fs_root set, so don't do
4875 * the async reclaim as we will panic.
4877 if (!root->fs_info->log_root_recovering &&
4878 need_do_async_reclaim(space_info, root->fs_info, used) &&
4879 !work_busy(&root->fs_info->async_reclaim_work))
4880 queue_work(system_unbound_wq,
4881 &root->fs_info->async_reclaim_work);
4883 spin_unlock(&space_info->lock);
4885 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4888 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4893 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4894 * would happen. So skip delalloc flush.
4896 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4897 (flush_state == FLUSH_DELALLOC ||
4898 flush_state == FLUSH_DELALLOC_WAIT))
4899 flush_state = ALLOC_CHUNK;
4903 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4904 flush_state < COMMIT_TRANS)
4906 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4907 flush_state <= COMMIT_TRANS)
4911 if (ret == -ENOSPC &&
4912 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4913 struct btrfs_block_rsv *global_rsv =
4914 &root->fs_info->global_block_rsv;
4916 if (block_rsv != global_rsv &&
4917 !block_rsv_use_bytes(global_rsv, orig_bytes))
4921 trace_btrfs_space_reservation(root->fs_info,
4922 "space_info:enospc",
4923 space_info->flags, orig_bytes, 1);
4925 spin_lock(&space_info->lock);
4926 space_info->flush = 0;
4927 wake_up_all(&space_info->wait);
4928 spin_unlock(&space_info->lock);
4933 static struct btrfs_block_rsv *get_block_rsv(
4934 const struct btrfs_trans_handle *trans,
4935 const struct btrfs_root *root)
4937 struct btrfs_block_rsv *block_rsv = NULL;
4939 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4940 (root == root->fs_info->csum_root && trans->adding_csums) ||
4941 (root == root->fs_info->uuid_root))
4942 block_rsv = trans->block_rsv;
4945 block_rsv = root->block_rsv;
4948 block_rsv = &root->fs_info->empty_block_rsv;
4953 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4957 spin_lock(&block_rsv->lock);
4958 if (block_rsv->reserved >= num_bytes) {
4959 block_rsv->reserved -= num_bytes;
4960 if (block_rsv->reserved < block_rsv->size)
4961 block_rsv->full = 0;
4964 spin_unlock(&block_rsv->lock);
4968 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4969 u64 num_bytes, int update_size)
4971 spin_lock(&block_rsv->lock);
4972 block_rsv->reserved += num_bytes;
4974 block_rsv->size += num_bytes;
4975 else if (block_rsv->reserved >= block_rsv->size)
4976 block_rsv->full = 1;
4977 spin_unlock(&block_rsv->lock);
4980 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4981 struct btrfs_block_rsv *dest, u64 num_bytes,
4984 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4987 if (global_rsv->space_info != dest->space_info)
4990 spin_lock(&global_rsv->lock);
4991 min_bytes = div_factor(global_rsv->size, min_factor);
4992 if (global_rsv->reserved < min_bytes + num_bytes) {
4993 spin_unlock(&global_rsv->lock);
4996 global_rsv->reserved -= num_bytes;
4997 if (global_rsv->reserved < global_rsv->size)
4998 global_rsv->full = 0;
4999 spin_unlock(&global_rsv->lock);
5001 block_rsv_add_bytes(dest, num_bytes, 1);
5005 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5006 struct btrfs_block_rsv *block_rsv,
5007 struct btrfs_block_rsv *dest, u64 num_bytes)
5009 struct btrfs_space_info *space_info = block_rsv->space_info;
5011 spin_lock(&block_rsv->lock);
5012 if (num_bytes == (u64)-1)
5013 num_bytes = block_rsv->size;
5014 block_rsv->size -= num_bytes;
5015 if (block_rsv->reserved >= block_rsv->size) {
5016 num_bytes = block_rsv->reserved - block_rsv->size;
5017 block_rsv->reserved = block_rsv->size;
5018 block_rsv->full = 1;
5022 spin_unlock(&block_rsv->lock);
5024 if (num_bytes > 0) {
5026 spin_lock(&dest->lock);
5030 bytes_to_add = dest->size - dest->reserved;
5031 bytes_to_add = min(num_bytes, bytes_to_add);
5032 dest->reserved += bytes_to_add;
5033 if (dest->reserved >= dest->size)
5035 num_bytes -= bytes_to_add;
5037 spin_unlock(&dest->lock);
5040 spin_lock(&space_info->lock);
5041 space_info->bytes_may_use -= num_bytes;
5042 trace_btrfs_space_reservation(fs_info, "space_info",
5043 space_info->flags, num_bytes, 0);
5044 spin_unlock(&space_info->lock);
5049 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5050 struct btrfs_block_rsv *dst, u64 num_bytes)
5054 ret = block_rsv_use_bytes(src, num_bytes);
5058 block_rsv_add_bytes(dst, num_bytes, 1);
5062 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5064 memset(rsv, 0, sizeof(*rsv));
5065 spin_lock_init(&rsv->lock);
5069 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5070 unsigned short type)
5072 struct btrfs_block_rsv *block_rsv;
5073 struct btrfs_fs_info *fs_info = root->fs_info;
5075 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5079 btrfs_init_block_rsv(block_rsv, type);
5080 block_rsv->space_info = __find_space_info(fs_info,
5081 BTRFS_BLOCK_GROUP_METADATA);
5085 void btrfs_free_block_rsv(struct btrfs_root *root,
5086 struct btrfs_block_rsv *rsv)
5090 btrfs_block_rsv_release(root, rsv, (u64)-1);
5094 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5099 int btrfs_block_rsv_add(struct btrfs_root *root,
5100 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5101 enum btrfs_reserve_flush_enum flush)
5108 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5110 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5117 int btrfs_block_rsv_check(struct btrfs_root *root,
5118 struct btrfs_block_rsv *block_rsv, int min_factor)
5126 spin_lock(&block_rsv->lock);
5127 num_bytes = div_factor(block_rsv->size, min_factor);
5128 if (block_rsv->reserved >= num_bytes)
5130 spin_unlock(&block_rsv->lock);
5135 int btrfs_block_rsv_refill(struct btrfs_root *root,
5136 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5137 enum btrfs_reserve_flush_enum flush)
5145 spin_lock(&block_rsv->lock);
5146 num_bytes = min_reserved;
5147 if (block_rsv->reserved >= num_bytes)
5150 num_bytes -= block_rsv->reserved;
5151 spin_unlock(&block_rsv->lock);
5156 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5158 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5165 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5166 struct btrfs_block_rsv *dst_rsv,
5169 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5172 void btrfs_block_rsv_release(struct btrfs_root *root,
5173 struct btrfs_block_rsv *block_rsv,
5176 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5177 if (global_rsv == block_rsv ||
5178 block_rsv->space_info != global_rsv->space_info)
5180 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5185 * helper to calculate size of global block reservation.
5186 * the desired value is sum of space used by extent tree,
5187 * checksum tree and root tree
5189 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5191 struct btrfs_space_info *sinfo;
5195 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5197 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5198 spin_lock(&sinfo->lock);
5199 data_used = sinfo->bytes_used;
5200 spin_unlock(&sinfo->lock);
5202 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5203 spin_lock(&sinfo->lock);
5204 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5206 meta_used = sinfo->bytes_used;
5207 spin_unlock(&sinfo->lock);
5209 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5211 num_bytes += div_u64(data_used + meta_used, 50);
5213 if (num_bytes * 3 > meta_used)
5214 num_bytes = div_u64(meta_used, 3);
5216 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5219 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5221 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5222 struct btrfs_space_info *sinfo = block_rsv->space_info;
5225 num_bytes = calc_global_metadata_size(fs_info);
5227 spin_lock(&sinfo->lock);
5228 spin_lock(&block_rsv->lock);
5230 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5232 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5233 sinfo->bytes_reserved + sinfo->bytes_readonly +
5234 sinfo->bytes_may_use;
5236 if (sinfo->total_bytes > num_bytes) {
5237 num_bytes = sinfo->total_bytes - num_bytes;
5238 block_rsv->reserved += num_bytes;
5239 sinfo->bytes_may_use += num_bytes;
5240 trace_btrfs_space_reservation(fs_info, "space_info",
5241 sinfo->flags, num_bytes, 1);
5244 if (block_rsv->reserved >= block_rsv->size) {
5245 num_bytes = block_rsv->reserved - block_rsv->size;
5246 sinfo->bytes_may_use -= num_bytes;
5247 trace_btrfs_space_reservation(fs_info, "space_info",
5248 sinfo->flags, num_bytes, 0);
5249 block_rsv->reserved = block_rsv->size;
5250 block_rsv->full = 1;
5253 spin_unlock(&block_rsv->lock);
5254 spin_unlock(&sinfo->lock);
5257 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5259 struct btrfs_space_info *space_info;
5261 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5262 fs_info->chunk_block_rsv.space_info = space_info;
5264 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5265 fs_info->global_block_rsv.space_info = space_info;
5266 fs_info->delalloc_block_rsv.space_info = space_info;
5267 fs_info->trans_block_rsv.space_info = space_info;
5268 fs_info->empty_block_rsv.space_info = space_info;
5269 fs_info->delayed_block_rsv.space_info = space_info;
5271 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5272 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5273 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5274 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5275 if (fs_info->quota_root)
5276 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5277 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5279 update_global_block_rsv(fs_info);
5282 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5284 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5286 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5287 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5288 WARN_ON(fs_info->trans_block_rsv.size > 0);
5289 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5290 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5291 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5292 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5293 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5296 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5297 struct btrfs_root *root)
5299 if (!trans->block_rsv)
5302 if (!trans->bytes_reserved)
5305 trace_btrfs_space_reservation(root->fs_info, "transaction",
5306 trans->transid, trans->bytes_reserved, 0);
5307 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5308 trans->bytes_reserved = 0;
5312 * To be called after all the new block groups attached to the transaction
5313 * handle have been created (btrfs_create_pending_block_groups()).
5315 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5317 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5319 if (!trans->chunk_bytes_reserved)
5322 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5324 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5325 trans->chunk_bytes_reserved);
5326 trans->chunk_bytes_reserved = 0;
5329 /* Can only return 0 or -ENOSPC */
5330 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5331 struct inode *inode)
5333 struct btrfs_root *root = BTRFS_I(inode)->root;
5334 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5335 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5338 * We need to hold space in order to delete our orphan item once we've
5339 * added it, so this takes the reservation so we can release it later
5340 * when we are truly done with the orphan item.
5342 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5343 trace_btrfs_space_reservation(root->fs_info, "orphan",
5344 btrfs_ino(inode), num_bytes, 1);
5345 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5348 void btrfs_orphan_release_metadata(struct inode *inode)
5350 struct btrfs_root *root = BTRFS_I(inode)->root;
5351 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5352 trace_btrfs_space_reservation(root->fs_info, "orphan",
5353 btrfs_ino(inode), num_bytes, 0);
5354 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5358 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5359 * root: the root of the parent directory
5360 * rsv: block reservation
5361 * items: the number of items that we need do reservation
5362 * qgroup_reserved: used to return the reserved size in qgroup
5364 * This function is used to reserve the space for snapshot/subvolume
5365 * creation and deletion. Those operations are different with the
5366 * common file/directory operations, they change two fs/file trees
5367 * and root tree, the number of items that the qgroup reserves is
5368 * different with the free space reservation. So we can not use
5369 * the space reseravtion mechanism in start_transaction().
5371 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5372 struct btrfs_block_rsv *rsv,
5374 u64 *qgroup_reserved,
5375 bool use_global_rsv)
5379 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5381 if (root->fs_info->quota_enabled) {
5382 /* One for parent inode, two for dir entries */
5383 num_bytes = 3 * root->nodesize;
5384 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5391 *qgroup_reserved = num_bytes;
5393 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5394 rsv->space_info = __find_space_info(root->fs_info,
5395 BTRFS_BLOCK_GROUP_METADATA);
5396 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5397 BTRFS_RESERVE_FLUSH_ALL);
5399 if (ret == -ENOSPC && use_global_rsv)
5400 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5402 if (ret && *qgroup_reserved)
5403 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5408 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5409 struct btrfs_block_rsv *rsv,
5410 u64 qgroup_reserved)
5412 btrfs_block_rsv_release(root, rsv, (u64)-1);
5416 * drop_outstanding_extent - drop an outstanding extent
5417 * @inode: the inode we're dropping the extent for
5418 * @num_bytes: the number of bytes we're relaseing.
5420 * This is called when we are freeing up an outstanding extent, either called
5421 * after an error or after an extent is written. This will return the number of
5422 * reserved extents that need to be freed. This must be called with
5423 * BTRFS_I(inode)->lock held.
5425 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5427 unsigned drop_inode_space = 0;
5428 unsigned dropped_extents = 0;
5429 unsigned num_extents = 0;
5431 num_extents = (unsigned)div64_u64(num_bytes +
5432 BTRFS_MAX_EXTENT_SIZE - 1,
5433 BTRFS_MAX_EXTENT_SIZE);
5434 ASSERT(num_extents);
5435 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5436 BTRFS_I(inode)->outstanding_extents -= num_extents;
5438 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5439 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5440 &BTRFS_I(inode)->runtime_flags))
5441 drop_inode_space = 1;
5444 * If we have more or the same amount of outsanding extents than we have
5445 * reserved then we need to leave the reserved extents count alone.
5447 if (BTRFS_I(inode)->outstanding_extents >=
5448 BTRFS_I(inode)->reserved_extents)
5449 return drop_inode_space;
5451 dropped_extents = BTRFS_I(inode)->reserved_extents -
5452 BTRFS_I(inode)->outstanding_extents;
5453 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5454 return dropped_extents + drop_inode_space;
5458 * calc_csum_metadata_size - return the amount of metada space that must be
5459 * reserved/free'd for the given bytes.
5460 * @inode: the inode we're manipulating
5461 * @num_bytes: the number of bytes in question
5462 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5464 * This adjusts the number of csum_bytes in the inode and then returns the
5465 * correct amount of metadata that must either be reserved or freed. We
5466 * calculate how many checksums we can fit into one leaf and then divide the
5467 * number of bytes that will need to be checksumed by this value to figure out
5468 * how many checksums will be required. If we are adding bytes then the number
5469 * may go up and we will return the number of additional bytes that must be
5470 * reserved. If it is going down we will return the number of bytes that must
5473 * This must be called with BTRFS_I(inode)->lock held.
5475 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5478 struct btrfs_root *root = BTRFS_I(inode)->root;
5479 u64 old_csums, num_csums;
5481 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5482 BTRFS_I(inode)->csum_bytes == 0)
5485 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5487 BTRFS_I(inode)->csum_bytes += num_bytes;
5489 BTRFS_I(inode)->csum_bytes -= num_bytes;
5490 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5492 /* No change, no need to reserve more */
5493 if (old_csums == num_csums)
5497 return btrfs_calc_trans_metadata_size(root,
5498 num_csums - old_csums);
5500 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5503 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5505 struct btrfs_root *root = BTRFS_I(inode)->root;
5506 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5509 unsigned nr_extents = 0;
5510 int extra_reserve = 0;
5511 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5513 bool delalloc_lock = true;
5517 /* If we are a free space inode we need to not flush since we will be in
5518 * the middle of a transaction commit. We also don't need the delalloc
5519 * mutex since we won't race with anybody. We need this mostly to make
5520 * lockdep shut its filthy mouth.
5522 if (btrfs_is_free_space_inode(inode)) {
5523 flush = BTRFS_RESERVE_NO_FLUSH;
5524 delalloc_lock = false;
5527 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5528 btrfs_transaction_in_commit(root->fs_info))
5529 schedule_timeout(1);
5532 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5534 num_bytes = ALIGN(num_bytes, root->sectorsize);
5536 spin_lock(&BTRFS_I(inode)->lock);
5537 nr_extents = (unsigned)div64_u64(num_bytes +
5538 BTRFS_MAX_EXTENT_SIZE - 1,
5539 BTRFS_MAX_EXTENT_SIZE);
5540 BTRFS_I(inode)->outstanding_extents += nr_extents;
5543 if (BTRFS_I(inode)->outstanding_extents >
5544 BTRFS_I(inode)->reserved_extents)
5545 nr_extents = BTRFS_I(inode)->outstanding_extents -
5546 BTRFS_I(inode)->reserved_extents;
5549 * Add an item to reserve for updating the inode when we complete the
5552 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5553 &BTRFS_I(inode)->runtime_flags)) {
5558 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5559 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5560 csum_bytes = BTRFS_I(inode)->csum_bytes;
5561 spin_unlock(&BTRFS_I(inode)->lock);
5563 if (root->fs_info->quota_enabled) {
5564 ret = btrfs_qgroup_reserve_meta(root,
5565 nr_extents * root->nodesize);
5570 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5571 if (unlikely(ret)) {
5572 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5576 spin_lock(&BTRFS_I(inode)->lock);
5577 if (extra_reserve) {
5578 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5579 &BTRFS_I(inode)->runtime_flags);
5582 BTRFS_I(inode)->reserved_extents += nr_extents;
5583 spin_unlock(&BTRFS_I(inode)->lock);
5586 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5589 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5590 btrfs_ino(inode), to_reserve, 1);
5591 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5596 spin_lock(&BTRFS_I(inode)->lock);
5597 dropped = drop_outstanding_extent(inode, num_bytes);
5599 * If the inodes csum_bytes is the same as the original
5600 * csum_bytes then we know we haven't raced with any free()ers
5601 * so we can just reduce our inodes csum bytes and carry on.
5603 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5604 calc_csum_metadata_size(inode, num_bytes, 0);
5606 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5610 * This is tricky, but first we need to figure out how much we
5611 * free'd from any free-ers that occured during this
5612 * reservation, so we reset ->csum_bytes to the csum_bytes
5613 * before we dropped our lock, and then call the free for the
5614 * number of bytes that were freed while we were trying our
5617 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5618 BTRFS_I(inode)->csum_bytes = csum_bytes;
5619 to_free = calc_csum_metadata_size(inode, bytes, 0);
5623 * Now we need to see how much we would have freed had we not
5624 * been making this reservation and our ->csum_bytes were not
5625 * artificially inflated.
5627 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5628 bytes = csum_bytes - orig_csum_bytes;
5629 bytes = calc_csum_metadata_size(inode, bytes, 0);
5632 * Now reset ->csum_bytes to what it should be. If bytes is
5633 * more than to_free then we would have free'd more space had we
5634 * not had an artificially high ->csum_bytes, so we need to free
5635 * the remainder. If bytes is the same or less then we don't
5636 * need to do anything, the other free-ers did the correct
5639 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5640 if (bytes > to_free)
5641 to_free = bytes - to_free;
5645 spin_unlock(&BTRFS_I(inode)->lock);
5647 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5650 btrfs_block_rsv_release(root, block_rsv, to_free);
5651 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5652 btrfs_ino(inode), to_free, 0);
5655 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5660 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5661 * @inode: the inode to release the reservation for
5662 * @num_bytes: the number of bytes we're releasing
5664 * This will release the metadata reservation for an inode. This can be called
5665 * once we complete IO for a given set of bytes to release their metadata
5668 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5670 struct btrfs_root *root = BTRFS_I(inode)->root;
5674 num_bytes = ALIGN(num_bytes, root->sectorsize);
5675 spin_lock(&BTRFS_I(inode)->lock);
5676 dropped = drop_outstanding_extent(inode, num_bytes);
5679 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5680 spin_unlock(&BTRFS_I(inode)->lock);
5682 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5684 if (btrfs_test_is_dummy_root(root))
5687 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5688 btrfs_ino(inode), to_free, 0);
5690 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5695 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5697 * @inode: inode we're writing to
5698 * @start: start range we are writing to
5699 * @len: how long the range we are writing to
5701 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5703 * This will do the following things
5705 * o reserve space in data space info for num bytes
5706 * and reserve precious corresponding qgroup space
5707 * (Done in check_data_free_space)
5709 * o reserve space for metadata space, based on the number of outstanding
5710 * extents and how much csums will be needed
5711 * also reserve metadata space in a per root over-reserve method.
5712 * o add to the inodes->delalloc_bytes
5713 * o add it to the fs_info's delalloc inodes list.
5714 * (Above 3 all done in delalloc_reserve_metadata)
5716 * Return 0 for success
5717 * Return <0 for error(-ENOSPC or -EQUOT)
5719 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5723 ret = btrfs_check_data_free_space(inode, start, len);
5726 ret = btrfs_delalloc_reserve_metadata(inode, len);
5728 btrfs_free_reserved_data_space(inode, start, len);
5733 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5734 * @inode: inode we're releasing space for
5735 * @start: start position of the space already reserved
5736 * @len: the len of the space already reserved
5738 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5739 * called in the case that we don't need the metadata AND data reservations
5740 * anymore. So if there is an error or we insert an inline extent.
5742 * This function will release the metadata space that was not used and will
5743 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5744 * list if there are no delalloc bytes left.
5745 * Also it will handle the qgroup reserved space.
5747 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5749 btrfs_delalloc_release_metadata(inode, len);
5750 btrfs_free_reserved_data_space(inode, start, len);
5753 static int update_block_group(struct btrfs_trans_handle *trans,
5754 struct btrfs_root *root, u64 bytenr,
5755 u64 num_bytes, int alloc)
5757 struct btrfs_block_group_cache *cache = NULL;
5758 struct btrfs_fs_info *info = root->fs_info;
5759 u64 total = num_bytes;
5764 /* block accounting for super block */
5765 spin_lock(&info->delalloc_root_lock);
5766 old_val = btrfs_super_bytes_used(info->super_copy);
5768 old_val += num_bytes;
5770 old_val -= num_bytes;
5771 btrfs_set_super_bytes_used(info->super_copy, old_val);
5772 spin_unlock(&info->delalloc_root_lock);
5775 cache = btrfs_lookup_block_group(info, bytenr);
5778 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5779 BTRFS_BLOCK_GROUP_RAID1 |
5780 BTRFS_BLOCK_GROUP_RAID10))
5785 * If this block group has free space cache written out, we
5786 * need to make sure to load it if we are removing space. This
5787 * is because we need the unpinning stage to actually add the
5788 * space back to the block group, otherwise we will leak space.
5790 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5791 cache_block_group(cache, 1);
5793 byte_in_group = bytenr - cache->key.objectid;
5794 WARN_ON(byte_in_group > cache->key.offset);
5796 spin_lock(&cache->space_info->lock);
5797 spin_lock(&cache->lock);
5799 if (btrfs_test_opt(root, SPACE_CACHE) &&
5800 cache->disk_cache_state < BTRFS_DC_CLEAR)
5801 cache->disk_cache_state = BTRFS_DC_CLEAR;
5803 old_val = btrfs_block_group_used(&cache->item);
5804 num_bytes = min(total, cache->key.offset - byte_in_group);
5806 old_val += num_bytes;
5807 btrfs_set_block_group_used(&cache->item, old_val);
5808 cache->reserved -= num_bytes;
5809 cache->space_info->bytes_reserved -= num_bytes;
5810 cache->space_info->bytes_used += num_bytes;
5811 cache->space_info->disk_used += num_bytes * factor;
5812 spin_unlock(&cache->lock);
5813 spin_unlock(&cache->space_info->lock);
5815 old_val -= num_bytes;
5816 btrfs_set_block_group_used(&cache->item, old_val);
5817 cache->pinned += num_bytes;
5818 cache->space_info->bytes_pinned += num_bytes;
5819 cache->space_info->bytes_used -= num_bytes;
5820 cache->space_info->disk_used -= num_bytes * factor;
5821 spin_unlock(&cache->lock);
5822 spin_unlock(&cache->space_info->lock);
5824 set_extent_dirty(info->pinned_extents,
5825 bytenr, bytenr + num_bytes - 1,
5826 GFP_NOFS | __GFP_NOFAIL);
5828 * No longer have used bytes in this block group, queue
5832 spin_lock(&info->unused_bgs_lock);
5833 if (list_empty(&cache->bg_list)) {
5834 btrfs_get_block_group(cache);
5835 list_add_tail(&cache->bg_list,
5838 spin_unlock(&info->unused_bgs_lock);
5842 spin_lock(&trans->transaction->dirty_bgs_lock);
5843 if (list_empty(&cache->dirty_list)) {
5844 list_add_tail(&cache->dirty_list,
5845 &trans->transaction->dirty_bgs);
5846 trans->transaction->num_dirty_bgs++;
5847 btrfs_get_block_group(cache);
5849 spin_unlock(&trans->transaction->dirty_bgs_lock);
5851 btrfs_put_block_group(cache);
5853 bytenr += num_bytes;
5858 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5860 struct btrfs_block_group_cache *cache;
5863 spin_lock(&root->fs_info->block_group_cache_lock);
5864 bytenr = root->fs_info->first_logical_byte;
5865 spin_unlock(&root->fs_info->block_group_cache_lock);
5867 if (bytenr < (u64)-1)
5870 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5874 bytenr = cache->key.objectid;
5875 btrfs_put_block_group(cache);
5880 static int pin_down_extent(struct btrfs_root *root,
5881 struct btrfs_block_group_cache *cache,
5882 u64 bytenr, u64 num_bytes, int reserved)
5884 spin_lock(&cache->space_info->lock);
5885 spin_lock(&cache->lock);
5886 cache->pinned += num_bytes;
5887 cache->space_info->bytes_pinned += num_bytes;
5889 cache->reserved -= num_bytes;
5890 cache->space_info->bytes_reserved -= num_bytes;
5892 spin_unlock(&cache->lock);
5893 spin_unlock(&cache->space_info->lock);
5895 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5896 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5898 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5903 * this function must be called within transaction
5905 int btrfs_pin_extent(struct btrfs_root *root,
5906 u64 bytenr, u64 num_bytes, int reserved)
5908 struct btrfs_block_group_cache *cache;
5910 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5911 BUG_ON(!cache); /* Logic error */
5913 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5915 btrfs_put_block_group(cache);
5920 * this function must be called within transaction
5922 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5923 u64 bytenr, u64 num_bytes)
5925 struct btrfs_block_group_cache *cache;
5928 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5933 * pull in the free space cache (if any) so that our pin
5934 * removes the free space from the cache. We have load_only set
5935 * to one because the slow code to read in the free extents does check
5936 * the pinned extents.
5938 cache_block_group(cache, 1);
5940 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5942 /* remove us from the free space cache (if we're there at all) */
5943 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5944 btrfs_put_block_group(cache);
5948 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5951 struct btrfs_block_group_cache *block_group;
5952 struct btrfs_caching_control *caching_ctl;
5954 block_group = btrfs_lookup_block_group(root->fs_info, start);
5958 cache_block_group(block_group, 0);
5959 caching_ctl = get_caching_control(block_group);
5963 BUG_ON(!block_group_cache_done(block_group));
5964 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5966 mutex_lock(&caching_ctl->mutex);
5968 if (start >= caching_ctl->progress) {
5969 ret = add_excluded_extent(root, start, num_bytes);
5970 } else if (start + num_bytes <= caching_ctl->progress) {
5971 ret = btrfs_remove_free_space(block_group,
5974 num_bytes = caching_ctl->progress - start;
5975 ret = btrfs_remove_free_space(block_group,
5980 num_bytes = (start + num_bytes) -
5981 caching_ctl->progress;
5982 start = caching_ctl->progress;
5983 ret = add_excluded_extent(root, start, num_bytes);
5986 mutex_unlock(&caching_ctl->mutex);
5987 put_caching_control(caching_ctl);
5989 btrfs_put_block_group(block_group);
5993 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5994 struct extent_buffer *eb)
5996 struct btrfs_file_extent_item *item;
5997 struct btrfs_key key;
6001 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6004 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6005 btrfs_item_key_to_cpu(eb, &key, i);
6006 if (key.type != BTRFS_EXTENT_DATA_KEY)
6008 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6009 found_type = btrfs_file_extent_type(eb, item);
6010 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6012 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6014 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6015 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6016 __exclude_logged_extent(log, key.objectid, key.offset);
6023 * btrfs_update_reserved_bytes - update the block_group and space info counters
6024 * @cache: The cache we are manipulating
6025 * @num_bytes: The number of bytes in question
6026 * @reserve: One of the reservation enums
6027 * @delalloc: The blocks are allocated for the delalloc write
6029 * This is called by the allocator when it reserves space, or by somebody who is
6030 * freeing space that was never actually used on disk. For example if you
6031 * reserve some space for a new leaf in transaction A and before transaction A
6032 * commits you free that leaf, you call this with reserve set to 0 in order to
6033 * clear the reservation.
6035 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6036 * ENOSPC accounting. For data we handle the reservation through clearing the
6037 * delalloc bits in the io_tree. We have to do this since we could end up
6038 * allocating less disk space for the amount of data we have reserved in the
6039 * case of compression.
6041 * If this is a reservation and the block group has become read only we cannot
6042 * make the reservation and return -EAGAIN, otherwise this function always
6045 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6046 u64 num_bytes, int reserve, int delalloc)
6048 struct btrfs_space_info *space_info = cache->space_info;
6051 spin_lock(&space_info->lock);
6052 spin_lock(&cache->lock);
6053 if (reserve != RESERVE_FREE) {
6057 cache->reserved += num_bytes;
6058 space_info->bytes_reserved += num_bytes;
6059 if (reserve == RESERVE_ALLOC) {
6060 trace_btrfs_space_reservation(cache->fs_info,
6061 "space_info", space_info->flags,
6063 space_info->bytes_may_use -= num_bytes;
6067 cache->delalloc_bytes += num_bytes;
6071 space_info->bytes_readonly += num_bytes;
6072 cache->reserved -= num_bytes;
6073 space_info->bytes_reserved -= num_bytes;
6076 cache->delalloc_bytes -= num_bytes;
6078 spin_unlock(&cache->lock);
6079 spin_unlock(&space_info->lock);
6083 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6084 struct btrfs_root *root)
6086 struct btrfs_fs_info *fs_info = root->fs_info;
6087 struct btrfs_caching_control *next;
6088 struct btrfs_caching_control *caching_ctl;
6089 struct btrfs_block_group_cache *cache;
6091 down_write(&fs_info->commit_root_sem);
6093 list_for_each_entry_safe(caching_ctl, next,
6094 &fs_info->caching_block_groups, list) {
6095 cache = caching_ctl->block_group;
6096 if (block_group_cache_done(cache)) {
6097 cache->last_byte_to_unpin = (u64)-1;
6098 list_del_init(&caching_ctl->list);
6099 put_caching_control(caching_ctl);
6101 cache->last_byte_to_unpin = caching_ctl->progress;
6105 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6106 fs_info->pinned_extents = &fs_info->freed_extents[1];
6108 fs_info->pinned_extents = &fs_info->freed_extents[0];
6110 up_write(&fs_info->commit_root_sem);
6112 update_global_block_rsv(fs_info);
6115 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6116 const bool return_free_space)
6118 struct btrfs_fs_info *fs_info = root->fs_info;
6119 struct btrfs_block_group_cache *cache = NULL;
6120 struct btrfs_space_info *space_info;
6121 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6125 while (start <= end) {
6128 start >= cache->key.objectid + cache->key.offset) {
6130 btrfs_put_block_group(cache);
6131 cache = btrfs_lookup_block_group(fs_info, start);
6132 BUG_ON(!cache); /* Logic error */
6135 len = cache->key.objectid + cache->key.offset - start;
6136 len = min(len, end + 1 - start);
6138 if (start < cache->last_byte_to_unpin) {
6139 len = min(len, cache->last_byte_to_unpin - start);
6140 if (return_free_space)
6141 btrfs_add_free_space(cache, start, len);
6145 space_info = cache->space_info;
6147 spin_lock(&space_info->lock);
6148 spin_lock(&cache->lock);
6149 cache->pinned -= len;
6150 space_info->bytes_pinned -= len;
6151 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6153 space_info->bytes_readonly += len;
6156 spin_unlock(&cache->lock);
6157 if (!readonly && global_rsv->space_info == space_info) {
6158 spin_lock(&global_rsv->lock);
6159 if (!global_rsv->full) {
6160 len = min(len, global_rsv->size -
6161 global_rsv->reserved);
6162 global_rsv->reserved += len;
6163 space_info->bytes_may_use += len;
6164 if (global_rsv->reserved >= global_rsv->size)
6165 global_rsv->full = 1;
6167 spin_unlock(&global_rsv->lock);
6169 spin_unlock(&space_info->lock);
6173 btrfs_put_block_group(cache);
6177 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6178 struct btrfs_root *root)
6180 struct btrfs_fs_info *fs_info = root->fs_info;
6181 struct btrfs_block_group_cache *block_group, *tmp;
6182 struct list_head *deleted_bgs;
6183 struct extent_io_tree *unpin;
6188 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6189 unpin = &fs_info->freed_extents[1];
6191 unpin = &fs_info->freed_extents[0];
6193 while (!trans->aborted) {
6194 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6195 ret = find_first_extent_bit(unpin, 0, &start, &end,
6196 EXTENT_DIRTY, NULL);
6198 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6202 if (btrfs_test_opt(root, DISCARD))
6203 ret = btrfs_discard_extent(root, start,
6204 end + 1 - start, NULL);
6206 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6207 unpin_extent_range(root, start, end, true);
6208 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6213 * Transaction is finished. We don't need the lock anymore. We
6214 * do need to clean up the block groups in case of a transaction
6217 deleted_bgs = &trans->transaction->deleted_bgs;
6218 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6222 if (!trans->aborted)
6223 ret = btrfs_discard_extent(root,
6224 block_group->key.objectid,
6225 block_group->key.offset,
6228 list_del_init(&block_group->bg_list);
6229 btrfs_put_block_group_trimming(block_group);
6230 btrfs_put_block_group(block_group);
6233 const char *errstr = btrfs_decode_error(ret);
6235 "Discard failed while removing blockgroup: errno=%d %s\n",
6243 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6244 u64 owner, u64 root_objectid)
6246 struct btrfs_space_info *space_info;
6249 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6250 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6251 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6253 flags = BTRFS_BLOCK_GROUP_METADATA;
6255 flags = BTRFS_BLOCK_GROUP_DATA;
6258 space_info = __find_space_info(fs_info, flags);
6259 BUG_ON(!space_info); /* Logic bug */
6260 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6264 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6265 struct btrfs_root *root,
6266 struct btrfs_delayed_ref_node *node, u64 parent,
6267 u64 root_objectid, u64 owner_objectid,
6268 u64 owner_offset, int refs_to_drop,
6269 struct btrfs_delayed_extent_op *extent_op)
6271 struct btrfs_key key;
6272 struct btrfs_path *path;
6273 struct btrfs_fs_info *info = root->fs_info;
6274 struct btrfs_root *extent_root = info->extent_root;
6275 struct extent_buffer *leaf;
6276 struct btrfs_extent_item *ei;
6277 struct btrfs_extent_inline_ref *iref;
6280 int extent_slot = 0;
6281 int found_extent = 0;
6283 int no_quota = node->no_quota;
6286 u64 bytenr = node->bytenr;
6287 u64 num_bytes = node->num_bytes;
6289 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6292 if (!info->quota_enabled || !is_fstree(root_objectid))
6295 path = btrfs_alloc_path();
6300 path->leave_spinning = 1;
6302 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6303 BUG_ON(!is_data && refs_to_drop != 1);
6306 skinny_metadata = 0;
6308 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6309 bytenr, num_bytes, parent,
6310 root_objectid, owner_objectid,
6313 extent_slot = path->slots[0];
6314 while (extent_slot >= 0) {
6315 btrfs_item_key_to_cpu(path->nodes[0], &key,
6317 if (key.objectid != bytenr)
6319 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6320 key.offset == num_bytes) {
6324 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6325 key.offset == owner_objectid) {
6329 if (path->slots[0] - extent_slot > 5)
6333 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6334 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6335 if (found_extent && item_size < sizeof(*ei))
6338 if (!found_extent) {
6340 ret = remove_extent_backref(trans, extent_root, path,
6342 is_data, &last_ref);
6344 btrfs_abort_transaction(trans, extent_root, ret);
6347 btrfs_release_path(path);
6348 path->leave_spinning = 1;
6350 key.objectid = bytenr;
6351 key.type = BTRFS_EXTENT_ITEM_KEY;
6352 key.offset = num_bytes;
6354 if (!is_data && skinny_metadata) {
6355 key.type = BTRFS_METADATA_ITEM_KEY;
6356 key.offset = owner_objectid;
6359 ret = btrfs_search_slot(trans, extent_root,
6361 if (ret > 0 && skinny_metadata && path->slots[0]) {
6363 * Couldn't find our skinny metadata item,
6364 * see if we have ye olde extent item.
6367 btrfs_item_key_to_cpu(path->nodes[0], &key,
6369 if (key.objectid == bytenr &&
6370 key.type == BTRFS_EXTENT_ITEM_KEY &&
6371 key.offset == num_bytes)
6375 if (ret > 0 && skinny_metadata) {
6376 skinny_metadata = false;
6377 key.objectid = bytenr;
6378 key.type = BTRFS_EXTENT_ITEM_KEY;
6379 key.offset = num_bytes;
6380 btrfs_release_path(path);
6381 ret = btrfs_search_slot(trans, extent_root,
6386 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6389 btrfs_print_leaf(extent_root,
6393 btrfs_abort_transaction(trans, extent_root, ret);
6396 extent_slot = path->slots[0];
6398 } else if (WARN_ON(ret == -ENOENT)) {
6399 btrfs_print_leaf(extent_root, path->nodes[0]);
6401 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6402 bytenr, parent, root_objectid, owner_objectid,
6404 btrfs_abort_transaction(trans, extent_root, ret);
6407 btrfs_abort_transaction(trans, extent_root, ret);
6411 leaf = path->nodes[0];
6412 item_size = btrfs_item_size_nr(leaf, extent_slot);
6413 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6414 if (item_size < sizeof(*ei)) {
6415 BUG_ON(found_extent || extent_slot != path->slots[0]);
6416 ret = convert_extent_item_v0(trans, extent_root, path,
6419 btrfs_abort_transaction(trans, extent_root, ret);
6423 btrfs_release_path(path);
6424 path->leave_spinning = 1;
6426 key.objectid = bytenr;
6427 key.type = BTRFS_EXTENT_ITEM_KEY;
6428 key.offset = num_bytes;
6430 ret = btrfs_search_slot(trans, extent_root, &key, path,
6433 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6435 btrfs_print_leaf(extent_root, path->nodes[0]);
6438 btrfs_abort_transaction(trans, extent_root, ret);
6442 extent_slot = path->slots[0];
6443 leaf = path->nodes[0];
6444 item_size = btrfs_item_size_nr(leaf, extent_slot);
6447 BUG_ON(item_size < sizeof(*ei));
6448 ei = btrfs_item_ptr(leaf, extent_slot,
6449 struct btrfs_extent_item);
6450 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6451 key.type == BTRFS_EXTENT_ITEM_KEY) {
6452 struct btrfs_tree_block_info *bi;
6453 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6454 bi = (struct btrfs_tree_block_info *)(ei + 1);
6455 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6458 refs = btrfs_extent_refs(leaf, ei);
6459 if (refs < refs_to_drop) {
6460 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6461 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6463 btrfs_abort_transaction(trans, extent_root, ret);
6466 refs -= refs_to_drop;
6470 __run_delayed_extent_op(extent_op, leaf, ei);
6472 * In the case of inline back ref, reference count will
6473 * be updated by remove_extent_backref
6476 BUG_ON(!found_extent);
6478 btrfs_set_extent_refs(leaf, ei, refs);
6479 btrfs_mark_buffer_dirty(leaf);
6482 ret = remove_extent_backref(trans, extent_root, path,
6484 is_data, &last_ref);
6486 btrfs_abort_transaction(trans, extent_root, ret);
6490 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6494 BUG_ON(is_data && refs_to_drop !=
6495 extent_data_ref_count(path, iref));
6497 BUG_ON(path->slots[0] != extent_slot);
6499 BUG_ON(path->slots[0] != extent_slot + 1);
6500 path->slots[0] = extent_slot;
6506 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6509 btrfs_abort_transaction(trans, extent_root, ret);
6512 btrfs_release_path(path);
6515 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6517 btrfs_abort_transaction(trans, extent_root, ret);
6522 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6524 btrfs_abort_transaction(trans, extent_root, ret);
6528 btrfs_release_path(path);
6531 btrfs_free_path(path);
6536 * when we free an block, it is possible (and likely) that we free the last
6537 * delayed ref for that extent as well. This searches the delayed ref tree for
6538 * a given extent, and if there are no other delayed refs to be processed, it
6539 * removes it from the tree.
6541 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6542 struct btrfs_root *root, u64 bytenr)
6544 struct btrfs_delayed_ref_head *head;
6545 struct btrfs_delayed_ref_root *delayed_refs;
6548 delayed_refs = &trans->transaction->delayed_refs;
6549 spin_lock(&delayed_refs->lock);
6550 head = btrfs_find_delayed_ref_head(trans, bytenr);
6552 goto out_delayed_unlock;
6554 spin_lock(&head->lock);
6555 if (!list_empty(&head->ref_list))
6558 if (head->extent_op) {
6559 if (!head->must_insert_reserved)
6561 btrfs_free_delayed_extent_op(head->extent_op);
6562 head->extent_op = NULL;
6566 * waiting for the lock here would deadlock. If someone else has it
6567 * locked they are already in the process of dropping it anyway
6569 if (!mutex_trylock(&head->mutex))
6573 * at this point we have a head with no other entries. Go
6574 * ahead and process it.
6576 head->node.in_tree = 0;
6577 rb_erase(&head->href_node, &delayed_refs->href_root);
6579 atomic_dec(&delayed_refs->num_entries);
6582 * we don't take a ref on the node because we're removing it from the
6583 * tree, so we just steal the ref the tree was holding.
6585 delayed_refs->num_heads--;
6586 if (head->processing == 0)
6587 delayed_refs->num_heads_ready--;
6588 head->processing = 0;
6589 spin_unlock(&head->lock);
6590 spin_unlock(&delayed_refs->lock);
6592 BUG_ON(head->extent_op);
6593 if (head->must_insert_reserved)
6596 mutex_unlock(&head->mutex);
6597 btrfs_put_delayed_ref(&head->node);
6600 spin_unlock(&head->lock);
6603 spin_unlock(&delayed_refs->lock);
6607 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6608 struct btrfs_root *root,
6609 struct extent_buffer *buf,
6610 u64 parent, int last_ref)
6615 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6616 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6617 buf->start, buf->len,
6618 parent, root->root_key.objectid,
6619 btrfs_header_level(buf),
6620 BTRFS_DROP_DELAYED_REF, NULL, 0);
6621 BUG_ON(ret); /* -ENOMEM */
6627 if (btrfs_header_generation(buf) == trans->transid) {
6628 struct btrfs_block_group_cache *cache;
6630 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6631 ret = check_ref_cleanup(trans, root, buf->start);
6636 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6638 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6639 pin_down_extent(root, cache, buf->start, buf->len, 1);
6640 btrfs_put_block_group(cache);
6644 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6646 btrfs_add_free_space(cache, buf->start, buf->len);
6647 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6648 btrfs_put_block_group(cache);
6649 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6654 add_pinned_bytes(root->fs_info, buf->len,
6655 btrfs_header_level(buf),
6656 root->root_key.objectid);
6659 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6662 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6665 /* Can return -ENOMEM */
6666 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6667 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6668 u64 owner, u64 offset, int no_quota)
6671 struct btrfs_fs_info *fs_info = root->fs_info;
6673 if (btrfs_test_is_dummy_root(root))
6676 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6679 * tree log blocks never actually go into the extent allocation
6680 * tree, just update pinning info and exit early.
6682 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6683 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6684 /* unlocks the pinned mutex */
6685 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6687 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6688 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6690 parent, root_objectid, (int)owner,
6691 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6693 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6695 parent, root_objectid, owner,
6696 offset, BTRFS_DROP_DELAYED_REF,
6703 * when we wait for progress in the block group caching, its because
6704 * our allocation attempt failed at least once. So, we must sleep
6705 * and let some progress happen before we try again.
6707 * This function will sleep at least once waiting for new free space to
6708 * show up, and then it will check the block group free space numbers
6709 * for our min num_bytes. Another option is to have it go ahead
6710 * and look in the rbtree for a free extent of a given size, but this
6713 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6714 * any of the information in this block group.
6716 static noinline void
6717 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6720 struct btrfs_caching_control *caching_ctl;
6722 caching_ctl = get_caching_control(cache);
6726 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6727 (cache->free_space_ctl->free_space >= num_bytes));
6729 put_caching_control(caching_ctl);
6733 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6735 struct btrfs_caching_control *caching_ctl;
6738 caching_ctl = get_caching_control(cache);
6740 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6742 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6743 if (cache->cached == BTRFS_CACHE_ERROR)
6745 put_caching_control(caching_ctl);
6749 int __get_raid_index(u64 flags)
6751 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6752 return BTRFS_RAID_RAID10;
6753 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6754 return BTRFS_RAID_RAID1;
6755 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6756 return BTRFS_RAID_DUP;
6757 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6758 return BTRFS_RAID_RAID0;
6759 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6760 return BTRFS_RAID_RAID5;
6761 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6762 return BTRFS_RAID_RAID6;
6764 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6767 int get_block_group_index(struct btrfs_block_group_cache *cache)
6769 return __get_raid_index(cache->flags);
6772 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6773 [BTRFS_RAID_RAID10] = "raid10",
6774 [BTRFS_RAID_RAID1] = "raid1",
6775 [BTRFS_RAID_DUP] = "dup",
6776 [BTRFS_RAID_RAID0] = "raid0",
6777 [BTRFS_RAID_SINGLE] = "single",
6778 [BTRFS_RAID_RAID5] = "raid5",
6779 [BTRFS_RAID_RAID6] = "raid6",
6782 static const char *get_raid_name(enum btrfs_raid_types type)
6784 if (type >= BTRFS_NR_RAID_TYPES)
6787 return btrfs_raid_type_names[type];
6790 enum btrfs_loop_type {
6791 LOOP_CACHING_NOWAIT = 0,
6792 LOOP_CACHING_WAIT = 1,
6793 LOOP_ALLOC_CHUNK = 2,
6794 LOOP_NO_EMPTY_SIZE = 3,
6798 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6802 down_read(&cache->data_rwsem);
6806 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6809 btrfs_get_block_group(cache);
6811 down_read(&cache->data_rwsem);
6814 static struct btrfs_block_group_cache *
6815 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6816 struct btrfs_free_cluster *cluster,
6819 struct btrfs_block_group_cache *used_bg;
6820 bool locked = false;
6822 spin_lock(&cluster->refill_lock);
6824 if (used_bg == cluster->block_group)
6827 up_read(&used_bg->data_rwsem);
6828 btrfs_put_block_group(used_bg);
6831 used_bg = cluster->block_group;
6835 if (used_bg == block_group)
6838 btrfs_get_block_group(used_bg);
6843 if (down_read_trylock(&used_bg->data_rwsem))
6846 spin_unlock(&cluster->refill_lock);
6847 down_read(&used_bg->data_rwsem);
6853 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6857 up_read(&cache->data_rwsem);
6858 btrfs_put_block_group(cache);
6862 * walks the btree of allocated extents and find a hole of a given size.
6863 * The key ins is changed to record the hole:
6864 * ins->objectid == start position
6865 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6866 * ins->offset == the size of the hole.
6867 * Any available blocks before search_start are skipped.
6869 * If there is no suitable free space, we will record the max size of
6870 * the free space extent currently.
6872 static noinline int find_free_extent(struct btrfs_root *orig_root,
6873 u64 num_bytes, u64 empty_size,
6874 u64 hint_byte, struct btrfs_key *ins,
6875 u64 flags, int delalloc)
6878 struct btrfs_root *root = orig_root->fs_info->extent_root;
6879 struct btrfs_free_cluster *last_ptr = NULL;
6880 struct btrfs_block_group_cache *block_group = NULL;
6881 u64 search_start = 0;
6882 u64 max_extent_size = 0;
6883 int empty_cluster = 2 * 1024 * 1024;
6884 struct btrfs_space_info *space_info;
6886 int index = __get_raid_index(flags);
6887 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6888 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6889 bool failed_cluster_refill = false;
6890 bool failed_alloc = false;
6891 bool use_cluster = true;
6892 bool have_caching_bg = false;
6894 WARN_ON(num_bytes < root->sectorsize);
6895 ins->type = BTRFS_EXTENT_ITEM_KEY;
6899 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6901 space_info = __find_space_info(root->fs_info, flags);
6903 btrfs_err(root->fs_info, "No space info for %llu", flags);
6908 * If the space info is for both data and metadata it means we have a
6909 * small filesystem and we can't use the clustering stuff.
6911 if (btrfs_mixed_space_info(space_info))
6912 use_cluster = false;
6914 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6915 last_ptr = &root->fs_info->meta_alloc_cluster;
6916 if (!btrfs_test_opt(root, SSD))
6917 empty_cluster = 64 * 1024;
6920 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6921 btrfs_test_opt(root, SSD)) {
6922 last_ptr = &root->fs_info->data_alloc_cluster;
6926 spin_lock(&last_ptr->lock);
6927 if (last_ptr->block_group)
6928 hint_byte = last_ptr->window_start;
6929 spin_unlock(&last_ptr->lock);
6932 search_start = max(search_start, first_logical_byte(root, 0));
6933 search_start = max(search_start, hint_byte);
6938 if (search_start == hint_byte) {
6939 block_group = btrfs_lookup_block_group(root->fs_info,
6942 * we don't want to use the block group if it doesn't match our
6943 * allocation bits, or if its not cached.
6945 * However if we are re-searching with an ideal block group
6946 * picked out then we don't care that the block group is cached.
6948 if (block_group && block_group_bits(block_group, flags) &&
6949 block_group->cached != BTRFS_CACHE_NO) {
6950 down_read(&space_info->groups_sem);
6951 if (list_empty(&block_group->list) ||
6954 * someone is removing this block group,
6955 * we can't jump into the have_block_group
6956 * target because our list pointers are not
6959 btrfs_put_block_group(block_group);
6960 up_read(&space_info->groups_sem);
6962 index = get_block_group_index(block_group);
6963 btrfs_lock_block_group(block_group, delalloc);
6964 goto have_block_group;
6966 } else if (block_group) {
6967 btrfs_put_block_group(block_group);
6971 have_caching_bg = false;
6972 down_read(&space_info->groups_sem);
6973 list_for_each_entry(block_group, &space_info->block_groups[index],
6978 btrfs_grab_block_group(block_group, delalloc);
6979 search_start = block_group->key.objectid;
6982 * this can happen if we end up cycling through all the
6983 * raid types, but we want to make sure we only allocate
6984 * for the proper type.
6986 if (!block_group_bits(block_group, flags)) {
6987 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6988 BTRFS_BLOCK_GROUP_RAID1 |
6989 BTRFS_BLOCK_GROUP_RAID5 |
6990 BTRFS_BLOCK_GROUP_RAID6 |
6991 BTRFS_BLOCK_GROUP_RAID10;
6994 * if they asked for extra copies and this block group
6995 * doesn't provide them, bail. This does allow us to
6996 * fill raid0 from raid1.
6998 if ((flags & extra) && !(block_group->flags & extra))
7003 cached = block_group_cache_done(block_group);
7004 if (unlikely(!cached)) {
7005 ret = cache_block_group(block_group, 0);
7010 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7012 if (unlikely(block_group->ro))
7016 * Ok we want to try and use the cluster allocator, so
7020 struct btrfs_block_group_cache *used_block_group;
7021 unsigned long aligned_cluster;
7023 * the refill lock keeps out other
7024 * people trying to start a new cluster
7026 used_block_group = btrfs_lock_cluster(block_group,
7029 if (!used_block_group)
7030 goto refill_cluster;
7032 if (used_block_group != block_group &&
7033 (used_block_group->ro ||
7034 !block_group_bits(used_block_group, flags)))
7035 goto release_cluster;
7037 offset = btrfs_alloc_from_cluster(used_block_group,
7040 used_block_group->key.objectid,
7043 /* we have a block, we're done */
7044 spin_unlock(&last_ptr->refill_lock);
7045 trace_btrfs_reserve_extent_cluster(root,
7047 search_start, num_bytes);
7048 if (used_block_group != block_group) {
7049 btrfs_release_block_group(block_group,
7051 block_group = used_block_group;
7056 WARN_ON(last_ptr->block_group != used_block_group);
7058 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7059 * set up a new clusters, so lets just skip it
7060 * and let the allocator find whatever block
7061 * it can find. If we reach this point, we
7062 * will have tried the cluster allocator
7063 * plenty of times and not have found
7064 * anything, so we are likely way too
7065 * fragmented for the clustering stuff to find
7068 * However, if the cluster is taken from the
7069 * current block group, release the cluster
7070 * first, so that we stand a better chance of
7071 * succeeding in the unclustered
7073 if (loop >= LOOP_NO_EMPTY_SIZE &&
7074 used_block_group != block_group) {
7075 spin_unlock(&last_ptr->refill_lock);
7076 btrfs_release_block_group(used_block_group,
7078 goto unclustered_alloc;
7082 * this cluster didn't work out, free it and
7085 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7087 if (used_block_group != block_group)
7088 btrfs_release_block_group(used_block_group,
7091 if (loop >= LOOP_NO_EMPTY_SIZE) {
7092 spin_unlock(&last_ptr->refill_lock);
7093 goto unclustered_alloc;
7096 aligned_cluster = max_t(unsigned long,
7097 empty_cluster + empty_size,
7098 block_group->full_stripe_len);
7100 /* allocate a cluster in this block group */
7101 ret = btrfs_find_space_cluster(root, block_group,
7102 last_ptr, search_start,
7107 * now pull our allocation out of this
7110 offset = btrfs_alloc_from_cluster(block_group,
7116 /* we found one, proceed */
7117 spin_unlock(&last_ptr->refill_lock);
7118 trace_btrfs_reserve_extent_cluster(root,
7119 block_group, search_start,
7123 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7124 && !failed_cluster_refill) {
7125 spin_unlock(&last_ptr->refill_lock);
7127 failed_cluster_refill = true;
7128 wait_block_group_cache_progress(block_group,
7129 num_bytes + empty_cluster + empty_size);
7130 goto have_block_group;
7134 * at this point we either didn't find a cluster
7135 * or we weren't able to allocate a block from our
7136 * cluster. Free the cluster we've been trying
7137 * to use, and go to the next block group
7139 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7140 spin_unlock(&last_ptr->refill_lock);
7145 spin_lock(&block_group->free_space_ctl->tree_lock);
7147 block_group->free_space_ctl->free_space <
7148 num_bytes + empty_cluster + empty_size) {
7149 if (block_group->free_space_ctl->free_space >
7152 block_group->free_space_ctl->free_space;
7153 spin_unlock(&block_group->free_space_ctl->tree_lock);
7156 spin_unlock(&block_group->free_space_ctl->tree_lock);
7158 offset = btrfs_find_space_for_alloc(block_group, search_start,
7159 num_bytes, empty_size,
7162 * If we didn't find a chunk, and we haven't failed on this
7163 * block group before, and this block group is in the middle of
7164 * caching and we are ok with waiting, then go ahead and wait
7165 * for progress to be made, and set failed_alloc to true.
7167 * If failed_alloc is true then we've already waited on this
7168 * block group once and should move on to the next block group.
7170 if (!offset && !failed_alloc && !cached &&
7171 loop > LOOP_CACHING_NOWAIT) {
7172 wait_block_group_cache_progress(block_group,
7173 num_bytes + empty_size);
7174 failed_alloc = true;
7175 goto have_block_group;
7176 } else if (!offset) {
7178 have_caching_bg = true;
7182 search_start = ALIGN(offset, root->stripesize);
7184 /* move on to the next group */
7185 if (search_start + num_bytes >
7186 block_group->key.objectid + block_group->key.offset) {
7187 btrfs_add_free_space(block_group, offset, num_bytes);
7191 if (offset < search_start)
7192 btrfs_add_free_space(block_group, offset,
7193 search_start - offset);
7194 BUG_ON(offset > search_start);
7196 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7197 alloc_type, delalloc);
7198 if (ret == -EAGAIN) {
7199 btrfs_add_free_space(block_group, offset, num_bytes);
7203 /* we are all good, lets return */
7204 ins->objectid = search_start;
7205 ins->offset = num_bytes;
7207 trace_btrfs_reserve_extent(orig_root, block_group,
7208 search_start, num_bytes);
7209 btrfs_release_block_group(block_group, delalloc);
7212 failed_cluster_refill = false;
7213 failed_alloc = false;
7214 BUG_ON(index != get_block_group_index(block_group));
7215 btrfs_release_block_group(block_group, delalloc);
7217 up_read(&space_info->groups_sem);
7219 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7222 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7226 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7227 * caching kthreads as we move along
7228 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7229 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7230 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7233 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7236 if (loop == LOOP_ALLOC_CHUNK) {
7237 struct btrfs_trans_handle *trans;
7240 trans = current->journal_info;
7244 trans = btrfs_join_transaction(root);
7246 if (IS_ERR(trans)) {
7247 ret = PTR_ERR(trans);
7251 ret = do_chunk_alloc(trans, root, flags,
7254 * Do not bail out on ENOSPC since we
7255 * can do more things.
7257 if (ret < 0 && ret != -ENOSPC)
7258 btrfs_abort_transaction(trans,
7263 btrfs_end_transaction(trans, root);
7268 if (loop == LOOP_NO_EMPTY_SIZE) {
7274 } else if (!ins->objectid) {
7276 } else if (ins->objectid) {
7281 ins->offset = max_extent_size;
7285 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7286 int dump_block_groups)
7288 struct btrfs_block_group_cache *cache;
7291 spin_lock(&info->lock);
7292 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7294 info->total_bytes - info->bytes_used - info->bytes_pinned -
7295 info->bytes_reserved - info->bytes_readonly,
7296 (info->full) ? "" : "not ");
7297 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7298 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7299 info->total_bytes, info->bytes_used, info->bytes_pinned,
7300 info->bytes_reserved, info->bytes_may_use,
7301 info->bytes_readonly);
7302 spin_unlock(&info->lock);
7304 if (!dump_block_groups)
7307 down_read(&info->groups_sem);
7309 list_for_each_entry(cache, &info->block_groups[index], list) {
7310 spin_lock(&cache->lock);
7311 printk(KERN_INFO "BTRFS: "
7312 "block group %llu has %llu bytes, "
7313 "%llu used %llu pinned %llu reserved %s\n",
7314 cache->key.objectid, cache->key.offset,
7315 btrfs_block_group_used(&cache->item), cache->pinned,
7316 cache->reserved, cache->ro ? "[readonly]" : "");
7317 btrfs_dump_free_space(cache, bytes);
7318 spin_unlock(&cache->lock);
7320 if (++index < BTRFS_NR_RAID_TYPES)
7322 up_read(&info->groups_sem);
7325 int btrfs_reserve_extent(struct btrfs_root *root,
7326 u64 num_bytes, u64 min_alloc_size,
7327 u64 empty_size, u64 hint_byte,
7328 struct btrfs_key *ins, int is_data, int delalloc)
7330 bool final_tried = false;
7334 flags = btrfs_get_alloc_profile(root, is_data);
7336 WARN_ON(num_bytes < root->sectorsize);
7337 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7340 if (ret == -ENOSPC) {
7341 if (!final_tried && ins->offset) {
7342 num_bytes = min(num_bytes >> 1, ins->offset);
7343 num_bytes = round_down(num_bytes, root->sectorsize);
7344 num_bytes = max(num_bytes, min_alloc_size);
7345 if (num_bytes == min_alloc_size)
7348 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7349 struct btrfs_space_info *sinfo;
7351 sinfo = __find_space_info(root->fs_info, flags);
7352 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7355 dump_space_info(sinfo, num_bytes, 1);
7362 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7364 int pin, int delalloc)
7366 struct btrfs_block_group_cache *cache;
7369 cache = btrfs_lookup_block_group(root->fs_info, start);
7371 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7377 pin_down_extent(root, cache, start, len, 1);
7379 if (btrfs_test_opt(root, DISCARD))
7380 ret = btrfs_discard_extent(root, start, len, NULL);
7381 btrfs_add_free_space(cache, start, len);
7382 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7385 btrfs_put_block_group(cache);
7387 trace_btrfs_reserved_extent_free(root, start, len);
7392 int btrfs_free_reserved_extent(struct btrfs_root *root,
7393 u64 start, u64 len, int delalloc)
7395 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7398 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7401 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7404 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7405 struct btrfs_root *root,
7406 u64 parent, u64 root_objectid,
7407 u64 flags, u64 owner, u64 offset,
7408 struct btrfs_key *ins, int ref_mod)
7411 struct btrfs_fs_info *fs_info = root->fs_info;
7412 struct btrfs_extent_item *extent_item;
7413 struct btrfs_extent_inline_ref *iref;
7414 struct btrfs_path *path;
7415 struct extent_buffer *leaf;
7420 type = BTRFS_SHARED_DATA_REF_KEY;
7422 type = BTRFS_EXTENT_DATA_REF_KEY;
7424 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7426 path = btrfs_alloc_path();
7430 path->leave_spinning = 1;
7431 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7434 btrfs_free_path(path);
7438 leaf = path->nodes[0];
7439 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7440 struct btrfs_extent_item);
7441 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7442 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7443 btrfs_set_extent_flags(leaf, extent_item,
7444 flags | BTRFS_EXTENT_FLAG_DATA);
7446 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7447 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7449 struct btrfs_shared_data_ref *ref;
7450 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7451 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7452 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7454 struct btrfs_extent_data_ref *ref;
7455 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7456 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7457 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7458 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7459 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7462 btrfs_mark_buffer_dirty(path->nodes[0]);
7463 btrfs_free_path(path);
7465 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7466 if (ret) { /* -ENOENT, logic error */
7467 btrfs_err(fs_info, "update block group failed for %llu %llu",
7468 ins->objectid, ins->offset);
7471 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7475 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7476 struct btrfs_root *root,
7477 u64 parent, u64 root_objectid,
7478 u64 flags, struct btrfs_disk_key *key,
7479 int level, struct btrfs_key *ins,
7483 struct btrfs_fs_info *fs_info = root->fs_info;
7484 struct btrfs_extent_item *extent_item;
7485 struct btrfs_tree_block_info *block_info;
7486 struct btrfs_extent_inline_ref *iref;
7487 struct btrfs_path *path;
7488 struct extent_buffer *leaf;
7489 u32 size = sizeof(*extent_item) + sizeof(*iref);
7490 u64 num_bytes = ins->offset;
7491 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7494 if (!skinny_metadata)
7495 size += sizeof(*block_info);
7497 path = btrfs_alloc_path();
7499 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7504 path->leave_spinning = 1;
7505 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7508 btrfs_free_path(path);
7509 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7514 leaf = path->nodes[0];
7515 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7516 struct btrfs_extent_item);
7517 btrfs_set_extent_refs(leaf, extent_item, 1);
7518 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7519 btrfs_set_extent_flags(leaf, extent_item,
7520 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7522 if (skinny_metadata) {
7523 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7524 num_bytes = root->nodesize;
7526 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7527 btrfs_set_tree_block_key(leaf, block_info, key);
7528 btrfs_set_tree_block_level(leaf, block_info, level);
7529 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7533 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7534 btrfs_set_extent_inline_ref_type(leaf, iref,
7535 BTRFS_SHARED_BLOCK_REF_KEY);
7536 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7538 btrfs_set_extent_inline_ref_type(leaf, iref,
7539 BTRFS_TREE_BLOCK_REF_KEY);
7540 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7543 btrfs_mark_buffer_dirty(leaf);
7544 btrfs_free_path(path);
7546 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7548 if (ret) { /* -ENOENT, logic error */
7549 btrfs_err(fs_info, "update block group failed for %llu %llu",
7550 ins->objectid, ins->offset);
7554 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7558 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7559 struct btrfs_root *root,
7560 u64 root_objectid, u64 owner,
7561 u64 offset, struct btrfs_key *ins)
7565 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7567 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7569 root_objectid, owner, offset,
7570 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7575 * this is used by the tree logging recovery code. It records that
7576 * an extent has been allocated and makes sure to clear the free
7577 * space cache bits as well
7579 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7580 struct btrfs_root *root,
7581 u64 root_objectid, u64 owner, u64 offset,
7582 struct btrfs_key *ins)
7585 struct btrfs_block_group_cache *block_group;
7588 * Mixed block groups will exclude before processing the log so we only
7589 * need to do the exlude dance if this fs isn't mixed.
7591 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7592 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7597 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7601 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7602 RESERVE_ALLOC_NO_ACCOUNT, 0);
7603 BUG_ON(ret); /* logic error */
7604 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7605 0, owner, offset, ins, 1);
7606 btrfs_put_block_group(block_group);
7610 static struct extent_buffer *
7611 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7612 u64 bytenr, int level)
7614 struct extent_buffer *buf;
7616 buf = btrfs_find_create_tree_block(root, bytenr);
7618 return ERR_PTR(-ENOMEM);
7619 btrfs_set_header_generation(buf, trans->transid);
7620 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7621 btrfs_tree_lock(buf);
7622 clean_tree_block(trans, root->fs_info, buf);
7623 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7625 btrfs_set_lock_blocking(buf);
7626 btrfs_set_buffer_uptodate(buf);
7628 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7629 buf->log_index = root->log_transid % 2;
7631 * we allow two log transactions at a time, use different
7632 * EXENT bit to differentiate dirty pages.
7634 if (buf->log_index == 0)
7635 set_extent_dirty(&root->dirty_log_pages, buf->start,
7636 buf->start + buf->len - 1, GFP_NOFS);
7638 set_extent_new(&root->dirty_log_pages, buf->start,
7639 buf->start + buf->len - 1, GFP_NOFS);
7641 buf->log_index = -1;
7642 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7643 buf->start + buf->len - 1, GFP_NOFS);
7645 trans->blocks_used++;
7646 /* this returns a buffer locked for blocking */
7650 static struct btrfs_block_rsv *
7651 use_block_rsv(struct btrfs_trans_handle *trans,
7652 struct btrfs_root *root, u32 blocksize)
7654 struct btrfs_block_rsv *block_rsv;
7655 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7657 bool global_updated = false;
7659 block_rsv = get_block_rsv(trans, root);
7661 if (unlikely(block_rsv->size == 0))
7664 ret = block_rsv_use_bytes(block_rsv, blocksize);
7668 if (block_rsv->failfast)
7669 return ERR_PTR(ret);
7671 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7672 global_updated = true;
7673 update_global_block_rsv(root->fs_info);
7677 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7678 static DEFINE_RATELIMIT_STATE(_rs,
7679 DEFAULT_RATELIMIT_INTERVAL * 10,
7680 /*DEFAULT_RATELIMIT_BURST*/ 1);
7681 if (__ratelimit(&_rs))
7683 "BTRFS: block rsv returned %d\n", ret);
7686 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7687 BTRFS_RESERVE_NO_FLUSH);
7691 * If we couldn't reserve metadata bytes try and use some from
7692 * the global reserve if its space type is the same as the global
7695 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7696 block_rsv->space_info == global_rsv->space_info) {
7697 ret = block_rsv_use_bytes(global_rsv, blocksize);
7701 return ERR_PTR(ret);
7704 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7705 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7707 block_rsv_add_bytes(block_rsv, blocksize, 0);
7708 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7712 * finds a free extent and does all the dirty work required for allocation
7713 * returns the tree buffer or an ERR_PTR on error.
7715 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7716 struct btrfs_root *root,
7717 u64 parent, u64 root_objectid,
7718 struct btrfs_disk_key *key, int level,
7719 u64 hint, u64 empty_size)
7721 struct btrfs_key ins;
7722 struct btrfs_block_rsv *block_rsv;
7723 struct extent_buffer *buf;
7724 struct btrfs_delayed_extent_op *extent_op;
7727 u32 blocksize = root->nodesize;
7728 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7731 if (btrfs_test_is_dummy_root(root)) {
7732 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7735 root->alloc_bytenr += blocksize;
7739 block_rsv = use_block_rsv(trans, root, blocksize);
7740 if (IS_ERR(block_rsv))
7741 return ERR_CAST(block_rsv);
7743 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7744 empty_size, hint, &ins, 0, 0);
7748 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7751 goto out_free_reserved;
7754 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7756 parent = ins.objectid;
7757 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7761 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7762 extent_op = btrfs_alloc_delayed_extent_op();
7768 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7770 memset(&extent_op->key, 0, sizeof(extent_op->key));
7771 extent_op->flags_to_set = flags;
7772 if (skinny_metadata)
7773 extent_op->update_key = 0;
7775 extent_op->update_key = 1;
7776 extent_op->update_flags = 1;
7777 extent_op->is_data = 0;
7778 extent_op->level = level;
7780 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7781 ins.objectid, ins.offset,
7782 parent, root_objectid, level,
7783 BTRFS_ADD_DELAYED_EXTENT,
7786 goto out_free_delayed;
7791 btrfs_free_delayed_extent_op(extent_op);
7793 free_extent_buffer(buf);
7795 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7797 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7798 return ERR_PTR(ret);
7801 struct walk_control {
7802 u64 refs[BTRFS_MAX_LEVEL];
7803 u64 flags[BTRFS_MAX_LEVEL];
7804 struct btrfs_key update_progress;
7815 #define DROP_REFERENCE 1
7816 #define UPDATE_BACKREF 2
7818 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7819 struct btrfs_root *root,
7820 struct walk_control *wc,
7821 struct btrfs_path *path)
7829 struct btrfs_key key;
7830 struct extent_buffer *eb;
7835 if (path->slots[wc->level] < wc->reada_slot) {
7836 wc->reada_count = wc->reada_count * 2 / 3;
7837 wc->reada_count = max(wc->reada_count, 2);
7839 wc->reada_count = wc->reada_count * 3 / 2;
7840 wc->reada_count = min_t(int, wc->reada_count,
7841 BTRFS_NODEPTRS_PER_BLOCK(root));
7844 eb = path->nodes[wc->level];
7845 nritems = btrfs_header_nritems(eb);
7846 blocksize = root->nodesize;
7848 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7849 if (nread >= wc->reada_count)
7853 bytenr = btrfs_node_blockptr(eb, slot);
7854 generation = btrfs_node_ptr_generation(eb, slot);
7856 if (slot == path->slots[wc->level])
7859 if (wc->stage == UPDATE_BACKREF &&
7860 generation <= root->root_key.offset)
7863 /* We don't lock the tree block, it's OK to be racy here */
7864 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7865 wc->level - 1, 1, &refs,
7867 /* We don't care about errors in readahead. */
7872 if (wc->stage == DROP_REFERENCE) {
7876 if (wc->level == 1 &&
7877 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7879 if (!wc->update_ref ||
7880 generation <= root->root_key.offset)
7882 btrfs_node_key_to_cpu(eb, &key, slot);
7883 ret = btrfs_comp_cpu_keys(&key,
7884 &wc->update_progress);
7888 if (wc->level == 1 &&
7889 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7893 readahead_tree_block(root, bytenr);
7896 wc->reada_slot = slot;
7900 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7901 * for later qgroup accounting.
7903 * Current, this function does nothing.
7905 static int account_leaf_items(struct btrfs_trans_handle *trans,
7906 struct btrfs_root *root,
7907 struct extent_buffer *eb)
7909 int nr = btrfs_header_nritems(eb);
7911 struct btrfs_key key;
7912 struct btrfs_file_extent_item *fi;
7913 u64 bytenr, num_bytes;
7915 for (i = 0; i < nr; i++) {
7916 btrfs_item_key_to_cpu(eb, &key, i);
7918 if (key.type != BTRFS_EXTENT_DATA_KEY)
7921 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7922 /* filter out non qgroup-accountable extents */
7923 extent_type = btrfs_file_extent_type(eb, fi);
7925 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7928 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7932 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7938 * Walk up the tree from the bottom, freeing leaves and any interior
7939 * nodes which have had all slots visited. If a node (leaf or
7940 * interior) is freed, the node above it will have it's slot
7941 * incremented. The root node will never be freed.
7943 * At the end of this function, we should have a path which has all
7944 * slots incremented to the next position for a search. If we need to
7945 * read a new node it will be NULL and the node above it will have the
7946 * correct slot selected for a later read.
7948 * If we increment the root nodes slot counter past the number of
7949 * elements, 1 is returned to signal completion of the search.
7951 static int adjust_slots_upwards(struct btrfs_root *root,
7952 struct btrfs_path *path, int root_level)
7956 struct extent_buffer *eb;
7958 if (root_level == 0)
7961 while (level <= root_level) {
7962 eb = path->nodes[level];
7963 nr = btrfs_header_nritems(eb);
7964 path->slots[level]++;
7965 slot = path->slots[level];
7966 if (slot >= nr || level == 0) {
7968 * Don't free the root - we will detect this
7969 * condition after our loop and return a
7970 * positive value for caller to stop walking the tree.
7972 if (level != root_level) {
7973 btrfs_tree_unlock_rw(eb, path->locks[level]);
7974 path->locks[level] = 0;
7976 free_extent_buffer(eb);
7977 path->nodes[level] = NULL;
7978 path->slots[level] = 0;
7982 * We have a valid slot to walk back down
7983 * from. Stop here so caller can process these
7992 eb = path->nodes[root_level];
7993 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8000 * root_eb is the subtree root and is locked before this function is called.
8001 * TODO: Modify this function to mark all (including complete shared node)
8002 * to dirty_extent_root to allow it get accounted in qgroup.
8004 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8005 struct btrfs_root *root,
8006 struct extent_buffer *root_eb,
8012 struct extent_buffer *eb = root_eb;
8013 struct btrfs_path *path = NULL;
8015 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8016 BUG_ON(root_eb == NULL);
8018 if (!root->fs_info->quota_enabled)
8021 if (!extent_buffer_uptodate(root_eb)) {
8022 ret = btrfs_read_buffer(root_eb, root_gen);
8027 if (root_level == 0) {
8028 ret = account_leaf_items(trans, root, root_eb);
8032 path = btrfs_alloc_path();
8037 * Walk down the tree. Missing extent blocks are filled in as
8038 * we go. Metadata is accounted every time we read a new
8041 * When we reach a leaf, we account for file extent items in it,
8042 * walk back up the tree (adjusting slot pointers as we go)
8043 * and restart the search process.
8045 extent_buffer_get(root_eb); /* For path */
8046 path->nodes[root_level] = root_eb;
8047 path->slots[root_level] = 0;
8048 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8051 while (level >= 0) {
8052 if (path->nodes[level] == NULL) {
8057 /* We need to get child blockptr/gen from
8058 * parent before we can read it. */
8059 eb = path->nodes[level + 1];
8060 parent_slot = path->slots[level + 1];
8061 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8062 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8064 eb = read_tree_block(root, child_bytenr, child_gen);
8068 } else if (!extent_buffer_uptodate(eb)) {
8069 free_extent_buffer(eb);
8074 path->nodes[level] = eb;
8075 path->slots[level] = 0;
8077 btrfs_tree_read_lock(eb);
8078 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8079 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8083 ret = account_leaf_items(trans, root, path->nodes[level]);
8087 /* Nonzero return here means we completed our search */
8088 ret = adjust_slots_upwards(root, path, root_level);
8092 /* Restart search with new slots */
8101 btrfs_free_path(path);
8107 * helper to process tree block while walking down the tree.
8109 * when wc->stage == UPDATE_BACKREF, this function updates
8110 * back refs for pointers in the block.
8112 * NOTE: return value 1 means we should stop walking down.
8114 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8115 struct btrfs_root *root,
8116 struct btrfs_path *path,
8117 struct walk_control *wc, int lookup_info)
8119 int level = wc->level;
8120 struct extent_buffer *eb = path->nodes[level];
8121 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8124 if (wc->stage == UPDATE_BACKREF &&
8125 btrfs_header_owner(eb) != root->root_key.objectid)
8129 * when reference count of tree block is 1, it won't increase
8130 * again. once full backref flag is set, we never clear it.
8133 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8134 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8135 BUG_ON(!path->locks[level]);
8136 ret = btrfs_lookup_extent_info(trans, root,
8137 eb->start, level, 1,
8140 BUG_ON(ret == -ENOMEM);
8143 BUG_ON(wc->refs[level] == 0);
8146 if (wc->stage == DROP_REFERENCE) {
8147 if (wc->refs[level] > 1)
8150 if (path->locks[level] && !wc->keep_locks) {
8151 btrfs_tree_unlock_rw(eb, path->locks[level]);
8152 path->locks[level] = 0;
8157 /* wc->stage == UPDATE_BACKREF */
8158 if (!(wc->flags[level] & flag)) {
8159 BUG_ON(!path->locks[level]);
8160 ret = btrfs_inc_ref(trans, root, eb, 1);
8161 BUG_ON(ret); /* -ENOMEM */
8162 ret = btrfs_dec_ref(trans, root, eb, 0);
8163 BUG_ON(ret); /* -ENOMEM */
8164 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8166 btrfs_header_level(eb), 0);
8167 BUG_ON(ret); /* -ENOMEM */
8168 wc->flags[level] |= flag;
8172 * the block is shared by multiple trees, so it's not good to
8173 * keep the tree lock
8175 if (path->locks[level] && level > 0) {
8176 btrfs_tree_unlock_rw(eb, path->locks[level]);
8177 path->locks[level] = 0;
8183 * helper to process tree block pointer.
8185 * when wc->stage == DROP_REFERENCE, this function checks
8186 * reference count of the block pointed to. if the block
8187 * is shared and we need update back refs for the subtree
8188 * rooted at the block, this function changes wc->stage to
8189 * UPDATE_BACKREF. if the block is shared and there is no
8190 * need to update back, this function drops the reference
8193 * NOTE: return value 1 means we should stop walking down.
8195 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8196 struct btrfs_root *root,
8197 struct btrfs_path *path,
8198 struct walk_control *wc, int *lookup_info)
8204 struct btrfs_key key;
8205 struct extent_buffer *next;
8206 int level = wc->level;
8209 bool need_account = false;
8211 generation = btrfs_node_ptr_generation(path->nodes[level],
8212 path->slots[level]);
8214 * if the lower level block was created before the snapshot
8215 * was created, we know there is no need to update back refs
8218 if (wc->stage == UPDATE_BACKREF &&
8219 generation <= root->root_key.offset) {
8224 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8225 blocksize = root->nodesize;
8227 next = btrfs_find_tree_block(root->fs_info, bytenr);
8229 next = btrfs_find_create_tree_block(root, bytenr);
8232 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8236 btrfs_tree_lock(next);
8237 btrfs_set_lock_blocking(next);
8239 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8240 &wc->refs[level - 1],
8241 &wc->flags[level - 1]);
8243 btrfs_tree_unlock(next);
8247 if (unlikely(wc->refs[level - 1] == 0)) {
8248 btrfs_err(root->fs_info, "Missing references.");
8253 if (wc->stage == DROP_REFERENCE) {
8254 if (wc->refs[level - 1] > 1) {
8255 need_account = true;
8257 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8260 if (!wc->update_ref ||
8261 generation <= root->root_key.offset)
8264 btrfs_node_key_to_cpu(path->nodes[level], &key,
8265 path->slots[level]);
8266 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8270 wc->stage = UPDATE_BACKREF;
8271 wc->shared_level = level - 1;
8275 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8279 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8280 btrfs_tree_unlock(next);
8281 free_extent_buffer(next);
8287 if (reada && level == 1)
8288 reada_walk_down(trans, root, wc, path);
8289 next = read_tree_block(root, bytenr, generation);
8291 return PTR_ERR(next);
8292 } else if (!extent_buffer_uptodate(next)) {
8293 free_extent_buffer(next);
8296 btrfs_tree_lock(next);
8297 btrfs_set_lock_blocking(next);
8301 BUG_ON(level != btrfs_header_level(next));
8302 path->nodes[level] = next;
8303 path->slots[level] = 0;
8304 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8310 wc->refs[level - 1] = 0;
8311 wc->flags[level - 1] = 0;
8312 if (wc->stage == DROP_REFERENCE) {
8313 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8314 parent = path->nodes[level]->start;
8316 BUG_ON(root->root_key.objectid !=
8317 btrfs_header_owner(path->nodes[level]));
8322 ret = account_shared_subtree(trans, root, next,
8323 generation, level - 1);
8325 btrfs_err_rl(root->fs_info,
8327 "%d accounting shared subtree. Quota "
8328 "is out of sync, rescan required.",
8332 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8333 root->root_key.objectid, level - 1, 0, 0);
8334 BUG_ON(ret); /* -ENOMEM */
8336 btrfs_tree_unlock(next);
8337 free_extent_buffer(next);
8343 * helper to process tree block while walking up the tree.
8345 * when wc->stage == DROP_REFERENCE, this function drops
8346 * reference count on the block.
8348 * when wc->stage == UPDATE_BACKREF, this function changes
8349 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8350 * to UPDATE_BACKREF previously while processing the block.
8352 * NOTE: return value 1 means we should stop walking up.
8354 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8355 struct btrfs_root *root,
8356 struct btrfs_path *path,
8357 struct walk_control *wc)
8360 int level = wc->level;
8361 struct extent_buffer *eb = path->nodes[level];
8364 if (wc->stage == UPDATE_BACKREF) {
8365 BUG_ON(wc->shared_level < level);
8366 if (level < wc->shared_level)
8369 ret = find_next_key(path, level + 1, &wc->update_progress);
8373 wc->stage = DROP_REFERENCE;
8374 wc->shared_level = -1;
8375 path->slots[level] = 0;
8378 * check reference count again if the block isn't locked.
8379 * we should start walking down the tree again if reference
8382 if (!path->locks[level]) {
8384 btrfs_tree_lock(eb);
8385 btrfs_set_lock_blocking(eb);
8386 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8388 ret = btrfs_lookup_extent_info(trans, root,
8389 eb->start, level, 1,
8393 btrfs_tree_unlock_rw(eb, path->locks[level]);
8394 path->locks[level] = 0;
8397 BUG_ON(wc->refs[level] == 0);
8398 if (wc->refs[level] == 1) {
8399 btrfs_tree_unlock_rw(eb, path->locks[level]);
8400 path->locks[level] = 0;
8406 /* wc->stage == DROP_REFERENCE */
8407 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8409 if (wc->refs[level] == 1) {
8411 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8412 ret = btrfs_dec_ref(trans, root, eb, 1);
8414 ret = btrfs_dec_ref(trans, root, eb, 0);
8415 BUG_ON(ret); /* -ENOMEM */
8416 ret = account_leaf_items(trans, root, eb);
8418 btrfs_err_rl(root->fs_info,
8420 "%d accounting leaf items. Quota "
8421 "is out of sync, rescan required.",
8425 /* make block locked assertion in clean_tree_block happy */
8426 if (!path->locks[level] &&
8427 btrfs_header_generation(eb) == trans->transid) {
8428 btrfs_tree_lock(eb);
8429 btrfs_set_lock_blocking(eb);
8430 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8432 clean_tree_block(trans, root->fs_info, eb);
8435 if (eb == root->node) {
8436 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8439 BUG_ON(root->root_key.objectid !=
8440 btrfs_header_owner(eb));
8442 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8443 parent = path->nodes[level + 1]->start;
8445 BUG_ON(root->root_key.objectid !=
8446 btrfs_header_owner(path->nodes[level + 1]));
8449 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8451 wc->refs[level] = 0;
8452 wc->flags[level] = 0;
8456 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8457 struct btrfs_root *root,
8458 struct btrfs_path *path,
8459 struct walk_control *wc)
8461 int level = wc->level;
8462 int lookup_info = 1;
8465 while (level >= 0) {
8466 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8473 if (path->slots[level] >=
8474 btrfs_header_nritems(path->nodes[level]))
8477 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8479 path->slots[level]++;
8488 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8489 struct btrfs_root *root,
8490 struct btrfs_path *path,
8491 struct walk_control *wc, int max_level)
8493 int level = wc->level;
8496 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8497 while (level < max_level && path->nodes[level]) {
8499 if (path->slots[level] + 1 <
8500 btrfs_header_nritems(path->nodes[level])) {
8501 path->slots[level]++;
8504 ret = walk_up_proc(trans, root, path, wc);
8508 if (path->locks[level]) {
8509 btrfs_tree_unlock_rw(path->nodes[level],
8510 path->locks[level]);
8511 path->locks[level] = 0;
8513 free_extent_buffer(path->nodes[level]);
8514 path->nodes[level] = NULL;
8522 * drop a subvolume tree.
8524 * this function traverses the tree freeing any blocks that only
8525 * referenced by the tree.
8527 * when a shared tree block is found. this function decreases its
8528 * reference count by one. if update_ref is true, this function
8529 * also make sure backrefs for the shared block and all lower level
8530 * blocks are properly updated.
8532 * If called with for_reloc == 0, may exit early with -EAGAIN
8534 int btrfs_drop_snapshot(struct btrfs_root *root,
8535 struct btrfs_block_rsv *block_rsv, int update_ref,
8538 struct btrfs_path *path;
8539 struct btrfs_trans_handle *trans;
8540 struct btrfs_root *tree_root = root->fs_info->tree_root;
8541 struct btrfs_root_item *root_item = &root->root_item;
8542 struct walk_control *wc;
8543 struct btrfs_key key;
8547 bool root_dropped = false;
8549 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8551 path = btrfs_alloc_path();
8557 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8559 btrfs_free_path(path);
8564 trans = btrfs_start_transaction(tree_root, 0);
8565 if (IS_ERR(trans)) {
8566 err = PTR_ERR(trans);
8571 trans->block_rsv = block_rsv;
8573 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8574 level = btrfs_header_level(root->node);
8575 path->nodes[level] = btrfs_lock_root_node(root);
8576 btrfs_set_lock_blocking(path->nodes[level]);
8577 path->slots[level] = 0;
8578 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8579 memset(&wc->update_progress, 0,
8580 sizeof(wc->update_progress));
8582 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8583 memcpy(&wc->update_progress, &key,
8584 sizeof(wc->update_progress));
8586 level = root_item->drop_level;
8588 path->lowest_level = level;
8589 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8590 path->lowest_level = 0;
8598 * unlock our path, this is safe because only this
8599 * function is allowed to delete this snapshot
8601 btrfs_unlock_up_safe(path, 0);
8603 level = btrfs_header_level(root->node);
8605 btrfs_tree_lock(path->nodes[level]);
8606 btrfs_set_lock_blocking(path->nodes[level]);
8607 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8609 ret = btrfs_lookup_extent_info(trans, root,
8610 path->nodes[level]->start,
8611 level, 1, &wc->refs[level],
8617 BUG_ON(wc->refs[level] == 0);
8619 if (level == root_item->drop_level)
8622 btrfs_tree_unlock(path->nodes[level]);
8623 path->locks[level] = 0;
8624 WARN_ON(wc->refs[level] != 1);
8630 wc->shared_level = -1;
8631 wc->stage = DROP_REFERENCE;
8632 wc->update_ref = update_ref;
8634 wc->for_reloc = for_reloc;
8635 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8639 ret = walk_down_tree(trans, root, path, wc);
8645 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8652 BUG_ON(wc->stage != DROP_REFERENCE);
8656 if (wc->stage == DROP_REFERENCE) {
8658 btrfs_node_key(path->nodes[level],
8659 &root_item->drop_progress,
8660 path->slots[level]);
8661 root_item->drop_level = level;
8664 BUG_ON(wc->level == 0);
8665 if (btrfs_should_end_transaction(trans, tree_root) ||
8666 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8667 ret = btrfs_update_root(trans, tree_root,
8671 btrfs_abort_transaction(trans, tree_root, ret);
8676 btrfs_end_transaction_throttle(trans, tree_root);
8677 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8678 pr_debug("BTRFS: drop snapshot early exit\n");
8683 trans = btrfs_start_transaction(tree_root, 0);
8684 if (IS_ERR(trans)) {
8685 err = PTR_ERR(trans);
8689 trans->block_rsv = block_rsv;
8692 btrfs_release_path(path);
8696 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8698 btrfs_abort_transaction(trans, tree_root, ret);
8702 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8703 ret = btrfs_find_root(tree_root, &root->root_key, path,
8706 btrfs_abort_transaction(trans, tree_root, ret);
8709 } else if (ret > 0) {
8710 /* if we fail to delete the orphan item this time
8711 * around, it'll get picked up the next time.
8713 * The most common failure here is just -ENOENT.
8715 btrfs_del_orphan_item(trans, tree_root,
8716 root->root_key.objectid);
8720 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8721 btrfs_add_dropped_root(trans, root);
8723 free_extent_buffer(root->node);
8724 free_extent_buffer(root->commit_root);
8725 btrfs_put_fs_root(root);
8727 root_dropped = true;
8729 btrfs_end_transaction_throttle(trans, tree_root);
8732 btrfs_free_path(path);
8735 * So if we need to stop dropping the snapshot for whatever reason we
8736 * need to make sure to add it back to the dead root list so that we
8737 * keep trying to do the work later. This also cleans up roots if we
8738 * don't have it in the radix (like when we recover after a power fail
8739 * or unmount) so we don't leak memory.
8741 if (!for_reloc && root_dropped == false)
8742 btrfs_add_dead_root(root);
8743 if (err && err != -EAGAIN)
8744 btrfs_std_error(root->fs_info, err, NULL);
8749 * drop subtree rooted at tree block 'node'.
8751 * NOTE: this function will unlock and release tree block 'node'
8752 * only used by relocation code
8754 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8755 struct btrfs_root *root,
8756 struct extent_buffer *node,
8757 struct extent_buffer *parent)
8759 struct btrfs_path *path;
8760 struct walk_control *wc;
8766 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8768 path = btrfs_alloc_path();
8772 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8774 btrfs_free_path(path);
8778 btrfs_assert_tree_locked(parent);
8779 parent_level = btrfs_header_level(parent);
8780 extent_buffer_get(parent);
8781 path->nodes[parent_level] = parent;
8782 path->slots[parent_level] = btrfs_header_nritems(parent);
8784 btrfs_assert_tree_locked(node);
8785 level = btrfs_header_level(node);
8786 path->nodes[level] = node;
8787 path->slots[level] = 0;
8788 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8790 wc->refs[parent_level] = 1;
8791 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8793 wc->shared_level = -1;
8794 wc->stage = DROP_REFERENCE;
8798 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8801 wret = walk_down_tree(trans, root, path, wc);
8807 wret = walk_up_tree(trans, root, path, wc, parent_level);
8815 btrfs_free_path(path);
8819 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8825 * if restripe for this chunk_type is on pick target profile and
8826 * return, otherwise do the usual balance
8828 stripped = get_restripe_target(root->fs_info, flags);
8830 return extended_to_chunk(stripped);
8832 num_devices = root->fs_info->fs_devices->rw_devices;
8834 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8835 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8836 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8838 if (num_devices == 1) {
8839 stripped |= BTRFS_BLOCK_GROUP_DUP;
8840 stripped = flags & ~stripped;
8842 /* turn raid0 into single device chunks */
8843 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8846 /* turn mirroring into duplication */
8847 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8848 BTRFS_BLOCK_GROUP_RAID10))
8849 return stripped | BTRFS_BLOCK_GROUP_DUP;
8851 /* they already had raid on here, just return */
8852 if (flags & stripped)
8855 stripped |= BTRFS_BLOCK_GROUP_DUP;
8856 stripped = flags & ~stripped;
8858 /* switch duplicated blocks with raid1 */
8859 if (flags & BTRFS_BLOCK_GROUP_DUP)
8860 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8862 /* this is drive concat, leave it alone */
8868 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8870 struct btrfs_space_info *sinfo = cache->space_info;
8872 u64 min_allocable_bytes;
8876 * We need some metadata space and system metadata space for
8877 * allocating chunks in some corner cases until we force to set
8878 * it to be readonly.
8881 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8883 min_allocable_bytes = 1 * 1024 * 1024;
8885 min_allocable_bytes = 0;
8887 spin_lock(&sinfo->lock);
8888 spin_lock(&cache->lock);
8896 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8897 cache->bytes_super - btrfs_block_group_used(&cache->item);
8899 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8900 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8901 min_allocable_bytes <= sinfo->total_bytes) {
8902 sinfo->bytes_readonly += num_bytes;
8904 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8908 spin_unlock(&cache->lock);
8909 spin_unlock(&sinfo->lock);
8913 int btrfs_inc_block_group_ro(struct btrfs_root *root,
8914 struct btrfs_block_group_cache *cache)
8917 struct btrfs_trans_handle *trans;
8922 trans = btrfs_join_transaction(root);
8924 return PTR_ERR(trans);
8927 * we're not allowed to set block groups readonly after the dirty
8928 * block groups cache has started writing. If it already started,
8929 * back off and let this transaction commit
8931 mutex_lock(&root->fs_info->ro_block_group_mutex);
8932 if (trans->transaction->dirty_bg_run) {
8933 u64 transid = trans->transid;
8935 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8936 btrfs_end_transaction(trans, root);
8938 ret = btrfs_wait_for_commit(root, transid);
8945 * if we are changing raid levels, try to allocate a corresponding
8946 * block group with the new raid level.
8948 alloc_flags = update_block_group_flags(root, cache->flags);
8949 if (alloc_flags != cache->flags) {
8950 ret = do_chunk_alloc(trans, root, alloc_flags,
8953 * ENOSPC is allowed here, we may have enough space
8954 * already allocated at the new raid level to
8963 ret = inc_block_group_ro(cache, 0);
8966 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8967 ret = do_chunk_alloc(trans, root, alloc_flags,
8971 ret = inc_block_group_ro(cache, 0);
8973 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8974 alloc_flags = update_block_group_flags(root, cache->flags);
8975 lock_chunks(root->fs_info->chunk_root);
8976 check_system_chunk(trans, root, alloc_flags);
8977 unlock_chunks(root->fs_info->chunk_root);
8979 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8981 btrfs_end_transaction(trans, root);
8985 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8986 struct btrfs_root *root, u64 type)
8988 u64 alloc_flags = get_alloc_profile(root, type);
8989 return do_chunk_alloc(trans, root, alloc_flags,
8994 * helper to account the unused space of all the readonly block group in the
8995 * space_info. takes mirrors into account.
8997 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8999 struct btrfs_block_group_cache *block_group;
9003 /* It's df, we don't care if it's racey */
9004 if (list_empty(&sinfo->ro_bgs))
9007 spin_lock(&sinfo->lock);
9008 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9009 spin_lock(&block_group->lock);
9011 if (!block_group->ro) {
9012 spin_unlock(&block_group->lock);
9016 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9017 BTRFS_BLOCK_GROUP_RAID10 |
9018 BTRFS_BLOCK_GROUP_DUP))
9023 free_bytes += (block_group->key.offset -
9024 btrfs_block_group_used(&block_group->item)) *
9027 spin_unlock(&block_group->lock);
9029 spin_unlock(&sinfo->lock);
9034 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9035 struct btrfs_block_group_cache *cache)
9037 struct btrfs_space_info *sinfo = cache->space_info;
9042 spin_lock(&sinfo->lock);
9043 spin_lock(&cache->lock);
9045 num_bytes = cache->key.offset - cache->reserved -
9046 cache->pinned - cache->bytes_super -
9047 btrfs_block_group_used(&cache->item);
9048 sinfo->bytes_readonly -= num_bytes;
9049 list_del_init(&cache->ro_list);
9051 spin_unlock(&cache->lock);
9052 spin_unlock(&sinfo->lock);
9056 * checks to see if its even possible to relocate this block group.
9058 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9059 * ok to go ahead and try.
9061 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9063 struct btrfs_block_group_cache *block_group;
9064 struct btrfs_space_info *space_info;
9065 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9066 struct btrfs_device *device;
9067 struct btrfs_trans_handle *trans;
9076 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9078 /* odd, couldn't find the block group, leave it alone */
9082 min_free = btrfs_block_group_used(&block_group->item);
9084 /* no bytes used, we're good */
9088 space_info = block_group->space_info;
9089 spin_lock(&space_info->lock);
9091 full = space_info->full;
9094 * if this is the last block group we have in this space, we can't
9095 * relocate it unless we're able to allocate a new chunk below.
9097 * Otherwise, we need to make sure we have room in the space to handle
9098 * all of the extents from this block group. If we can, we're good
9100 if ((space_info->total_bytes != block_group->key.offset) &&
9101 (space_info->bytes_used + space_info->bytes_reserved +
9102 space_info->bytes_pinned + space_info->bytes_readonly +
9103 min_free < space_info->total_bytes)) {
9104 spin_unlock(&space_info->lock);
9107 spin_unlock(&space_info->lock);
9110 * ok we don't have enough space, but maybe we have free space on our
9111 * devices to allocate new chunks for relocation, so loop through our
9112 * alloc devices and guess if we have enough space. if this block
9113 * group is going to be restriped, run checks against the target
9114 * profile instead of the current one.
9126 target = get_restripe_target(root->fs_info, block_group->flags);
9128 index = __get_raid_index(extended_to_chunk(target));
9131 * this is just a balance, so if we were marked as full
9132 * we know there is no space for a new chunk
9137 index = get_block_group_index(block_group);
9140 if (index == BTRFS_RAID_RAID10) {
9144 } else if (index == BTRFS_RAID_RAID1) {
9146 } else if (index == BTRFS_RAID_DUP) {
9149 } else if (index == BTRFS_RAID_RAID0) {
9150 dev_min = fs_devices->rw_devices;
9151 min_free = div64_u64(min_free, dev_min);
9154 /* We need to do this so that we can look at pending chunks */
9155 trans = btrfs_join_transaction(root);
9156 if (IS_ERR(trans)) {
9157 ret = PTR_ERR(trans);
9161 mutex_lock(&root->fs_info->chunk_mutex);
9162 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9166 * check to make sure we can actually find a chunk with enough
9167 * space to fit our block group in.
9169 if (device->total_bytes > device->bytes_used + min_free &&
9170 !device->is_tgtdev_for_dev_replace) {
9171 ret = find_free_dev_extent(trans, device, min_free,
9176 if (dev_nr >= dev_min)
9182 mutex_unlock(&root->fs_info->chunk_mutex);
9183 btrfs_end_transaction(trans, root);
9185 btrfs_put_block_group(block_group);
9189 static int find_first_block_group(struct btrfs_root *root,
9190 struct btrfs_path *path, struct btrfs_key *key)
9193 struct btrfs_key found_key;
9194 struct extent_buffer *leaf;
9197 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9202 slot = path->slots[0];
9203 leaf = path->nodes[0];
9204 if (slot >= btrfs_header_nritems(leaf)) {
9205 ret = btrfs_next_leaf(root, path);
9212 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9214 if (found_key.objectid >= key->objectid &&
9215 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9225 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9227 struct btrfs_block_group_cache *block_group;
9231 struct inode *inode;
9233 block_group = btrfs_lookup_first_block_group(info, last);
9234 while (block_group) {
9235 spin_lock(&block_group->lock);
9236 if (block_group->iref)
9238 spin_unlock(&block_group->lock);
9239 block_group = next_block_group(info->tree_root,
9249 inode = block_group->inode;
9250 block_group->iref = 0;
9251 block_group->inode = NULL;
9252 spin_unlock(&block_group->lock);
9254 last = block_group->key.objectid + block_group->key.offset;
9255 btrfs_put_block_group(block_group);
9259 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9261 struct btrfs_block_group_cache *block_group;
9262 struct btrfs_space_info *space_info;
9263 struct btrfs_caching_control *caching_ctl;
9266 down_write(&info->commit_root_sem);
9267 while (!list_empty(&info->caching_block_groups)) {
9268 caching_ctl = list_entry(info->caching_block_groups.next,
9269 struct btrfs_caching_control, list);
9270 list_del(&caching_ctl->list);
9271 put_caching_control(caching_ctl);
9273 up_write(&info->commit_root_sem);
9275 spin_lock(&info->unused_bgs_lock);
9276 while (!list_empty(&info->unused_bgs)) {
9277 block_group = list_first_entry(&info->unused_bgs,
9278 struct btrfs_block_group_cache,
9280 list_del_init(&block_group->bg_list);
9281 btrfs_put_block_group(block_group);
9283 spin_unlock(&info->unused_bgs_lock);
9285 spin_lock(&info->block_group_cache_lock);
9286 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9287 block_group = rb_entry(n, struct btrfs_block_group_cache,
9289 rb_erase(&block_group->cache_node,
9290 &info->block_group_cache_tree);
9291 RB_CLEAR_NODE(&block_group->cache_node);
9292 spin_unlock(&info->block_group_cache_lock);
9294 down_write(&block_group->space_info->groups_sem);
9295 list_del(&block_group->list);
9296 up_write(&block_group->space_info->groups_sem);
9298 if (block_group->cached == BTRFS_CACHE_STARTED)
9299 wait_block_group_cache_done(block_group);
9302 * We haven't cached this block group, which means we could
9303 * possibly have excluded extents on this block group.
9305 if (block_group->cached == BTRFS_CACHE_NO ||
9306 block_group->cached == BTRFS_CACHE_ERROR)
9307 free_excluded_extents(info->extent_root, block_group);
9309 btrfs_remove_free_space_cache(block_group);
9310 btrfs_put_block_group(block_group);
9312 spin_lock(&info->block_group_cache_lock);
9314 spin_unlock(&info->block_group_cache_lock);
9316 /* now that all the block groups are freed, go through and
9317 * free all the space_info structs. This is only called during
9318 * the final stages of unmount, and so we know nobody is
9319 * using them. We call synchronize_rcu() once before we start,
9320 * just to be on the safe side.
9324 release_global_block_rsv(info);
9326 while (!list_empty(&info->space_info)) {
9329 space_info = list_entry(info->space_info.next,
9330 struct btrfs_space_info,
9332 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9333 if (WARN_ON(space_info->bytes_pinned > 0 ||
9334 space_info->bytes_reserved > 0 ||
9335 space_info->bytes_may_use > 0)) {
9336 dump_space_info(space_info, 0, 0);
9339 list_del(&space_info->list);
9340 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9341 struct kobject *kobj;
9342 kobj = space_info->block_group_kobjs[i];
9343 space_info->block_group_kobjs[i] = NULL;
9349 kobject_del(&space_info->kobj);
9350 kobject_put(&space_info->kobj);
9355 static void __link_block_group(struct btrfs_space_info *space_info,
9356 struct btrfs_block_group_cache *cache)
9358 int index = get_block_group_index(cache);
9361 down_write(&space_info->groups_sem);
9362 if (list_empty(&space_info->block_groups[index]))
9364 list_add_tail(&cache->list, &space_info->block_groups[index]);
9365 up_write(&space_info->groups_sem);
9368 struct raid_kobject *rkobj;
9371 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9374 rkobj->raid_type = index;
9375 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9376 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9377 "%s", get_raid_name(index));
9379 kobject_put(&rkobj->kobj);
9382 space_info->block_group_kobjs[index] = &rkobj->kobj;
9387 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9390 static struct btrfs_block_group_cache *
9391 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9393 struct btrfs_block_group_cache *cache;
9395 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9399 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9401 if (!cache->free_space_ctl) {
9406 cache->key.objectid = start;
9407 cache->key.offset = size;
9408 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9410 cache->sectorsize = root->sectorsize;
9411 cache->fs_info = root->fs_info;
9412 cache->full_stripe_len = btrfs_full_stripe_len(root,
9413 &root->fs_info->mapping_tree,
9415 atomic_set(&cache->count, 1);
9416 spin_lock_init(&cache->lock);
9417 init_rwsem(&cache->data_rwsem);
9418 INIT_LIST_HEAD(&cache->list);
9419 INIT_LIST_HEAD(&cache->cluster_list);
9420 INIT_LIST_HEAD(&cache->bg_list);
9421 INIT_LIST_HEAD(&cache->ro_list);
9422 INIT_LIST_HEAD(&cache->dirty_list);
9423 INIT_LIST_HEAD(&cache->io_list);
9424 btrfs_init_free_space_ctl(cache);
9425 atomic_set(&cache->trimming, 0);
9430 int btrfs_read_block_groups(struct btrfs_root *root)
9432 struct btrfs_path *path;
9434 struct btrfs_block_group_cache *cache;
9435 struct btrfs_fs_info *info = root->fs_info;
9436 struct btrfs_space_info *space_info;
9437 struct btrfs_key key;
9438 struct btrfs_key found_key;
9439 struct extent_buffer *leaf;
9443 root = info->extent_root;
9446 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9447 path = btrfs_alloc_path();
9452 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9453 if (btrfs_test_opt(root, SPACE_CACHE) &&
9454 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9456 if (btrfs_test_opt(root, CLEAR_CACHE))
9460 ret = find_first_block_group(root, path, &key);
9466 leaf = path->nodes[0];
9467 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9469 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9478 * When we mount with old space cache, we need to
9479 * set BTRFS_DC_CLEAR and set dirty flag.
9481 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9482 * truncate the old free space cache inode and
9484 * b) Setting 'dirty flag' makes sure that we flush
9485 * the new space cache info onto disk.
9487 if (btrfs_test_opt(root, SPACE_CACHE))
9488 cache->disk_cache_state = BTRFS_DC_CLEAR;
9491 read_extent_buffer(leaf, &cache->item,
9492 btrfs_item_ptr_offset(leaf, path->slots[0]),
9493 sizeof(cache->item));
9494 cache->flags = btrfs_block_group_flags(&cache->item);
9496 key.objectid = found_key.objectid + found_key.offset;
9497 btrfs_release_path(path);
9500 * We need to exclude the super stripes now so that the space
9501 * info has super bytes accounted for, otherwise we'll think
9502 * we have more space than we actually do.
9504 ret = exclude_super_stripes(root, cache);
9507 * We may have excluded something, so call this just in
9510 free_excluded_extents(root, cache);
9511 btrfs_put_block_group(cache);
9516 * check for two cases, either we are full, and therefore
9517 * don't need to bother with the caching work since we won't
9518 * find any space, or we are empty, and we can just add all
9519 * the space in and be done with it. This saves us _alot_ of
9520 * time, particularly in the full case.
9522 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9523 cache->last_byte_to_unpin = (u64)-1;
9524 cache->cached = BTRFS_CACHE_FINISHED;
9525 free_excluded_extents(root, cache);
9526 } else if (btrfs_block_group_used(&cache->item) == 0) {
9527 cache->last_byte_to_unpin = (u64)-1;
9528 cache->cached = BTRFS_CACHE_FINISHED;
9529 add_new_free_space(cache, root->fs_info,
9531 found_key.objectid +
9533 free_excluded_extents(root, cache);
9536 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9538 btrfs_remove_free_space_cache(cache);
9539 btrfs_put_block_group(cache);
9543 ret = update_space_info(info, cache->flags, found_key.offset,
9544 btrfs_block_group_used(&cache->item),
9547 btrfs_remove_free_space_cache(cache);
9548 spin_lock(&info->block_group_cache_lock);
9549 rb_erase(&cache->cache_node,
9550 &info->block_group_cache_tree);
9551 RB_CLEAR_NODE(&cache->cache_node);
9552 spin_unlock(&info->block_group_cache_lock);
9553 btrfs_put_block_group(cache);
9557 cache->space_info = space_info;
9558 spin_lock(&cache->space_info->lock);
9559 cache->space_info->bytes_readonly += cache->bytes_super;
9560 spin_unlock(&cache->space_info->lock);
9562 __link_block_group(space_info, cache);
9564 set_avail_alloc_bits(root->fs_info, cache->flags);
9565 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9566 inc_block_group_ro(cache, 1);
9567 } else if (btrfs_block_group_used(&cache->item) == 0) {
9568 spin_lock(&info->unused_bgs_lock);
9569 /* Should always be true but just in case. */
9570 if (list_empty(&cache->bg_list)) {
9571 btrfs_get_block_group(cache);
9572 list_add_tail(&cache->bg_list,
9575 spin_unlock(&info->unused_bgs_lock);
9579 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9580 if (!(get_alloc_profile(root, space_info->flags) &
9581 (BTRFS_BLOCK_GROUP_RAID10 |
9582 BTRFS_BLOCK_GROUP_RAID1 |
9583 BTRFS_BLOCK_GROUP_RAID5 |
9584 BTRFS_BLOCK_GROUP_RAID6 |
9585 BTRFS_BLOCK_GROUP_DUP)))
9588 * avoid allocating from un-mirrored block group if there are
9589 * mirrored block groups.
9591 list_for_each_entry(cache,
9592 &space_info->block_groups[BTRFS_RAID_RAID0],
9594 inc_block_group_ro(cache, 1);
9595 list_for_each_entry(cache,
9596 &space_info->block_groups[BTRFS_RAID_SINGLE],
9598 inc_block_group_ro(cache, 1);
9601 init_global_block_rsv(info);
9604 btrfs_free_path(path);
9608 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9609 struct btrfs_root *root)
9611 struct btrfs_block_group_cache *block_group, *tmp;
9612 struct btrfs_root *extent_root = root->fs_info->extent_root;
9613 struct btrfs_block_group_item item;
9614 struct btrfs_key key;
9616 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9618 trans->can_flush_pending_bgs = false;
9619 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9623 spin_lock(&block_group->lock);
9624 memcpy(&item, &block_group->item, sizeof(item));
9625 memcpy(&key, &block_group->key, sizeof(key));
9626 spin_unlock(&block_group->lock);
9628 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9631 btrfs_abort_transaction(trans, extent_root, ret);
9632 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9633 key.objectid, key.offset);
9635 btrfs_abort_transaction(trans, extent_root, ret);
9637 list_del_init(&block_group->bg_list);
9639 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9642 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9643 struct btrfs_root *root, u64 bytes_used,
9644 u64 type, u64 chunk_objectid, u64 chunk_offset,
9648 struct btrfs_root *extent_root;
9649 struct btrfs_block_group_cache *cache;
9651 extent_root = root->fs_info->extent_root;
9653 btrfs_set_log_full_commit(root->fs_info, trans);
9655 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9659 btrfs_set_block_group_used(&cache->item, bytes_used);
9660 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9661 btrfs_set_block_group_flags(&cache->item, type);
9663 cache->flags = type;
9664 cache->last_byte_to_unpin = (u64)-1;
9665 cache->cached = BTRFS_CACHE_FINISHED;
9666 ret = exclude_super_stripes(root, cache);
9669 * We may have excluded something, so call this just in
9672 free_excluded_extents(root, cache);
9673 btrfs_put_block_group(cache);
9677 add_new_free_space(cache, root->fs_info, chunk_offset,
9678 chunk_offset + size);
9680 free_excluded_extents(root, cache);
9683 * Call to ensure the corresponding space_info object is created and
9684 * assigned to our block group, but don't update its counters just yet.
9685 * We want our bg to be added to the rbtree with its ->space_info set.
9687 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9688 &cache->space_info);
9690 btrfs_remove_free_space_cache(cache);
9691 btrfs_put_block_group(cache);
9695 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9697 btrfs_remove_free_space_cache(cache);
9698 btrfs_put_block_group(cache);
9703 * Now that our block group has its ->space_info set and is inserted in
9704 * the rbtree, update the space info's counters.
9706 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9707 &cache->space_info);
9709 btrfs_remove_free_space_cache(cache);
9710 spin_lock(&root->fs_info->block_group_cache_lock);
9711 rb_erase(&cache->cache_node,
9712 &root->fs_info->block_group_cache_tree);
9713 RB_CLEAR_NODE(&cache->cache_node);
9714 spin_unlock(&root->fs_info->block_group_cache_lock);
9715 btrfs_put_block_group(cache);
9718 update_global_block_rsv(root->fs_info);
9720 spin_lock(&cache->space_info->lock);
9721 cache->space_info->bytes_readonly += cache->bytes_super;
9722 spin_unlock(&cache->space_info->lock);
9724 __link_block_group(cache->space_info, cache);
9726 list_add_tail(&cache->bg_list, &trans->new_bgs);
9728 set_avail_alloc_bits(extent_root->fs_info, type);
9733 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9735 u64 extra_flags = chunk_to_extended(flags) &
9736 BTRFS_EXTENDED_PROFILE_MASK;
9738 write_seqlock(&fs_info->profiles_lock);
9739 if (flags & BTRFS_BLOCK_GROUP_DATA)
9740 fs_info->avail_data_alloc_bits &= ~extra_flags;
9741 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9742 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9743 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9744 fs_info->avail_system_alloc_bits &= ~extra_flags;
9745 write_sequnlock(&fs_info->profiles_lock);
9748 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9749 struct btrfs_root *root, u64 group_start,
9750 struct extent_map *em)
9752 struct btrfs_path *path;
9753 struct btrfs_block_group_cache *block_group;
9754 struct btrfs_free_cluster *cluster;
9755 struct btrfs_root *tree_root = root->fs_info->tree_root;
9756 struct btrfs_key key;
9757 struct inode *inode;
9758 struct kobject *kobj = NULL;
9762 struct btrfs_caching_control *caching_ctl = NULL;
9765 root = root->fs_info->extent_root;
9767 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9768 BUG_ON(!block_group);
9769 BUG_ON(!block_group->ro);
9772 * Free the reserved super bytes from this block group before
9775 free_excluded_extents(root, block_group);
9777 memcpy(&key, &block_group->key, sizeof(key));
9778 index = get_block_group_index(block_group);
9779 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9780 BTRFS_BLOCK_GROUP_RAID1 |
9781 BTRFS_BLOCK_GROUP_RAID10))
9786 /* make sure this block group isn't part of an allocation cluster */
9787 cluster = &root->fs_info->data_alloc_cluster;
9788 spin_lock(&cluster->refill_lock);
9789 btrfs_return_cluster_to_free_space(block_group, cluster);
9790 spin_unlock(&cluster->refill_lock);
9793 * make sure this block group isn't part of a metadata
9794 * allocation cluster
9796 cluster = &root->fs_info->meta_alloc_cluster;
9797 spin_lock(&cluster->refill_lock);
9798 btrfs_return_cluster_to_free_space(block_group, cluster);
9799 spin_unlock(&cluster->refill_lock);
9801 path = btrfs_alloc_path();
9808 * get the inode first so any iput calls done for the io_list
9809 * aren't the final iput (no unlinks allowed now)
9811 inode = lookup_free_space_inode(tree_root, block_group, path);
9813 mutex_lock(&trans->transaction->cache_write_mutex);
9815 * make sure our free spache cache IO is done before remove the
9818 spin_lock(&trans->transaction->dirty_bgs_lock);
9819 if (!list_empty(&block_group->io_list)) {
9820 list_del_init(&block_group->io_list);
9822 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9824 spin_unlock(&trans->transaction->dirty_bgs_lock);
9825 btrfs_wait_cache_io(root, trans, block_group,
9826 &block_group->io_ctl, path,
9827 block_group->key.objectid);
9828 btrfs_put_block_group(block_group);
9829 spin_lock(&trans->transaction->dirty_bgs_lock);
9832 if (!list_empty(&block_group->dirty_list)) {
9833 list_del_init(&block_group->dirty_list);
9834 btrfs_put_block_group(block_group);
9836 spin_unlock(&trans->transaction->dirty_bgs_lock);
9837 mutex_unlock(&trans->transaction->cache_write_mutex);
9839 if (!IS_ERR(inode)) {
9840 ret = btrfs_orphan_add(trans, inode);
9842 btrfs_add_delayed_iput(inode);
9846 /* One for the block groups ref */
9847 spin_lock(&block_group->lock);
9848 if (block_group->iref) {
9849 block_group->iref = 0;
9850 block_group->inode = NULL;
9851 spin_unlock(&block_group->lock);
9854 spin_unlock(&block_group->lock);
9856 /* One for our lookup ref */
9857 btrfs_add_delayed_iput(inode);
9860 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9861 key.offset = block_group->key.objectid;
9864 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9868 btrfs_release_path(path);
9870 ret = btrfs_del_item(trans, tree_root, path);
9873 btrfs_release_path(path);
9876 spin_lock(&root->fs_info->block_group_cache_lock);
9877 rb_erase(&block_group->cache_node,
9878 &root->fs_info->block_group_cache_tree);
9879 RB_CLEAR_NODE(&block_group->cache_node);
9881 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9882 root->fs_info->first_logical_byte = (u64)-1;
9883 spin_unlock(&root->fs_info->block_group_cache_lock);
9885 down_write(&block_group->space_info->groups_sem);
9887 * we must use list_del_init so people can check to see if they
9888 * are still on the list after taking the semaphore
9890 list_del_init(&block_group->list);
9891 if (list_empty(&block_group->space_info->block_groups[index])) {
9892 kobj = block_group->space_info->block_group_kobjs[index];
9893 block_group->space_info->block_group_kobjs[index] = NULL;
9894 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9896 up_write(&block_group->space_info->groups_sem);
9902 if (block_group->has_caching_ctl)
9903 caching_ctl = get_caching_control(block_group);
9904 if (block_group->cached == BTRFS_CACHE_STARTED)
9905 wait_block_group_cache_done(block_group);
9906 if (block_group->has_caching_ctl) {
9907 down_write(&root->fs_info->commit_root_sem);
9909 struct btrfs_caching_control *ctl;
9911 list_for_each_entry(ctl,
9912 &root->fs_info->caching_block_groups, list)
9913 if (ctl->block_group == block_group) {
9915 atomic_inc(&caching_ctl->count);
9920 list_del_init(&caching_ctl->list);
9921 up_write(&root->fs_info->commit_root_sem);
9923 /* Once for the caching bgs list and once for us. */
9924 put_caching_control(caching_ctl);
9925 put_caching_control(caching_ctl);
9929 spin_lock(&trans->transaction->dirty_bgs_lock);
9930 if (!list_empty(&block_group->dirty_list)) {
9933 if (!list_empty(&block_group->io_list)) {
9936 spin_unlock(&trans->transaction->dirty_bgs_lock);
9937 btrfs_remove_free_space_cache(block_group);
9939 spin_lock(&block_group->space_info->lock);
9940 list_del_init(&block_group->ro_list);
9942 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9943 WARN_ON(block_group->space_info->total_bytes
9944 < block_group->key.offset);
9945 WARN_ON(block_group->space_info->bytes_readonly
9946 < block_group->key.offset);
9947 WARN_ON(block_group->space_info->disk_total
9948 < block_group->key.offset * factor);
9950 block_group->space_info->total_bytes -= block_group->key.offset;
9951 block_group->space_info->bytes_readonly -= block_group->key.offset;
9952 block_group->space_info->disk_total -= block_group->key.offset * factor;
9954 spin_unlock(&block_group->space_info->lock);
9956 memcpy(&key, &block_group->key, sizeof(key));
9959 if (!list_empty(&em->list)) {
9960 /* We're in the transaction->pending_chunks list. */
9961 free_extent_map(em);
9963 spin_lock(&block_group->lock);
9964 block_group->removed = 1;
9966 * At this point trimming can't start on this block group, because we
9967 * removed the block group from the tree fs_info->block_group_cache_tree
9968 * so no one can't find it anymore and even if someone already got this
9969 * block group before we removed it from the rbtree, they have already
9970 * incremented block_group->trimming - if they didn't, they won't find
9971 * any free space entries because we already removed them all when we
9972 * called btrfs_remove_free_space_cache().
9974 * And we must not remove the extent map from the fs_info->mapping_tree
9975 * to prevent the same logical address range and physical device space
9976 * ranges from being reused for a new block group. This is because our
9977 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9978 * completely transactionless, so while it is trimming a range the
9979 * currently running transaction might finish and a new one start,
9980 * allowing for new block groups to be created that can reuse the same
9981 * physical device locations unless we take this special care.
9983 * There may also be an implicit trim operation if the file system
9984 * is mounted with -odiscard. The same protections must remain
9985 * in place until the extents have been discarded completely when
9986 * the transaction commit has completed.
9988 remove_em = (atomic_read(&block_group->trimming) == 0);
9990 * Make sure a trimmer task always sees the em in the pinned_chunks list
9991 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9992 * before checking block_group->removed).
9996 * Our em might be in trans->transaction->pending_chunks which
9997 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9998 * and so is the fs_info->pinned_chunks list.
10000 * So at this point we must be holding the chunk_mutex to avoid
10001 * any races with chunk allocation (more specifically at
10002 * volumes.c:contains_pending_extent()), to ensure it always
10003 * sees the em, either in the pending_chunks list or in the
10004 * pinned_chunks list.
10006 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10008 spin_unlock(&block_group->lock);
10011 struct extent_map_tree *em_tree;
10013 em_tree = &root->fs_info->mapping_tree.map_tree;
10014 write_lock(&em_tree->lock);
10016 * The em might be in the pending_chunks list, so make sure the
10017 * chunk mutex is locked, since remove_extent_mapping() will
10018 * delete us from that list.
10020 remove_extent_mapping(em_tree, em);
10021 write_unlock(&em_tree->lock);
10022 /* once for the tree */
10023 free_extent_map(em);
10026 unlock_chunks(root);
10028 btrfs_put_block_group(block_group);
10029 btrfs_put_block_group(block_group);
10031 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10037 ret = btrfs_del_item(trans, root, path);
10039 btrfs_free_path(path);
10044 * Process the unused_bgs list and remove any that don't have any allocated
10045 * space inside of them.
10047 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10049 struct btrfs_block_group_cache *block_group;
10050 struct btrfs_space_info *space_info;
10051 struct btrfs_root *root = fs_info->extent_root;
10052 struct btrfs_trans_handle *trans;
10055 if (!fs_info->open)
10058 spin_lock(&fs_info->unused_bgs_lock);
10059 while (!list_empty(&fs_info->unused_bgs)) {
10063 block_group = list_first_entry(&fs_info->unused_bgs,
10064 struct btrfs_block_group_cache,
10066 space_info = block_group->space_info;
10067 list_del_init(&block_group->bg_list);
10068 if (ret || btrfs_mixed_space_info(space_info)) {
10069 btrfs_put_block_group(block_group);
10072 spin_unlock(&fs_info->unused_bgs_lock);
10074 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10076 /* Don't want to race with allocators so take the groups_sem */
10077 down_write(&space_info->groups_sem);
10078 spin_lock(&block_group->lock);
10079 if (block_group->reserved ||
10080 btrfs_block_group_used(&block_group->item) ||
10083 * We want to bail if we made new allocations or have
10084 * outstanding allocations in this block group. We do
10085 * the ro check in case balance is currently acting on
10086 * this block group.
10088 spin_unlock(&block_group->lock);
10089 up_write(&space_info->groups_sem);
10092 spin_unlock(&block_group->lock);
10094 /* We don't want to force the issue, only flip if it's ok. */
10095 ret = inc_block_group_ro(block_group, 0);
10096 up_write(&space_info->groups_sem);
10103 * Want to do this before we do anything else so we can recover
10104 * properly if we fail to join the transaction.
10106 /* 1 for btrfs_orphan_reserve_metadata() */
10107 trans = btrfs_start_transaction(root, 1);
10108 if (IS_ERR(trans)) {
10109 btrfs_dec_block_group_ro(root, block_group);
10110 ret = PTR_ERR(trans);
10115 * We could have pending pinned extents for this block group,
10116 * just delete them, we don't care about them anymore.
10118 start = block_group->key.objectid;
10119 end = start + block_group->key.offset - 1;
10121 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10122 * btrfs_finish_extent_commit(). If we are at transaction N,
10123 * another task might be running finish_extent_commit() for the
10124 * previous transaction N - 1, and have seen a range belonging
10125 * to the block group in freed_extents[] before we were able to
10126 * clear the whole block group range from freed_extents[]. This
10127 * means that task can lookup for the block group after we
10128 * unpinned it from freed_extents[] and removed it, leading to
10129 * a BUG_ON() at btrfs_unpin_extent_range().
10131 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10132 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10133 EXTENT_DIRTY, GFP_NOFS);
10135 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10136 btrfs_dec_block_group_ro(root, block_group);
10139 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10140 EXTENT_DIRTY, GFP_NOFS);
10142 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10143 btrfs_dec_block_group_ro(root, block_group);
10146 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10148 /* Reset pinned so btrfs_put_block_group doesn't complain */
10149 spin_lock(&space_info->lock);
10150 spin_lock(&block_group->lock);
10152 space_info->bytes_pinned -= block_group->pinned;
10153 space_info->bytes_readonly += block_group->pinned;
10154 percpu_counter_add(&space_info->total_bytes_pinned,
10155 -block_group->pinned);
10156 block_group->pinned = 0;
10158 spin_unlock(&block_group->lock);
10159 spin_unlock(&space_info->lock);
10161 /* DISCARD can flip during remount */
10162 trimming = btrfs_test_opt(root, DISCARD);
10164 /* Implicit trim during transaction commit. */
10166 btrfs_get_block_group_trimming(block_group);
10169 * Btrfs_remove_chunk will abort the transaction if things go
10172 ret = btrfs_remove_chunk(trans, root,
10173 block_group->key.objectid);
10177 btrfs_put_block_group_trimming(block_group);
10182 * If we're not mounted with -odiscard, we can just forget
10183 * about this block group. Otherwise we'll need to wait
10184 * until transaction commit to do the actual discard.
10187 WARN_ON(!list_empty(&block_group->bg_list));
10188 spin_lock(&trans->transaction->deleted_bgs_lock);
10189 list_move(&block_group->bg_list,
10190 &trans->transaction->deleted_bgs);
10191 spin_unlock(&trans->transaction->deleted_bgs_lock);
10192 btrfs_get_block_group(block_group);
10195 btrfs_end_transaction(trans, root);
10197 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10198 btrfs_put_block_group(block_group);
10199 spin_lock(&fs_info->unused_bgs_lock);
10201 spin_unlock(&fs_info->unused_bgs_lock);
10204 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10206 struct btrfs_space_info *space_info;
10207 struct btrfs_super_block *disk_super;
10213 disk_super = fs_info->super_copy;
10214 if (!btrfs_super_root(disk_super))
10217 features = btrfs_super_incompat_flags(disk_super);
10218 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10221 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10222 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10227 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10228 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10230 flags = BTRFS_BLOCK_GROUP_METADATA;
10231 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10235 flags = BTRFS_BLOCK_GROUP_DATA;
10236 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10242 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10244 return unpin_extent_range(root, start, end, false);
10248 * It used to be that old block groups would be left around forever.
10249 * Iterating over them would be enough to trim unused space. Since we
10250 * now automatically remove them, we also need to iterate over unallocated
10253 * We don't want a transaction for this since the discard may take a
10254 * substantial amount of time. We don't require that a transaction be
10255 * running, but we do need to take a running transaction into account
10256 * to ensure that we're not discarding chunks that were released in
10257 * the current transaction.
10259 * Holding the chunks lock will prevent other threads from allocating
10260 * or releasing chunks, but it won't prevent a running transaction
10261 * from committing and releasing the memory that the pending chunks
10262 * list head uses. For that, we need to take a reference to the
10265 static int btrfs_trim_free_extents(struct btrfs_device *device,
10266 u64 minlen, u64 *trimmed)
10268 u64 start = 0, len = 0;
10273 /* Not writeable = nothing to do. */
10274 if (!device->writeable)
10277 /* No free space = nothing to do. */
10278 if (device->total_bytes <= device->bytes_used)
10284 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10285 struct btrfs_transaction *trans;
10288 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10292 down_read(&fs_info->commit_root_sem);
10294 spin_lock(&fs_info->trans_lock);
10295 trans = fs_info->running_transaction;
10297 atomic_inc(&trans->use_count);
10298 spin_unlock(&fs_info->trans_lock);
10300 ret = find_free_dev_extent_start(trans, device, minlen, start,
10303 btrfs_put_transaction(trans);
10306 up_read(&fs_info->commit_root_sem);
10307 mutex_unlock(&fs_info->chunk_mutex);
10308 if (ret == -ENOSPC)
10313 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10314 up_read(&fs_info->commit_root_sem);
10315 mutex_unlock(&fs_info->chunk_mutex);
10323 if (fatal_signal_pending(current)) {
10324 ret = -ERESTARTSYS;
10334 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10336 struct btrfs_fs_info *fs_info = root->fs_info;
10337 struct btrfs_block_group_cache *cache = NULL;
10338 struct btrfs_device *device;
10339 struct list_head *devices;
10344 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10348 * try to trim all FS space, our block group may start from non-zero.
10350 if (range->len == total_bytes)
10351 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10353 cache = btrfs_lookup_block_group(fs_info, range->start);
10356 if (cache->key.objectid >= (range->start + range->len)) {
10357 btrfs_put_block_group(cache);
10361 start = max(range->start, cache->key.objectid);
10362 end = min(range->start + range->len,
10363 cache->key.objectid + cache->key.offset);
10365 if (end - start >= range->minlen) {
10366 if (!block_group_cache_done(cache)) {
10367 ret = cache_block_group(cache, 0);
10369 btrfs_put_block_group(cache);
10372 ret = wait_block_group_cache_done(cache);
10374 btrfs_put_block_group(cache);
10378 ret = btrfs_trim_block_group(cache,
10384 trimmed += group_trimmed;
10386 btrfs_put_block_group(cache);
10391 cache = next_block_group(fs_info->tree_root, cache);
10394 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10395 devices = &root->fs_info->fs_devices->alloc_list;
10396 list_for_each_entry(device, devices, dev_alloc_list) {
10397 ret = btrfs_trim_free_extents(device, range->minlen,
10402 trimmed += group_trimmed;
10404 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10406 range->len = trimmed;
10411 * btrfs_{start,end}_write_no_snapshoting() are similar to
10412 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10413 * data into the page cache through nocow before the subvolume is snapshoted,
10414 * but flush the data into disk after the snapshot creation, or to prevent
10415 * operations while snapshoting is ongoing and that cause the snapshot to be
10416 * inconsistent (writes followed by expanding truncates for example).
10418 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10420 percpu_counter_dec(&root->subv_writers->counter);
10422 * Make sure counter is updated before we wake up waiters.
10425 if (waitqueue_active(&root->subv_writers->wait))
10426 wake_up(&root->subv_writers->wait);
10429 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10431 if (atomic_read(&root->will_be_snapshoted))
10434 percpu_counter_inc(&root->subv_writers->counter);
10436 * Make sure counter is updated before we check for snapshot creation.
10439 if (atomic_read(&root->will_be_snapshoted)) {
10440 btrfs_end_write_no_snapshoting(root);