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"
32 #include "transaction.h"
36 #include "free-space-cache.h"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE = 0,
57 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 static int update_block_group(struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
108 int btrfs_pin_extent(struct btrfs_root *root,
109 u64 bytenr, u64 num_bytes, int reserved);
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 if (block_group->key.objectid < cache->key.objectid) {
159 } else if (block_group->key.objectid > cache->key.objectid) {
162 spin_unlock(&info->block_group_cache_lock);
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
174 spin_unlock(&info->block_group_cache_lock);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
187 struct btrfs_block_group_cache *cache, *ret = NULL;
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
195 cache = rb_entry(n, struct btrfs_block_group_cache,
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
220 spin_unlock(&info->block_group_cache_lock);
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (cache->cached != BTRFS_CACHE_STARTED) {
315 spin_unlock(&cache->lock);
319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache->caching_ctl) {
321 spin_unlock(&cache->lock);
325 ctl = cache->caching_ctl;
326 atomic_inc(&ctl->count);
327 spin_unlock(&cache->lock);
331 static void put_caching_control(struct btrfs_caching_control *ctl)
333 if (atomic_dec_and_test(&ctl->count))
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
343 struct btrfs_fs_info *info, u64 start, u64 end)
345 u64 extent_start, extent_end, size, total_added = 0;
348 while (start < end) {
349 ret = find_first_extent_bit(info->pinned_extents, start,
350 &extent_start, &extent_end,
351 EXTENT_DIRTY | EXTENT_UPTODATE,
356 if (extent_start <= start) {
357 start = extent_end + 1;
358 } else if (extent_start > start && extent_start < end) {
359 size = extent_start - start;
361 ret = btrfs_add_free_space(block_group, start,
363 BUG_ON(ret); /* -ENOMEM or logic error */
364 start = extent_end + 1;
373 ret = btrfs_add_free_space(block_group, start, size);
374 BUG_ON(ret); /* -ENOMEM or logic error */
380 static noinline void caching_thread(struct btrfs_work *work)
382 struct btrfs_block_group_cache *block_group;
383 struct btrfs_fs_info *fs_info;
384 struct btrfs_caching_control *caching_ctl;
385 struct btrfs_root *extent_root;
386 struct btrfs_path *path;
387 struct extent_buffer *leaf;
388 struct btrfs_key key;
394 caching_ctl = container_of(work, struct btrfs_caching_control, work);
395 block_group = caching_ctl->block_group;
396 fs_info = block_group->fs_info;
397 extent_root = fs_info->extent_root;
399 path = btrfs_alloc_path();
403 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
411 path->skip_locking = 1;
412 path->search_commit_root = 1;
417 key.type = BTRFS_EXTENT_ITEM_KEY;
419 mutex_lock(&caching_ctl->mutex);
420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info->extent_commit_sem);
424 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
428 leaf = path->nodes[0];
429 nritems = btrfs_header_nritems(leaf);
432 if (btrfs_fs_closing(fs_info) > 1) {
437 if (path->slots[0] < nritems) {
438 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
440 ret = find_next_key(path, 0, &key);
444 if (need_resched()) {
445 caching_ctl->progress = last;
446 btrfs_release_path(path);
447 up_read(&fs_info->extent_commit_sem);
448 mutex_unlock(&caching_ctl->mutex);
453 ret = btrfs_next_leaf(extent_root, path);
458 leaf = path->nodes[0];
459 nritems = btrfs_header_nritems(leaf);
463 if (key.objectid < last) {
466 key.type = BTRFS_EXTENT_ITEM_KEY;
468 caching_ctl->progress = last;
469 btrfs_release_path(path);
473 if (key.objectid < block_group->key.objectid) {
478 if (key.objectid >= block_group->key.objectid +
479 block_group->key.offset)
482 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 key.type == BTRFS_METADATA_ITEM_KEY) {
484 total_found += add_new_free_space(block_group,
487 if (key.type == BTRFS_METADATA_ITEM_KEY)
488 last = key.objectid +
489 fs_info->tree_root->leafsize;
491 last = key.objectid + key.offset;
493 if (total_found > (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl->wait);
502 total_found += add_new_free_space(block_group, fs_info, last,
503 block_group->key.objectid +
504 block_group->key.offset);
505 caching_ctl->progress = (u64)-1;
507 spin_lock(&block_group->lock);
508 block_group->caching_ctl = NULL;
509 block_group->cached = BTRFS_CACHE_FINISHED;
510 spin_unlock(&block_group->lock);
513 btrfs_free_path(path);
514 up_read(&fs_info->extent_commit_sem);
516 free_excluded_extents(extent_root, block_group);
518 mutex_unlock(&caching_ctl->mutex);
521 spin_lock(&block_group->lock);
522 block_group->caching_ctl = NULL;
523 block_group->cached = BTRFS_CACHE_ERROR;
524 spin_unlock(&block_group->lock);
526 wake_up(&caching_ctl->wait);
528 put_caching_control(caching_ctl);
529 btrfs_put_block_group(block_group);
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
536 struct btrfs_fs_info *fs_info = cache->fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
544 INIT_LIST_HEAD(&caching_ctl->list);
545 mutex_init(&caching_ctl->mutex);
546 init_waitqueue_head(&caching_ctl->wait);
547 caching_ctl->block_group = cache;
548 caching_ctl->progress = cache->key.objectid;
549 atomic_set(&caching_ctl->count, 1);
550 caching_ctl->work.func = caching_thread;
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 ret = load_free_space_cache(fs_info, cache);
593 spin_lock(&cache->lock);
595 cache->caching_ctl = NULL;
596 cache->cached = BTRFS_CACHE_FINISHED;
597 cache->last_byte_to_unpin = (u64)-1;
599 if (load_cache_only) {
600 cache->caching_ctl = NULL;
601 cache->cached = BTRFS_CACHE_NO;
603 cache->cached = BTRFS_CACHE_STARTED;
606 spin_unlock(&cache->lock);
607 wake_up(&caching_ctl->wait);
609 put_caching_control(caching_ctl);
610 free_excluded_extents(fs_info->extent_root, cache);
615 * We are not going to do the fast caching, set cached to the
616 * appropriate value and wakeup any waiters.
618 spin_lock(&cache->lock);
619 if (load_cache_only) {
620 cache->caching_ctl = NULL;
621 cache->cached = BTRFS_CACHE_NO;
623 cache->cached = BTRFS_CACHE_STARTED;
625 spin_unlock(&cache->lock);
626 wake_up(&caching_ctl->wait);
629 if (load_cache_only) {
630 put_caching_control(caching_ctl);
634 down_write(&fs_info->extent_commit_sem);
635 atomic_inc(&caching_ctl->count);
636 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
637 up_write(&fs_info->extent_commit_sem);
639 btrfs_get_block_group(cache);
641 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
647 * return the block group that starts at or after bytenr
649 static struct btrfs_block_group_cache *
650 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
652 struct btrfs_block_group_cache *cache;
654 cache = block_group_cache_tree_search(info, bytenr, 0);
660 * return the block group that contains the given bytenr
662 struct btrfs_block_group_cache *btrfs_lookup_block_group(
663 struct btrfs_fs_info *info,
666 struct btrfs_block_group_cache *cache;
668 cache = block_group_cache_tree_search(info, bytenr, 1);
673 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
676 struct list_head *head = &info->space_info;
677 struct btrfs_space_info *found;
679 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
682 list_for_each_entry_rcu(found, head, list) {
683 if (found->flags & flags) {
693 * after adding space to the filesystem, we need to clear the full flags
694 * on all the space infos.
696 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
698 struct list_head *head = &info->space_info;
699 struct btrfs_space_info *found;
702 list_for_each_entry_rcu(found, head, list)
707 /* simple helper to search for an existing extent at a given offset */
708 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
711 struct btrfs_key key;
712 struct btrfs_path *path;
714 path = btrfs_alloc_path();
718 key.objectid = start;
720 key.type = BTRFS_EXTENT_ITEM_KEY;
721 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
724 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
725 if (key.objectid == start &&
726 key.type == BTRFS_METADATA_ITEM_KEY)
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->leafsize;
766 path = btrfs_alloc_path();
771 key.objectid = bytenr;
772 key.type = BTRFS_METADATA_ITEM_KEY;
775 key.objectid = bytenr;
776 key.type = BTRFS_EXTENT_ITEM_KEY;
781 path->skip_locking = 1;
782 path->search_commit_root = 1;
785 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
790 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
792 if (path->slots[0]) {
794 btrfs_item_key_to_cpu(path->nodes[0], &key,
796 if (key.objectid == bytenr &&
797 key.type == BTRFS_EXTENT_ITEM_KEY &&
798 key.offset == root->leafsize)
802 key.objectid = bytenr;
803 key.type = BTRFS_EXTENT_ITEM_KEY;
804 key.offset = root->leafsize;
805 btrfs_release_path(path);
811 leaf = path->nodes[0];
812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
813 if (item_size >= sizeof(*ei)) {
814 ei = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_extent_item);
816 num_refs = btrfs_extent_refs(leaf, ei);
817 extent_flags = btrfs_extent_flags(leaf, ei);
819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
820 struct btrfs_extent_item_v0 *ei0;
821 BUG_ON(item_size != sizeof(*ei0));
822 ei0 = btrfs_item_ptr(leaf, path->slots[0],
823 struct btrfs_extent_item_v0);
824 num_refs = btrfs_extent_refs_v0(leaf, ei0);
825 /* FIXME: this isn't correct for data */
826 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
831 BUG_ON(num_refs == 0);
841 delayed_refs = &trans->transaction->delayed_refs;
842 spin_lock(&delayed_refs->lock);
843 head = btrfs_find_delayed_ref_head(trans, bytenr);
845 if (!mutex_trylock(&head->mutex)) {
846 atomic_inc(&head->node.refs);
847 spin_unlock(&delayed_refs->lock);
849 btrfs_release_path(path);
852 * Mutex was contended, block until it's released and try
855 mutex_lock(&head->mutex);
856 mutex_unlock(&head->mutex);
857 btrfs_put_delayed_ref(&head->node);
860 if (head->extent_op && head->extent_op->update_flags)
861 extent_flags |= head->extent_op->flags_to_set;
863 BUG_ON(num_refs == 0);
865 num_refs += head->node.ref_mod;
866 mutex_unlock(&head->mutex);
868 spin_unlock(&delayed_refs->lock);
870 WARN_ON(num_refs == 0);
874 *flags = extent_flags;
876 btrfs_free_path(path);
881 * Back reference rules. Back refs have three main goals:
883 * 1) differentiate between all holders of references to an extent so that
884 * when a reference is dropped we can make sure it was a valid reference
885 * before freeing the extent.
887 * 2) Provide enough information to quickly find the holders of an extent
888 * if we notice a given block is corrupted or bad.
890 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
891 * maintenance. This is actually the same as #2, but with a slightly
892 * different use case.
894 * There are two kinds of back refs. The implicit back refs is optimized
895 * for pointers in non-shared tree blocks. For a given pointer in a block,
896 * back refs of this kind provide information about the block's owner tree
897 * and the pointer's key. These information allow us to find the block by
898 * b-tree searching. The full back refs is for pointers in tree blocks not
899 * referenced by their owner trees. The location of tree block is recorded
900 * in the back refs. Actually the full back refs is generic, and can be
901 * used in all cases the implicit back refs is used. The major shortcoming
902 * of the full back refs is its overhead. Every time a tree block gets
903 * COWed, we have to update back refs entry for all pointers in it.
905 * For a newly allocated tree block, we use implicit back refs for
906 * pointers in it. This means most tree related operations only involve
907 * implicit back refs. For a tree block created in old transaction, the
908 * only way to drop a reference to it is COW it. So we can detect the
909 * event that tree block loses its owner tree's reference and do the
910 * back refs conversion.
912 * When a tree block is COW'd through a tree, there are four cases:
914 * The reference count of the block is one and the tree is the block's
915 * owner tree. Nothing to do in this case.
917 * The reference count of the block is one and the tree is not the
918 * block's owner tree. In this case, full back refs is used for pointers
919 * in the block. Remove these full back refs, add implicit back refs for
920 * every pointers in the new block.
922 * The reference count of the block is greater than one and the tree is
923 * the block's owner tree. In this case, implicit back refs is used for
924 * pointers in the block. Add full back refs for every pointers in the
925 * block, increase lower level extents' reference counts. The original
926 * implicit back refs are entailed to the new block.
928 * The reference count of the block is greater than one and the tree is
929 * not the block's owner tree. Add implicit back refs for every pointer in
930 * the new block, increase lower level extents' reference count.
932 * Back Reference Key composing:
934 * The key objectid corresponds to the first byte in the extent,
935 * The key type is used to differentiate between types of back refs.
936 * There are different meanings of the key offset for different types
939 * File extents can be referenced by:
941 * - multiple snapshots, subvolumes, or different generations in one subvol
942 * - different files inside a single subvolume
943 * - different offsets inside a file (bookend extents in file.c)
945 * The extent ref structure for the implicit back refs has fields for:
947 * - Objectid of the subvolume root
948 * - objectid of the file holding the reference
949 * - original offset in the file
950 * - how many bookend extents
952 * The key offset for the implicit back refs is hash of the first
955 * The extent ref structure for the full back refs has field for:
957 * - number of pointers in the tree leaf
959 * The key offset for the implicit back refs is the first byte of
962 * When a file extent is allocated, The implicit back refs is used.
963 * the fields are filled in:
965 * (root_key.objectid, inode objectid, offset in file, 1)
967 * When a file extent is removed file truncation, we find the
968 * corresponding implicit back refs and check the following fields:
970 * (btrfs_header_owner(leaf), inode objectid, offset in file)
972 * Btree extents can be referenced by:
974 * - Different subvolumes
976 * Both the implicit back refs and the full back refs for tree blocks
977 * only consist of key. The key offset for the implicit back refs is
978 * objectid of block's owner tree. The key offset for the full back refs
979 * is the first byte of parent block.
981 * When implicit back refs is used, information about the lowest key and
982 * level of the tree block are required. These information are stored in
983 * tree block info structure.
986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
987 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
989 struct btrfs_path *path,
990 u64 owner, u32 extra_size)
992 struct btrfs_extent_item *item;
993 struct btrfs_extent_item_v0 *ei0;
994 struct btrfs_extent_ref_v0 *ref0;
995 struct btrfs_tree_block_info *bi;
996 struct extent_buffer *leaf;
997 struct btrfs_key key;
998 struct btrfs_key found_key;
999 u32 new_size = sizeof(*item);
1003 leaf = path->nodes[0];
1004 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1006 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1007 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1008 struct btrfs_extent_item_v0);
1009 refs = btrfs_extent_refs_v0(leaf, ei0);
1011 if (owner == (u64)-1) {
1013 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1014 ret = btrfs_next_leaf(root, path);
1017 BUG_ON(ret > 0); /* Corruption */
1018 leaf = path->nodes[0];
1020 btrfs_item_key_to_cpu(leaf, &found_key,
1022 BUG_ON(key.objectid != found_key.objectid);
1023 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1027 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1028 struct btrfs_extent_ref_v0);
1029 owner = btrfs_ref_objectid_v0(leaf, ref0);
1033 btrfs_release_path(path);
1035 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1036 new_size += sizeof(*bi);
1038 new_size -= sizeof(*ei0);
1039 ret = btrfs_search_slot(trans, root, &key, path,
1040 new_size + extra_size, 1);
1043 BUG_ON(ret); /* Corruption */
1045 btrfs_extend_item(root, path, new_size);
1047 leaf = path->nodes[0];
1048 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1049 btrfs_set_extent_refs(leaf, item, refs);
1050 /* FIXME: get real generation */
1051 btrfs_set_extent_generation(leaf, item, 0);
1052 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1053 btrfs_set_extent_flags(leaf, item,
1054 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1055 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1056 bi = (struct btrfs_tree_block_info *)(item + 1);
1057 /* FIXME: get first key of the block */
1058 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1059 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1061 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1063 btrfs_mark_buffer_dirty(leaf);
1068 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1070 u32 high_crc = ~(u32)0;
1071 u32 low_crc = ~(u32)0;
1074 lenum = cpu_to_le64(root_objectid);
1075 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1076 lenum = cpu_to_le64(owner);
1077 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1078 lenum = cpu_to_le64(offset);
1079 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1081 return ((u64)high_crc << 31) ^ (u64)low_crc;
1084 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref)
1087 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1088 btrfs_extent_data_ref_objectid(leaf, ref),
1089 btrfs_extent_data_ref_offset(leaf, ref));
1092 static int match_extent_data_ref(struct extent_buffer *leaf,
1093 struct btrfs_extent_data_ref *ref,
1094 u64 root_objectid, u64 owner, u64 offset)
1096 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1097 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1098 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1103 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1104 struct btrfs_root *root,
1105 struct btrfs_path *path,
1106 u64 bytenr, u64 parent,
1108 u64 owner, u64 offset)
1110 struct btrfs_key key;
1111 struct btrfs_extent_data_ref *ref;
1112 struct extent_buffer *leaf;
1118 key.objectid = bytenr;
1120 key.type = BTRFS_SHARED_DATA_REF_KEY;
1121 key.offset = parent;
1123 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1124 key.offset = hash_extent_data_ref(root_objectid,
1129 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1139 key.type = BTRFS_EXTENT_REF_V0_KEY;
1140 btrfs_release_path(path);
1141 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1152 leaf = path->nodes[0];
1153 nritems = btrfs_header_nritems(leaf);
1155 if (path->slots[0] >= nritems) {
1156 ret = btrfs_next_leaf(root, path);
1162 leaf = path->nodes[0];
1163 nritems = btrfs_header_nritems(leaf);
1167 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1168 if (key.objectid != bytenr ||
1169 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1172 ref = btrfs_item_ptr(leaf, path->slots[0],
1173 struct btrfs_extent_data_ref);
1175 if (match_extent_data_ref(leaf, ref, root_objectid,
1178 btrfs_release_path(path);
1190 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1191 struct btrfs_root *root,
1192 struct btrfs_path *path,
1193 u64 bytenr, u64 parent,
1194 u64 root_objectid, u64 owner,
1195 u64 offset, int refs_to_add)
1197 struct btrfs_key key;
1198 struct extent_buffer *leaf;
1203 key.objectid = bytenr;
1205 key.type = BTRFS_SHARED_DATA_REF_KEY;
1206 key.offset = parent;
1207 size = sizeof(struct btrfs_shared_data_ref);
1209 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1210 key.offset = hash_extent_data_ref(root_objectid,
1212 size = sizeof(struct btrfs_extent_data_ref);
1215 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1216 if (ret && ret != -EEXIST)
1219 leaf = path->nodes[0];
1221 struct btrfs_shared_data_ref *ref;
1222 ref = btrfs_item_ptr(leaf, path->slots[0],
1223 struct btrfs_shared_data_ref);
1225 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1227 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1228 num_refs += refs_to_add;
1229 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1232 struct btrfs_extent_data_ref *ref;
1233 while (ret == -EEXIST) {
1234 ref = btrfs_item_ptr(leaf, path->slots[0],
1235 struct btrfs_extent_data_ref);
1236 if (match_extent_data_ref(leaf, ref, root_objectid,
1239 btrfs_release_path(path);
1241 ret = btrfs_insert_empty_item(trans, root, path, &key,
1243 if (ret && ret != -EEXIST)
1246 leaf = path->nodes[0];
1248 ref = btrfs_item_ptr(leaf, path->slots[0],
1249 struct btrfs_extent_data_ref);
1251 btrfs_set_extent_data_ref_root(leaf, ref,
1253 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1254 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1255 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1258 num_refs += refs_to_add;
1259 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1262 btrfs_mark_buffer_dirty(leaf);
1265 btrfs_release_path(path);
1269 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root,
1271 struct btrfs_path *path,
1274 struct btrfs_key key;
1275 struct btrfs_extent_data_ref *ref1 = NULL;
1276 struct btrfs_shared_data_ref *ref2 = NULL;
1277 struct extent_buffer *leaf;
1281 leaf = path->nodes[0];
1282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1284 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1285 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_data_ref);
1287 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1288 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1289 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_shared_data_ref);
1291 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1292 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1293 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1294 struct btrfs_extent_ref_v0 *ref0;
1295 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1296 struct btrfs_extent_ref_v0);
1297 num_refs = btrfs_ref_count_v0(leaf, ref0);
1303 BUG_ON(num_refs < refs_to_drop);
1304 num_refs -= refs_to_drop;
1306 if (num_refs == 0) {
1307 ret = btrfs_del_item(trans, root, path);
1309 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1310 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1311 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1312 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1313 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1315 struct btrfs_extent_ref_v0 *ref0;
1316 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_extent_ref_v0);
1318 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1321 btrfs_mark_buffer_dirty(leaf);
1326 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1327 struct btrfs_path *path,
1328 struct btrfs_extent_inline_ref *iref)
1330 struct btrfs_key key;
1331 struct extent_buffer *leaf;
1332 struct btrfs_extent_data_ref *ref1;
1333 struct btrfs_shared_data_ref *ref2;
1336 leaf = path->nodes[0];
1337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1340 BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1345 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1348 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_extent_data_ref);
1350 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1351 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1352 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_shared_data_ref);
1354 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1355 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1356 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1357 struct btrfs_extent_ref_v0 *ref0;
1358 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1359 struct btrfs_extent_ref_v0);
1360 num_refs = btrfs_ref_count_v0(leaf, ref0);
1368 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root,
1370 struct btrfs_path *path,
1371 u64 bytenr, u64 parent,
1374 struct btrfs_key key;
1377 key.objectid = bytenr;
1379 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380 key.offset = parent;
1382 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383 key.offset = root_objectid;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1389 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1390 if (ret == -ENOENT && parent) {
1391 btrfs_release_path(path);
1392 key.type = BTRFS_EXTENT_REF_V0_KEY;
1393 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1401 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_path *path,
1404 u64 bytenr, u64 parent,
1407 struct btrfs_key key;
1410 key.objectid = bytenr;
1412 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1413 key.offset = parent;
1415 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1416 key.offset = root_objectid;
1419 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1420 btrfs_release_path(path);
1424 static inline int extent_ref_type(u64 parent, u64 owner)
1427 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1429 type = BTRFS_SHARED_BLOCK_REF_KEY;
1431 type = BTRFS_TREE_BLOCK_REF_KEY;
1434 type = BTRFS_SHARED_DATA_REF_KEY;
1436 type = BTRFS_EXTENT_DATA_REF_KEY;
1441 static int find_next_key(struct btrfs_path *path, int level,
1442 struct btrfs_key *key)
1445 for (; level < BTRFS_MAX_LEVEL; level++) {
1446 if (!path->nodes[level])
1448 if (path->slots[level] + 1 >=
1449 btrfs_header_nritems(path->nodes[level]))
1452 btrfs_item_key_to_cpu(path->nodes[level], key,
1453 path->slots[level] + 1);
1455 btrfs_node_key_to_cpu(path->nodes[level], key,
1456 path->slots[level] + 1);
1463 * look for inline back ref. if back ref is found, *ref_ret is set
1464 * to the address of inline back ref, and 0 is returned.
1466 * if back ref isn't found, *ref_ret is set to the address where it
1467 * should be inserted, and -ENOENT is returned.
1469 * if insert is true and there are too many inline back refs, the path
1470 * points to the extent item, and -EAGAIN is returned.
1472 * NOTE: inline back refs are ordered in the same way that back ref
1473 * items in the tree are ordered.
1475 static noinline_for_stack
1476 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1477 struct btrfs_root *root,
1478 struct btrfs_path *path,
1479 struct btrfs_extent_inline_ref **ref_ret,
1480 u64 bytenr, u64 num_bytes,
1481 u64 parent, u64 root_objectid,
1482 u64 owner, u64 offset, int insert)
1484 struct btrfs_key key;
1485 struct extent_buffer *leaf;
1486 struct btrfs_extent_item *ei;
1487 struct btrfs_extent_inline_ref *iref;
1497 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1500 key.objectid = bytenr;
1501 key.type = BTRFS_EXTENT_ITEM_KEY;
1502 key.offset = num_bytes;
1504 want = extent_ref_type(parent, owner);
1506 extra_size = btrfs_extent_inline_ref_size(want);
1507 path->keep_locks = 1;
1512 * Owner is our parent level, so we can just add one to get the level
1513 * for the block we are interested in.
1515 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1516 key.type = BTRFS_METADATA_ITEM_KEY;
1521 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1528 * We may be a newly converted file system which still has the old fat
1529 * extent entries for metadata, so try and see if we have one of those.
1531 if (ret > 0 && skinny_metadata) {
1532 skinny_metadata = false;
1533 if (path->slots[0]) {
1535 btrfs_item_key_to_cpu(path->nodes[0], &key,
1537 if (key.objectid == bytenr &&
1538 key.type == BTRFS_EXTENT_ITEM_KEY &&
1539 key.offset == num_bytes)
1543 key.type = BTRFS_EXTENT_ITEM_KEY;
1544 key.offset = num_bytes;
1545 btrfs_release_path(path);
1550 if (ret && !insert) {
1559 leaf = path->nodes[0];
1560 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1561 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1562 if (item_size < sizeof(*ei)) {
1567 ret = convert_extent_item_v0(trans, root, path, owner,
1573 leaf = path->nodes[0];
1574 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1577 BUG_ON(item_size < sizeof(*ei));
1579 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1580 flags = btrfs_extent_flags(leaf, ei);
1582 ptr = (unsigned long)(ei + 1);
1583 end = (unsigned long)ei + item_size;
1585 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1586 ptr += sizeof(struct btrfs_tree_block_info);
1596 iref = (struct btrfs_extent_inline_ref *)ptr;
1597 type = btrfs_extent_inline_ref_type(leaf, iref);
1601 ptr += btrfs_extent_inline_ref_size(type);
1605 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1606 struct btrfs_extent_data_ref *dref;
1607 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1608 if (match_extent_data_ref(leaf, dref, root_objectid,
1613 if (hash_extent_data_ref_item(leaf, dref) <
1614 hash_extent_data_ref(root_objectid, owner, offset))
1618 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1620 if (parent == ref_offset) {
1624 if (ref_offset < parent)
1627 if (root_objectid == ref_offset) {
1631 if (ref_offset < root_objectid)
1635 ptr += btrfs_extent_inline_ref_size(type);
1637 if (err == -ENOENT && insert) {
1638 if (item_size + extra_size >=
1639 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1644 * To add new inline back ref, we have to make sure
1645 * there is no corresponding back ref item.
1646 * For simplicity, we just do not add new inline back
1647 * ref if there is any kind of item for this block
1649 if (find_next_key(path, 0, &key) == 0 &&
1650 key.objectid == bytenr &&
1651 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1656 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1659 path->keep_locks = 0;
1660 btrfs_unlock_up_safe(path, 1);
1666 * helper to add new inline back ref
1668 static noinline_for_stack
1669 void setup_inline_extent_backref(struct btrfs_root *root,
1670 struct btrfs_path *path,
1671 struct btrfs_extent_inline_ref *iref,
1672 u64 parent, u64 root_objectid,
1673 u64 owner, u64 offset, int refs_to_add,
1674 struct btrfs_delayed_extent_op *extent_op)
1676 struct extent_buffer *leaf;
1677 struct btrfs_extent_item *ei;
1680 unsigned long item_offset;
1685 leaf = path->nodes[0];
1686 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1687 item_offset = (unsigned long)iref - (unsigned long)ei;
1689 type = extent_ref_type(parent, owner);
1690 size = btrfs_extent_inline_ref_size(type);
1692 btrfs_extend_item(root, path, size);
1694 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1695 refs = btrfs_extent_refs(leaf, ei);
1696 refs += refs_to_add;
1697 btrfs_set_extent_refs(leaf, ei, refs);
1699 __run_delayed_extent_op(extent_op, leaf, ei);
1701 ptr = (unsigned long)ei + item_offset;
1702 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1703 if (ptr < end - size)
1704 memmove_extent_buffer(leaf, ptr + size, ptr,
1707 iref = (struct btrfs_extent_inline_ref *)ptr;
1708 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1709 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1710 struct btrfs_extent_data_ref *dref;
1711 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1712 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1713 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1714 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1715 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1716 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1717 struct btrfs_shared_data_ref *sref;
1718 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1719 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1720 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1721 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1722 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1724 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1726 btrfs_mark_buffer_dirty(leaf);
1729 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1730 struct btrfs_root *root,
1731 struct btrfs_path *path,
1732 struct btrfs_extent_inline_ref **ref_ret,
1733 u64 bytenr, u64 num_bytes, u64 parent,
1734 u64 root_objectid, u64 owner, u64 offset)
1738 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1739 bytenr, num_bytes, parent,
1740 root_objectid, owner, offset, 0);
1744 btrfs_release_path(path);
1747 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1748 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1751 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1752 root_objectid, owner, offset);
1758 * helper to update/remove inline back ref
1760 static noinline_for_stack
1761 void update_inline_extent_backref(struct btrfs_root *root,
1762 struct btrfs_path *path,
1763 struct btrfs_extent_inline_ref *iref,
1765 struct btrfs_delayed_extent_op *extent_op)
1767 struct extent_buffer *leaf;
1768 struct btrfs_extent_item *ei;
1769 struct btrfs_extent_data_ref *dref = NULL;
1770 struct btrfs_shared_data_ref *sref = NULL;
1778 leaf = path->nodes[0];
1779 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1780 refs = btrfs_extent_refs(leaf, ei);
1781 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1782 refs += refs_to_mod;
1783 btrfs_set_extent_refs(leaf, ei, refs);
1785 __run_delayed_extent_op(extent_op, leaf, ei);
1787 type = btrfs_extent_inline_ref_type(leaf, iref);
1789 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1790 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1791 refs = btrfs_extent_data_ref_count(leaf, dref);
1792 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1793 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1794 refs = btrfs_shared_data_ref_count(leaf, sref);
1797 BUG_ON(refs_to_mod != -1);
1800 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1801 refs += refs_to_mod;
1804 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1805 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1807 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1809 size = btrfs_extent_inline_ref_size(type);
1810 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1811 ptr = (unsigned long)iref;
1812 end = (unsigned long)ei + item_size;
1813 if (ptr + size < end)
1814 memmove_extent_buffer(leaf, ptr, ptr + size,
1817 btrfs_truncate_item(root, path, item_size, 1);
1819 btrfs_mark_buffer_dirty(leaf);
1822 static noinline_for_stack
1823 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *root,
1825 struct btrfs_path *path,
1826 u64 bytenr, u64 num_bytes, u64 parent,
1827 u64 root_objectid, u64 owner,
1828 u64 offset, int refs_to_add,
1829 struct btrfs_delayed_extent_op *extent_op)
1831 struct btrfs_extent_inline_ref *iref;
1834 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1835 bytenr, num_bytes, parent,
1836 root_objectid, owner, offset, 1);
1838 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1839 update_inline_extent_backref(root, path, iref,
1840 refs_to_add, extent_op);
1841 } else if (ret == -ENOENT) {
1842 setup_inline_extent_backref(root, path, iref, parent,
1843 root_objectid, owner, offset,
1844 refs_to_add, extent_op);
1850 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1851 struct btrfs_root *root,
1852 struct btrfs_path *path,
1853 u64 bytenr, u64 parent, u64 root_objectid,
1854 u64 owner, u64 offset, int refs_to_add)
1857 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1858 BUG_ON(refs_to_add != 1);
1859 ret = insert_tree_block_ref(trans, root, path, bytenr,
1860 parent, root_objectid);
1862 ret = insert_extent_data_ref(trans, root, path, bytenr,
1863 parent, root_objectid,
1864 owner, offset, refs_to_add);
1869 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1870 struct btrfs_root *root,
1871 struct btrfs_path *path,
1872 struct btrfs_extent_inline_ref *iref,
1873 int refs_to_drop, int is_data)
1877 BUG_ON(!is_data && refs_to_drop != 1);
1879 update_inline_extent_backref(root, path, iref,
1880 -refs_to_drop, NULL);
1881 } else if (is_data) {
1882 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1884 ret = btrfs_del_item(trans, root, path);
1889 static int btrfs_issue_discard(struct block_device *bdev,
1892 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1895 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1896 u64 num_bytes, u64 *actual_bytes)
1899 u64 discarded_bytes = 0;
1900 struct btrfs_bio *bbio = NULL;
1903 /* Tell the block device(s) that the sectors can be discarded */
1904 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1905 bytenr, &num_bytes, &bbio, 0);
1906 /* Error condition is -ENOMEM */
1908 struct btrfs_bio_stripe *stripe = bbio->stripes;
1912 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1913 if (!stripe->dev->can_discard)
1916 ret = btrfs_issue_discard(stripe->dev->bdev,
1920 discarded_bytes += stripe->length;
1921 else if (ret != -EOPNOTSUPP)
1922 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1925 * Just in case we get back EOPNOTSUPP for some reason,
1926 * just ignore the return value so we don't screw up
1927 * people calling discard_extent.
1935 *actual_bytes = discarded_bytes;
1938 if (ret == -EOPNOTSUPP)
1943 /* Can return -ENOMEM */
1944 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1945 struct btrfs_root *root,
1946 u64 bytenr, u64 num_bytes, u64 parent,
1947 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1950 struct btrfs_fs_info *fs_info = root->fs_info;
1952 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1953 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1955 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1956 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1958 parent, root_objectid, (int)owner,
1959 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1961 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1963 parent, root_objectid, owner, offset,
1964 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1969 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1970 struct btrfs_root *root,
1971 u64 bytenr, u64 num_bytes,
1972 u64 parent, u64 root_objectid,
1973 u64 owner, u64 offset, int refs_to_add,
1974 struct btrfs_delayed_extent_op *extent_op)
1976 struct btrfs_path *path;
1977 struct extent_buffer *leaf;
1978 struct btrfs_extent_item *item;
1982 path = btrfs_alloc_path();
1987 path->leave_spinning = 1;
1988 /* this will setup the path even if it fails to insert the back ref */
1989 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1990 path, bytenr, num_bytes, parent,
1991 root_objectid, owner, offset,
1992 refs_to_add, extent_op);
1996 leaf = path->nodes[0];
1997 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998 refs = btrfs_extent_refs(leaf, item);
1999 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2001 __run_delayed_extent_op(extent_op, leaf, item);
2003 btrfs_mark_buffer_dirty(leaf);
2004 btrfs_release_path(path);
2007 path->leave_spinning = 1;
2009 /* now insert the actual backref */
2010 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2011 path, bytenr, parent, root_objectid,
2012 owner, offset, refs_to_add);
2014 btrfs_abort_transaction(trans, root, ret);
2016 btrfs_free_path(path);
2020 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2021 struct btrfs_root *root,
2022 struct btrfs_delayed_ref_node *node,
2023 struct btrfs_delayed_extent_op *extent_op,
2024 int insert_reserved)
2027 struct btrfs_delayed_data_ref *ref;
2028 struct btrfs_key ins;
2033 ins.objectid = node->bytenr;
2034 ins.offset = node->num_bytes;
2035 ins.type = BTRFS_EXTENT_ITEM_KEY;
2037 ref = btrfs_delayed_node_to_data_ref(node);
2038 trace_run_delayed_data_ref(node, ref, node->action);
2040 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2041 parent = ref->parent;
2043 ref_root = ref->root;
2045 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2047 flags |= extent_op->flags_to_set;
2048 ret = alloc_reserved_file_extent(trans, root,
2049 parent, ref_root, flags,
2050 ref->objectid, ref->offset,
2051 &ins, node->ref_mod);
2052 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2053 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2054 node->num_bytes, parent,
2055 ref_root, ref->objectid,
2056 ref->offset, node->ref_mod,
2058 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2059 ret = __btrfs_free_extent(trans, root, node->bytenr,
2060 node->num_bytes, parent,
2061 ref_root, ref->objectid,
2062 ref->offset, node->ref_mod,
2070 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2071 struct extent_buffer *leaf,
2072 struct btrfs_extent_item *ei)
2074 u64 flags = btrfs_extent_flags(leaf, ei);
2075 if (extent_op->update_flags) {
2076 flags |= extent_op->flags_to_set;
2077 btrfs_set_extent_flags(leaf, ei, flags);
2080 if (extent_op->update_key) {
2081 struct btrfs_tree_block_info *bi;
2082 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2083 bi = (struct btrfs_tree_block_info *)(ei + 1);
2084 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2088 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2089 struct btrfs_root *root,
2090 struct btrfs_delayed_ref_node *node,
2091 struct btrfs_delayed_extent_op *extent_op)
2093 struct btrfs_key key;
2094 struct btrfs_path *path;
2095 struct btrfs_extent_item *ei;
2096 struct extent_buffer *leaf;
2100 int metadata = !extent_op->is_data;
2105 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2108 path = btrfs_alloc_path();
2112 key.objectid = node->bytenr;
2115 key.type = BTRFS_METADATA_ITEM_KEY;
2116 key.offset = extent_op->level;
2118 key.type = BTRFS_EXTENT_ITEM_KEY;
2119 key.offset = node->num_bytes;
2124 path->leave_spinning = 1;
2125 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2133 btrfs_release_path(path);
2136 key.offset = node->num_bytes;
2137 key.type = BTRFS_EXTENT_ITEM_KEY;
2144 leaf = path->nodes[0];
2145 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2146 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2147 if (item_size < sizeof(*ei)) {
2148 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2154 leaf = path->nodes[0];
2155 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2158 BUG_ON(item_size < sizeof(*ei));
2159 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2160 __run_delayed_extent_op(extent_op, leaf, ei);
2162 btrfs_mark_buffer_dirty(leaf);
2164 btrfs_free_path(path);
2168 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2169 struct btrfs_root *root,
2170 struct btrfs_delayed_ref_node *node,
2171 struct btrfs_delayed_extent_op *extent_op,
2172 int insert_reserved)
2175 struct btrfs_delayed_tree_ref *ref;
2176 struct btrfs_key ins;
2179 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2182 ref = btrfs_delayed_node_to_tree_ref(node);
2183 trace_run_delayed_tree_ref(node, ref, node->action);
2185 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2186 parent = ref->parent;
2188 ref_root = ref->root;
2190 ins.objectid = node->bytenr;
2191 if (skinny_metadata) {
2192 ins.offset = ref->level;
2193 ins.type = BTRFS_METADATA_ITEM_KEY;
2195 ins.offset = node->num_bytes;
2196 ins.type = BTRFS_EXTENT_ITEM_KEY;
2199 BUG_ON(node->ref_mod != 1);
2200 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2201 BUG_ON(!extent_op || !extent_op->update_flags);
2202 ret = alloc_reserved_tree_block(trans, root,
2204 extent_op->flags_to_set,
2207 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2208 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2209 node->num_bytes, parent, ref_root,
2210 ref->level, 0, 1, extent_op);
2211 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2212 ret = __btrfs_free_extent(trans, root, node->bytenr,
2213 node->num_bytes, parent, ref_root,
2214 ref->level, 0, 1, extent_op);
2221 /* helper function to actually process a single delayed ref entry */
2222 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2223 struct btrfs_root *root,
2224 struct btrfs_delayed_ref_node *node,
2225 struct btrfs_delayed_extent_op *extent_op,
2226 int insert_reserved)
2230 if (trans->aborted) {
2231 if (insert_reserved)
2232 btrfs_pin_extent(root, node->bytenr,
2233 node->num_bytes, 1);
2237 if (btrfs_delayed_ref_is_head(node)) {
2238 struct btrfs_delayed_ref_head *head;
2240 * we've hit the end of the chain and we were supposed
2241 * to insert this extent into the tree. But, it got
2242 * deleted before we ever needed to insert it, so all
2243 * we have to do is clean up the accounting
2246 head = btrfs_delayed_node_to_head(node);
2247 trace_run_delayed_ref_head(node, head, node->action);
2249 if (insert_reserved) {
2250 btrfs_pin_extent(root, node->bytenr,
2251 node->num_bytes, 1);
2252 if (head->is_data) {
2253 ret = btrfs_del_csums(trans, root,
2261 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2262 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2263 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2265 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2266 node->type == BTRFS_SHARED_DATA_REF_KEY)
2267 ret = run_delayed_data_ref(trans, root, node, extent_op,
2274 static noinline struct btrfs_delayed_ref_node *
2275 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2277 struct rb_node *node;
2278 struct btrfs_delayed_ref_node *ref;
2279 int action = BTRFS_ADD_DELAYED_REF;
2282 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2283 * this prevents ref count from going down to zero when
2284 * there still are pending delayed ref.
2286 node = rb_prev(&head->node.rb_node);
2290 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2292 if (ref->bytenr != head->node.bytenr)
2294 if (ref->action == action)
2296 node = rb_prev(node);
2298 if (action == BTRFS_ADD_DELAYED_REF) {
2299 action = BTRFS_DROP_DELAYED_REF;
2306 * Returns 0 on success or if called with an already aborted transaction.
2307 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2309 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2310 struct btrfs_root *root,
2311 struct list_head *cluster)
2313 struct btrfs_delayed_ref_root *delayed_refs;
2314 struct btrfs_delayed_ref_node *ref;
2315 struct btrfs_delayed_ref_head *locked_ref = NULL;
2316 struct btrfs_delayed_extent_op *extent_op;
2317 struct btrfs_fs_info *fs_info = root->fs_info;
2320 int must_insert_reserved = 0;
2322 delayed_refs = &trans->transaction->delayed_refs;
2325 /* pick a new head ref from the cluster list */
2326 if (list_empty(cluster))
2329 locked_ref = list_entry(cluster->next,
2330 struct btrfs_delayed_ref_head, cluster);
2332 /* grab the lock that says we are going to process
2333 * all the refs for this head */
2334 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2337 * we may have dropped the spin lock to get the head
2338 * mutex lock, and that might have given someone else
2339 * time to free the head. If that's true, it has been
2340 * removed from our list and we can move on.
2342 if (ret == -EAGAIN) {
2350 * We need to try and merge add/drops of the same ref since we
2351 * can run into issues with relocate dropping the implicit ref
2352 * and then it being added back again before the drop can
2353 * finish. If we merged anything we need to re-loop so we can
2356 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2360 * locked_ref is the head node, so we have to go one
2361 * node back for any delayed ref updates
2363 ref = select_delayed_ref(locked_ref);
2365 if (ref && ref->seq &&
2366 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2368 * there are still refs with lower seq numbers in the
2369 * process of being added. Don't run this ref yet.
2371 list_del_init(&locked_ref->cluster);
2372 btrfs_delayed_ref_unlock(locked_ref);
2374 delayed_refs->num_heads_ready++;
2375 spin_unlock(&delayed_refs->lock);
2377 spin_lock(&delayed_refs->lock);
2382 * record the must insert reserved flag before we
2383 * drop the spin lock.
2385 must_insert_reserved = locked_ref->must_insert_reserved;
2386 locked_ref->must_insert_reserved = 0;
2388 extent_op = locked_ref->extent_op;
2389 locked_ref->extent_op = NULL;
2392 /* All delayed refs have been processed, Go ahead
2393 * and send the head node to run_one_delayed_ref,
2394 * so that any accounting fixes can happen
2396 ref = &locked_ref->node;
2398 if (extent_op && must_insert_reserved) {
2399 btrfs_free_delayed_extent_op(extent_op);
2404 spin_unlock(&delayed_refs->lock);
2406 ret = run_delayed_extent_op(trans, root,
2408 btrfs_free_delayed_extent_op(extent_op);
2412 * Need to reset must_insert_reserved if
2413 * there was an error so the abort stuff
2414 * can cleanup the reserved space
2417 if (must_insert_reserved)
2418 locked_ref->must_insert_reserved = 1;
2419 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2420 spin_lock(&delayed_refs->lock);
2421 btrfs_delayed_ref_unlock(locked_ref);
2430 rb_erase(&ref->rb_node, &delayed_refs->root);
2431 delayed_refs->num_entries--;
2432 if (!btrfs_delayed_ref_is_head(ref)) {
2434 * when we play the delayed ref, also correct the
2437 switch (ref->action) {
2438 case BTRFS_ADD_DELAYED_REF:
2439 case BTRFS_ADD_DELAYED_EXTENT:
2440 locked_ref->node.ref_mod -= ref->ref_mod;
2442 case BTRFS_DROP_DELAYED_REF:
2443 locked_ref->node.ref_mod += ref->ref_mod;
2449 list_del_init(&locked_ref->cluster);
2451 spin_unlock(&delayed_refs->lock);
2453 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2454 must_insert_reserved);
2456 btrfs_free_delayed_extent_op(extent_op);
2458 btrfs_delayed_ref_unlock(locked_ref);
2459 btrfs_put_delayed_ref(ref);
2460 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2461 spin_lock(&delayed_refs->lock);
2466 * If this node is a head, that means all the refs in this head
2467 * have been dealt with, and we will pick the next head to deal
2468 * with, so we must unlock the head and drop it from the cluster
2469 * list before we release it.
2471 if (btrfs_delayed_ref_is_head(ref)) {
2472 btrfs_delayed_ref_unlock(locked_ref);
2475 btrfs_put_delayed_ref(ref);
2479 spin_lock(&delayed_refs->lock);
2484 #ifdef SCRAMBLE_DELAYED_REFS
2486 * Normally delayed refs get processed in ascending bytenr order. This
2487 * correlates in most cases to the order added. To expose dependencies on this
2488 * order, we start to process the tree in the middle instead of the beginning
2490 static u64 find_middle(struct rb_root *root)
2492 struct rb_node *n = root->rb_node;
2493 struct btrfs_delayed_ref_node *entry;
2496 u64 first = 0, last = 0;
2500 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501 first = entry->bytenr;
2505 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2506 last = entry->bytenr;
2511 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2512 WARN_ON(!entry->in_tree);
2514 middle = entry->bytenr;
2527 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2528 struct btrfs_fs_info *fs_info)
2530 struct qgroup_update *qgroup_update;
2533 if (list_empty(&trans->qgroup_ref_list) !=
2534 !trans->delayed_ref_elem.seq) {
2535 /* list without seq or seq without list */
2537 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2538 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2539 (u32)(trans->delayed_ref_elem.seq >> 32),
2540 (u32)trans->delayed_ref_elem.seq);
2544 if (!trans->delayed_ref_elem.seq)
2547 while (!list_empty(&trans->qgroup_ref_list)) {
2548 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2549 struct qgroup_update, list);
2550 list_del(&qgroup_update->list);
2552 ret = btrfs_qgroup_account_ref(
2553 trans, fs_info, qgroup_update->node,
2554 qgroup_update->extent_op);
2555 kfree(qgroup_update);
2558 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2563 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2566 int val = atomic_read(&delayed_refs->ref_seq);
2568 if (val < seq || val >= seq + count)
2573 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2577 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2578 sizeof(struct btrfs_extent_inline_ref));
2579 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2580 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2583 * We don't ever fill up leaves all the way so multiply by 2 just to be
2584 * closer to what we're really going to want to ouse.
2586 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2589 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2590 struct btrfs_root *root)
2592 struct btrfs_block_rsv *global_rsv;
2593 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2597 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2598 num_heads = heads_to_leaves(root, num_heads);
2600 num_bytes += (num_heads - 1) * root->leafsize;
2602 global_rsv = &root->fs_info->global_block_rsv;
2605 * If we can't allocate any more chunks lets make sure we have _lots_ of
2606 * wiggle room since running delayed refs can create more delayed refs.
2608 if (global_rsv->space_info->full)
2611 spin_lock(&global_rsv->lock);
2612 if (global_rsv->reserved <= num_bytes)
2614 spin_unlock(&global_rsv->lock);
2619 * this starts processing the delayed reference count updates and
2620 * extent insertions we have queued up so far. count can be
2621 * 0, which means to process everything in the tree at the start
2622 * of the run (but not newly added entries), or it can be some target
2623 * number you'd like to process.
2625 * Returns 0 on success or if called with an aborted transaction
2626 * Returns <0 on error and aborts the transaction
2628 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root, unsigned long count)
2631 struct rb_node *node;
2632 struct btrfs_delayed_ref_root *delayed_refs;
2633 struct btrfs_delayed_ref_node *ref;
2634 struct list_head cluster;
2637 int run_all = count == (unsigned long)-1;
2641 /* We'll clean this up in btrfs_cleanup_transaction */
2645 if (root == root->fs_info->extent_root)
2646 root = root->fs_info->tree_root;
2648 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2650 delayed_refs = &trans->transaction->delayed_refs;
2651 INIT_LIST_HEAD(&cluster);
2653 count = delayed_refs->num_entries * 2;
2657 if (!run_all && !run_most) {
2659 int seq = atomic_read(&delayed_refs->ref_seq);
2662 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2664 DEFINE_WAIT(__wait);
2665 if (delayed_refs->flushing ||
2666 !btrfs_should_throttle_delayed_refs(trans, root))
2669 prepare_to_wait(&delayed_refs->wait, &__wait,
2670 TASK_UNINTERRUPTIBLE);
2672 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2675 finish_wait(&delayed_refs->wait, &__wait);
2677 if (!refs_newer(delayed_refs, seq, 256))
2682 finish_wait(&delayed_refs->wait, &__wait);
2688 atomic_inc(&delayed_refs->procs_running_refs);
2693 spin_lock(&delayed_refs->lock);
2695 #ifdef SCRAMBLE_DELAYED_REFS
2696 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2700 if (!(run_all || run_most) &&
2701 !btrfs_should_throttle_delayed_refs(trans, root))
2705 * go find something we can process in the rbtree. We start at
2706 * the beginning of the tree, and then build a cluster
2707 * of refs to process starting at the first one we are able to
2710 delayed_start = delayed_refs->run_delayed_start;
2711 ret = btrfs_find_ref_cluster(trans, &cluster,
2712 delayed_refs->run_delayed_start);
2716 ret = run_clustered_refs(trans, root, &cluster);
2718 btrfs_release_ref_cluster(&cluster);
2719 spin_unlock(&delayed_refs->lock);
2720 btrfs_abort_transaction(trans, root, ret);
2721 atomic_dec(&delayed_refs->procs_running_refs);
2722 wake_up(&delayed_refs->wait);
2726 atomic_add(ret, &delayed_refs->ref_seq);
2728 count -= min_t(unsigned long, ret, count);
2733 if (delayed_start >= delayed_refs->run_delayed_start) {
2736 * btrfs_find_ref_cluster looped. let's do one
2737 * more cycle. if we don't run any delayed ref
2738 * during that cycle (because we can't because
2739 * all of them are blocked), bail out.
2744 * no runnable refs left, stop trying
2751 /* refs were run, let's reset staleness detection */
2757 if (!list_empty(&trans->new_bgs)) {
2758 spin_unlock(&delayed_refs->lock);
2759 btrfs_create_pending_block_groups(trans, root);
2760 spin_lock(&delayed_refs->lock);
2763 node = rb_first(&delayed_refs->root);
2766 count = (unsigned long)-1;
2769 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2771 if (btrfs_delayed_ref_is_head(ref)) {
2772 struct btrfs_delayed_ref_head *head;
2774 head = btrfs_delayed_node_to_head(ref);
2775 atomic_inc(&ref->refs);
2777 spin_unlock(&delayed_refs->lock);
2779 * Mutex was contended, block until it's
2780 * released and try again
2782 mutex_lock(&head->mutex);
2783 mutex_unlock(&head->mutex);
2785 btrfs_put_delayed_ref(ref);
2789 node = rb_next(node);
2791 spin_unlock(&delayed_refs->lock);
2792 schedule_timeout(1);
2796 atomic_dec(&delayed_refs->procs_running_refs);
2798 if (waitqueue_active(&delayed_refs->wait))
2799 wake_up(&delayed_refs->wait);
2801 spin_unlock(&delayed_refs->lock);
2802 assert_qgroups_uptodate(trans);
2806 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2807 struct btrfs_root *root,
2808 u64 bytenr, u64 num_bytes, u64 flags,
2809 int level, int is_data)
2811 struct btrfs_delayed_extent_op *extent_op;
2814 extent_op = btrfs_alloc_delayed_extent_op();
2818 extent_op->flags_to_set = flags;
2819 extent_op->update_flags = 1;
2820 extent_op->update_key = 0;
2821 extent_op->is_data = is_data ? 1 : 0;
2822 extent_op->level = level;
2824 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2825 num_bytes, extent_op);
2827 btrfs_free_delayed_extent_op(extent_op);
2831 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2832 struct btrfs_root *root,
2833 struct btrfs_path *path,
2834 u64 objectid, u64 offset, u64 bytenr)
2836 struct btrfs_delayed_ref_head *head;
2837 struct btrfs_delayed_ref_node *ref;
2838 struct btrfs_delayed_data_ref *data_ref;
2839 struct btrfs_delayed_ref_root *delayed_refs;
2840 struct rb_node *node;
2844 delayed_refs = &trans->transaction->delayed_refs;
2845 spin_lock(&delayed_refs->lock);
2846 head = btrfs_find_delayed_ref_head(trans, bytenr);
2850 if (!mutex_trylock(&head->mutex)) {
2851 atomic_inc(&head->node.refs);
2852 spin_unlock(&delayed_refs->lock);
2854 btrfs_release_path(path);
2857 * Mutex was contended, block until it's released and let
2860 mutex_lock(&head->mutex);
2861 mutex_unlock(&head->mutex);
2862 btrfs_put_delayed_ref(&head->node);
2866 node = rb_prev(&head->node.rb_node);
2870 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2872 if (ref->bytenr != bytenr)
2876 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2879 data_ref = btrfs_delayed_node_to_data_ref(ref);
2881 node = rb_prev(node);
2885 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2886 if (ref->bytenr == bytenr && ref->seq == seq)
2890 if (data_ref->root != root->root_key.objectid ||
2891 data_ref->objectid != objectid || data_ref->offset != offset)
2896 mutex_unlock(&head->mutex);
2898 spin_unlock(&delayed_refs->lock);
2902 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2903 struct btrfs_root *root,
2904 struct btrfs_path *path,
2905 u64 objectid, u64 offset, u64 bytenr)
2907 struct btrfs_root *extent_root = root->fs_info->extent_root;
2908 struct extent_buffer *leaf;
2909 struct btrfs_extent_data_ref *ref;
2910 struct btrfs_extent_inline_ref *iref;
2911 struct btrfs_extent_item *ei;
2912 struct btrfs_key key;
2916 key.objectid = bytenr;
2917 key.offset = (u64)-1;
2918 key.type = BTRFS_EXTENT_ITEM_KEY;
2920 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2923 BUG_ON(ret == 0); /* Corruption */
2926 if (path->slots[0] == 0)
2930 leaf = path->nodes[0];
2931 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2933 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2937 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2938 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2939 if (item_size < sizeof(*ei)) {
2940 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2944 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2946 if (item_size != sizeof(*ei) +
2947 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2950 if (btrfs_extent_generation(leaf, ei) <=
2951 btrfs_root_last_snapshot(&root->root_item))
2954 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2955 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2956 BTRFS_EXTENT_DATA_REF_KEY)
2959 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2960 if (btrfs_extent_refs(leaf, ei) !=
2961 btrfs_extent_data_ref_count(leaf, ref) ||
2962 btrfs_extent_data_ref_root(leaf, ref) !=
2963 root->root_key.objectid ||
2964 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2965 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2973 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2974 struct btrfs_root *root,
2975 u64 objectid, u64 offset, u64 bytenr)
2977 struct btrfs_path *path;
2981 path = btrfs_alloc_path();
2986 ret = check_committed_ref(trans, root, path, objectid,
2988 if (ret && ret != -ENOENT)
2991 ret2 = check_delayed_ref(trans, root, path, objectid,
2993 } while (ret2 == -EAGAIN);
2995 if (ret2 && ret2 != -ENOENT) {
3000 if (ret != -ENOENT || ret2 != -ENOENT)
3003 btrfs_free_path(path);
3004 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3009 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3010 struct btrfs_root *root,
3011 struct extent_buffer *buf,
3012 int full_backref, int inc, int for_cow)
3019 struct btrfs_key key;
3020 struct btrfs_file_extent_item *fi;
3024 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3025 u64, u64, u64, u64, u64, u64, int);
3027 ref_root = btrfs_header_owner(buf);
3028 nritems = btrfs_header_nritems(buf);
3029 level = btrfs_header_level(buf);
3031 if (!root->ref_cows && level == 0)
3035 process_func = btrfs_inc_extent_ref;
3037 process_func = btrfs_free_extent;
3040 parent = buf->start;
3044 for (i = 0; i < nritems; i++) {
3046 btrfs_item_key_to_cpu(buf, &key, i);
3047 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3049 fi = btrfs_item_ptr(buf, i,
3050 struct btrfs_file_extent_item);
3051 if (btrfs_file_extent_type(buf, fi) ==
3052 BTRFS_FILE_EXTENT_INLINE)
3054 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3058 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3059 key.offset -= btrfs_file_extent_offset(buf, fi);
3060 ret = process_func(trans, root, bytenr, num_bytes,
3061 parent, ref_root, key.objectid,
3062 key.offset, for_cow);
3066 bytenr = btrfs_node_blockptr(buf, i);
3067 num_bytes = btrfs_level_size(root, level - 1);
3068 ret = process_func(trans, root, bytenr, num_bytes,
3069 parent, ref_root, level - 1, 0,
3080 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3081 struct extent_buffer *buf, int full_backref, int for_cow)
3083 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3086 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3087 struct extent_buffer *buf, int full_backref, int for_cow)
3089 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3092 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3093 struct btrfs_root *root,
3094 struct btrfs_path *path,
3095 struct btrfs_block_group_cache *cache)
3098 struct btrfs_root *extent_root = root->fs_info->extent_root;
3100 struct extent_buffer *leaf;
3102 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3105 BUG_ON(ret); /* Corruption */
3107 leaf = path->nodes[0];
3108 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3109 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3110 btrfs_mark_buffer_dirty(leaf);
3111 btrfs_release_path(path);
3114 btrfs_abort_transaction(trans, root, ret);
3121 static struct btrfs_block_group_cache *
3122 next_block_group(struct btrfs_root *root,
3123 struct btrfs_block_group_cache *cache)
3125 struct rb_node *node;
3126 spin_lock(&root->fs_info->block_group_cache_lock);
3127 node = rb_next(&cache->cache_node);
3128 btrfs_put_block_group(cache);
3130 cache = rb_entry(node, struct btrfs_block_group_cache,
3132 btrfs_get_block_group(cache);
3135 spin_unlock(&root->fs_info->block_group_cache_lock);
3139 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3140 struct btrfs_trans_handle *trans,
3141 struct btrfs_path *path)
3143 struct btrfs_root *root = block_group->fs_info->tree_root;
3144 struct inode *inode = NULL;
3146 int dcs = BTRFS_DC_ERROR;
3152 * If this block group is smaller than 100 megs don't bother caching the
3155 if (block_group->key.offset < (100 * 1024 * 1024)) {
3156 spin_lock(&block_group->lock);
3157 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3158 spin_unlock(&block_group->lock);
3163 inode = lookup_free_space_inode(root, block_group, path);
3164 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3165 ret = PTR_ERR(inode);
3166 btrfs_release_path(path);
3170 if (IS_ERR(inode)) {
3174 if (block_group->ro)
3177 ret = create_free_space_inode(root, trans, block_group, path);
3183 /* We've already setup this transaction, go ahead and exit */
3184 if (block_group->cache_generation == trans->transid &&
3185 i_size_read(inode)) {
3186 dcs = BTRFS_DC_SETUP;
3191 * We want to set the generation to 0, that way if anything goes wrong
3192 * from here on out we know not to trust this cache when we load up next
3195 BTRFS_I(inode)->generation = 0;
3196 ret = btrfs_update_inode(trans, root, inode);
3199 if (i_size_read(inode) > 0) {
3200 ret = btrfs_check_trunc_cache_free_space(root,
3201 &root->fs_info->global_block_rsv);
3205 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3210 spin_lock(&block_group->lock);
3211 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3212 !btrfs_test_opt(root, SPACE_CACHE)) {
3214 * don't bother trying to write stuff out _if_
3215 * a) we're not cached,
3216 * b) we're with nospace_cache mount option.
3218 dcs = BTRFS_DC_WRITTEN;
3219 spin_unlock(&block_group->lock);
3222 spin_unlock(&block_group->lock);
3225 * Try to preallocate enough space based on how big the block group is.
3226 * Keep in mind this has to include any pinned space which could end up
3227 * taking up quite a bit since it's not folded into the other space
3230 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3235 num_pages *= PAGE_CACHE_SIZE;
3237 ret = btrfs_check_data_free_space(inode, num_pages);
3241 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3242 num_pages, num_pages,
3245 dcs = BTRFS_DC_SETUP;
3246 btrfs_free_reserved_data_space(inode, num_pages);
3251 btrfs_release_path(path);
3253 spin_lock(&block_group->lock);
3254 if (!ret && dcs == BTRFS_DC_SETUP)
3255 block_group->cache_generation = trans->transid;
3256 block_group->disk_cache_state = dcs;
3257 spin_unlock(&block_group->lock);
3262 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3263 struct btrfs_root *root)
3265 struct btrfs_block_group_cache *cache;
3267 struct btrfs_path *path;
3270 path = btrfs_alloc_path();
3276 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3278 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3280 cache = next_block_group(root, cache);
3288 err = cache_save_setup(cache, trans, path);
3289 last = cache->key.objectid + cache->key.offset;
3290 btrfs_put_block_group(cache);
3295 err = btrfs_run_delayed_refs(trans, root,
3297 if (err) /* File system offline */
3301 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3303 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3304 btrfs_put_block_group(cache);
3310 cache = next_block_group(root, cache);
3319 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3320 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3322 last = cache->key.objectid + cache->key.offset;
3324 err = write_one_cache_group(trans, root, path, cache);
3325 btrfs_put_block_group(cache);
3326 if (err) /* File system offline */
3332 * I don't think this is needed since we're just marking our
3333 * preallocated extent as written, but just in case it can't
3337 err = btrfs_run_delayed_refs(trans, root,
3339 if (err) /* File system offline */
3343 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3346 * Really this shouldn't happen, but it could if we
3347 * couldn't write the entire preallocated extent and
3348 * splitting the extent resulted in a new block.
3351 btrfs_put_block_group(cache);
3354 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3356 cache = next_block_group(root, cache);
3365 err = btrfs_write_out_cache(root, trans, cache, path);
3368 * If we didn't have an error then the cache state is still
3369 * NEED_WRITE, so we can set it to WRITTEN.
3371 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3372 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3373 last = cache->key.objectid + cache->key.offset;
3374 btrfs_put_block_group(cache);
3378 btrfs_free_path(path);
3382 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3384 struct btrfs_block_group_cache *block_group;
3387 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3388 if (!block_group || block_group->ro)
3391 btrfs_put_block_group(block_group);
3395 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3396 u64 total_bytes, u64 bytes_used,
3397 struct btrfs_space_info **space_info)
3399 struct btrfs_space_info *found;
3404 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3405 BTRFS_BLOCK_GROUP_RAID10))
3410 found = __find_space_info(info, flags);
3412 spin_lock(&found->lock);
3413 found->total_bytes += total_bytes;
3414 found->disk_total += total_bytes * factor;
3415 found->bytes_used += bytes_used;
3416 found->disk_used += bytes_used * factor;
3418 spin_unlock(&found->lock);
3419 *space_info = found;
3422 found = kzalloc(sizeof(*found), GFP_NOFS);
3426 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3432 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3433 INIT_LIST_HEAD(&found->block_groups[i]);
3434 init_rwsem(&found->groups_sem);
3435 spin_lock_init(&found->lock);
3436 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3437 found->total_bytes = total_bytes;
3438 found->disk_total = total_bytes * factor;
3439 found->bytes_used = bytes_used;
3440 found->disk_used = bytes_used * factor;
3441 found->bytes_pinned = 0;
3442 found->bytes_reserved = 0;
3443 found->bytes_readonly = 0;
3444 found->bytes_may_use = 0;
3446 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3447 found->chunk_alloc = 0;
3449 init_waitqueue_head(&found->wait);
3450 *space_info = found;
3451 list_add_rcu(&found->list, &info->space_info);
3452 if (flags & BTRFS_BLOCK_GROUP_DATA)
3453 info->data_sinfo = found;
3457 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3459 u64 extra_flags = chunk_to_extended(flags) &
3460 BTRFS_EXTENDED_PROFILE_MASK;
3462 write_seqlock(&fs_info->profiles_lock);
3463 if (flags & BTRFS_BLOCK_GROUP_DATA)
3464 fs_info->avail_data_alloc_bits |= extra_flags;
3465 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3466 fs_info->avail_metadata_alloc_bits |= extra_flags;
3467 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3468 fs_info->avail_system_alloc_bits |= extra_flags;
3469 write_sequnlock(&fs_info->profiles_lock);
3473 * returns target flags in extended format or 0 if restripe for this
3474 * chunk_type is not in progress
3476 * should be called with either volume_mutex or balance_lock held
3478 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3480 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3486 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3487 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3488 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3489 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3490 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3491 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3492 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3493 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3494 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3501 * @flags: available profiles in extended format (see ctree.h)
3503 * Returns reduced profile in chunk format. If profile changing is in
3504 * progress (either running or paused) picks the target profile (if it's
3505 * already available), otherwise falls back to plain reducing.
3507 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3510 * we add in the count of missing devices because we want
3511 * to make sure that any RAID levels on a degraded FS
3512 * continue to be honored.
3514 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3515 root->fs_info->fs_devices->missing_devices;
3520 * see if restripe for this chunk_type is in progress, if so
3521 * try to reduce to the target profile
3523 spin_lock(&root->fs_info->balance_lock);
3524 target = get_restripe_target(root->fs_info, flags);
3526 /* pick target profile only if it's already available */
3527 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3528 spin_unlock(&root->fs_info->balance_lock);
3529 return extended_to_chunk(target);
3532 spin_unlock(&root->fs_info->balance_lock);
3534 /* First, mask out the RAID levels which aren't possible */
3535 if (num_devices == 1)
3536 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3537 BTRFS_BLOCK_GROUP_RAID5);
3538 if (num_devices < 3)
3539 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3540 if (num_devices < 4)
3541 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3543 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3544 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3545 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3548 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3549 tmp = BTRFS_BLOCK_GROUP_RAID6;
3550 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3551 tmp = BTRFS_BLOCK_GROUP_RAID5;
3552 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3553 tmp = BTRFS_BLOCK_GROUP_RAID10;
3554 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3555 tmp = BTRFS_BLOCK_GROUP_RAID1;
3556 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3557 tmp = BTRFS_BLOCK_GROUP_RAID0;
3559 return extended_to_chunk(flags | tmp);
3562 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3567 seq = read_seqbegin(&root->fs_info->profiles_lock);
3569 if (flags & BTRFS_BLOCK_GROUP_DATA)
3570 flags |= root->fs_info->avail_data_alloc_bits;
3571 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3572 flags |= root->fs_info->avail_system_alloc_bits;
3573 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3574 flags |= root->fs_info->avail_metadata_alloc_bits;
3575 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3577 return btrfs_reduce_alloc_profile(root, flags);
3580 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3586 flags = BTRFS_BLOCK_GROUP_DATA;
3587 else if (root == root->fs_info->chunk_root)
3588 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3590 flags = BTRFS_BLOCK_GROUP_METADATA;
3592 ret = get_alloc_profile(root, flags);
3597 * This will check the space that the inode allocates from to make sure we have
3598 * enough space for bytes.
3600 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3602 struct btrfs_space_info *data_sinfo;
3603 struct btrfs_root *root = BTRFS_I(inode)->root;
3604 struct btrfs_fs_info *fs_info = root->fs_info;
3606 int ret = 0, committed = 0, alloc_chunk = 1;
3608 /* make sure bytes are sectorsize aligned */
3609 bytes = ALIGN(bytes, root->sectorsize);
3611 if (root == root->fs_info->tree_root ||
3612 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3617 data_sinfo = fs_info->data_sinfo;
3622 /* make sure we have enough space to handle the data first */
3623 spin_lock(&data_sinfo->lock);
3624 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3625 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3626 data_sinfo->bytes_may_use;
3628 if (used + bytes > data_sinfo->total_bytes) {
3629 struct btrfs_trans_handle *trans;
3632 * if we don't have enough free bytes in this space then we need
3633 * to alloc a new chunk.
3635 if (!data_sinfo->full && alloc_chunk) {
3638 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3639 spin_unlock(&data_sinfo->lock);
3641 alloc_target = btrfs_get_alloc_profile(root, 1);
3642 trans = btrfs_join_transaction(root);
3644 return PTR_ERR(trans);
3646 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3648 CHUNK_ALLOC_NO_FORCE);
3649 btrfs_end_transaction(trans, root);
3658 data_sinfo = fs_info->data_sinfo;
3664 * If we don't have enough pinned space to deal with this
3665 * allocation don't bother committing the transaction.
3667 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3670 spin_unlock(&data_sinfo->lock);
3672 /* commit the current transaction and try again */
3675 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3678 trans = btrfs_join_transaction(root);
3680 return PTR_ERR(trans);
3681 ret = btrfs_commit_transaction(trans, root);
3689 data_sinfo->bytes_may_use += bytes;
3690 trace_btrfs_space_reservation(root->fs_info, "space_info",
3691 data_sinfo->flags, bytes, 1);
3692 spin_unlock(&data_sinfo->lock);
3698 * Called if we need to clear a data reservation for this inode.
3700 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3702 struct btrfs_root *root = BTRFS_I(inode)->root;
3703 struct btrfs_space_info *data_sinfo;
3705 /* make sure bytes are sectorsize aligned */
3706 bytes = ALIGN(bytes, root->sectorsize);
3708 data_sinfo = root->fs_info->data_sinfo;
3709 spin_lock(&data_sinfo->lock);
3710 WARN_ON(data_sinfo->bytes_may_use < bytes);
3711 data_sinfo->bytes_may_use -= bytes;
3712 trace_btrfs_space_reservation(root->fs_info, "space_info",
3713 data_sinfo->flags, bytes, 0);
3714 spin_unlock(&data_sinfo->lock);
3717 static void force_metadata_allocation(struct btrfs_fs_info *info)
3719 struct list_head *head = &info->space_info;
3720 struct btrfs_space_info *found;
3723 list_for_each_entry_rcu(found, head, list) {
3724 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3725 found->force_alloc = CHUNK_ALLOC_FORCE;
3730 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3732 return (global->size << 1);
3735 static int should_alloc_chunk(struct btrfs_root *root,
3736 struct btrfs_space_info *sinfo, int force)
3738 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3739 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3740 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3743 if (force == CHUNK_ALLOC_FORCE)
3747 * We need to take into account the global rsv because for all intents
3748 * and purposes it's used space. Don't worry about locking the
3749 * global_rsv, it doesn't change except when the transaction commits.
3751 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3752 num_allocated += calc_global_rsv_need_space(global_rsv);
3755 * in limited mode, we want to have some free space up to
3756 * about 1% of the FS size.
3758 if (force == CHUNK_ALLOC_LIMITED) {
3759 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3760 thresh = max_t(u64, 64 * 1024 * 1024,
3761 div_factor_fine(thresh, 1));
3763 if (num_bytes - num_allocated < thresh)
3767 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3772 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3776 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3777 BTRFS_BLOCK_GROUP_RAID0 |
3778 BTRFS_BLOCK_GROUP_RAID5 |
3779 BTRFS_BLOCK_GROUP_RAID6))
3780 num_dev = root->fs_info->fs_devices->rw_devices;
3781 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3784 num_dev = 1; /* DUP or single */
3786 /* metadata for updaing devices and chunk tree */
3787 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3790 static void check_system_chunk(struct btrfs_trans_handle *trans,
3791 struct btrfs_root *root, u64 type)
3793 struct btrfs_space_info *info;
3797 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3798 spin_lock(&info->lock);
3799 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3800 info->bytes_reserved - info->bytes_readonly;
3801 spin_unlock(&info->lock);
3803 thresh = get_system_chunk_thresh(root, type);
3804 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3805 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3806 left, thresh, type);
3807 dump_space_info(info, 0, 0);
3810 if (left < thresh) {
3813 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3814 btrfs_alloc_chunk(trans, root, flags);
3818 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3819 struct btrfs_root *extent_root, u64 flags, int force)
3821 struct btrfs_space_info *space_info;
3822 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3823 int wait_for_alloc = 0;
3826 /* Don't re-enter if we're already allocating a chunk */
3827 if (trans->allocating_chunk)
3830 space_info = __find_space_info(extent_root->fs_info, flags);
3832 ret = update_space_info(extent_root->fs_info, flags,
3834 BUG_ON(ret); /* -ENOMEM */
3836 BUG_ON(!space_info); /* Logic error */
3839 spin_lock(&space_info->lock);
3840 if (force < space_info->force_alloc)
3841 force = space_info->force_alloc;
3842 if (space_info->full) {
3843 if (should_alloc_chunk(extent_root, space_info, force))
3847 spin_unlock(&space_info->lock);
3851 if (!should_alloc_chunk(extent_root, space_info, force)) {
3852 spin_unlock(&space_info->lock);
3854 } else if (space_info->chunk_alloc) {
3857 space_info->chunk_alloc = 1;
3860 spin_unlock(&space_info->lock);
3862 mutex_lock(&fs_info->chunk_mutex);
3865 * The chunk_mutex is held throughout the entirety of a chunk
3866 * allocation, so once we've acquired the chunk_mutex we know that the
3867 * other guy is done and we need to recheck and see if we should
3870 if (wait_for_alloc) {
3871 mutex_unlock(&fs_info->chunk_mutex);
3876 trans->allocating_chunk = true;
3879 * If we have mixed data/metadata chunks we want to make sure we keep
3880 * allocating mixed chunks instead of individual chunks.
3882 if (btrfs_mixed_space_info(space_info))
3883 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3886 * if we're doing a data chunk, go ahead and make sure that
3887 * we keep a reasonable number of metadata chunks allocated in the
3890 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3891 fs_info->data_chunk_allocations++;
3892 if (!(fs_info->data_chunk_allocations %
3893 fs_info->metadata_ratio))
3894 force_metadata_allocation(fs_info);
3898 * Check if we have enough space in SYSTEM chunk because we may need
3899 * to update devices.
3901 check_system_chunk(trans, extent_root, flags);
3903 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3904 trans->allocating_chunk = false;
3906 spin_lock(&space_info->lock);
3907 if (ret < 0 && ret != -ENOSPC)
3910 space_info->full = 1;
3914 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3916 space_info->chunk_alloc = 0;
3917 spin_unlock(&space_info->lock);
3918 mutex_unlock(&fs_info->chunk_mutex);
3922 static int can_overcommit(struct btrfs_root *root,
3923 struct btrfs_space_info *space_info, u64 bytes,
3924 enum btrfs_reserve_flush_enum flush)
3926 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3927 u64 profile = btrfs_get_alloc_profile(root, 0);
3932 used = space_info->bytes_used + space_info->bytes_reserved +
3933 space_info->bytes_pinned + space_info->bytes_readonly;
3936 * We only want to allow over committing if we have lots of actual space
3937 * free, but if we don't have enough space to handle the global reserve
3938 * space then we could end up having a real enospc problem when trying
3939 * to allocate a chunk or some other such important allocation.
3941 spin_lock(&global_rsv->lock);
3942 space_size = calc_global_rsv_need_space(global_rsv);
3943 spin_unlock(&global_rsv->lock);
3944 if (used + space_size >= space_info->total_bytes)
3947 used += space_info->bytes_may_use;
3949 spin_lock(&root->fs_info->free_chunk_lock);
3950 avail = root->fs_info->free_chunk_space;
3951 spin_unlock(&root->fs_info->free_chunk_lock);
3954 * If we have dup, raid1 or raid10 then only half of the free
3955 * space is actually useable. For raid56, the space info used
3956 * doesn't include the parity drive, so we don't have to
3959 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3960 BTRFS_BLOCK_GROUP_RAID1 |
3961 BTRFS_BLOCK_GROUP_RAID10))
3965 * If we aren't flushing all things, let us overcommit up to
3966 * 1/2th of the space. If we can flush, don't let us overcommit
3967 * too much, let it overcommit up to 1/8 of the space.
3969 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3974 if (used + bytes < space_info->total_bytes + avail)
3979 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3980 unsigned long nr_pages)
3982 struct super_block *sb = root->fs_info->sb;
3984 if (down_read_trylock(&sb->s_umount)) {
3985 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3986 up_read(&sb->s_umount);
3989 * We needn't worry the filesystem going from r/w to r/o though
3990 * we don't acquire ->s_umount mutex, because the filesystem
3991 * should guarantee the delalloc inodes list be empty after
3992 * the filesystem is readonly(all dirty pages are written to
3995 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
3996 if (!current->journal_info)
3997 btrfs_wait_all_ordered_extents(root->fs_info);
4002 * shrink metadata reservation for delalloc
4004 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4007 struct btrfs_block_rsv *block_rsv;
4008 struct btrfs_space_info *space_info;
4009 struct btrfs_trans_handle *trans;
4013 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
4015 enum btrfs_reserve_flush_enum flush;
4017 trans = (struct btrfs_trans_handle *)current->journal_info;
4018 block_rsv = &root->fs_info->delalloc_block_rsv;
4019 space_info = block_rsv->space_info;
4022 delalloc_bytes = percpu_counter_sum_positive(
4023 &root->fs_info->delalloc_bytes);
4024 if (delalloc_bytes == 0) {
4027 btrfs_wait_all_ordered_extents(root->fs_info);
4031 while (delalloc_bytes && loops < 3) {
4032 max_reclaim = min(delalloc_bytes, to_reclaim);
4033 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4034 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4036 * We need to wait for the async pages to actually start before
4039 wait_event(root->fs_info->async_submit_wait,
4040 !atomic_read(&root->fs_info->async_delalloc_pages));
4043 flush = BTRFS_RESERVE_FLUSH_ALL;
4045 flush = BTRFS_RESERVE_NO_FLUSH;
4046 spin_lock(&space_info->lock);
4047 if (can_overcommit(root, space_info, orig, flush)) {
4048 spin_unlock(&space_info->lock);
4051 spin_unlock(&space_info->lock);
4054 if (wait_ordered && !trans) {
4055 btrfs_wait_all_ordered_extents(root->fs_info);
4057 time_left = schedule_timeout_killable(1);
4062 delalloc_bytes = percpu_counter_sum_positive(
4063 &root->fs_info->delalloc_bytes);
4068 * maybe_commit_transaction - possibly commit the transaction if its ok to
4069 * @root - the root we're allocating for
4070 * @bytes - the number of bytes we want to reserve
4071 * @force - force the commit
4073 * This will check to make sure that committing the transaction will actually
4074 * get us somewhere and then commit the transaction if it does. Otherwise it
4075 * will return -ENOSPC.
4077 static int may_commit_transaction(struct btrfs_root *root,
4078 struct btrfs_space_info *space_info,
4079 u64 bytes, int force)
4081 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4082 struct btrfs_trans_handle *trans;
4084 trans = (struct btrfs_trans_handle *)current->journal_info;
4091 /* See if there is enough pinned space to make this reservation */
4092 spin_lock(&space_info->lock);
4093 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4095 spin_unlock(&space_info->lock);
4098 spin_unlock(&space_info->lock);
4101 * See if there is some space in the delayed insertion reservation for
4104 if (space_info != delayed_rsv->space_info)
4107 spin_lock(&space_info->lock);
4108 spin_lock(&delayed_rsv->lock);
4109 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4110 bytes - delayed_rsv->size) >= 0) {
4111 spin_unlock(&delayed_rsv->lock);
4112 spin_unlock(&space_info->lock);
4115 spin_unlock(&delayed_rsv->lock);
4116 spin_unlock(&space_info->lock);
4119 trans = btrfs_join_transaction(root);
4123 return btrfs_commit_transaction(trans, root);
4127 FLUSH_DELAYED_ITEMS_NR = 1,
4128 FLUSH_DELAYED_ITEMS = 2,
4130 FLUSH_DELALLOC_WAIT = 4,
4135 static int flush_space(struct btrfs_root *root,
4136 struct btrfs_space_info *space_info, u64 num_bytes,
4137 u64 orig_bytes, int state)
4139 struct btrfs_trans_handle *trans;
4144 case FLUSH_DELAYED_ITEMS_NR:
4145 case FLUSH_DELAYED_ITEMS:
4146 if (state == FLUSH_DELAYED_ITEMS_NR) {
4147 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4149 nr = (int)div64_u64(num_bytes, bytes);
4156 trans = btrfs_join_transaction(root);
4157 if (IS_ERR(trans)) {
4158 ret = PTR_ERR(trans);
4161 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4162 btrfs_end_transaction(trans, root);
4164 case FLUSH_DELALLOC:
4165 case FLUSH_DELALLOC_WAIT:
4166 shrink_delalloc(root, num_bytes, orig_bytes,
4167 state == FLUSH_DELALLOC_WAIT);
4170 trans = btrfs_join_transaction(root);
4171 if (IS_ERR(trans)) {
4172 ret = PTR_ERR(trans);
4175 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4176 btrfs_get_alloc_profile(root, 0),
4177 CHUNK_ALLOC_NO_FORCE);
4178 btrfs_end_transaction(trans, root);
4183 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4193 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4194 * @root - the root we're allocating for
4195 * @block_rsv - the block_rsv we're allocating for
4196 * @orig_bytes - the number of bytes we want
4197 * @flush - whether or not we can flush to make our reservation
4199 * This will reserve orgi_bytes number of bytes from the space info associated
4200 * with the block_rsv. If there is not enough space it will make an attempt to
4201 * flush out space to make room. It will do this by flushing delalloc if
4202 * possible or committing the transaction. If flush is 0 then no attempts to
4203 * regain reservations will be made and this will fail if there is not enough
4206 static int reserve_metadata_bytes(struct btrfs_root *root,
4207 struct btrfs_block_rsv *block_rsv,
4209 enum btrfs_reserve_flush_enum flush)
4211 struct btrfs_space_info *space_info = block_rsv->space_info;
4213 u64 num_bytes = orig_bytes;
4214 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4216 bool flushing = false;
4220 spin_lock(&space_info->lock);
4222 * We only want to wait if somebody other than us is flushing and we
4223 * are actually allowed to flush all things.
4225 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4226 space_info->flush) {
4227 spin_unlock(&space_info->lock);
4229 * If we have a trans handle we can't wait because the flusher
4230 * may have to commit the transaction, which would mean we would
4231 * deadlock since we are waiting for the flusher to finish, but
4232 * hold the current transaction open.
4234 if (current->journal_info)
4236 ret = wait_event_killable(space_info->wait, !space_info->flush);
4237 /* Must have been killed, return */
4241 spin_lock(&space_info->lock);
4245 used = space_info->bytes_used + space_info->bytes_reserved +
4246 space_info->bytes_pinned + space_info->bytes_readonly +
4247 space_info->bytes_may_use;
4250 * The idea here is that we've not already over-reserved the block group
4251 * then we can go ahead and save our reservation first and then start
4252 * flushing if we need to. Otherwise if we've already overcommitted
4253 * lets start flushing stuff first and then come back and try to make
4256 if (used <= space_info->total_bytes) {
4257 if (used + orig_bytes <= space_info->total_bytes) {
4258 space_info->bytes_may_use += orig_bytes;
4259 trace_btrfs_space_reservation(root->fs_info,
4260 "space_info", space_info->flags, orig_bytes, 1);
4264 * Ok set num_bytes to orig_bytes since we aren't
4265 * overocmmitted, this way we only try and reclaim what
4268 num_bytes = orig_bytes;
4272 * Ok we're over committed, set num_bytes to the overcommitted
4273 * amount plus the amount of bytes that we need for this
4276 num_bytes = used - space_info->total_bytes +
4280 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4281 space_info->bytes_may_use += orig_bytes;
4282 trace_btrfs_space_reservation(root->fs_info, "space_info",
4283 space_info->flags, orig_bytes,
4289 * Couldn't make our reservation, save our place so while we're trying
4290 * to reclaim space we can actually use it instead of somebody else
4291 * stealing it from us.
4293 * We make the other tasks wait for the flush only when we can flush
4296 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4298 space_info->flush = 1;
4301 spin_unlock(&space_info->lock);
4303 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4306 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4311 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4312 * would happen. So skip delalloc flush.
4314 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4315 (flush_state == FLUSH_DELALLOC ||
4316 flush_state == FLUSH_DELALLOC_WAIT))
4317 flush_state = ALLOC_CHUNK;
4321 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4322 flush_state < COMMIT_TRANS)
4324 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4325 flush_state <= COMMIT_TRANS)
4329 if (ret == -ENOSPC &&
4330 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4331 struct btrfs_block_rsv *global_rsv =
4332 &root->fs_info->global_block_rsv;
4334 if (block_rsv != global_rsv &&
4335 !block_rsv_use_bytes(global_rsv, orig_bytes))
4339 spin_lock(&space_info->lock);
4340 space_info->flush = 0;
4341 wake_up_all(&space_info->wait);
4342 spin_unlock(&space_info->lock);
4347 static struct btrfs_block_rsv *get_block_rsv(
4348 const struct btrfs_trans_handle *trans,
4349 const struct btrfs_root *root)
4351 struct btrfs_block_rsv *block_rsv = NULL;
4354 block_rsv = trans->block_rsv;
4356 if (root == root->fs_info->csum_root && trans->adding_csums)
4357 block_rsv = trans->block_rsv;
4359 if (root == root->fs_info->uuid_root)
4360 block_rsv = trans->block_rsv;
4363 block_rsv = root->block_rsv;
4366 block_rsv = &root->fs_info->empty_block_rsv;
4371 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4375 spin_lock(&block_rsv->lock);
4376 if (block_rsv->reserved >= num_bytes) {
4377 block_rsv->reserved -= num_bytes;
4378 if (block_rsv->reserved < block_rsv->size)
4379 block_rsv->full = 0;
4382 spin_unlock(&block_rsv->lock);
4386 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4387 u64 num_bytes, int update_size)
4389 spin_lock(&block_rsv->lock);
4390 block_rsv->reserved += num_bytes;
4392 block_rsv->size += num_bytes;
4393 else if (block_rsv->reserved >= block_rsv->size)
4394 block_rsv->full = 1;
4395 spin_unlock(&block_rsv->lock);
4398 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4399 struct btrfs_block_rsv *dest, u64 num_bytes,
4402 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4405 if (global_rsv->space_info != dest->space_info)
4408 spin_lock(&global_rsv->lock);
4409 min_bytes = div_factor(global_rsv->size, min_factor);
4410 if (global_rsv->reserved < min_bytes + num_bytes) {
4411 spin_unlock(&global_rsv->lock);
4414 global_rsv->reserved -= num_bytes;
4415 if (global_rsv->reserved < global_rsv->size)
4416 global_rsv->full = 0;
4417 spin_unlock(&global_rsv->lock);
4419 block_rsv_add_bytes(dest, num_bytes, 1);
4423 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4424 struct btrfs_block_rsv *block_rsv,
4425 struct btrfs_block_rsv *dest, u64 num_bytes)
4427 struct btrfs_space_info *space_info = block_rsv->space_info;
4429 spin_lock(&block_rsv->lock);
4430 if (num_bytes == (u64)-1)
4431 num_bytes = block_rsv->size;
4432 block_rsv->size -= num_bytes;
4433 if (block_rsv->reserved >= block_rsv->size) {
4434 num_bytes = block_rsv->reserved - block_rsv->size;
4435 block_rsv->reserved = block_rsv->size;
4436 block_rsv->full = 1;
4440 spin_unlock(&block_rsv->lock);
4442 if (num_bytes > 0) {
4444 spin_lock(&dest->lock);
4448 bytes_to_add = dest->size - dest->reserved;
4449 bytes_to_add = min(num_bytes, bytes_to_add);
4450 dest->reserved += bytes_to_add;
4451 if (dest->reserved >= dest->size)
4453 num_bytes -= bytes_to_add;
4455 spin_unlock(&dest->lock);
4458 spin_lock(&space_info->lock);
4459 space_info->bytes_may_use -= num_bytes;
4460 trace_btrfs_space_reservation(fs_info, "space_info",
4461 space_info->flags, num_bytes, 0);
4462 spin_unlock(&space_info->lock);
4467 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4468 struct btrfs_block_rsv *dst, u64 num_bytes)
4472 ret = block_rsv_use_bytes(src, num_bytes);
4476 block_rsv_add_bytes(dst, num_bytes, 1);
4480 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4482 memset(rsv, 0, sizeof(*rsv));
4483 spin_lock_init(&rsv->lock);
4487 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4488 unsigned short type)
4490 struct btrfs_block_rsv *block_rsv;
4491 struct btrfs_fs_info *fs_info = root->fs_info;
4493 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4497 btrfs_init_block_rsv(block_rsv, type);
4498 block_rsv->space_info = __find_space_info(fs_info,
4499 BTRFS_BLOCK_GROUP_METADATA);
4503 void btrfs_free_block_rsv(struct btrfs_root *root,
4504 struct btrfs_block_rsv *rsv)
4508 btrfs_block_rsv_release(root, rsv, (u64)-1);
4512 int btrfs_block_rsv_add(struct btrfs_root *root,
4513 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4514 enum btrfs_reserve_flush_enum flush)
4521 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4523 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4530 int btrfs_block_rsv_check(struct btrfs_root *root,
4531 struct btrfs_block_rsv *block_rsv, int min_factor)
4539 spin_lock(&block_rsv->lock);
4540 num_bytes = div_factor(block_rsv->size, min_factor);
4541 if (block_rsv->reserved >= num_bytes)
4543 spin_unlock(&block_rsv->lock);
4548 int btrfs_block_rsv_refill(struct btrfs_root *root,
4549 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4550 enum btrfs_reserve_flush_enum flush)
4558 spin_lock(&block_rsv->lock);
4559 num_bytes = min_reserved;
4560 if (block_rsv->reserved >= num_bytes)
4563 num_bytes -= block_rsv->reserved;
4564 spin_unlock(&block_rsv->lock);
4569 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4571 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4578 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4579 struct btrfs_block_rsv *dst_rsv,
4582 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4585 void btrfs_block_rsv_release(struct btrfs_root *root,
4586 struct btrfs_block_rsv *block_rsv,
4589 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4590 if (global_rsv->full || global_rsv == block_rsv ||
4591 block_rsv->space_info != global_rsv->space_info)
4593 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4598 * helper to calculate size of global block reservation.
4599 * the desired value is sum of space used by extent tree,
4600 * checksum tree and root tree
4602 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4604 struct btrfs_space_info *sinfo;
4608 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4610 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4611 spin_lock(&sinfo->lock);
4612 data_used = sinfo->bytes_used;
4613 spin_unlock(&sinfo->lock);
4615 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4616 spin_lock(&sinfo->lock);
4617 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4619 meta_used = sinfo->bytes_used;
4620 spin_unlock(&sinfo->lock);
4622 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4624 num_bytes += div64_u64(data_used + meta_used, 50);
4626 if (num_bytes * 3 > meta_used)
4627 num_bytes = div64_u64(meta_used, 3);
4629 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4632 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4634 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4635 struct btrfs_space_info *sinfo = block_rsv->space_info;
4638 num_bytes = calc_global_metadata_size(fs_info);
4640 spin_lock(&sinfo->lock);
4641 spin_lock(&block_rsv->lock);
4643 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4645 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4646 sinfo->bytes_reserved + sinfo->bytes_readonly +
4647 sinfo->bytes_may_use;
4649 if (sinfo->total_bytes > num_bytes) {
4650 num_bytes = sinfo->total_bytes - num_bytes;
4651 block_rsv->reserved += num_bytes;
4652 sinfo->bytes_may_use += num_bytes;
4653 trace_btrfs_space_reservation(fs_info, "space_info",
4654 sinfo->flags, num_bytes, 1);
4657 if (block_rsv->reserved >= block_rsv->size) {
4658 num_bytes = block_rsv->reserved - block_rsv->size;
4659 sinfo->bytes_may_use -= num_bytes;
4660 trace_btrfs_space_reservation(fs_info, "space_info",
4661 sinfo->flags, num_bytes, 0);
4662 block_rsv->reserved = block_rsv->size;
4663 block_rsv->full = 1;
4666 spin_unlock(&block_rsv->lock);
4667 spin_unlock(&sinfo->lock);
4670 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4672 struct btrfs_space_info *space_info;
4674 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4675 fs_info->chunk_block_rsv.space_info = space_info;
4677 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4678 fs_info->global_block_rsv.space_info = space_info;
4679 fs_info->delalloc_block_rsv.space_info = space_info;
4680 fs_info->trans_block_rsv.space_info = space_info;
4681 fs_info->empty_block_rsv.space_info = space_info;
4682 fs_info->delayed_block_rsv.space_info = space_info;
4684 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4685 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4686 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4687 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4688 if (fs_info->quota_root)
4689 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4690 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4692 update_global_block_rsv(fs_info);
4695 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4697 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4699 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4700 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4701 WARN_ON(fs_info->trans_block_rsv.size > 0);
4702 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4703 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4704 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4705 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4706 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4709 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4710 struct btrfs_root *root)
4712 if (!trans->block_rsv)
4715 if (!trans->bytes_reserved)
4718 trace_btrfs_space_reservation(root->fs_info, "transaction",
4719 trans->transid, trans->bytes_reserved, 0);
4720 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4721 trans->bytes_reserved = 0;
4724 /* Can only return 0 or -ENOSPC */
4725 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4726 struct inode *inode)
4728 struct btrfs_root *root = BTRFS_I(inode)->root;
4729 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4730 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4733 * We need to hold space in order to delete our orphan item once we've
4734 * added it, so this takes the reservation so we can release it later
4735 * when we are truly done with the orphan item.
4737 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4738 trace_btrfs_space_reservation(root->fs_info, "orphan",
4739 btrfs_ino(inode), num_bytes, 1);
4740 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4743 void btrfs_orphan_release_metadata(struct inode *inode)
4745 struct btrfs_root *root = BTRFS_I(inode)->root;
4746 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4747 trace_btrfs_space_reservation(root->fs_info, "orphan",
4748 btrfs_ino(inode), num_bytes, 0);
4749 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4753 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4754 * root: the root of the parent directory
4755 * rsv: block reservation
4756 * items: the number of items that we need do reservation
4757 * qgroup_reserved: used to return the reserved size in qgroup
4759 * This function is used to reserve the space for snapshot/subvolume
4760 * creation and deletion. Those operations are different with the
4761 * common file/directory operations, they change two fs/file trees
4762 * and root tree, the number of items that the qgroup reserves is
4763 * different with the free space reservation. So we can not use
4764 * the space reseravtion mechanism in start_transaction().
4766 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4767 struct btrfs_block_rsv *rsv,
4769 u64 *qgroup_reserved,
4770 bool use_global_rsv)
4774 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4776 if (root->fs_info->quota_enabled) {
4777 /* One for parent inode, two for dir entries */
4778 num_bytes = 3 * root->leafsize;
4779 ret = btrfs_qgroup_reserve(root, num_bytes);
4786 *qgroup_reserved = num_bytes;
4788 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4789 rsv->space_info = __find_space_info(root->fs_info,
4790 BTRFS_BLOCK_GROUP_METADATA);
4791 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4792 BTRFS_RESERVE_FLUSH_ALL);
4794 if (ret == -ENOSPC && use_global_rsv)
4795 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4798 if (*qgroup_reserved)
4799 btrfs_qgroup_free(root, *qgroup_reserved);
4805 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4806 struct btrfs_block_rsv *rsv,
4807 u64 qgroup_reserved)
4809 btrfs_block_rsv_release(root, rsv, (u64)-1);
4810 if (qgroup_reserved)
4811 btrfs_qgroup_free(root, qgroup_reserved);
4815 * drop_outstanding_extent - drop an outstanding extent
4816 * @inode: the inode we're dropping the extent for
4818 * This is called when we are freeing up an outstanding extent, either called
4819 * after an error or after an extent is written. This will return the number of
4820 * reserved extents that need to be freed. This must be called with
4821 * BTRFS_I(inode)->lock held.
4823 static unsigned drop_outstanding_extent(struct inode *inode)
4825 unsigned drop_inode_space = 0;
4826 unsigned dropped_extents = 0;
4828 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4829 BTRFS_I(inode)->outstanding_extents--;
4831 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4832 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4833 &BTRFS_I(inode)->runtime_flags))
4834 drop_inode_space = 1;
4837 * If we have more or the same amount of outsanding extents than we have
4838 * reserved then we need to leave the reserved extents count alone.
4840 if (BTRFS_I(inode)->outstanding_extents >=
4841 BTRFS_I(inode)->reserved_extents)
4842 return drop_inode_space;
4844 dropped_extents = BTRFS_I(inode)->reserved_extents -
4845 BTRFS_I(inode)->outstanding_extents;
4846 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4847 return dropped_extents + drop_inode_space;
4851 * calc_csum_metadata_size - return the amount of metada space that must be
4852 * reserved/free'd for the given bytes.
4853 * @inode: the inode we're manipulating
4854 * @num_bytes: the number of bytes in question
4855 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4857 * This adjusts the number of csum_bytes in the inode and then returns the
4858 * correct amount of metadata that must either be reserved or freed. We
4859 * calculate how many checksums we can fit into one leaf and then divide the
4860 * number of bytes that will need to be checksumed by this value to figure out
4861 * how many checksums will be required. If we are adding bytes then the number
4862 * may go up and we will return the number of additional bytes that must be
4863 * reserved. If it is going down we will return the number of bytes that must
4866 * This must be called with BTRFS_I(inode)->lock held.
4868 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4871 struct btrfs_root *root = BTRFS_I(inode)->root;
4873 int num_csums_per_leaf;
4877 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4878 BTRFS_I(inode)->csum_bytes == 0)
4881 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4883 BTRFS_I(inode)->csum_bytes += num_bytes;
4885 BTRFS_I(inode)->csum_bytes -= num_bytes;
4886 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4887 num_csums_per_leaf = (int)div64_u64(csum_size,
4888 sizeof(struct btrfs_csum_item) +
4889 sizeof(struct btrfs_disk_key));
4890 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4891 num_csums = num_csums + num_csums_per_leaf - 1;
4892 num_csums = num_csums / num_csums_per_leaf;
4894 old_csums = old_csums + num_csums_per_leaf - 1;
4895 old_csums = old_csums / num_csums_per_leaf;
4897 /* No change, no need to reserve more */
4898 if (old_csums == num_csums)
4902 return btrfs_calc_trans_metadata_size(root,
4903 num_csums - old_csums);
4905 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4908 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4910 struct btrfs_root *root = BTRFS_I(inode)->root;
4911 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4914 unsigned nr_extents = 0;
4915 int extra_reserve = 0;
4916 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4918 bool delalloc_lock = true;
4922 /* If we are a free space inode we need to not flush since we will be in
4923 * the middle of a transaction commit. We also don't need the delalloc
4924 * mutex since we won't race with anybody. We need this mostly to make
4925 * lockdep shut its filthy mouth.
4927 if (btrfs_is_free_space_inode(inode)) {
4928 flush = BTRFS_RESERVE_NO_FLUSH;
4929 delalloc_lock = false;
4932 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4933 btrfs_transaction_in_commit(root->fs_info))
4934 schedule_timeout(1);
4937 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4939 num_bytes = ALIGN(num_bytes, root->sectorsize);
4941 spin_lock(&BTRFS_I(inode)->lock);
4942 BTRFS_I(inode)->outstanding_extents++;
4944 if (BTRFS_I(inode)->outstanding_extents >
4945 BTRFS_I(inode)->reserved_extents)
4946 nr_extents = BTRFS_I(inode)->outstanding_extents -
4947 BTRFS_I(inode)->reserved_extents;
4950 * Add an item to reserve for updating the inode when we complete the
4953 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4954 &BTRFS_I(inode)->runtime_flags)) {
4959 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4960 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4961 csum_bytes = BTRFS_I(inode)->csum_bytes;
4962 spin_unlock(&BTRFS_I(inode)->lock);
4964 if (root->fs_info->quota_enabled) {
4965 ret = btrfs_qgroup_reserve(root, num_bytes +
4966 nr_extents * root->leafsize);
4971 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4972 if (unlikely(ret)) {
4973 if (root->fs_info->quota_enabled)
4974 btrfs_qgroup_free(root, num_bytes +
4975 nr_extents * root->leafsize);
4979 spin_lock(&BTRFS_I(inode)->lock);
4980 if (extra_reserve) {
4981 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4982 &BTRFS_I(inode)->runtime_flags);
4985 BTRFS_I(inode)->reserved_extents += nr_extents;
4986 spin_unlock(&BTRFS_I(inode)->lock);
4989 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4992 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4993 btrfs_ino(inode), to_reserve, 1);
4994 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4999 spin_lock(&BTRFS_I(inode)->lock);
5000 dropped = drop_outstanding_extent(inode);
5002 * If the inodes csum_bytes is the same as the original
5003 * csum_bytes then we know we haven't raced with any free()ers
5004 * so we can just reduce our inodes csum bytes and carry on.
5006 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5007 calc_csum_metadata_size(inode, num_bytes, 0);
5009 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5013 * This is tricky, but first we need to figure out how much we
5014 * free'd from any free-ers that occured during this
5015 * reservation, so we reset ->csum_bytes to the csum_bytes
5016 * before we dropped our lock, and then call the free for the
5017 * number of bytes that were freed while we were trying our
5020 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5021 BTRFS_I(inode)->csum_bytes = csum_bytes;
5022 to_free = calc_csum_metadata_size(inode, bytes, 0);
5026 * Now we need to see how much we would have freed had we not
5027 * been making this reservation and our ->csum_bytes were not
5028 * artificially inflated.
5030 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5031 bytes = csum_bytes - orig_csum_bytes;
5032 bytes = calc_csum_metadata_size(inode, bytes, 0);
5035 * Now reset ->csum_bytes to what it should be. If bytes is
5036 * more than to_free then we would have free'd more space had we
5037 * not had an artificially high ->csum_bytes, so we need to free
5038 * the remainder. If bytes is the same or less then we don't
5039 * need to do anything, the other free-ers did the correct
5042 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5043 if (bytes > to_free)
5044 to_free = bytes - to_free;
5048 spin_unlock(&BTRFS_I(inode)->lock);
5050 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5053 btrfs_block_rsv_release(root, block_rsv, to_free);
5054 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5055 btrfs_ino(inode), to_free, 0);
5058 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5063 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5064 * @inode: the inode to release the reservation for
5065 * @num_bytes: the number of bytes we're releasing
5067 * This will release the metadata reservation for an inode. This can be called
5068 * once we complete IO for a given set of bytes to release their metadata
5071 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5073 struct btrfs_root *root = BTRFS_I(inode)->root;
5077 num_bytes = ALIGN(num_bytes, root->sectorsize);
5078 spin_lock(&BTRFS_I(inode)->lock);
5079 dropped = drop_outstanding_extent(inode);
5082 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5083 spin_unlock(&BTRFS_I(inode)->lock);
5085 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5087 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5088 btrfs_ino(inode), to_free, 0);
5089 if (root->fs_info->quota_enabled) {
5090 btrfs_qgroup_free(root, num_bytes +
5091 dropped * root->leafsize);
5094 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5099 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5100 * @inode: inode we're writing to
5101 * @num_bytes: the number of bytes we want to allocate
5103 * This will do the following things
5105 * o reserve space in the data space info for num_bytes
5106 * o reserve space in the metadata space info based on number of outstanding
5107 * extents and how much csums will be needed
5108 * o add to the inodes ->delalloc_bytes
5109 * o add it to the fs_info's delalloc inodes list.
5111 * This will return 0 for success and -ENOSPC if there is no space left.
5113 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5117 ret = btrfs_check_data_free_space(inode, num_bytes);
5121 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5123 btrfs_free_reserved_data_space(inode, num_bytes);
5131 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5132 * @inode: inode we're releasing space for
5133 * @num_bytes: the number of bytes we want to free up
5135 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5136 * called in the case that we don't need the metadata AND data reservations
5137 * anymore. So if there is an error or we insert an inline extent.
5139 * This function will release the metadata space that was not used and will
5140 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5141 * list if there are no delalloc bytes left.
5143 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5145 btrfs_delalloc_release_metadata(inode, num_bytes);
5146 btrfs_free_reserved_data_space(inode, num_bytes);
5149 static int update_block_group(struct btrfs_root *root,
5150 u64 bytenr, u64 num_bytes, int alloc)
5152 struct btrfs_block_group_cache *cache = NULL;
5153 struct btrfs_fs_info *info = root->fs_info;
5154 u64 total = num_bytes;
5159 /* block accounting for super block */
5160 spin_lock(&info->delalloc_root_lock);
5161 old_val = btrfs_super_bytes_used(info->super_copy);
5163 old_val += num_bytes;
5165 old_val -= num_bytes;
5166 btrfs_set_super_bytes_used(info->super_copy, old_val);
5167 spin_unlock(&info->delalloc_root_lock);
5170 cache = btrfs_lookup_block_group(info, bytenr);
5173 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5174 BTRFS_BLOCK_GROUP_RAID1 |
5175 BTRFS_BLOCK_GROUP_RAID10))
5180 * If this block group has free space cache written out, we
5181 * need to make sure to load it if we are removing space. This
5182 * is because we need the unpinning stage to actually add the
5183 * space back to the block group, otherwise we will leak space.
5185 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5186 cache_block_group(cache, 1);
5188 byte_in_group = bytenr - cache->key.objectid;
5189 WARN_ON(byte_in_group > cache->key.offset);
5191 spin_lock(&cache->space_info->lock);
5192 spin_lock(&cache->lock);
5194 if (btrfs_test_opt(root, SPACE_CACHE) &&
5195 cache->disk_cache_state < BTRFS_DC_CLEAR)
5196 cache->disk_cache_state = BTRFS_DC_CLEAR;
5199 old_val = btrfs_block_group_used(&cache->item);
5200 num_bytes = min(total, cache->key.offset - byte_in_group);
5202 old_val += num_bytes;
5203 btrfs_set_block_group_used(&cache->item, old_val);
5204 cache->reserved -= num_bytes;
5205 cache->space_info->bytes_reserved -= num_bytes;
5206 cache->space_info->bytes_used += num_bytes;
5207 cache->space_info->disk_used += num_bytes * factor;
5208 spin_unlock(&cache->lock);
5209 spin_unlock(&cache->space_info->lock);
5211 old_val -= num_bytes;
5212 btrfs_set_block_group_used(&cache->item, old_val);
5213 cache->pinned += num_bytes;
5214 cache->space_info->bytes_pinned += num_bytes;
5215 cache->space_info->bytes_used -= num_bytes;
5216 cache->space_info->disk_used -= num_bytes * factor;
5217 spin_unlock(&cache->lock);
5218 spin_unlock(&cache->space_info->lock);
5220 set_extent_dirty(info->pinned_extents,
5221 bytenr, bytenr + num_bytes - 1,
5222 GFP_NOFS | __GFP_NOFAIL);
5224 btrfs_put_block_group(cache);
5226 bytenr += num_bytes;
5231 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5233 struct btrfs_block_group_cache *cache;
5236 spin_lock(&root->fs_info->block_group_cache_lock);
5237 bytenr = root->fs_info->first_logical_byte;
5238 spin_unlock(&root->fs_info->block_group_cache_lock);
5240 if (bytenr < (u64)-1)
5243 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5247 bytenr = cache->key.objectid;
5248 btrfs_put_block_group(cache);
5253 static int pin_down_extent(struct btrfs_root *root,
5254 struct btrfs_block_group_cache *cache,
5255 u64 bytenr, u64 num_bytes, int reserved)
5257 spin_lock(&cache->space_info->lock);
5258 spin_lock(&cache->lock);
5259 cache->pinned += num_bytes;
5260 cache->space_info->bytes_pinned += num_bytes;
5262 cache->reserved -= num_bytes;
5263 cache->space_info->bytes_reserved -= num_bytes;
5265 spin_unlock(&cache->lock);
5266 spin_unlock(&cache->space_info->lock);
5268 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5269 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5271 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5276 * this function must be called within transaction
5278 int btrfs_pin_extent(struct btrfs_root *root,
5279 u64 bytenr, u64 num_bytes, int reserved)
5281 struct btrfs_block_group_cache *cache;
5283 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5284 BUG_ON(!cache); /* Logic error */
5286 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5288 btrfs_put_block_group(cache);
5293 * this function must be called within transaction
5295 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5296 u64 bytenr, u64 num_bytes)
5298 struct btrfs_block_group_cache *cache;
5301 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5306 * pull in the free space cache (if any) so that our pin
5307 * removes the free space from the cache. We have load_only set
5308 * to one because the slow code to read in the free extents does check
5309 * the pinned extents.
5311 cache_block_group(cache, 1);
5313 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5315 /* remove us from the free space cache (if we're there at all) */
5316 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5317 btrfs_put_block_group(cache);
5321 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5324 struct btrfs_block_group_cache *block_group;
5325 struct btrfs_caching_control *caching_ctl;
5327 block_group = btrfs_lookup_block_group(root->fs_info, start);
5331 cache_block_group(block_group, 0);
5332 caching_ctl = get_caching_control(block_group);
5336 BUG_ON(!block_group_cache_done(block_group));
5337 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5339 mutex_lock(&caching_ctl->mutex);
5341 if (start >= caching_ctl->progress) {
5342 ret = add_excluded_extent(root, start, num_bytes);
5343 } else if (start + num_bytes <= caching_ctl->progress) {
5344 ret = btrfs_remove_free_space(block_group,
5347 num_bytes = caching_ctl->progress - start;
5348 ret = btrfs_remove_free_space(block_group,
5353 num_bytes = (start + num_bytes) -
5354 caching_ctl->progress;
5355 start = caching_ctl->progress;
5356 ret = add_excluded_extent(root, start, num_bytes);
5359 mutex_unlock(&caching_ctl->mutex);
5360 put_caching_control(caching_ctl);
5362 btrfs_put_block_group(block_group);
5366 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5367 struct extent_buffer *eb)
5369 struct btrfs_file_extent_item *item;
5370 struct btrfs_key key;
5374 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5377 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5378 btrfs_item_key_to_cpu(eb, &key, i);
5379 if (key.type != BTRFS_EXTENT_DATA_KEY)
5381 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5382 found_type = btrfs_file_extent_type(eb, item);
5383 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5385 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5387 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5388 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5389 __exclude_logged_extent(log, key.objectid, key.offset);
5396 * btrfs_update_reserved_bytes - update the block_group and space info counters
5397 * @cache: The cache we are manipulating
5398 * @num_bytes: The number of bytes in question
5399 * @reserve: One of the reservation enums
5401 * This is called by the allocator when it reserves space, or by somebody who is
5402 * freeing space that was never actually used on disk. For example if you
5403 * reserve some space for a new leaf in transaction A and before transaction A
5404 * commits you free that leaf, you call this with reserve set to 0 in order to
5405 * clear the reservation.
5407 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5408 * ENOSPC accounting. For data we handle the reservation through clearing the
5409 * delalloc bits in the io_tree. We have to do this since we could end up
5410 * allocating less disk space for the amount of data we have reserved in the
5411 * case of compression.
5413 * If this is a reservation and the block group has become read only we cannot
5414 * make the reservation and return -EAGAIN, otherwise this function always
5417 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5418 u64 num_bytes, int reserve)
5420 struct btrfs_space_info *space_info = cache->space_info;
5423 spin_lock(&space_info->lock);
5424 spin_lock(&cache->lock);
5425 if (reserve != RESERVE_FREE) {
5429 cache->reserved += num_bytes;
5430 space_info->bytes_reserved += num_bytes;
5431 if (reserve == RESERVE_ALLOC) {
5432 trace_btrfs_space_reservation(cache->fs_info,
5433 "space_info", space_info->flags,
5435 space_info->bytes_may_use -= num_bytes;
5440 space_info->bytes_readonly += num_bytes;
5441 cache->reserved -= num_bytes;
5442 space_info->bytes_reserved -= num_bytes;
5444 spin_unlock(&cache->lock);
5445 spin_unlock(&space_info->lock);
5449 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5450 struct btrfs_root *root)
5452 struct btrfs_fs_info *fs_info = root->fs_info;
5453 struct btrfs_caching_control *next;
5454 struct btrfs_caching_control *caching_ctl;
5455 struct btrfs_block_group_cache *cache;
5456 struct btrfs_space_info *space_info;
5458 down_write(&fs_info->extent_commit_sem);
5460 list_for_each_entry_safe(caching_ctl, next,
5461 &fs_info->caching_block_groups, list) {
5462 cache = caching_ctl->block_group;
5463 if (block_group_cache_done(cache)) {
5464 cache->last_byte_to_unpin = (u64)-1;
5465 list_del_init(&caching_ctl->list);
5466 put_caching_control(caching_ctl);
5468 cache->last_byte_to_unpin = caching_ctl->progress;
5472 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5473 fs_info->pinned_extents = &fs_info->freed_extents[1];
5475 fs_info->pinned_extents = &fs_info->freed_extents[0];
5477 up_write(&fs_info->extent_commit_sem);
5479 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5480 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5482 update_global_block_rsv(fs_info);
5485 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5487 struct btrfs_fs_info *fs_info = root->fs_info;
5488 struct btrfs_block_group_cache *cache = NULL;
5489 struct btrfs_space_info *space_info;
5490 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5494 while (start <= end) {
5497 start >= cache->key.objectid + cache->key.offset) {
5499 btrfs_put_block_group(cache);
5500 cache = btrfs_lookup_block_group(fs_info, start);
5501 BUG_ON(!cache); /* Logic error */
5504 len = cache->key.objectid + cache->key.offset - start;
5505 len = min(len, end + 1 - start);
5507 if (start < cache->last_byte_to_unpin) {
5508 len = min(len, cache->last_byte_to_unpin - start);
5509 btrfs_add_free_space(cache, start, len);
5513 space_info = cache->space_info;
5515 spin_lock(&space_info->lock);
5516 spin_lock(&cache->lock);
5517 cache->pinned -= len;
5518 space_info->bytes_pinned -= len;
5520 space_info->bytes_readonly += len;
5523 spin_unlock(&cache->lock);
5524 if (!readonly && global_rsv->space_info == space_info) {
5525 spin_lock(&global_rsv->lock);
5526 if (!global_rsv->full) {
5527 len = min(len, global_rsv->size -
5528 global_rsv->reserved);
5529 global_rsv->reserved += len;
5530 space_info->bytes_may_use += len;
5531 if (global_rsv->reserved >= global_rsv->size)
5532 global_rsv->full = 1;
5534 spin_unlock(&global_rsv->lock);
5536 spin_unlock(&space_info->lock);
5540 btrfs_put_block_group(cache);
5544 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5545 struct btrfs_root *root)
5547 struct btrfs_fs_info *fs_info = root->fs_info;
5548 struct extent_io_tree *unpin;
5556 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5557 unpin = &fs_info->freed_extents[1];
5559 unpin = &fs_info->freed_extents[0];
5562 ret = find_first_extent_bit(unpin, 0, &start, &end,
5563 EXTENT_DIRTY, NULL);
5567 if (btrfs_test_opt(root, DISCARD))
5568 ret = btrfs_discard_extent(root, start,
5569 end + 1 - start, NULL);
5571 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5572 unpin_extent_range(root, start, end);
5579 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5580 u64 owner, u64 root_objectid)
5582 struct btrfs_space_info *space_info;
5585 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5586 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5587 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5589 flags = BTRFS_BLOCK_GROUP_METADATA;
5591 flags = BTRFS_BLOCK_GROUP_DATA;
5594 space_info = __find_space_info(fs_info, flags);
5595 BUG_ON(!space_info); /* Logic bug */
5596 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5600 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5601 struct btrfs_root *root,
5602 u64 bytenr, u64 num_bytes, u64 parent,
5603 u64 root_objectid, u64 owner_objectid,
5604 u64 owner_offset, int refs_to_drop,
5605 struct btrfs_delayed_extent_op *extent_op)
5607 struct btrfs_key key;
5608 struct btrfs_path *path;
5609 struct btrfs_fs_info *info = root->fs_info;
5610 struct btrfs_root *extent_root = info->extent_root;
5611 struct extent_buffer *leaf;
5612 struct btrfs_extent_item *ei;
5613 struct btrfs_extent_inline_ref *iref;
5616 int extent_slot = 0;
5617 int found_extent = 0;
5621 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5624 path = btrfs_alloc_path();
5629 path->leave_spinning = 1;
5631 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5632 BUG_ON(!is_data && refs_to_drop != 1);
5635 skinny_metadata = 0;
5637 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5638 bytenr, num_bytes, parent,
5639 root_objectid, owner_objectid,
5642 extent_slot = path->slots[0];
5643 while (extent_slot >= 0) {
5644 btrfs_item_key_to_cpu(path->nodes[0], &key,
5646 if (key.objectid != bytenr)
5648 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5649 key.offset == num_bytes) {
5653 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5654 key.offset == owner_objectid) {
5658 if (path->slots[0] - extent_slot > 5)
5662 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5663 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5664 if (found_extent && item_size < sizeof(*ei))
5667 if (!found_extent) {
5669 ret = remove_extent_backref(trans, extent_root, path,
5673 btrfs_abort_transaction(trans, extent_root, ret);
5676 btrfs_release_path(path);
5677 path->leave_spinning = 1;
5679 key.objectid = bytenr;
5680 key.type = BTRFS_EXTENT_ITEM_KEY;
5681 key.offset = num_bytes;
5683 if (!is_data && skinny_metadata) {
5684 key.type = BTRFS_METADATA_ITEM_KEY;
5685 key.offset = owner_objectid;
5688 ret = btrfs_search_slot(trans, extent_root,
5690 if (ret > 0 && skinny_metadata && path->slots[0]) {
5692 * Couldn't find our skinny metadata item,
5693 * see if we have ye olde extent item.
5696 btrfs_item_key_to_cpu(path->nodes[0], &key,
5698 if (key.objectid == bytenr &&
5699 key.type == BTRFS_EXTENT_ITEM_KEY &&
5700 key.offset == num_bytes)
5704 if (ret > 0 && skinny_metadata) {
5705 skinny_metadata = false;
5706 key.type = BTRFS_EXTENT_ITEM_KEY;
5707 key.offset = num_bytes;
5708 btrfs_release_path(path);
5709 ret = btrfs_search_slot(trans, extent_root,
5714 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5717 btrfs_print_leaf(extent_root,
5721 btrfs_abort_transaction(trans, extent_root, ret);
5724 extent_slot = path->slots[0];
5726 } else if (ret == -ENOENT) {
5727 btrfs_print_leaf(extent_root, path->nodes[0]);
5730 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5731 bytenr, parent, root_objectid, owner_objectid,
5734 btrfs_abort_transaction(trans, extent_root, ret);
5738 leaf = path->nodes[0];
5739 item_size = btrfs_item_size_nr(leaf, extent_slot);
5740 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5741 if (item_size < sizeof(*ei)) {
5742 BUG_ON(found_extent || extent_slot != path->slots[0]);
5743 ret = convert_extent_item_v0(trans, extent_root, path,
5746 btrfs_abort_transaction(trans, extent_root, ret);
5750 btrfs_release_path(path);
5751 path->leave_spinning = 1;
5753 key.objectid = bytenr;
5754 key.type = BTRFS_EXTENT_ITEM_KEY;
5755 key.offset = num_bytes;
5757 ret = btrfs_search_slot(trans, extent_root, &key, path,
5760 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5762 btrfs_print_leaf(extent_root, path->nodes[0]);
5765 btrfs_abort_transaction(trans, extent_root, ret);
5769 extent_slot = path->slots[0];
5770 leaf = path->nodes[0];
5771 item_size = btrfs_item_size_nr(leaf, extent_slot);
5774 BUG_ON(item_size < sizeof(*ei));
5775 ei = btrfs_item_ptr(leaf, extent_slot,
5776 struct btrfs_extent_item);
5777 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5778 key.type == BTRFS_EXTENT_ITEM_KEY) {
5779 struct btrfs_tree_block_info *bi;
5780 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5781 bi = (struct btrfs_tree_block_info *)(ei + 1);
5782 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5785 refs = btrfs_extent_refs(leaf, ei);
5786 if (refs < refs_to_drop) {
5787 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5788 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5790 btrfs_abort_transaction(trans, extent_root, ret);
5793 refs -= refs_to_drop;
5797 __run_delayed_extent_op(extent_op, leaf, ei);
5799 * In the case of inline back ref, reference count will
5800 * be updated by remove_extent_backref
5803 BUG_ON(!found_extent);
5805 btrfs_set_extent_refs(leaf, ei, refs);
5806 btrfs_mark_buffer_dirty(leaf);
5809 ret = remove_extent_backref(trans, extent_root, path,
5813 btrfs_abort_transaction(trans, extent_root, ret);
5817 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5821 BUG_ON(is_data && refs_to_drop !=
5822 extent_data_ref_count(root, path, iref));
5824 BUG_ON(path->slots[0] != extent_slot);
5826 BUG_ON(path->slots[0] != extent_slot + 1);
5827 path->slots[0] = extent_slot;
5832 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5835 btrfs_abort_transaction(trans, extent_root, ret);
5838 btrfs_release_path(path);
5841 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5843 btrfs_abort_transaction(trans, extent_root, ret);
5848 ret = update_block_group(root, bytenr, num_bytes, 0);
5850 btrfs_abort_transaction(trans, extent_root, ret);
5855 btrfs_free_path(path);
5860 * when we free an block, it is possible (and likely) that we free the last
5861 * delayed ref for that extent as well. This searches the delayed ref tree for
5862 * a given extent, and if there are no other delayed refs to be processed, it
5863 * removes it from the tree.
5865 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5866 struct btrfs_root *root, u64 bytenr)
5868 struct btrfs_delayed_ref_head *head;
5869 struct btrfs_delayed_ref_root *delayed_refs;
5870 struct btrfs_delayed_ref_node *ref;
5871 struct rb_node *node;
5874 delayed_refs = &trans->transaction->delayed_refs;
5875 spin_lock(&delayed_refs->lock);
5876 head = btrfs_find_delayed_ref_head(trans, bytenr);
5880 node = rb_prev(&head->node.rb_node);
5884 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5886 /* there are still entries for this ref, we can't drop it */
5887 if (ref->bytenr == bytenr)
5890 if (head->extent_op) {
5891 if (!head->must_insert_reserved)
5893 btrfs_free_delayed_extent_op(head->extent_op);
5894 head->extent_op = NULL;
5898 * waiting for the lock here would deadlock. If someone else has it
5899 * locked they are already in the process of dropping it anyway
5901 if (!mutex_trylock(&head->mutex))
5905 * at this point we have a head with no other entries. Go
5906 * ahead and process it.
5908 head->node.in_tree = 0;
5909 rb_erase(&head->node.rb_node, &delayed_refs->root);
5911 delayed_refs->num_entries--;
5914 * we don't take a ref on the node because we're removing it from the
5915 * tree, so we just steal the ref the tree was holding.
5917 delayed_refs->num_heads--;
5918 if (list_empty(&head->cluster))
5919 delayed_refs->num_heads_ready--;
5921 list_del_init(&head->cluster);
5922 spin_unlock(&delayed_refs->lock);
5924 BUG_ON(head->extent_op);
5925 if (head->must_insert_reserved)
5928 mutex_unlock(&head->mutex);
5929 btrfs_put_delayed_ref(&head->node);
5932 spin_unlock(&delayed_refs->lock);
5936 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5937 struct btrfs_root *root,
5938 struct extent_buffer *buf,
5939 u64 parent, int last_ref)
5941 struct btrfs_block_group_cache *cache = NULL;
5945 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5946 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5947 buf->start, buf->len,
5948 parent, root->root_key.objectid,
5949 btrfs_header_level(buf),
5950 BTRFS_DROP_DELAYED_REF, NULL, 0);
5951 BUG_ON(ret); /* -ENOMEM */
5957 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5959 if (btrfs_header_generation(buf) == trans->transid) {
5960 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5961 ret = check_ref_cleanup(trans, root, buf->start);
5966 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5967 pin_down_extent(root, cache, buf->start, buf->len, 1);
5971 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5973 btrfs_add_free_space(cache, buf->start, buf->len);
5974 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5975 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5980 add_pinned_bytes(root->fs_info, buf->len,
5981 btrfs_header_level(buf),
5982 root->root_key.objectid);
5985 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5988 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5989 btrfs_put_block_group(cache);
5992 /* Can return -ENOMEM */
5993 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5994 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5995 u64 owner, u64 offset, int for_cow)
5998 struct btrfs_fs_info *fs_info = root->fs_info;
6000 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6003 * tree log blocks never actually go into the extent allocation
6004 * tree, just update pinning info and exit early.
6006 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6007 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6008 /* unlocks the pinned mutex */
6009 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6011 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6012 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6014 parent, root_objectid, (int)owner,
6015 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6017 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6019 parent, root_objectid, owner,
6020 offset, BTRFS_DROP_DELAYED_REF,
6026 static u64 stripe_align(struct btrfs_root *root,
6027 struct btrfs_block_group_cache *cache,
6028 u64 val, u64 num_bytes)
6030 u64 ret = ALIGN(val, root->stripesize);
6035 * when we wait for progress in the block group caching, its because
6036 * our allocation attempt failed at least once. So, we must sleep
6037 * and let some progress happen before we try again.
6039 * This function will sleep at least once waiting for new free space to
6040 * show up, and then it will check the block group free space numbers
6041 * for our min num_bytes. Another option is to have it go ahead
6042 * and look in the rbtree for a free extent of a given size, but this
6045 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6046 * any of the information in this block group.
6048 static noinline void
6049 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6052 struct btrfs_caching_control *caching_ctl;
6054 caching_ctl = get_caching_control(cache);
6058 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6059 (cache->free_space_ctl->free_space >= num_bytes));
6061 put_caching_control(caching_ctl);
6065 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6067 struct btrfs_caching_control *caching_ctl;
6070 caching_ctl = get_caching_control(cache);
6072 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6074 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6075 if (cache->cached == BTRFS_CACHE_ERROR)
6077 put_caching_control(caching_ctl);
6081 int __get_raid_index(u64 flags)
6083 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6084 return BTRFS_RAID_RAID10;
6085 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6086 return BTRFS_RAID_RAID1;
6087 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6088 return BTRFS_RAID_DUP;
6089 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6090 return BTRFS_RAID_RAID0;
6091 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6092 return BTRFS_RAID_RAID5;
6093 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6094 return BTRFS_RAID_RAID6;
6096 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6099 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6101 return __get_raid_index(cache->flags);
6104 enum btrfs_loop_type {
6105 LOOP_CACHING_NOWAIT = 0,
6106 LOOP_CACHING_WAIT = 1,
6107 LOOP_ALLOC_CHUNK = 2,
6108 LOOP_NO_EMPTY_SIZE = 3,
6112 * walks the btree of allocated extents and find a hole of a given size.
6113 * The key ins is changed to record the hole:
6114 * ins->objectid == start position
6115 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6116 * ins->offset == the size of the hole.
6117 * Any available blocks before search_start are skipped.
6119 * If there is no suitable free space, we will record the max size of
6120 * the free space extent currently.
6122 static noinline int find_free_extent(struct btrfs_root *orig_root,
6123 u64 num_bytes, u64 empty_size,
6124 u64 hint_byte, struct btrfs_key *ins,
6128 struct btrfs_root *root = orig_root->fs_info->extent_root;
6129 struct btrfs_free_cluster *last_ptr = NULL;
6130 struct btrfs_block_group_cache *block_group = NULL;
6131 struct btrfs_block_group_cache *used_block_group;
6132 u64 search_start = 0;
6133 u64 max_extent_size = 0;
6134 int empty_cluster = 2 * 1024 * 1024;
6135 struct btrfs_space_info *space_info;
6137 int index = __get_raid_index(flags);
6138 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6139 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6140 bool found_uncached_bg = false;
6141 bool failed_cluster_refill = false;
6142 bool failed_alloc = false;
6143 bool use_cluster = true;
6144 bool have_caching_bg = false;
6146 WARN_ON(num_bytes < root->sectorsize);
6147 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6151 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6153 space_info = __find_space_info(root->fs_info, flags);
6155 btrfs_err(root->fs_info, "No space info for %llu", flags);
6160 * If the space info is for both data and metadata it means we have a
6161 * small filesystem and we can't use the clustering stuff.
6163 if (btrfs_mixed_space_info(space_info))
6164 use_cluster = false;
6166 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6167 last_ptr = &root->fs_info->meta_alloc_cluster;
6168 if (!btrfs_test_opt(root, SSD))
6169 empty_cluster = 64 * 1024;
6172 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6173 btrfs_test_opt(root, SSD)) {
6174 last_ptr = &root->fs_info->data_alloc_cluster;
6178 spin_lock(&last_ptr->lock);
6179 if (last_ptr->block_group)
6180 hint_byte = last_ptr->window_start;
6181 spin_unlock(&last_ptr->lock);
6184 search_start = max(search_start, first_logical_byte(root, 0));
6185 search_start = max(search_start, hint_byte);
6190 if (search_start == hint_byte) {
6191 block_group = btrfs_lookup_block_group(root->fs_info,
6193 used_block_group = block_group;
6195 * we don't want to use the block group if it doesn't match our
6196 * allocation bits, or if its not cached.
6198 * However if we are re-searching with an ideal block group
6199 * picked out then we don't care that the block group is cached.
6201 if (block_group && block_group_bits(block_group, flags) &&
6202 block_group->cached != BTRFS_CACHE_NO) {
6203 down_read(&space_info->groups_sem);
6204 if (list_empty(&block_group->list) ||
6207 * someone is removing this block group,
6208 * we can't jump into the have_block_group
6209 * target because our list pointers are not
6212 btrfs_put_block_group(block_group);
6213 up_read(&space_info->groups_sem);
6215 index = get_block_group_index(block_group);
6216 goto have_block_group;
6218 } else if (block_group) {
6219 btrfs_put_block_group(block_group);
6223 have_caching_bg = false;
6224 down_read(&space_info->groups_sem);
6225 list_for_each_entry(block_group, &space_info->block_groups[index],
6230 used_block_group = block_group;
6231 btrfs_get_block_group(block_group);
6232 search_start = block_group->key.objectid;
6235 * this can happen if we end up cycling through all the
6236 * raid types, but we want to make sure we only allocate
6237 * for the proper type.
6239 if (!block_group_bits(block_group, flags)) {
6240 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6241 BTRFS_BLOCK_GROUP_RAID1 |
6242 BTRFS_BLOCK_GROUP_RAID5 |
6243 BTRFS_BLOCK_GROUP_RAID6 |
6244 BTRFS_BLOCK_GROUP_RAID10;
6247 * if they asked for extra copies and this block group
6248 * doesn't provide them, bail. This does allow us to
6249 * fill raid0 from raid1.
6251 if ((flags & extra) && !(block_group->flags & extra))
6256 cached = block_group_cache_done(block_group);
6257 if (unlikely(!cached)) {
6258 found_uncached_bg = true;
6259 ret = cache_block_group(block_group, 0);
6264 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6266 if (unlikely(block_group->ro))
6270 * Ok we want to try and use the cluster allocator, so
6274 unsigned long aligned_cluster;
6276 * the refill lock keeps out other
6277 * people trying to start a new cluster
6279 spin_lock(&last_ptr->refill_lock);
6280 used_block_group = last_ptr->block_group;
6281 if (used_block_group != block_group &&
6282 (!used_block_group ||
6283 used_block_group->ro ||
6284 !block_group_bits(used_block_group, flags))) {
6285 used_block_group = block_group;
6286 goto refill_cluster;
6289 if (used_block_group != block_group)
6290 btrfs_get_block_group(used_block_group);
6292 offset = btrfs_alloc_from_cluster(used_block_group,
6295 used_block_group->key.objectid,
6298 /* we have a block, we're done */
6299 spin_unlock(&last_ptr->refill_lock);
6300 trace_btrfs_reserve_extent_cluster(root,
6301 block_group, search_start, num_bytes);
6305 WARN_ON(last_ptr->block_group != used_block_group);
6306 if (used_block_group != block_group) {
6307 btrfs_put_block_group(used_block_group);
6308 used_block_group = block_group;
6311 BUG_ON(used_block_group != block_group);
6312 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6313 * set up a new clusters, so lets just skip it
6314 * and let the allocator find whatever block
6315 * it can find. If we reach this point, we
6316 * will have tried the cluster allocator
6317 * plenty of times and not have found
6318 * anything, so we are likely way too
6319 * fragmented for the clustering stuff to find
6322 * However, if the cluster is taken from the
6323 * current block group, release the cluster
6324 * first, so that we stand a better chance of
6325 * succeeding in the unclustered
6327 if (loop >= LOOP_NO_EMPTY_SIZE &&
6328 last_ptr->block_group != block_group) {
6329 spin_unlock(&last_ptr->refill_lock);
6330 goto unclustered_alloc;
6334 * this cluster didn't work out, free it and
6337 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6339 if (loop >= LOOP_NO_EMPTY_SIZE) {
6340 spin_unlock(&last_ptr->refill_lock);
6341 goto unclustered_alloc;
6344 aligned_cluster = max_t(unsigned long,
6345 empty_cluster + empty_size,
6346 block_group->full_stripe_len);
6348 /* allocate a cluster in this block group */
6349 ret = btrfs_find_space_cluster(root, block_group,
6350 last_ptr, search_start,
6355 * now pull our allocation out of this
6358 offset = btrfs_alloc_from_cluster(block_group,
6364 /* we found one, proceed */
6365 spin_unlock(&last_ptr->refill_lock);
6366 trace_btrfs_reserve_extent_cluster(root,
6367 block_group, search_start,
6371 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6372 && !failed_cluster_refill) {
6373 spin_unlock(&last_ptr->refill_lock);
6375 failed_cluster_refill = true;
6376 wait_block_group_cache_progress(block_group,
6377 num_bytes + empty_cluster + empty_size);
6378 goto have_block_group;
6382 * at this point we either didn't find a cluster
6383 * or we weren't able to allocate a block from our
6384 * cluster. Free the cluster we've been trying
6385 * to use, and go to the next block group
6387 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6388 spin_unlock(&last_ptr->refill_lock);
6393 spin_lock(&block_group->free_space_ctl->tree_lock);
6395 block_group->free_space_ctl->free_space <
6396 num_bytes + empty_cluster + empty_size) {
6397 if (block_group->free_space_ctl->free_space >
6400 block_group->free_space_ctl->free_space;
6401 spin_unlock(&block_group->free_space_ctl->tree_lock);
6404 spin_unlock(&block_group->free_space_ctl->tree_lock);
6406 offset = btrfs_find_space_for_alloc(block_group, search_start,
6407 num_bytes, empty_size,
6410 * If we didn't find a chunk, and we haven't failed on this
6411 * block group before, and this block group is in the middle of
6412 * caching and we are ok with waiting, then go ahead and wait
6413 * for progress to be made, and set failed_alloc to true.
6415 * If failed_alloc is true then we've already waited on this
6416 * block group once and should move on to the next block group.
6418 if (!offset && !failed_alloc && !cached &&
6419 loop > LOOP_CACHING_NOWAIT) {
6420 wait_block_group_cache_progress(block_group,
6421 num_bytes + empty_size);
6422 failed_alloc = true;
6423 goto have_block_group;
6424 } else if (!offset) {
6426 have_caching_bg = true;
6430 search_start = stripe_align(root, used_block_group,
6433 /* move on to the next group */
6434 if (search_start + num_bytes >
6435 used_block_group->key.objectid + used_block_group->key.offset) {
6436 btrfs_add_free_space(used_block_group, offset, num_bytes);
6440 if (offset < search_start)
6441 btrfs_add_free_space(used_block_group, offset,
6442 search_start - offset);
6443 BUG_ON(offset > search_start);
6445 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6447 if (ret == -EAGAIN) {
6448 btrfs_add_free_space(used_block_group, offset, num_bytes);
6452 /* we are all good, lets return */
6453 ins->objectid = search_start;
6454 ins->offset = num_bytes;
6456 trace_btrfs_reserve_extent(orig_root, block_group,
6457 search_start, num_bytes);
6458 if (used_block_group != block_group)
6459 btrfs_put_block_group(used_block_group);
6460 btrfs_put_block_group(block_group);
6463 failed_cluster_refill = false;
6464 failed_alloc = false;
6465 BUG_ON(index != get_block_group_index(block_group));
6466 if (used_block_group != block_group)
6467 btrfs_put_block_group(used_block_group);
6468 btrfs_put_block_group(block_group);
6470 up_read(&space_info->groups_sem);
6472 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6475 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6479 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6480 * caching kthreads as we move along
6481 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6482 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6483 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6486 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6489 if (loop == LOOP_ALLOC_CHUNK) {
6490 struct btrfs_trans_handle *trans;
6492 trans = btrfs_join_transaction(root);
6493 if (IS_ERR(trans)) {
6494 ret = PTR_ERR(trans);
6498 ret = do_chunk_alloc(trans, root, flags,
6501 * Do not bail out on ENOSPC since we
6502 * can do more things.
6504 if (ret < 0 && ret != -ENOSPC)
6505 btrfs_abort_transaction(trans,
6509 btrfs_end_transaction(trans, root);
6514 if (loop == LOOP_NO_EMPTY_SIZE) {
6520 } else if (!ins->objectid) {
6522 } else if (ins->objectid) {
6527 ins->offset = max_extent_size;
6531 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6532 int dump_block_groups)
6534 struct btrfs_block_group_cache *cache;
6537 spin_lock(&info->lock);
6538 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6540 info->total_bytes - info->bytes_used - info->bytes_pinned -
6541 info->bytes_reserved - info->bytes_readonly,
6542 (info->full) ? "" : "not ");
6543 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6544 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6545 info->total_bytes, info->bytes_used, info->bytes_pinned,
6546 info->bytes_reserved, info->bytes_may_use,
6547 info->bytes_readonly);
6548 spin_unlock(&info->lock);
6550 if (!dump_block_groups)
6553 down_read(&info->groups_sem);
6555 list_for_each_entry(cache, &info->block_groups[index], list) {
6556 spin_lock(&cache->lock);
6557 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6558 cache->key.objectid, cache->key.offset,
6559 btrfs_block_group_used(&cache->item), cache->pinned,
6560 cache->reserved, cache->ro ? "[readonly]" : "");
6561 btrfs_dump_free_space(cache, bytes);
6562 spin_unlock(&cache->lock);
6564 if (++index < BTRFS_NR_RAID_TYPES)
6566 up_read(&info->groups_sem);
6569 int btrfs_reserve_extent(struct btrfs_root *root,
6570 u64 num_bytes, u64 min_alloc_size,
6571 u64 empty_size, u64 hint_byte,
6572 struct btrfs_key *ins, int is_data)
6574 bool final_tried = false;
6578 flags = btrfs_get_alloc_profile(root, is_data);
6580 WARN_ON(num_bytes < root->sectorsize);
6581 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6584 if (ret == -ENOSPC) {
6585 if (!final_tried && ins->offset) {
6586 num_bytes = min(num_bytes >> 1, ins->offset);
6587 num_bytes = round_down(num_bytes, root->sectorsize);
6588 num_bytes = max(num_bytes, min_alloc_size);
6589 if (num_bytes == min_alloc_size)
6592 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6593 struct btrfs_space_info *sinfo;
6595 sinfo = __find_space_info(root->fs_info, flags);
6596 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6599 dump_space_info(sinfo, num_bytes, 1);
6606 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6607 u64 start, u64 len, int pin)
6609 struct btrfs_block_group_cache *cache;
6612 cache = btrfs_lookup_block_group(root->fs_info, start);
6614 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6619 if (btrfs_test_opt(root, DISCARD))
6620 ret = btrfs_discard_extent(root, start, len, NULL);
6623 pin_down_extent(root, cache, start, len, 1);
6625 btrfs_add_free_space(cache, start, len);
6626 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6628 btrfs_put_block_group(cache);
6630 trace_btrfs_reserved_extent_free(root, start, len);
6635 int btrfs_free_reserved_extent(struct btrfs_root *root,
6638 return __btrfs_free_reserved_extent(root, start, len, 0);
6641 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6644 return __btrfs_free_reserved_extent(root, start, len, 1);
6647 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6648 struct btrfs_root *root,
6649 u64 parent, u64 root_objectid,
6650 u64 flags, u64 owner, u64 offset,
6651 struct btrfs_key *ins, int ref_mod)
6654 struct btrfs_fs_info *fs_info = root->fs_info;
6655 struct btrfs_extent_item *extent_item;
6656 struct btrfs_extent_inline_ref *iref;
6657 struct btrfs_path *path;
6658 struct extent_buffer *leaf;
6663 type = BTRFS_SHARED_DATA_REF_KEY;
6665 type = BTRFS_EXTENT_DATA_REF_KEY;
6667 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6669 path = btrfs_alloc_path();
6673 path->leave_spinning = 1;
6674 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6677 btrfs_free_path(path);
6681 leaf = path->nodes[0];
6682 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6683 struct btrfs_extent_item);
6684 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6685 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6686 btrfs_set_extent_flags(leaf, extent_item,
6687 flags | BTRFS_EXTENT_FLAG_DATA);
6689 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6690 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6692 struct btrfs_shared_data_ref *ref;
6693 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6694 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6695 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6697 struct btrfs_extent_data_ref *ref;
6698 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6699 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6700 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6701 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6702 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6705 btrfs_mark_buffer_dirty(path->nodes[0]);
6706 btrfs_free_path(path);
6708 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6709 if (ret) { /* -ENOENT, logic error */
6710 btrfs_err(fs_info, "update block group failed for %llu %llu",
6711 ins->objectid, ins->offset);
6714 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6718 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6719 struct btrfs_root *root,
6720 u64 parent, u64 root_objectid,
6721 u64 flags, struct btrfs_disk_key *key,
6722 int level, struct btrfs_key *ins)
6725 struct btrfs_fs_info *fs_info = root->fs_info;
6726 struct btrfs_extent_item *extent_item;
6727 struct btrfs_tree_block_info *block_info;
6728 struct btrfs_extent_inline_ref *iref;
6729 struct btrfs_path *path;
6730 struct extent_buffer *leaf;
6731 u32 size = sizeof(*extent_item) + sizeof(*iref);
6732 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6735 if (!skinny_metadata)
6736 size += sizeof(*block_info);
6738 path = btrfs_alloc_path();
6740 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6745 path->leave_spinning = 1;
6746 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6749 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6751 btrfs_free_path(path);
6755 leaf = path->nodes[0];
6756 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6757 struct btrfs_extent_item);
6758 btrfs_set_extent_refs(leaf, extent_item, 1);
6759 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6760 btrfs_set_extent_flags(leaf, extent_item,
6761 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6763 if (skinny_metadata) {
6764 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6766 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6767 btrfs_set_tree_block_key(leaf, block_info, key);
6768 btrfs_set_tree_block_level(leaf, block_info, level);
6769 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6773 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6774 btrfs_set_extent_inline_ref_type(leaf, iref,
6775 BTRFS_SHARED_BLOCK_REF_KEY);
6776 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6778 btrfs_set_extent_inline_ref_type(leaf, iref,
6779 BTRFS_TREE_BLOCK_REF_KEY);
6780 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6783 btrfs_mark_buffer_dirty(leaf);
6784 btrfs_free_path(path);
6786 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6787 if (ret) { /* -ENOENT, logic error */
6788 btrfs_err(fs_info, "update block group failed for %llu %llu",
6789 ins->objectid, ins->offset);
6793 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6797 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6798 struct btrfs_root *root,
6799 u64 root_objectid, u64 owner,
6800 u64 offset, struct btrfs_key *ins)
6804 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6806 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6808 root_objectid, owner, offset,
6809 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6814 * this is used by the tree logging recovery code. It records that
6815 * an extent has been allocated and makes sure to clear the free
6816 * space cache bits as well
6818 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6819 struct btrfs_root *root,
6820 u64 root_objectid, u64 owner, u64 offset,
6821 struct btrfs_key *ins)
6824 struct btrfs_block_group_cache *block_group;
6827 * Mixed block groups will exclude before processing the log so we only
6828 * need to do the exlude dance if this fs isn't mixed.
6830 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6831 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6836 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6840 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6841 RESERVE_ALLOC_NO_ACCOUNT);
6842 BUG_ON(ret); /* logic error */
6843 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6844 0, owner, offset, ins, 1);
6845 btrfs_put_block_group(block_group);
6849 static struct extent_buffer *
6850 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6851 u64 bytenr, u32 blocksize, int level)
6853 struct extent_buffer *buf;
6855 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6857 return ERR_PTR(-ENOMEM);
6858 btrfs_set_header_generation(buf, trans->transid);
6859 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6860 btrfs_tree_lock(buf);
6861 clean_tree_block(trans, root, buf);
6862 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6864 btrfs_set_lock_blocking(buf);
6865 btrfs_set_buffer_uptodate(buf);
6867 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6869 * we allow two log transactions at a time, use different
6870 * EXENT bit to differentiate dirty pages.
6872 if (root->log_transid % 2 == 0)
6873 set_extent_dirty(&root->dirty_log_pages, buf->start,
6874 buf->start + buf->len - 1, GFP_NOFS);
6876 set_extent_new(&root->dirty_log_pages, buf->start,
6877 buf->start + buf->len - 1, GFP_NOFS);
6879 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6880 buf->start + buf->len - 1, GFP_NOFS);
6882 trans->blocks_used++;
6883 /* this returns a buffer locked for blocking */
6887 static struct btrfs_block_rsv *
6888 use_block_rsv(struct btrfs_trans_handle *trans,
6889 struct btrfs_root *root, u32 blocksize)
6891 struct btrfs_block_rsv *block_rsv;
6892 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6894 bool global_updated = false;
6896 block_rsv = get_block_rsv(trans, root);
6898 if (unlikely(block_rsv->size == 0))
6901 ret = block_rsv_use_bytes(block_rsv, blocksize);
6905 if (block_rsv->failfast)
6906 return ERR_PTR(ret);
6908 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6909 global_updated = true;
6910 update_global_block_rsv(root->fs_info);
6914 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6915 static DEFINE_RATELIMIT_STATE(_rs,
6916 DEFAULT_RATELIMIT_INTERVAL * 10,
6917 /*DEFAULT_RATELIMIT_BURST*/ 1);
6918 if (__ratelimit(&_rs))
6920 "btrfs: block rsv returned %d\n", ret);
6923 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6924 BTRFS_RESERVE_NO_FLUSH);
6928 * If we couldn't reserve metadata bytes try and use some from
6929 * the global reserve if its space type is the same as the global
6932 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6933 block_rsv->space_info == global_rsv->space_info) {
6934 ret = block_rsv_use_bytes(global_rsv, blocksize);
6938 return ERR_PTR(ret);
6941 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6942 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6944 block_rsv_add_bytes(block_rsv, blocksize, 0);
6945 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6949 * finds a free extent and does all the dirty work required for allocation
6950 * returns the key for the extent through ins, and a tree buffer for
6951 * the first block of the extent through buf.
6953 * returns the tree buffer or NULL.
6955 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6956 struct btrfs_root *root, u32 blocksize,
6957 u64 parent, u64 root_objectid,
6958 struct btrfs_disk_key *key, int level,
6959 u64 hint, u64 empty_size)
6961 struct btrfs_key ins;
6962 struct btrfs_block_rsv *block_rsv;
6963 struct extent_buffer *buf;
6966 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6969 block_rsv = use_block_rsv(trans, root, blocksize);
6970 if (IS_ERR(block_rsv))
6971 return ERR_CAST(block_rsv);
6973 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6974 empty_size, hint, &ins, 0);
6976 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6977 return ERR_PTR(ret);
6980 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6982 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6984 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6986 parent = ins.objectid;
6987 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6991 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6992 struct btrfs_delayed_extent_op *extent_op;
6993 extent_op = btrfs_alloc_delayed_extent_op();
6994 BUG_ON(!extent_op); /* -ENOMEM */
6996 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6998 memset(&extent_op->key, 0, sizeof(extent_op->key));
6999 extent_op->flags_to_set = flags;
7000 if (skinny_metadata)
7001 extent_op->update_key = 0;
7003 extent_op->update_key = 1;
7004 extent_op->update_flags = 1;
7005 extent_op->is_data = 0;
7006 extent_op->level = level;
7008 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7010 ins.offset, parent, root_objectid,
7011 level, BTRFS_ADD_DELAYED_EXTENT,
7013 BUG_ON(ret); /* -ENOMEM */
7018 struct walk_control {
7019 u64 refs[BTRFS_MAX_LEVEL];
7020 u64 flags[BTRFS_MAX_LEVEL];
7021 struct btrfs_key update_progress;
7032 #define DROP_REFERENCE 1
7033 #define UPDATE_BACKREF 2
7035 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7036 struct btrfs_root *root,
7037 struct walk_control *wc,
7038 struct btrfs_path *path)
7046 struct btrfs_key key;
7047 struct extent_buffer *eb;
7052 if (path->slots[wc->level] < wc->reada_slot) {
7053 wc->reada_count = wc->reada_count * 2 / 3;
7054 wc->reada_count = max(wc->reada_count, 2);
7056 wc->reada_count = wc->reada_count * 3 / 2;
7057 wc->reada_count = min_t(int, wc->reada_count,
7058 BTRFS_NODEPTRS_PER_BLOCK(root));
7061 eb = path->nodes[wc->level];
7062 nritems = btrfs_header_nritems(eb);
7063 blocksize = btrfs_level_size(root, wc->level - 1);
7065 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7066 if (nread >= wc->reada_count)
7070 bytenr = btrfs_node_blockptr(eb, slot);
7071 generation = btrfs_node_ptr_generation(eb, slot);
7073 if (slot == path->slots[wc->level])
7076 if (wc->stage == UPDATE_BACKREF &&
7077 generation <= root->root_key.offset)
7080 /* We don't lock the tree block, it's OK to be racy here */
7081 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7082 wc->level - 1, 1, &refs,
7084 /* We don't care about errors in readahead. */
7089 if (wc->stage == DROP_REFERENCE) {
7093 if (wc->level == 1 &&
7094 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7096 if (!wc->update_ref ||
7097 generation <= root->root_key.offset)
7099 btrfs_node_key_to_cpu(eb, &key, slot);
7100 ret = btrfs_comp_cpu_keys(&key,
7101 &wc->update_progress);
7105 if (wc->level == 1 &&
7106 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7110 ret = readahead_tree_block(root, bytenr, blocksize,
7116 wc->reada_slot = slot;
7120 * helper to process tree block while walking down the tree.
7122 * when wc->stage == UPDATE_BACKREF, this function updates
7123 * back refs for pointers in the block.
7125 * NOTE: return value 1 means we should stop walking down.
7127 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7128 struct btrfs_root *root,
7129 struct btrfs_path *path,
7130 struct walk_control *wc, int lookup_info)
7132 int level = wc->level;
7133 struct extent_buffer *eb = path->nodes[level];
7134 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7137 if (wc->stage == UPDATE_BACKREF &&
7138 btrfs_header_owner(eb) != root->root_key.objectid)
7142 * when reference count of tree block is 1, it won't increase
7143 * again. once full backref flag is set, we never clear it.
7146 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7147 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7148 BUG_ON(!path->locks[level]);
7149 ret = btrfs_lookup_extent_info(trans, root,
7150 eb->start, level, 1,
7153 BUG_ON(ret == -ENOMEM);
7156 BUG_ON(wc->refs[level] == 0);
7159 if (wc->stage == DROP_REFERENCE) {
7160 if (wc->refs[level] > 1)
7163 if (path->locks[level] && !wc->keep_locks) {
7164 btrfs_tree_unlock_rw(eb, path->locks[level]);
7165 path->locks[level] = 0;
7170 /* wc->stage == UPDATE_BACKREF */
7171 if (!(wc->flags[level] & flag)) {
7172 BUG_ON(!path->locks[level]);
7173 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7174 BUG_ON(ret); /* -ENOMEM */
7175 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7176 BUG_ON(ret); /* -ENOMEM */
7177 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7179 btrfs_header_level(eb), 0);
7180 BUG_ON(ret); /* -ENOMEM */
7181 wc->flags[level] |= flag;
7185 * the block is shared by multiple trees, so it's not good to
7186 * keep the tree lock
7188 if (path->locks[level] && level > 0) {
7189 btrfs_tree_unlock_rw(eb, path->locks[level]);
7190 path->locks[level] = 0;
7196 * helper to process tree block pointer.
7198 * when wc->stage == DROP_REFERENCE, this function checks
7199 * reference count of the block pointed to. if the block
7200 * is shared and we need update back refs for the subtree
7201 * rooted at the block, this function changes wc->stage to
7202 * UPDATE_BACKREF. if the block is shared and there is no
7203 * need to update back, this function drops the reference
7206 * NOTE: return value 1 means we should stop walking down.
7208 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7209 struct btrfs_root *root,
7210 struct btrfs_path *path,
7211 struct walk_control *wc, int *lookup_info)
7217 struct btrfs_key key;
7218 struct extent_buffer *next;
7219 int level = wc->level;
7223 generation = btrfs_node_ptr_generation(path->nodes[level],
7224 path->slots[level]);
7226 * if the lower level block was created before the snapshot
7227 * was created, we know there is no need to update back refs
7230 if (wc->stage == UPDATE_BACKREF &&
7231 generation <= root->root_key.offset) {
7236 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7237 blocksize = btrfs_level_size(root, level - 1);
7239 next = btrfs_find_tree_block(root, bytenr, blocksize);
7241 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7244 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7248 btrfs_tree_lock(next);
7249 btrfs_set_lock_blocking(next);
7251 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7252 &wc->refs[level - 1],
7253 &wc->flags[level - 1]);
7255 btrfs_tree_unlock(next);
7259 if (unlikely(wc->refs[level - 1] == 0)) {
7260 btrfs_err(root->fs_info, "Missing references.");
7265 if (wc->stage == DROP_REFERENCE) {
7266 if (wc->refs[level - 1] > 1) {
7268 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7271 if (!wc->update_ref ||
7272 generation <= root->root_key.offset)
7275 btrfs_node_key_to_cpu(path->nodes[level], &key,
7276 path->slots[level]);
7277 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7281 wc->stage = UPDATE_BACKREF;
7282 wc->shared_level = level - 1;
7286 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7290 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7291 btrfs_tree_unlock(next);
7292 free_extent_buffer(next);
7298 if (reada && level == 1)
7299 reada_walk_down(trans, root, wc, path);
7300 next = read_tree_block(root, bytenr, blocksize, generation);
7301 if (!next || !extent_buffer_uptodate(next)) {
7302 free_extent_buffer(next);
7305 btrfs_tree_lock(next);
7306 btrfs_set_lock_blocking(next);
7310 BUG_ON(level != btrfs_header_level(next));
7311 path->nodes[level] = next;
7312 path->slots[level] = 0;
7313 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7319 wc->refs[level - 1] = 0;
7320 wc->flags[level - 1] = 0;
7321 if (wc->stage == DROP_REFERENCE) {
7322 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7323 parent = path->nodes[level]->start;
7325 BUG_ON(root->root_key.objectid !=
7326 btrfs_header_owner(path->nodes[level]));
7330 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7331 root->root_key.objectid, level - 1, 0, 0);
7332 BUG_ON(ret); /* -ENOMEM */
7334 btrfs_tree_unlock(next);
7335 free_extent_buffer(next);
7341 * helper to process tree block while walking up the tree.
7343 * when wc->stage == DROP_REFERENCE, this function drops
7344 * reference count on the block.
7346 * when wc->stage == UPDATE_BACKREF, this function changes
7347 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7348 * to UPDATE_BACKREF previously while processing the block.
7350 * NOTE: return value 1 means we should stop walking up.
7352 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7353 struct btrfs_root *root,
7354 struct btrfs_path *path,
7355 struct walk_control *wc)
7358 int level = wc->level;
7359 struct extent_buffer *eb = path->nodes[level];
7362 if (wc->stage == UPDATE_BACKREF) {
7363 BUG_ON(wc->shared_level < level);
7364 if (level < wc->shared_level)
7367 ret = find_next_key(path, level + 1, &wc->update_progress);
7371 wc->stage = DROP_REFERENCE;
7372 wc->shared_level = -1;
7373 path->slots[level] = 0;
7376 * check reference count again if the block isn't locked.
7377 * we should start walking down the tree again if reference
7380 if (!path->locks[level]) {
7382 btrfs_tree_lock(eb);
7383 btrfs_set_lock_blocking(eb);
7384 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7386 ret = btrfs_lookup_extent_info(trans, root,
7387 eb->start, level, 1,
7391 btrfs_tree_unlock_rw(eb, path->locks[level]);
7392 path->locks[level] = 0;
7395 BUG_ON(wc->refs[level] == 0);
7396 if (wc->refs[level] == 1) {
7397 btrfs_tree_unlock_rw(eb, path->locks[level]);
7398 path->locks[level] = 0;
7404 /* wc->stage == DROP_REFERENCE */
7405 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7407 if (wc->refs[level] == 1) {
7409 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7410 ret = btrfs_dec_ref(trans, root, eb, 1,
7413 ret = btrfs_dec_ref(trans, root, eb, 0,
7415 BUG_ON(ret); /* -ENOMEM */
7417 /* make block locked assertion in clean_tree_block happy */
7418 if (!path->locks[level] &&
7419 btrfs_header_generation(eb) == trans->transid) {
7420 btrfs_tree_lock(eb);
7421 btrfs_set_lock_blocking(eb);
7422 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7424 clean_tree_block(trans, root, eb);
7427 if (eb == root->node) {
7428 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7431 BUG_ON(root->root_key.objectid !=
7432 btrfs_header_owner(eb));
7434 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7435 parent = path->nodes[level + 1]->start;
7437 BUG_ON(root->root_key.objectid !=
7438 btrfs_header_owner(path->nodes[level + 1]));
7441 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7443 wc->refs[level] = 0;
7444 wc->flags[level] = 0;
7448 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7449 struct btrfs_root *root,
7450 struct btrfs_path *path,
7451 struct walk_control *wc)
7453 int level = wc->level;
7454 int lookup_info = 1;
7457 while (level >= 0) {
7458 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7465 if (path->slots[level] >=
7466 btrfs_header_nritems(path->nodes[level]))
7469 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7471 path->slots[level]++;
7480 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7481 struct btrfs_root *root,
7482 struct btrfs_path *path,
7483 struct walk_control *wc, int max_level)
7485 int level = wc->level;
7488 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7489 while (level < max_level && path->nodes[level]) {
7491 if (path->slots[level] + 1 <
7492 btrfs_header_nritems(path->nodes[level])) {
7493 path->slots[level]++;
7496 ret = walk_up_proc(trans, root, path, wc);
7500 if (path->locks[level]) {
7501 btrfs_tree_unlock_rw(path->nodes[level],
7502 path->locks[level]);
7503 path->locks[level] = 0;
7505 free_extent_buffer(path->nodes[level]);
7506 path->nodes[level] = NULL;
7514 * drop a subvolume tree.
7516 * this function traverses the tree freeing any blocks that only
7517 * referenced by the tree.
7519 * when a shared tree block is found. this function decreases its
7520 * reference count by one. if update_ref is true, this function
7521 * also make sure backrefs for the shared block and all lower level
7522 * blocks are properly updated.
7524 * If called with for_reloc == 0, may exit early with -EAGAIN
7526 int btrfs_drop_snapshot(struct btrfs_root *root,
7527 struct btrfs_block_rsv *block_rsv, int update_ref,
7530 struct btrfs_path *path;
7531 struct btrfs_trans_handle *trans;
7532 struct btrfs_root *tree_root = root->fs_info->tree_root;
7533 struct btrfs_root_item *root_item = &root->root_item;
7534 struct walk_control *wc;
7535 struct btrfs_key key;
7539 bool root_dropped = false;
7541 path = btrfs_alloc_path();
7547 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7549 btrfs_free_path(path);
7554 trans = btrfs_start_transaction(tree_root, 0);
7555 if (IS_ERR(trans)) {
7556 err = PTR_ERR(trans);
7561 trans->block_rsv = block_rsv;
7563 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7564 level = btrfs_header_level(root->node);
7565 path->nodes[level] = btrfs_lock_root_node(root);
7566 btrfs_set_lock_blocking(path->nodes[level]);
7567 path->slots[level] = 0;
7568 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7569 memset(&wc->update_progress, 0,
7570 sizeof(wc->update_progress));
7572 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7573 memcpy(&wc->update_progress, &key,
7574 sizeof(wc->update_progress));
7576 level = root_item->drop_level;
7578 path->lowest_level = level;
7579 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7580 path->lowest_level = 0;
7588 * unlock our path, this is safe because only this
7589 * function is allowed to delete this snapshot
7591 btrfs_unlock_up_safe(path, 0);
7593 level = btrfs_header_level(root->node);
7595 btrfs_tree_lock(path->nodes[level]);
7596 btrfs_set_lock_blocking(path->nodes[level]);
7597 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7599 ret = btrfs_lookup_extent_info(trans, root,
7600 path->nodes[level]->start,
7601 level, 1, &wc->refs[level],
7607 BUG_ON(wc->refs[level] == 0);
7609 if (level == root_item->drop_level)
7612 btrfs_tree_unlock(path->nodes[level]);
7613 path->locks[level] = 0;
7614 WARN_ON(wc->refs[level] != 1);
7620 wc->shared_level = -1;
7621 wc->stage = DROP_REFERENCE;
7622 wc->update_ref = update_ref;
7624 wc->for_reloc = for_reloc;
7625 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7629 ret = walk_down_tree(trans, root, path, wc);
7635 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7642 BUG_ON(wc->stage != DROP_REFERENCE);
7646 if (wc->stage == DROP_REFERENCE) {
7648 btrfs_node_key(path->nodes[level],
7649 &root_item->drop_progress,
7650 path->slots[level]);
7651 root_item->drop_level = level;
7654 BUG_ON(wc->level == 0);
7655 if (btrfs_should_end_transaction(trans, tree_root) ||
7656 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7657 ret = btrfs_update_root(trans, tree_root,
7661 btrfs_abort_transaction(trans, tree_root, ret);
7666 btrfs_end_transaction_throttle(trans, tree_root);
7667 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7668 pr_debug("btrfs: drop snapshot early exit\n");
7673 trans = btrfs_start_transaction(tree_root, 0);
7674 if (IS_ERR(trans)) {
7675 err = PTR_ERR(trans);
7679 trans->block_rsv = block_rsv;
7682 btrfs_release_path(path);
7686 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7688 btrfs_abort_transaction(trans, tree_root, ret);
7692 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7693 ret = btrfs_find_root(tree_root, &root->root_key, path,
7696 btrfs_abort_transaction(trans, tree_root, ret);
7699 } else if (ret > 0) {
7700 /* if we fail to delete the orphan item this time
7701 * around, it'll get picked up the next time.
7703 * The most common failure here is just -ENOENT.
7705 btrfs_del_orphan_item(trans, tree_root,
7706 root->root_key.objectid);
7710 if (root->in_radix) {
7711 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7713 free_extent_buffer(root->node);
7714 free_extent_buffer(root->commit_root);
7715 btrfs_put_fs_root(root);
7717 root_dropped = true;
7719 btrfs_end_transaction_throttle(trans, tree_root);
7722 btrfs_free_path(path);
7725 * So if we need to stop dropping the snapshot for whatever reason we
7726 * need to make sure to add it back to the dead root list so that we
7727 * keep trying to do the work later. This also cleans up roots if we
7728 * don't have it in the radix (like when we recover after a power fail
7729 * or unmount) so we don't leak memory.
7731 if (!for_reloc && root_dropped == false)
7732 btrfs_add_dead_root(root);
7734 btrfs_std_error(root->fs_info, err);
7739 * drop subtree rooted at tree block 'node'.
7741 * NOTE: this function will unlock and release tree block 'node'
7742 * only used by relocation code
7744 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7745 struct btrfs_root *root,
7746 struct extent_buffer *node,
7747 struct extent_buffer *parent)
7749 struct btrfs_path *path;
7750 struct walk_control *wc;
7756 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7758 path = btrfs_alloc_path();
7762 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7764 btrfs_free_path(path);
7768 btrfs_assert_tree_locked(parent);
7769 parent_level = btrfs_header_level(parent);
7770 extent_buffer_get(parent);
7771 path->nodes[parent_level] = parent;
7772 path->slots[parent_level] = btrfs_header_nritems(parent);
7774 btrfs_assert_tree_locked(node);
7775 level = btrfs_header_level(node);
7776 path->nodes[level] = node;
7777 path->slots[level] = 0;
7778 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7780 wc->refs[parent_level] = 1;
7781 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7783 wc->shared_level = -1;
7784 wc->stage = DROP_REFERENCE;
7788 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7791 wret = walk_down_tree(trans, root, path, wc);
7797 wret = walk_up_tree(trans, root, path, wc, parent_level);
7805 btrfs_free_path(path);
7809 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7815 * if restripe for this chunk_type is on pick target profile and
7816 * return, otherwise do the usual balance
7818 stripped = get_restripe_target(root->fs_info, flags);
7820 return extended_to_chunk(stripped);
7823 * we add in the count of missing devices because we want
7824 * to make sure that any RAID levels on a degraded FS
7825 * continue to be honored.
7827 num_devices = root->fs_info->fs_devices->rw_devices +
7828 root->fs_info->fs_devices->missing_devices;
7830 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7831 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7832 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7834 if (num_devices == 1) {
7835 stripped |= BTRFS_BLOCK_GROUP_DUP;
7836 stripped = flags & ~stripped;
7838 /* turn raid0 into single device chunks */
7839 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7842 /* turn mirroring into duplication */
7843 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7844 BTRFS_BLOCK_GROUP_RAID10))
7845 return stripped | BTRFS_BLOCK_GROUP_DUP;
7847 /* they already had raid on here, just return */
7848 if (flags & stripped)
7851 stripped |= BTRFS_BLOCK_GROUP_DUP;
7852 stripped = flags & ~stripped;
7854 /* switch duplicated blocks with raid1 */
7855 if (flags & BTRFS_BLOCK_GROUP_DUP)
7856 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7858 /* this is drive concat, leave it alone */
7864 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7866 struct btrfs_space_info *sinfo = cache->space_info;
7868 u64 min_allocable_bytes;
7873 * We need some metadata space and system metadata space for
7874 * allocating chunks in some corner cases until we force to set
7875 * it to be readonly.
7878 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7880 min_allocable_bytes = 1 * 1024 * 1024;
7882 min_allocable_bytes = 0;
7884 spin_lock(&sinfo->lock);
7885 spin_lock(&cache->lock);
7892 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7893 cache->bytes_super - btrfs_block_group_used(&cache->item);
7895 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7896 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7897 min_allocable_bytes <= sinfo->total_bytes) {
7898 sinfo->bytes_readonly += num_bytes;
7903 spin_unlock(&cache->lock);
7904 spin_unlock(&sinfo->lock);
7908 int btrfs_set_block_group_ro(struct btrfs_root *root,
7909 struct btrfs_block_group_cache *cache)
7912 struct btrfs_trans_handle *trans;
7918 trans = btrfs_join_transaction(root);
7920 return PTR_ERR(trans);
7922 alloc_flags = update_block_group_flags(root, cache->flags);
7923 if (alloc_flags != cache->flags) {
7924 ret = do_chunk_alloc(trans, root, alloc_flags,
7930 ret = set_block_group_ro(cache, 0);
7933 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7934 ret = do_chunk_alloc(trans, root, alloc_flags,
7938 ret = set_block_group_ro(cache, 0);
7940 btrfs_end_transaction(trans, root);
7944 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7945 struct btrfs_root *root, u64 type)
7947 u64 alloc_flags = get_alloc_profile(root, type);
7948 return do_chunk_alloc(trans, root, alloc_flags,
7953 * helper to account the unused space of all the readonly block group in the
7954 * list. takes mirrors into account.
7956 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7958 struct btrfs_block_group_cache *block_group;
7962 list_for_each_entry(block_group, groups_list, list) {
7963 spin_lock(&block_group->lock);
7965 if (!block_group->ro) {
7966 spin_unlock(&block_group->lock);
7970 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7971 BTRFS_BLOCK_GROUP_RAID10 |
7972 BTRFS_BLOCK_GROUP_DUP))
7977 free_bytes += (block_group->key.offset -
7978 btrfs_block_group_used(&block_group->item)) *
7981 spin_unlock(&block_group->lock);
7988 * helper to account the unused space of all the readonly block group in the
7989 * space_info. takes mirrors into account.
7991 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7996 spin_lock(&sinfo->lock);
7998 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7999 if (!list_empty(&sinfo->block_groups[i]))
8000 free_bytes += __btrfs_get_ro_block_group_free_space(
8001 &sinfo->block_groups[i]);
8003 spin_unlock(&sinfo->lock);
8008 void btrfs_set_block_group_rw(struct btrfs_root *root,
8009 struct btrfs_block_group_cache *cache)
8011 struct btrfs_space_info *sinfo = cache->space_info;
8016 spin_lock(&sinfo->lock);
8017 spin_lock(&cache->lock);
8018 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8019 cache->bytes_super - btrfs_block_group_used(&cache->item);
8020 sinfo->bytes_readonly -= num_bytes;
8022 spin_unlock(&cache->lock);
8023 spin_unlock(&sinfo->lock);
8027 * checks to see if its even possible to relocate this block group.
8029 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8030 * ok to go ahead and try.
8032 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8034 struct btrfs_block_group_cache *block_group;
8035 struct btrfs_space_info *space_info;
8036 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8037 struct btrfs_device *device;
8038 struct btrfs_trans_handle *trans;
8047 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8049 /* odd, couldn't find the block group, leave it alone */
8053 min_free = btrfs_block_group_used(&block_group->item);
8055 /* no bytes used, we're good */
8059 space_info = block_group->space_info;
8060 spin_lock(&space_info->lock);
8062 full = space_info->full;
8065 * if this is the last block group we have in this space, we can't
8066 * relocate it unless we're able to allocate a new chunk below.
8068 * Otherwise, we need to make sure we have room in the space to handle
8069 * all of the extents from this block group. If we can, we're good
8071 if ((space_info->total_bytes != block_group->key.offset) &&
8072 (space_info->bytes_used + space_info->bytes_reserved +
8073 space_info->bytes_pinned + space_info->bytes_readonly +
8074 min_free < space_info->total_bytes)) {
8075 spin_unlock(&space_info->lock);
8078 spin_unlock(&space_info->lock);
8081 * ok we don't have enough space, but maybe we have free space on our
8082 * devices to allocate new chunks for relocation, so loop through our
8083 * alloc devices and guess if we have enough space. if this block
8084 * group is going to be restriped, run checks against the target
8085 * profile instead of the current one.
8097 target = get_restripe_target(root->fs_info, block_group->flags);
8099 index = __get_raid_index(extended_to_chunk(target));
8102 * this is just a balance, so if we were marked as full
8103 * we know there is no space for a new chunk
8108 index = get_block_group_index(block_group);
8111 if (index == BTRFS_RAID_RAID10) {
8115 } else if (index == BTRFS_RAID_RAID1) {
8117 } else if (index == BTRFS_RAID_DUP) {
8120 } else if (index == BTRFS_RAID_RAID0) {
8121 dev_min = fs_devices->rw_devices;
8122 do_div(min_free, dev_min);
8125 /* We need to do this so that we can look at pending chunks */
8126 trans = btrfs_join_transaction(root);
8127 if (IS_ERR(trans)) {
8128 ret = PTR_ERR(trans);
8132 mutex_lock(&root->fs_info->chunk_mutex);
8133 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8137 * check to make sure we can actually find a chunk with enough
8138 * space to fit our block group in.
8140 if (device->total_bytes > device->bytes_used + min_free &&
8141 !device->is_tgtdev_for_dev_replace) {
8142 ret = find_free_dev_extent(trans, device, min_free,
8147 if (dev_nr >= dev_min)
8153 mutex_unlock(&root->fs_info->chunk_mutex);
8154 btrfs_end_transaction(trans, root);
8156 btrfs_put_block_group(block_group);
8160 static int find_first_block_group(struct btrfs_root *root,
8161 struct btrfs_path *path, struct btrfs_key *key)
8164 struct btrfs_key found_key;
8165 struct extent_buffer *leaf;
8168 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8173 slot = path->slots[0];
8174 leaf = path->nodes[0];
8175 if (slot >= btrfs_header_nritems(leaf)) {
8176 ret = btrfs_next_leaf(root, path);
8183 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8185 if (found_key.objectid >= key->objectid &&
8186 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8196 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8198 struct btrfs_block_group_cache *block_group;
8202 struct inode *inode;
8204 block_group = btrfs_lookup_first_block_group(info, last);
8205 while (block_group) {
8206 spin_lock(&block_group->lock);
8207 if (block_group->iref)
8209 spin_unlock(&block_group->lock);
8210 block_group = next_block_group(info->tree_root,
8220 inode = block_group->inode;
8221 block_group->iref = 0;
8222 block_group->inode = NULL;
8223 spin_unlock(&block_group->lock);
8225 last = block_group->key.objectid + block_group->key.offset;
8226 btrfs_put_block_group(block_group);
8230 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8232 struct btrfs_block_group_cache *block_group;
8233 struct btrfs_space_info *space_info;
8234 struct btrfs_caching_control *caching_ctl;
8237 down_write(&info->extent_commit_sem);
8238 while (!list_empty(&info->caching_block_groups)) {
8239 caching_ctl = list_entry(info->caching_block_groups.next,
8240 struct btrfs_caching_control, list);
8241 list_del(&caching_ctl->list);
8242 put_caching_control(caching_ctl);
8244 up_write(&info->extent_commit_sem);
8246 spin_lock(&info->block_group_cache_lock);
8247 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8248 block_group = rb_entry(n, struct btrfs_block_group_cache,
8250 rb_erase(&block_group->cache_node,
8251 &info->block_group_cache_tree);
8252 spin_unlock(&info->block_group_cache_lock);
8254 down_write(&block_group->space_info->groups_sem);
8255 list_del(&block_group->list);
8256 up_write(&block_group->space_info->groups_sem);
8258 if (block_group->cached == BTRFS_CACHE_STARTED)
8259 wait_block_group_cache_done(block_group);
8262 * We haven't cached this block group, which means we could
8263 * possibly have excluded extents on this block group.
8265 if (block_group->cached == BTRFS_CACHE_NO ||
8266 block_group->cached == BTRFS_CACHE_ERROR)
8267 free_excluded_extents(info->extent_root, block_group);
8269 btrfs_remove_free_space_cache(block_group);
8270 btrfs_put_block_group(block_group);
8272 spin_lock(&info->block_group_cache_lock);
8274 spin_unlock(&info->block_group_cache_lock);
8276 /* now that all the block groups are freed, go through and
8277 * free all the space_info structs. This is only called during
8278 * the final stages of unmount, and so we know nobody is
8279 * using them. We call synchronize_rcu() once before we start,
8280 * just to be on the safe side.
8284 release_global_block_rsv(info);
8286 while(!list_empty(&info->space_info)) {
8287 space_info = list_entry(info->space_info.next,
8288 struct btrfs_space_info,
8290 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8291 if (space_info->bytes_pinned > 0 ||
8292 space_info->bytes_reserved > 0 ||
8293 space_info->bytes_may_use > 0) {
8295 dump_space_info(space_info, 0, 0);
8298 percpu_counter_destroy(&space_info->total_bytes_pinned);
8299 list_del(&space_info->list);
8305 static void __link_block_group(struct btrfs_space_info *space_info,
8306 struct btrfs_block_group_cache *cache)
8308 int index = get_block_group_index(cache);
8310 down_write(&space_info->groups_sem);
8311 list_add_tail(&cache->list, &space_info->block_groups[index]);
8312 up_write(&space_info->groups_sem);
8315 int btrfs_read_block_groups(struct btrfs_root *root)
8317 struct btrfs_path *path;
8319 struct btrfs_block_group_cache *cache;
8320 struct btrfs_fs_info *info = root->fs_info;
8321 struct btrfs_space_info *space_info;
8322 struct btrfs_key key;
8323 struct btrfs_key found_key;
8324 struct extent_buffer *leaf;
8328 root = info->extent_root;
8331 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8332 path = btrfs_alloc_path();
8337 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8338 if (btrfs_test_opt(root, SPACE_CACHE) &&
8339 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8341 if (btrfs_test_opt(root, CLEAR_CACHE))
8345 ret = find_first_block_group(root, path, &key);
8350 leaf = path->nodes[0];
8351 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8352 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8357 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8359 if (!cache->free_space_ctl) {
8365 atomic_set(&cache->count, 1);
8366 spin_lock_init(&cache->lock);
8367 cache->fs_info = info;
8368 INIT_LIST_HEAD(&cache->list);
8369 INIT_LIST_HEAD(&cache->cluster_list);
8373 * When we mount with old space cache, we need to
8374 * set BTRFS_DC_CLEAR and set dirty flag.
8376 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8377 * truncate the old free space cache inode and
8379 * b) Setting 'dirty flag' makes sure that we flush
8380 * the new space cache info onto disk.
8382 cache->disk_cache_state = BTRFS_DC_CLEAR;
8383 if (btrfs_test_opt(root, SPACE_CACHE))
8387 read_extent_buffer(leaf, &cache->item,
8388 btrfs_item_ptr_offset(leaf, path->slots[0]),
8389 sizeof(cache->item));
8390 memcpy(&cache->key, &found_key, sizeof(found_key));
8392 key.objectid = found_key.objectid + found_key.offset;
8393 btrfs_release_path(path);
8394 cache->flags = btrfs_block_group_flags(&cache->item);
8395 cache->sectorsize = root->sectorsize;
8396 cache->full_stripe_len = btrfs_full_stripe_len(root,
8397 &root->fs_info->mapping_tree,
8398 found_key.objectid);
8399 btrfs_init_free_space_ctl(cache);
8402 * We need to exclude the super stripes now so that the space
8403 * info has super bytes accounted for, otherwise we'll think
8404 * we have more space than we actually do.
8406 ret = exclude_super_stripes(root, cache);
8409 * We may have excluded something, so call this just in
8412 free_excluded_extents(root, cache);
8413 kfree(cache->free_space_ctl);
8419 * check for two cases, either we are full, and therefore
8420 * don't need to bother with the caching work since we won't
8421 * find any space, or we are empty, and we can just add all
8422 * the space in and be done with it. This saves us _alot_ of
8423 * time, particularly in the full case.
8425 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8426 cache->last_byte_to_unpin = (u64)-1;
8427 cache->cached = BTRFS_CACHE_FINISHED;
8428 free_excluded_extents(root, cache);
8429 } else if (btrfs_block_group_used(&cache->item) == 0) {
8430 cache->last_byte_to_unpin = (u64)-1;
8431 cache->cached = BTRFS_CACHE_FINISHED;
8432 add_new_free_space(cache, root->fs_info,
8434 found_key.objectid +
8436 free_excluded_extents(root, cache);
8439 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8441 btrfs_remove_free_space_cache(cache);
8442 btrfs_put_block_group(cache);
8446 ret = update_space_info(info, cache->flags, found_key.offset,
8447 btrfs_block_group_used(&cache->item),
8450 btrfs_remove_free_space_cache(cache);
8451 spin_lock(&info->block_group_cache_lock);
8452 rb_erase(&cache->cache_node,
8453 &info->block_group_cache_tree);
8454 spin_unlock(&info->block_group_cache_lock);
8455 btrfs_put_block_group(cache);
8459 cache->space_info = space_info;
8460 spin_lock(&cache->space_info->lock);
8461 cache->space_info->bytes_readonly += cache->bytes_super;
8462 spin_unlock(&cache->space_info->lock);
8464 __link_block_group(space_info, cache);
8466 set_avail_alloc_bits(root->fs_info, cache->flags);
8467 if (btrfs_chunk_readonly(root, cache->key.objectid))
8468 set_block_group_ro(cache, 1);
8471 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8472 if (!(get_alloc_profile(root, space_info->flags) &
8473 (BTRFS_BLOCK_GROUP_RAID10 |
8474 BTRFS_BLOCK_GROUP_RAID1 |
8475 BTRFS_BLOCK_GROUP_RAID5 |
8476 BTRFS_BLOCK_GROUP_RAID6 |
8477 BTRFS_BLOCK_GROUP_DUP)))
8480 * avoid allocating from un-mirrored block group if there are
8481 * mirrored block groups.
8483 list_for_each_entry(cache,
8484 &space_info->block_groups[BTRFS_RAID_RAID0],
8486 set_block_group_ro(cache, 1);
8487 list_for_each_entry(cache,
8488 &space_info->block_groups[BTRFS_RAID_SINGLE],
8490 set_block_group_ro(cache, 1);
8493 init_global_block_rsv(info);
8496 btrfs_free_path(path);
8500 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8501 struct btrfs_root *root)
8503 struct btrfs_block_group_cache *block_group, *tmp;
8504 struct btrfs_root *extent_root = root->fs_info->extent_root;
8505 struct btrfs_block_group_item item;
8506 struct btrfs_key key;
8509 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8511 list_del_init(&block_group->new_bg_list);
8516 spin_lock(&block_group->lock);
8517 memcpy(&item, &block_group->item, sizeof(item));
8518 memcpy(&key, &block_group->key, sizeof(key));
8519 spin_unlock(&block_group->lock);
8521 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8524 btrfs_abort_transaction(trans, extent_root, ret);
8525 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8526 key.objectid, key.offset);
8528 btrfs_abort_transaction(trans, extent_root, ret);
8532 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8533 struct btrfs_root *root, u64 bytes_used,
8534 u64 type, u64 chunk_objectid, u64 chunk_offset,
8538 struct btrfs_root *extent_root;
8539 struct btrfs_block_group_cache *cache;
8541 extent_root = root->fs_info->extent_root;
8543 root->fs_info->last_trans_log_full_commit = trans->transid;
8545 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8548 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8550 if (!cache->free_space_ctl) {
8555 cache->key.objectid = chunk_offset;
8556 cache->key.offset = size;
8557 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8558 cache->sectorsize = root->sectorsize;
8559 cache->fs_info = root->fs_info;
8560 cache->full_stripe_len = btrfs_full_stripe_len(root,
8561 &root->fs_info->mapping_tree,
8564 atomic_set(&cache->count, 1);
8565 spin_lock_init(&cache->lock);
8566 INIT_LIST_HEAD(&cache->list);
8567 INIT_LIST_HEAD(&cache->cluster_list);
8568 INIT_LIST_HEAD(&cache->new_bg_list);
8570 btrfs_init_free_space_ctl(cache);
8572 btrfs_set_block_group_used(&cache->item, bytes_used);
8573 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8574 cache->flags = type;
8575 btrfs_set_block_group_flags(&cache->item, type);
8577 cache->last_byte_to_unpin = (u64)-1;
8578 cache->cached = BTRFS_CACHE_FINISHED;
8579 ret = exclude_super_stripes(root, cache);
8582 * We may have excluded something, so call this just in
8585 free_excluded_extents(root, cache);
8586 kfree(cache->free_space_ctl);
8591 add_new_free_space(cache, root->fs_info, chunk_offset,
8592 chunk_offset + size);
8594 free_excluded_extents(root, cache);
8596 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8598 btrfs_remove_free_space_cache(cache);
8599 btrfs_put_block_group(cache);
8603 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8604 &cache->space_info);
8606 btrfs_remove_free_space_cache(cache);
8607 spin_lock(&root->fs_info->block_group_cache_lock);
8608 rb_erase(&cache->cache_node,
8609 &root->fs_info->block_group_cache_tree);
8610 spin_unlock(&root->fs_info->block_group_cache_lock);
8611 btrfs_put_block_group(cache);
8614 update_global_block_rsv(root->fs_info);
8616 spin_lock(&cache->space_info->lock);
8617 cache->space_info->bytes_readonly += cache->bytes_super;
8618 spin_unlock(&cache->space_info->lock);
8620 __link_block_group(cache->space_info, cache);
8622 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8624 set_avail_alloc_bits(extent_root->fs_info, type);
8629 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8631 u64 extra_flags = chunk_to_extended(flags) &
8632 BTRFS_EXTENDED_PROFILE_MASK;
8634 write_seqlock(&fs_info->profiles_lock);
8635 if (flags & BTRFS_BLOCK_GROUP_DATA)
8636 fs_info->avail_data_alloc_bits &= ~extra_flags;
8637 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8638 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8639 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8640 fs_info->avail_system_alloc_bits &= ~extra_flags;
8641 write_sequnlock(&fs_info->profiles_lock);
8644 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8645 struct btrfs_root *root, u64 group_start)
8647 struct btrfs_path *path;
8648 struct btrfs_block_group_cache *block_group;
8649 struct btrfs_free_cluster *cluster;
8650 struct btrfs_root *tree_root = root->fs_info->tree_root;
8651 struct btrfs_key key;
8652 struct inode *inode;
8657 root = root->fs_info->extent_root;
8659 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8660 BUG_ON(!block_group);
8661 BUG_ON(!block_group->ro);
8664 * Free the reserved super bytes from this block group before
8667 free_excluded_extents(root, block_group);
8669 memcpy(&key, &block_group->key, sizeof(key));
8670 index = get_block_group_index(block_group);
8671 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8672 BTRFS_BLOCK_GROUP_RAID1 |
8673 BTRFS_BLOCK_GROUP_RAID10))
8678 /* make sure this block group isn't part of an allocation cluster */
8679 cluster = &root->fs_info->data_alloc_cluster;
8680 spin_lock(&cluster->refill_lock);
8681 btrfs_return_cluster_to_free_space(block_group, cluster);
8682 spin_unlock(&cluster->refill_lock);
8685 * make sure this block group isn't part of a metadata
8686 * allocation cluster
8688 cluster = &root->fs_info->meta_alloc_cluster;
8689 spin_lock(&cluster->refill_lock);
8690 btrfs_return_cluster_to_free_space(block_group, cluster);
8691 spin_unlock(&cluster->refill_lock);
8693 path = btrfs_alloc_path();
8699 inode = lookup_free_space_inode(tree_root, block_group, path);
8700 if (!IS_ERR(inode)) {
8701 ret = btrfs_orphan_add(trans, inode);
8703 btrfs_add_delayed_iput(inode);
8707 /* One for the block groups ref */
8708 spin_lock(&block_group->lock);
8709 if (block_group->iref) {
8710 block_group->iref = 0;
8711 block_group->inode = NULL;
8712 spin_unlock(&block_group->lock);
8715 spin_unlock(&block_group->lock);
8717 /* One for our lookup ref */
8718 btrfs_add_delayed_iput(inode);
8721 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8722 key.offset = block_group->key.objectid;
8725 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8729 btrfs_release_path(path);
8731 ret = btrfs_del_item(trans, tree_root, path);
8734 btrfs_release_path(path);
8737 spin_lock(&root->fs_info->block_group_cache_lock);
8738 rb_erase(&block_group->cache_node,
8739 &root->fs_info->block_group_cache_tree);
8741 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8742 root->fs_info->first_logical_byte = (u64)-1;
8743 spin_unlock(&root->fs_info->block_group_cache_lock);
8745 down_write(&block_group->space_info->groups_sem);
8747 * we must use list_del_init so people can check to see if they
8748 * are still on the list after taking the semaphore
8750 list_del_init(&block_group->list);
8751 if (list_empty(&block_group->space_info->block_groups[index]))
8752 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8753 up_write(&block_group->space_info->groups_sem);
8755 if (block_group->cached == BTRFS_CACHE_STARTED)
8756 wait_block_group_cache_done(block_group);
8758 btrfs_remove_free_space_cache(block_group);
8760 spin_lock(&block_group->space_info->lock);
8761 block_group->space_info->total_bytes -= block_group->key.offset;
8762 block_group->space_info->bytes_readonly -= block_group->key.offset;
8763 block_group->space_info->disk_total -= block_group->key.offset * factor;
8764 spin_unlock(&block_group->space_info->lock);
8766 memcpy(&key, &block_group->key, sizeof(key));
8768 btrfs_clear_space_info_full(root->fs_info);
8770 btrfs_put_block_group(block_group);
8771 btrfs_put_block_group(block_group);
8773 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8779 ret = btrfs_del_item(trans, root, path);
8781 btrfs_free_path(path);
8785 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8787 struct btrfs_space_info *space_info;
8788 struct btrfs_super_block *disk_super;
8794 disk_super = fs_info->super_copy;
8795 if (!btrfs_super_root(disk_super))
8798 features = btrfs_super_incompat_flags(disk_super);
8799 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8802 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8803 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8808 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8809 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8811 flags = BTRFS_BLOCK_GROUP_METADATA;
8812 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8816 flags = BTRFS_BLOCK_GROUP_DATA;
8817 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8823 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8825 return unpin_extent_range(root, start, end);
8828 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8829 u64 num_bytes, u64 *actual_bytes)
8831 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8834 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8836 struct btrfs_fs_info *fs_info = root->fs_info;
8837 struct btrfs_block_group_cache *cache = NULL;
8842 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8846 * try to trim all FS space, our block group may start from non-zero.
8848 if (range->len == total_bytes)
8849 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8851 cache = btrfs_lookup_block_group(fs_info, range->start);
8854 if (cache->key.objectid >= (range->start + range->len)) {
8855 btrfs_put_block_group(cache);
8859 start = max(range->start, cache->key.objectid);
8860 end = min(range->start + range->len,
8861 cache->key.objectid + cache->key.offset);
8863 if (end - start >= range->minlen) {
8864 if (!block_group_cache_done(cache)) {
8865 ret = cache_block_group(cache, 0);
8867 btrfs_put_block_group(cache);
8870 ret = wait_block_group_cache_done(cache);
8872 btrfs_put_block_group(cache);
8876 ret = btrfs_trim_block_group(cache,
8882 trimmed += group_trimmed;
8884 btrfs_put_block_group(cache);
8889 cache = next_block_group(fs_info->tree_root, cache);
8892 range->len = trimmed;