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>
32 #include "print-tree.h"
33 #include "transaction.h"
37 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
109 int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
135 kfree(cache->free_space_ctl);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
158 if (block_group->key.objectid < cache->key.objectid) {
160 } else if (block_group->key.objectid > cache->key.objectid) {
163 spin_unlock(&info->block_group_cache_lock);
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
188 struct btrfs_block_group_cache *cache, *ret = NULL;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
196 cache = rb_entry(n, struct btrfs_block_group_cache,
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
226 static int add_excluded_extent(struct btrfs_root *root,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&root->fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 set_extent_bits(&root->fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static void free_excluded_extents(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&root->fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 clear_extent_bits(&root->fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 static int exclude_super_stripes(struct btrfs_root *root,
252 struct btrfs_block_group_cache *cache)
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
271 cache->key.objectid, bytenr,
272 0, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(root, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (cache->cached != BTRFS_CACHE_STARTED) {
316 spin_unlock(&cache->lock);
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache->caching_ctl) {
322 spin_unlock(&cache->lock);
326 ctl = cache->caching_ctl;
327 atomic_inc(&ctl->count);
328 spin_unlock(&cache->lock);
332 static void put_caching_control(struct btrfs_caching_control *ctl)
334 if (atomic_dec_and_test(&ctl->count))
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
344 struct btrfs_fs_info *info, u64 start, u64 end)
346 u64 extent_start, extent_end, size, total_added = 0;
349 while (start < end) {
350 ret = find_first_extent_bit(info->pinned_extents, start,
351 &extent_start, &extent_end,
352 EXTENT_DIRTY | EXTENT_UPTODATE,
357 if (extent_start <= start) {
358 start = extent_end + 1;
359 } else if (extent_start > start && extent_start < end) {
360 size = extent_start - start;
362 ret = btrfs_add_free_space(block_group, start,
364 BUG_ON(ret); /* -ENOMEM or logic error */
365 start = extent_end + 1;
374 ret = btrfs_add_free_space(block_group, start, size);
375 BUG_ON(ret); /* -ENOMEM or logic error */
381 static noinline void caching_thread(struct btrfs_work *work)
383 struct btrfs_block_group_cache *block_group;
384 struct btrfs_fs_info *fs_info;
385 struct btrfs_caching_control *caching_ctl;
386 struct btrfs_root *extent_root;
387 struct btrfs_path *path;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
395 caching_ctl = container_of(work, struct btrfs_caching_control, work);
396 block_group = caching_ctl->block_group;
397 fs_info = block_group->fs_info;
398 extent_root = fs_info->extent_root;
400 path = btrfs_alloc_path();
404 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path->skip_locking = 1;
413 path->search_commit_root = 1;
418 key.type = BTRFS_EXTENT_ITEM_KEY;
420 mutex_lock(&caching_ctl->mutex);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info->extent_commit_sem);
425 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
429 leaf = path->nodes[0];
430 nritems = btrfs_header_nritems(leaf);
433 if (btrfs_fs_closing(fs_info) > 1) {
438 if (path->slots[0] < nritems) {
439 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
441 ret = find_next_key(path, 0, &key);
445 if (need_resched()) {
446 caching_ctl->progress = last;
447 btrfs_release_path(path);
448 up_read(&fs_info->extent_commit_sem);
449 mutex_unlock(&caching_ctl->mutex);
454 ret = btrfs_next_leaf(extent_root, path);
459 leaf = path->nodes[0];
460 nritems = btrfs_header_nritems(leaf);
464 if (key.objectid < last) {
467 key.type = BTRFS_EXTENT_ITEM_KEY;
469 caching_ctl->progress = last;
470 btrfs_release_path(path);
474 if (key.objectid < block_group->key.objectid) {
479 if (key.objectid >= block_group->key.objectid +
480 block_group->key.offset)
483 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
484 key.type == BTRFS_METADATA_ITEM_KEY) {
485 total_found += add_new_free_space(block_group,
488 if (key.type == BTRFS_METADATA_ITEM_KEY)
489 last = key.objectid +
490 fs_info->tree_root->leafsize;
492 last = key.objectid + key.offset;
494 if (total_found > (1024 * 1024 * 2)) {
496 wake_up(&caching_ctl->wait);
503 total_found += add_new_free_space(block_group, fs_info, last,
504 block_group->key.objectid +
505 block_group->key.offset);
506 caching_ctl->progress = (u64)-1;
508 spin_lock(&block_group->lock);
509 block_group->caching_ctl = NULL;
510 block_group->cached = BTRFS_CACHE_FINISHED;
511 spin_unlock(&block_group->lock);
514 btrfs_free_path(path);
515 up_read(&fs_info->extent_commit_sem);
517 free_excluded_extents(extent_root, block_group);
519 mutex_unlock(&caching_ctl->mutex);
522 spin_lock(&block_group->lock);
523 block_group->caching_ctl = NULL;
524 block_group->cached = BTRFS_CACHE_ERROR;
525 spin_unlock(&block_group->lock);
527 wake_up(&caching_ctl->wait);
529 put_caching_control(caching_ctl);
530 btrfs_put_block_group(block_group);
533 static int cache_block_group(struct btrfs_block_group_cache *cache,
537 struct btrfs_fs_info *fs_info = cache->fs_info;
538 struct btrfs_caching_control *caching_ctl;
541 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
545 INIT_LIST_HEAD(&caching_ctl->list);
546 mutex_init(&caching_ctl->mutex);
547 init_waitqueue_head(&caching_ctl->wait);
548 caching_ctl->block_group = cache;
549 caching_ctl->progress = cache->key.objectid;
550 atomic_set(&caching_ctl->count, 1);
551 caching_ctl->work.func = caching_thread;
553 spin_lock(&cache->lock);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache->cached == BTRFS_CACHE_FAST) {
567 struct btrfs_caching_control *ctl;
569 ctl = cache->caching_ctl;
570 atomic_inc(&ctl->count);
571 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 spin_unlock(&cache->lock);
576 finish_wait(&ctl->wait, &wait);
577 put_caching_control(ctl);
578 spin_lock(&cache->lock);
581 if (cache->cached != BTRFS_CACHE_NO) {
582 spin_unlock(&cache->lock);
586 WARN_ON(cache->caching_ctl);
587 cache->caching_ctl = caching_ctl;
588 cache->cached = BTRFS_CACHE_FAST;
589 spin_unlock(&cache->lock);
591 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
600 if (load_cache_only) {
601 cache->caching_ctl = NULL;
602 cache->cached = BTRFS_CACHE_NO;
604 cache->cached = BTRFS_CACHE_STARTED;
607 spin_unlock(&cache->lock);
608 wake_up(&caching_ctl->wait);
610 put_caching_control(caching_ctl);
611 free_excluded_extents(fs_info->extent_root, cache);
616 * We are not going to do the fast caching, set cached to the
617 * appropriate value and wakeup any waiters.
619 spin_lock(&cache->lock);
620 if (load_cache_only) {
621 cache->caching_ctl = NULL;
622 cache->cached = BTRFS_CACHE_NO;
624 cache->cached = BTRFS_CACHE_STARTED;
626 spin_unlock(&cache->lock);
627 wake_up(&caching_ctl->wait);
630 if (load_cache_only) {
631 put_caching_control(caching_ctl);
635 down_write(&fs_info->extent_commit_sem);
636 atomic_inc(&caching_ctl->count);
637 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
638 up_write(&fs_info->extent_commit_sem);
640 btrfs_get_block_group(cache);
642 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
648 * return the block group that starts at or after bytenr
650 static struct btrfs_block_group_cache *
651 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
653 struct btrfs_block_group_cache *cache;
655 cache = block_group_cache_tree_search(info, bytenr, 0);
661 * return the block group that contains the given bytenr
663 struct btrfs_block_group_cache *btrfs_lookup_block_group(
664 struct btrfs_fs_info *info,
667 struct btrfs_block_group_cache *cache;
669 cache = block_group_cache_tree_search(info, bytenr, 1);
674 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
677 struct list_head *head = &info->space_info;
678 struct btrfs_space_info *found;
680 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
683 list_for_each_entry_rcu(found, head, list) {
684 if (found->flags & flags) {
694 * after adding space to the filesystem, we need to clear the full flags
695 * on all the space infos.
697 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
699 struct list_head *head = &info->space_info;
700 struct btrfs_space_info *found;
703 list_for_each_entry_rcu(found, head, list)
708 /* simple helper to search for an existing extent at a given offset */
709 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
712 struct btrfs_key key;
713 struct btrfs_path *path;
715 path = btrfs_alloc_path();
719 key.objectid = start;
721 key.type = BTRFS_EXTENT_ITEM_KEY;
722 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
725 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
726 if (key.objectid == start &&
727 key.type == BTRFS_METADATA_ITEM_KEY)
730 btrfs_free_path(path);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 offset, int metadata, u64 *refs, u64 *flags)
747 struct btrfs_delayed_ref_head *head;
748 struct btrfs_delayed_ref_root *delayed_refs;
749 struct btrfs_path *path;
750 struct btrfs_extent_item *ei;
751 struct extent_buffer *leaf;
752 struct btrfs_key key;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 offset = root->leafsize;
767 path = btrfs_alloc_path();
772 key.objectid = bytenr;
773 key.type = BTRFS_METADATA_ITEM_KEY;
776 key.objectid = bytenr;
777 key.type = BTRFS_EXTENT_ITEM_KEY;
782 path->skip_locking = 1;
783 path->search_commit_root = 1;
786 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
791 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
793 if (path->slots[0]) {
795 btrfs_item_key_to_cpu(path->nodes[0], &key,
797 if (key.objectid == bytenr &&
798 key.type == BTRFS_EXTENT_ITEM_KEY &&
799 key.offset == root->leafsize)
803 key.objectid = bytenr;
804 key.type = BTRFS_EXTENT_ITEM_KEY;
805 key.offset = root->leafsize;
806 btrfs_release_path(path);
812 leaf = path->nodes[0];
813 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
814 if (item_size >= sizeof(*ei)) {
815 ei = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_extent_item);
817 num_refs = btrfs_extent_refs(leaf, ei);
818 extent_flags = btrfs_extent_flags(leaf, ei);
820 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
821 struct btrfs_extent_item_v0 *ei0;
822 BUG_ON(item_size != sizeof(*ei0));
823 ei0 = btrfs_item_ptr(leaf, path->slots[0],
824 struct btrfs_extent_item_v0);
825 num_refs = btrfs_extent_refs_v0(leaf, ei0);
826 /* FIXME: this isn't correct for data */
827 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
832 BUG_ON(num_refs == 0);
842 delayed_refs = &trans->transaction->delayed_refs;
843 spin_lock(&delayed_refs->lock);
844 head = btrfs_find_delayed_ref_head(trans, bytenr);
846 if (!mutex_trylock(&head->mutex)) {
847 atomic_inc(&head->node.refs);
848 spin_unlock(&delayed_refs->lock);
850 btrfs_release_path(path);
853 * Mutex was contended, block until it's released and try
856 mutex_lock(&head->mutex);
857 mutex_unlock(&head->mutex);
858 btrfs_put_delayed_ref(&head->node);
861 if (head->extent_op && head->extent_op->update_flags)
862 extent_flags |= head->extent_op->flags_to_set;
864 BUG_ON(num_refs == 0);
866 num_refs += head->node.ref_mod;
867 mutex_unlock(&head->mutex);
869 spin_unlock(&delayed_refs->lock);
871 WARN_ON(num_refs == 0);
875 *flags = extent_flags;
877 btrfs_free_path(path);
882 * Back reference rules. Back refs have three main goals:
884 * 1) differentiate between all holders of references to an extent so that
885 * when a reference is dropped we can make sure it was a valid reference
886 * before freeing the extent.
888 * 2) Provide enough information to quickly find the holders of an extent
889 * if we notice a given block is corrupted or bad.
891 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
892 * maintenance. This is actually the same as #2, but with a slightly
893 * different use case.
895 * There are two kinds of back refs. The implicit back refs is optimized
896 * for pointers in non-shared tree blocks. For a given pointer in a block,
897 * back refs of this kind provide information about the block's owner tree
898 * and the pointer's key. These information allow us to find the block by
899 * b-tree searching. The full back refs is for pointers in tree blocks not
900 * referenced by their owner trees. The location of tree block is recorded
901 * in the back refs. Actually the full back refs is generic, and can be
902 * used in all cases the implicit back refs is used. The major shortcoming
903 * of the full back refs is its overhead. Every time a tree block gets
904 * COWed, we have to update back refs entry for all pointers in it.
906 * For a newly allocated tree block, we use implicit back refs for
907 * pointers in it. This means most tree related operations only involve
908 * implicit back refs. For a tree block created in old transaction, the
909 * only way to drop a reference to it is COW it. So we can detect the
910 * event that tree block loses its owner tree's reference and do the
911 * back refs conversion.
913 * When a tree block is COW'd through a tree, there are four cases:
915 * The reference count of the block is one and the tree is the block's
916 * owner tree. Nothing to do in this case.
918 * The reference count of the block is one and the tree is not the
919 * block's owner tree. In this case, full back refs is used for pointers
920 * in the block. Remove these full back refs, add implicit back refs for
921 * every pointers in the new block.
923 * The reference count of the block is greater than one and the tree is
924 * the block's owner tree. In this case, implicit back refs is used for
925 * pointers in the block. Add full back refs for every pointers in the
926 * block, increase lower level extents' reference counts. The original
927 * implicit back refs are entailed to the new block.
929 * The reference count of the block is greater than one and the tree is
930 * not the block's owner tree. Add implicit back refs for every pointer in
931 * the new block, increase lower level extents' reference count.
933 * Back Reference Key composing:
935 * The key objectid corresponds to the first byte in the extent,
936 * The key type is used to differentiate between types of back refs.
937 * There are different meanings of the key offset for different types
940 * File extents can be referenced by:
942 * - multiple snapshots, subvolumes, or different generations in one subvol
943 * - different files inside a single subvolume
944 * - different offsets inside a file (bookend extents in file.c)
946 * The extent ref structure for the implicit back refs has fields for:
948 * - Objectid of the subvolume root
949 * - objectid of the file holding the reference
950 * - original offset in the file
951 * - how many bookend extents
953 * The key offset for the implicit back refs is hash of the first
956 * The extent ref structure for the full back refs has field for:
958 * - number of pointers in the tree leaf
960 * The key offset for the implicit back refs is the first byte of
963 * When a file extent is allocated, The implicit back refs is used.
964 * the fields are filled in:
966 * (root_key.objectid, inode objectid, offset in file, 1)
968 * When a file extent is removed file truncation, we find the
969 * corresponding implicit back refs and check the following fields:
971 * (btrfs_header_owner(leaf), inode objectid, offset in file)
973 * Btree extents can be referenced by:
975 * - Different subvolumes
977 * Both the implicit back refs and the full back refs for tree blocks
978 * only consist of key. The key offset for the implicit back refs is
979 * objectid of block's owner tree. The key offset for the full back refs
980 * is the first byte of parent block.
982 * When implicit back refs is used, information about the lowest key and
983 * level of the tree block are required. These information are stored in
984 * tree block info structure.
987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
988 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root,
990 struct btrfs_path *path,
991 u64 owner, u32 extra_size)
993 struct btrfs_extent_item *item;
994 struct btrfs_extent_item_v0 *ei0;
995 struct btrfs_extent_ref_v0 *ref0;
996 struct btrfs_tree_block_info *bi;
997 struct extent_buffer *leaf;
998 struct btrfs_key key;
999 struct btrfs_key found_key;
1000 u32 new_size = sizeof(*item);
1004 leaf = path->nodes[0];
1005 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1008 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1009 struct btrfs_extent_item_v0);
1010 refs = btrfs_extent_refs_v0(leaf, ei0);
1012 if (owner == (u64)-1) {
1014 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1015 ret = btrfs_next_leaf(root, path);
1018 BUG_ON(ret > 0); /* Corruption */
1019 leaf = path->nodes[0];
1021 btrfs_item_key_to_cpu(leaf, &found_key,
1023 BUG_ON(key.objectid != found_key.objectid);
1024 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1028 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1029 struct btrfs_extent_ref_v0);
1030 owner = btrfs_ref_objectid_v0(leaf, ref0);
1034 btrfs_release_path(path);
1036 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1037 new_size += sizeof(*bi);
1039 new_size -= sizeof(*ei0);
1040 ret = btrfs_search_slot(trans, root, &key, path,
1041 new_size + extra_size, 1);
1044 BUG_ON(ret); /* Corruption */
1046 btrfs_extend_item(root, path, new_size);
1048 leaf = path->nodes[0];
1049 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1050 btrfs_set_extent_refs(leaf, item, refs);
1051 /* FIXME: get real generation */
1052 btrfs_set_extent_generation(leaf, item, 0);
1053 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1054 btrfs_set_extent_flags(leaf, item,
1055 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1056 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1057 bi = (struct btrfs_tree_block_info *)(item + 1);
1058 /* FIXME: get first key of the block */
1059 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1060 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1062 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1064 btrfs_mark_buffer_dirty(leaf);
1069 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1071 u32 high_crc = ~(u32)0;
1072 u32 low_crc = ~(u32)0;
1075 lenum = cpu_to_le64(root_objectid);
1076 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1077 lenum = cpu_to_le64(owner);
1078 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1079 lenum = cpu_to_le64(offset);
1080 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1082 return ((u64)high_crc << 31) ^ (u64)low_crc;
1085 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref)
1088 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1089 btrfs_extent_data_ref_objectid(leaf, ref),
1090 btrfs_extent_data_ref_offset(leaf, ref));
1093 static int match_extent_data_ref(struct extent_buffer *leaf,
1094 struct btrfs_extent_data_ref *ref,
1095 u64 root_objectid, u64 owner, u64 offset)
1097 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1098 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1099 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1104 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1105 struct btrfs_root *root,
1106 struct btrfs_path *path,
1107 u64 bytenr, u64 parent,
1109 u64 owner, u64 offset)
1111 struct btrfs_key key;
1112 struct btrfs_extent_data_ref *ref;
1113 struct extent_buffer *leaf;
1119 key.objectid = bytenr;
1121 key.type = BTRFS_SHARED_DATA_REF_KEY;
1122 key.offset = parent;
1124 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1125 key.offset = hash_extent_data_ref(root_objectid,
1130 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1139 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1140 key.type = BTRFS_EXTENT_REF_V0_KEY;
1141 btrfs_release_path(path);
1142 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1153 leaf = path->nodes[0];
1154 nritems = btrfs_header_nritems(leaf);
1156 if (path->slots[0] >= nritems) {
1157 ret = btrfs_next_leaf(root, path);
1163 leaf = path->nodes[0];
1164 nritems = btrfs_header_nritems(leaf);
1168 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1169 if (key.objectid != bytenr ||
1170 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1173 ref = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_extent_data_ref);
1176 if (match_extent_data_ref(leaf, ref, root_objectid,
1179 btrfs_release_path(path);
1191 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1192 struct btrfs_root *root,
1193 struct btrfs_path *path,
1194 u64 bytenr, u64 parent,
1195 u64 root_objectid, u64 owner,
1196 u64 offset, int refs_to_add)
1198 struct btrfs_key key;
1199 struct extent_buffer *leaf;
1204 key.objectid = bytenr;
1206 key.type = BTRFS_SHARED_DATA_REF_KEY;
1207 key.offset = parent;
1208 size = sizeof(struct btrfs_shared_data_ref);
1210 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1211 key.offset = hash_extent_data_ref(root_objectid,
1213 size = sizeof(struct btrfs_extent_data_ref);
1216 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1217 if (ret && ret != -EEXIST)
1220 leaf = path->nodes[0];
1222 struct btrfs_shared_data_ref *ref;
1223 ref = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_shared_data_ref);
1226 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1228 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1229 num_refs += refs_to_add;
1230 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1233 struct btrfs_extent_data_ref *ref;
1234 while (ret == -EEXIST) {
1235 ref = btrfs_item_ptr(leaf, path->slots[0],
1236 struct btrfs_extent_data_ref);
1237 if (match_extent_data_ref(leaf, ref, root_objectid,
1240 btrfs_release_path(path);
1242 ret = btrfs_insert_empty_item(trans, root, path, &key,
1244 if (ret && ret != -EEXIST)
1247 leaf = path->nodes[0];
1249 ref = btrfs_item_ptr(leaf, path->slots[0],
1250 struct btrfs_extent_data_ref);
1252 btrfs_set_extent_data_ref_root(leaf, ref,
1254 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1255 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1256 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1258 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1259 num_refs += refs_to_add;
1260 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1263 btrfs_mark_buffer_dirty(leaf);
1266 btrfs_release_path(path);
1270 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1271 struct btrfs_root *root,
1272 struct btrfs_path *path,
1275 struct btrfs_key key;
1276 struct btrfs_extent_data_ref *ref1 = NULL;
1277 struct btrfs_shared_data_ref *ref2 = NULL;
1278 struct extent_buffer *leaf;
1282 leaf = path->nodes[0];
1283 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1285 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1286 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_extent_data_ref);
1288 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1289 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1290 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_shared_data_ref);
1292 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1294 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1295 struct btrfs_extent_ref_v0 *ref0;
1296 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1297 struct btrfs_extent_ref_v0);
1298 num_refs = btrfs_ref_count_v0(leaf, ref0);
1304 BUG_ON(num_refs < refs_to_drop);
1305 num_refs -= refs_to_drop;
1307 if (num_refs == 0) {
1308 ret = btrfs_del_item(trans, root, path);
1310 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1311 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1312 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1313 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1314 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316 struct btrfs_extent_ref_v0 *ref0;
1317 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_ref_v0);
1319 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1322 btrfs_mark_buffer_dirty(leaf);
1327 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1328 struct btrfs_path *path,
1329 struct btrfs_extent_inline_ref *iref)
1331 struct btrfs_key key;
1332 struct extent_buffer *leaf;
1333 struct btrfs_extent_data_ref *ref1;
1334 struct btrfs_shared_data_ref *ref2;
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1340 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1341 BTRFS_EXTENT_DATA_REF_KEY) {
1342 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1349 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1350 struct btrfs_extent_data_ref);
1351 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1352 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1353 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_shared_data_ref);
1355 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1358 struct btrfs_extent_ref_v0 *ref0;
1359 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1360 struct btrfs_extent_ref_v0);
1361 num_refs = btrfs_ref_count_v0(leaf, ref0);
1369 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 struct btrfs_path *path,
1372 u64 bytenr, u64 parent,
1375 struct btrfs_key key;
1378 key.objectid = bytenr;
1380 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381 key.offset = parent;
1383 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384 key.offset = root_objectid;
1387 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1390 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1391 if (ret == -ENOENT && parent) {
1392 btrfs_release_path(path);
1393 key.type = BTRFS_EXTENT_REF_V0_KEY;
1394 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1402 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root,
1404 struct btrfs_path *path,
1405 u64 bytenr, u64 parent,
1408 struct btrfs_key key;
1411 key.objectid = bytenr;
1413 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1414 key.offset = parent;
1416 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1417 key.offset = root_objectid;
1420 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1421 btrfs_release_path(path);
1425 static inline int extent_ref_type(u64 parent, u64 owner)
1428 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1430 type = BTRFS_SHARED_BLOCK_REF_KEY;
1432 type = BTRFS_TREE_BLOCK_REF_KEY;
1435 type = BTRFS_SHARED_DATA_REF_KEY;
1437 type = BTRFS_EXTENT_DATA_REF_KEY;
1442 static int find_next_key(struct btrfs_path *path, int level,
1443 struct btrfs_key *key)
1446 for (; level < BTRFS_MAX_LEVEL; level++) {
1447 if (!path->nodes[level])
1449 if (path->slots[level] + 1 >=
1450 btrfs_header_nritems(path->nodes[level]))
1453 btrfs_item_key_to_cpu(path->nodes[level], key,
1454 path->slots[level] + 1);
1456 btrfs_node_key_to_cpu(path->nodes[level], key,
1457 path->slots[level] + 1);
1464 * look for inline back ref. if back ref is found, *ref_ret is set
1465 * to the address of inline back ref, and 0 is returned.
1467 * if back ref isn't found, *ref_ret is set to the address where it
1468 * should be inserted, and -ENOENT is returned.
1470 * if insert is true and there are too many inline back refs, the path
1471 * points to the extent item, and -EAGAIN is returned.
1473 * NOTE: inline back refs are ordered in the same way that back ref
1474 * items in the tree are ordered.
1476 static noinline_for_stack
1477 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1478 struct btrfs_root *root,
1479 struct btrfs_path *path,
1480 struct btrfs_extent_inline_ref **ref_ret,
1481 u64 bytenr, u64 num_bytes,
1482 u64 parent, u64 root_objectid,
1483 u64 owner, u64 offset, int insert)
1485 struct btrfs_key key;
1486 struct extent_buffer *leaf;
1487 struct btrfs_extent_item *ei;
1488 struct btrfs_extent_inline_ref *iref;
1498 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1501 key.objectid = bytenr;
1502 key.type = BTRFS_EXTENT_ITEM_KEY;
1503 key.offset = num_bytes;
1505 want = extent_ref_type(parent, owner);
1507 extra_size = btrfs_extent_inline_ref_size(want);
1508 path->keep_locks = 1;
1513 * Owner is our parent level, so we can just add one to get the level
1514 * for the block we are interested in.
1516 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1517 key.type = BTRFS_METADATA_ITEM_KEY;
1522 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1529 * We may be a newly converted file system which still has the old fat
1530 * extent entries for metadata, so try and see if we have one of those.
1532 if (ret > 0 && skinny_metadata) {
1533 skinny_metadata = false;
1534 if (path->slots[0]) {
1536 btrfs_item_key_to_cpu(path->nodes[0], &key,
1538 if (key.objectid == bytenr &&
1539 key.type == BTRFS_EXTENT_ITEM_KEY &&
1540 key.offset == num_bytes)
1544 key.type = BTRFS_EXTENT_ITEM_KEY;
1545 key.offset = num_bytes;
1546 btrfs_release_path(path);
1551 if (ret && !insert) {
1560 leaf = path->nodes[0];
1561 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size < sizeof(*ei)) {
1568 ret = convert_extent_item_v0(trans, root, path, owner,
1574 leaf = path->nodes[0];
1575 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1578 BUG_ON(item_size < sizeof(*ei));
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 flags = btrfs_extent_flags(leaf, ei);
1583 ptr = (unsigned long)(ei + 1);
1584 end = (unsigned long)ei + item_size;
1586 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587 ptr += sizeof(struct btrfs_tree_block_info);
1597 iref = (struct btrfs_extent_inline_ref *)ptr;
1598 type = btrfs_extent_inline_ref_type(leaf, iref);
1602 ptr += btrfs_extent_inline_ref_size(type);
1606 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607 struct btrfs_extent_data_ref *dref;
1608 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609 if (match_extent_data_ref(leaf, dref, root_objectid,
1614 if (hash_extent_data_ref_item(leaf, dref) <
1615 hash_extent_data_ref(root_objectid, owner, offset))
1619 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1621 if (parent == ref_offset) {
1625 if (ref_offset < parent)
1628 if (root_objectid == ref_offset) {
1632 if (ref_offset < root_objectid)
1636 ptr += btrfs_extent_inline_ref_size(type);
1638 if (err == -ENOENT && insert) {
1639 if (item_size + extra_size >=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path, 0, &key) == 0 &&
1651 key.objectid == bytenr &&
1652 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1657 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1660 path->keep_locks = 0;
1661 btrfs_unlock_up_safe(path, 1);
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671 struct btrfs_path *path,
1672 struct btrfs_extent_inline_ref *iref,
1673 u64 parent, u64 root_objectid,
1674 u64 owner, u64 offset, int refs_to_add,
1675 struct btrfs_delayed_extent_op *extent_op)
1677 struct extent_buffer *leaf;
1678 struct btrfs_extent_item *ei;
1681 unsigned long item_offset;
1686 leaf = path->nodes[0];
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 item_offset = (unsigned long)iref - (unsigned long)ei;
1690 type = extent_ref_type(parent, owner);
1691 size = btrfs_extent_inline_ref_size(type);
1693 btrfs_extend_item(root, path, size);
1695 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696 refs = btrfs_extent_refs(leaf, ei);
1697 refs += refs_to_add;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 ptr = (unsigned long)ei + item_offset;
1703 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704 if (ptr < end - size)
1705 memmove_extent_buffer(leaf, ptr + size, ptr,
1708 iref = (struct btrfs_extent_inline_ref *)ptr;
1709 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711 struct btrfs_extent_data_ref *dref;
1712 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718 struct btrfs_shared_data_ref *sref;
1719 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1725 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1727 btrfs_mark_buffer_dirty(leaf);
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref **ref_ret,
1734 u64 bytenr, u64 num_bytes, u64 parent,
1735 u64 root_objectid, u64 owner, u64 offset)
1739 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740 bytenr, num_bytes, parent,
1741 root_objectid, owner, offset, 0);
1745 btrfs_release_path(path);
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1752 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753 root_objectid, owner, offset);
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763 struct btrfs_path *path,
1764 struct btrfs_extent_inline_ref *iref,
1766 struct btrfs_delayed_extent_op *extent_op)
1768 struct extent_buffer *leaf;
1769 struct btrfs_extent_item *ei;
1770 struct btrfs_extent_data_ref *dref = NULL;
1771 struct btrfs_shared_data_ref *sref = NULL;
1779 leaf = path->nodes[0];
1780 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781 refs = btrfs_extent_refs(leaf, ei);
1782 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783 refs += refs_to_mod;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 type = btrfs_extent_inline_ref_type(leaf, iref);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1810 size = btrfs_extent_inline_ref_size(type);
1811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812 ptr = (unsigned long)iref;
1813 end = (unsigned long)ei + item_size;
1814 if (ptr + size < end)
1815 memmove_extent_buffer(leaf, ptr, ptr + size,
1818 btrfs_truncate_item(root, path, item_size, 1);
1820 btrfs_mark_buffer_dirty(leaf);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1835 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836 bytenr, num_bytes, parent,
1837 root_objectid, owner, offset, 1);
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(root, path, iref,
1841 refs_to_add, extent_op);
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(root, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 u64 bytenr, u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add)
1858 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859 BUG_ON(refs_to_add != 1);
1860 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861 parent, root_objectid);
1863 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864 parent, root_objectid,
1865 owner, offset, refs_to_add);
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root,
1872 struct btrfs_path *path,
1873 struct btrfs_extent_inline_ref *iref,
1874 int refs_to_drop, int is_data)
1878 BUG_ON(!is_data && refs_to_drop != 1);
1880 update_inline_extent_backref(root, path, iref,
1881 -refs_to_drop, NULL);
1882 } else if (is_data) {
1883 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1885 ret = btrfs_del_item(trans, root, path);
1890 static int btrfs_issue_discard(struct block_device *bdev,
1893 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897 u64 num_bytes, u64 *actual_bytes)
1900 u64 discarded_bytes = 0;
1901 struct btrfs_bio *bbio = NULL;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906 bytenr, &num_bytes, &bbio, 0);
1907 /* Error condition is -ENOMEM */
1909 struct btrfs_bio_stripe *stripe = bbio->stripes;
1913 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914 if (!stripe->dev->can_discard)
1917 ret = btrfs_issue_discard(stripe->dev->bdev,
1921 discarded_bytes += stripe->length;
1922 else if (ret != -EOPNOTSUPP)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1936 *actual_bytes = discarded_bytes;
1939 if (ret == -EOPNOTSUPP)
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root,
1947 u64 bytenr, u64 num_bytes, u64 parent,
1948 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1951 struct btrfs_fs_info *fs_info = root->fs_info;
1953 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1956 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1959 parent, root_objectid, (int)owner,
1960 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1962 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1964 parent, root_objectid, owner, offset,
1965 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes,
1973 u64 parent, u64 root_objectid,
1974 u64 owner, u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1984 path = btrfs_alloc_path();
1989 path->leave_spinning = 1;
1990 /* this will setup the path even if it fails to insert the back ref */
1991 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1992 path, bytenr, num_bytes, parent,
1993 root_objectid, owner, offset,
1994 refs_to_add, extent_op);
1998 if (ret != -EAGAIN) {
2003 leaf = path->nodes[0];
2004 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2005 refs = btrfs_extent_refs(leaf, item);
2006 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2008 __run_delayed_extent_op(extent_op, leaf, item);
2010 btrfs_mark_buffer_dirty(leaf);
2011 btrfs_release_path(path);
2014 path->leave_spinning = 1;
2016 /* now insert the actual backref */
2017 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2018 path, bytenr, parent, root_objectid,
2019 owner, offset, refs_to_add);
2021 btrfs_abort_transaction(trans, root, ret);
2023 btrfs_free_path(path);
2027 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2028 struct btrfs_root *root,
2029 struct btrfs_delayed_ref_node *node,
2030 struct btrfs_delayed_extent_op *extent_op,
2031 int insert_reserved)
2034 struct btrfs_delayed_data_ref *ref;
2035 struct btrfs_key ins;
2040 ins.objectid = node->bytenr;
2041 ins.offset = node->num_bytes;
2042 ins.type = BTRFS_EXTENT_ITEM_KEY;
2044 ref = btrfs_delayed_node_to_data_ref(node);
2045 trace_run_delayed_data_ref(node, ref, node->action);
2047 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2048 parent = ref->parent;
2050 ref_root = ref->root;
2052 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2054 flags |= extent_op->flags_to_set;
2055 ret = alloc_reserved_file_extent(trans, root,
2056 parent, ref_root, flags,
2057 ref->objectid, ref->offset,
2058 &ins, node->ref_mod);
2059 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2060 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2061 node->num_bytes, parent,
2062 ref_root, ref->objectid,
2063 ref->offset, node->ref_mod,
2065 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2066 ret = __btrfs_free_extent(trans, root, node->bytenr,
2067 node->num_bytes, parent,
2068 ref_root, ref->objectid,
2069 ref->offset, node->ref_mod,
2077 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2078 struct extent_buffer *leaf,
2079 struct btrfs_extent_item *ei)
2081 u64 flags = btrfs_extent_flags(leaf, ei);
2082 if (extent_op->update_flags) {
2083 flags |= extent_op->flags_to_set;
2084 btrfs_set_extent_flags(leaf, ei, flags);
2087 if (extent_op->update_key) {
2088 struct btrfs_tree_block_info *bi;
2089 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2090 bi = (struct btrfs_tree_block_info *)(ei + 1);
2091 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2095 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2096 struct btrfs_root *root,
2097 struct btrfs_delayed_ref_node *node,
2098 struct btrfs_delayed_extent_op *extent_op)
2100 struct btrfs_key key;
2101 struct btrfs_path *path;
2102 struct btrfs_extent_item *ei;
2103 struct extent_buffer *leaf;
2107 int metadata = !extent_op->is_data;
2112 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2115 path = btrfs_alloc_path();
2119 key.objectid = node->bytenr;
2122 key.type = BTRFS_METADATA_ITEM_KEY;
2123 key.offset = extent_op->level;
2125 key.type = BTRFS_EXTENT_ITEM_KEY;
2126 key.offset = node->num_bytes;
2131 path->leave_spinning = 1;
2132 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2140 btrfs_release_path(path);
2143 key.offset = node->num_bytes;
2144 key.type = BTRFS_EXTENT_ITEM_KEY;
2151 leaf = path->nodes[0];
2152 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2154 if (item_size < sizeof(*ei)) {
2155 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2161 leaf = path->nodes[0];
2162 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2165 BUG_ON(item_size < sizeof(*ei));
2166 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2167 __run_delayed_extent_op(extent_op, leaf, ei);
2169 btrfs_mark_buffer_dirty(leaf);
2171 btrfs_free_path(path);
2175 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2176 struct btrfs_root *root,
2177 struct btrfs_delayed_ref_node *node,
2178 struct btrfs_delayed_extent_op *extent_op,
2179 int insert_reserved)
2182 struct btrfs_delayed_tree_ref *ref;
2183 struct btrfs_key ins;
2186 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2189 ref = btrfs_delayed_node_to_tree_ref(node);
2190 trace_run_delayed_tree_ref(node, ref, node->action);
2192 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2193 parent = ref->parent;
2195 ref_root = ref->root;
2197 ins.objectid = node->bytenr;
2198 if (skinny_metadata) {
2199 ins.offset = ref->level;
2200 ins.type = BTRFS_METADATA_ITEM_KEY;
2202 ins.offset = node->num_bytes;
2203 ins.type = BTRFS_EXTENT_ITEM_KEY;
2206 BUG_ON(node->ref_mod != 1);
2207 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2208 BUG_ON(!extent_op || !extent_op->update_flags);
2209 ret = alloc_reserved_tree_block(trans, root,
2211 extent_op->flags_to_set,
2214 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2215 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2216 node->num_bytes, parent, ref_root,
2217 ref->level, 0, 1, extent_op);
2218 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2219 ret = __btrfs_free_extent(trans, root, node->bytenr,
2220 node->num_bytes, parent, ref_root,
2221 ref->level, 0, 1, extent_op);
2228 /* helper function to actually process a single delayed ref entry */
2229 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2230 struct btrfs_root *root,
2231 struct btrfs_delayed_ref_node *node,
2232 struct btrfs_delayed_extent_op *extent_op,
2233 int insert_reserved)
2237 if (trans->aborted) {
2238 if (insert_reserved)
2239 btrfs_pin_extent(root, node->bytenr,
2240 node->num_bytes, 1);
2244 if (btrfs_delayed_ref_is_head(node)) {
2245 struct btrfs_delayed_ref_head *head;
2247 * we've hit the end of the chain and we were supposed
2248 * to insert this extent into the tree. But, it got
2249 * deleted before we ever needed to insert it, so all
2250 * we have to do is clean up the accounting
2253 head = btrfs_delayed_node_to_head(node);
2254 trace_run_delayed_ref_head(node, head, node->action);
2256 if (insert_reserved) {
2257 btrfs_pin_extent(root, node->bytenr,
2258 node->num_bytes, 1);
2259 if (head->is_data) {
2260 ret = btrfs_del_csums(trans, root,
2268 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2269 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2270 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2272 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2273 node->type == BTRFS_SHARED_DATA_REF_KEY)
2274 ret = run_delayed_data_ref(trans, root, node, extent_op,
2281 static noinline struct btrfs_delayed_ref_node *
2282 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2284 struct rb_node *node;
2285 struct btrfs_delayed_ref_node *ref;
2286 int action = BTRFS_ADD_DELAYED_REF;
2289 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2290 * this prevents ref count from going down to zero when
2291 * there still are pending delayed ref.
2293 node = rb_prev(&head->node.rb_node);
2297 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2299 if (ref->bytenr != head->node.bytenr)
2301 if (ref->action == action)
2303 node = rb_prev(node);
2305 if (action == BTRFS_ADD_DELAYED_REF) {
2306 action = BTRFS_DROP_DELAYED_REF;
2313 * Returns 0 on success or if called with an already aborted transaction.
2314 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2316 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2317 struct btrfs_root *root,
2318 struct list_head *cluster)
2320 struct btrfs_delayed_ref_root *delayed_refs;
2321 struct btrfs_delayed_ref_node *ref;
2322 struct btrfs_delayed_ref_head *locked_ref = NULL;
2323 struct btrfs_delayed_extent_op *extent_op;
2324 struct btrfs_fs_info *fs_info = root->fs_info;
2327 int must_insert_reserved = 0;
2329 delayed_refs = &trans->transaction->delayed_refs;
2332 /* pick a new head ref from the cluster list */
2333 if (list_empty(cluster))
2336 locked_ref = list_entry(cluster->next,
2337 struct btrfs_delayed_ref_head, cluster);
2339 /* grab the lock that says we are going to process
2340 * all the refs for this head */
2341 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2344 * we may have dropped the spin lock to get the head
2345 * mutex lock, and that might have given someone else
2346 * time to free the head. If that's true, it has been
2347 * removed from our list and we can move on.
2349 if (ret == -EAGAIN) {
2357 * We need to try and merge add/drops of the same ref since we
2358 * can run into issues with relocate dropping the implicit ref
2359 * and then it being added back again before the drop can
2360 * finish. If we merged anything we need to re-loop so we can
2363 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2367 * locked_ref is the head node, so we have to go one
2368 * node back for any delayed ref updates
2370 ref = select_delayed_ref(locked_ref);
2372 if (ref && ref->seq &&
2373 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2375 * there are still refs with lower seq numbers in the
2376 * process of being added. Don't run this ref yet.
2378 list_del_init(&locked_ref->cluster);
2379 btrfs_delayed_ref_unlock(locked_ref);
2381 delayed_refs->num_heads_ready++;
2382 spin_unlock(&delayed_refs->lock);
2384 spin_lock(&delayed_refs->lock);
2389 * record the must insert reserved flag before we
2390 * drop the spin lock.
2392 must_insert_reserved = locked_ref->must_insert_reserved;
2393 locked_ref->must_insert_reserved = 0;
2395 extent_op = locked_ref->extent_op;
2396 locked_ref->extent_op = NULL;
2399 /* All delayed refs have been processed, Go ahead
2400 * and send the head node to run_one_delayed_ref,
2401 * so that any accounting fixes can happen
2403 ref = &locked_ref->node;
2405 if (extent_op && must_insert_reserved) {
2406 btrfs_free_delayed_extent_op(extent_op);
2411 spin_unlock(&delayed_refs->lock);
2413 ret = run_delayed_extent_op(trans, root,
2415 btrfs_free_delayed_extent_op(extent_op);
2419 * Need to reset must_insert_reserved if
2420 * there was an error so the abort stuff
2421 * can cleanup the reserved space
2424 if (must_insert_reserved)
2425 locked_ref->must_insert_reserved = 1;
2426 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2427 spin_lock(&delayed_refs->lock);
2428 btrfs_delayed_ref_unlock(locked_ref);
2437 rb_erase(&ref->rb_node, &delayed_refs->root);
2438 delayed_refs->num_entries--;
2439 if (!btrfs_delayed_ref_is_head(ref)) {
2441 * when we play the delayed ref, also correct the
2444 switch (ref->action) {
2445 case BTRFS_ADD_DELAYED_REF:
2446 case BTRFS_ADD_DELAYED_EXTENT:
2447 locked_ref->node.ref_mod -= ref->ref_mod;
2449 case BTRFS_DROP_DELAYED_REF:
2450 locked_ref->node.ref_mod += ref->ref_mod;
2456 list_del_init(&locked_ref->cluster);
2458 spin_unlock(&delayed_refs->lock);
2460 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2461 must_insert_reserved);
2463 btrfs_free_delayed_extent_op(extent_op);
2465 btrfs_delayed_ref_unlock(locked_ref);
2466 btrfs_put_delayed_ref(ref);
2467 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2468 spin_lock(&delayed_refs->lock);
2473 * If this node is a head, that means all the refs in this head
2474 * have been dealt with, and we will pick the next head to deal
2475 * with, so we must unlock the head and drop it from the cluster
2476 * list before we release it.
2478 if (btrfs_delayed_ref_is_head(ref)) {
2479 btrfs_delayed_ref_unlock(locked_ref);
2482 btrfs_put_delayed_ref(ref);
2486 spin_lock(&delayed_refs->lock);
2491 #ifdef SCRAMBLE_DELAYED_REFS
2493 * Normally delayed refs get processed in ascending bytenr order. This
2494 * correlates in most cases to the order added. To expose dependencies on this
2495 * order, we start to process the tree in the middle instead of the beginning
2497 static u64 find_middle(struct rb_root *root)
2499 struct rb_node *n = root->rb_node;
2500 struct btrfs_delayed_ref_node *entry;
2503 u64 first = 0, last = 0;
2507 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2508 first = entry->bytenr;
2512 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2513 last = entry->bytenr;
2518 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2519 WARN_ON(!entry->in_tree);
2521 middle = entry->bytenr;
2534 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2535 struct btrfs_fs_info *fs_info)
2537 struct qgroup_update *qgroup_update;
2540 if (list_empty(&trans->qgroup_ref_list) !=
2541 !trans->delayed_ref_elem.seq) {
2542 /* list without seq or seq without list */
2544 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2545 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2546 (u32)(trans->delayed_ref_elem.seq >> 32),
2547 (u32)trans->delayed_ref_elem.seq);
2551 if (!trans->delayed_ref_elem.seq)
2554 while (!list_empty(&trans->qgroup_ref_list)) {
2555 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2556 struct qgroup_update, list);
2557 list_del(&qgroup_update->list);
2559 ret = btrfs_qgroup_account_ref(
2560 trans, fs_info, qgroup_update->node,
2561 qgroup_update->extent_op);
2562 kfree(qgroup_update);
2565 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2570 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2573 int val = atomic_read(&delayed_refs->ref_seq);
2575 if (val < seq || val >= seq + count)
2580 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2584 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2585 sizeof(struct btrfs_extent_inline_ref));
2586 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2587 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2590 * We don't ever fill up leaves all the way so multiply by 2 just to be
2591 * closer to what we're really going to want to ouse.
2593 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2596 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2597 struct btrfs_root *root)
2599 struct btrfs_block_rsv *global_rsv;
2600 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2604 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2605 num_heads = heads_to_leaves(root, num_heads);
2607 num_bytes += (num_heads - 1) * root->leafsize;
2609 global_rsv = &root->fs_info->global_block_rsv;
2612 * If we can't allocate any more chunks lets make sure we have _lots_ of
2613 * wiggle room since running delayed refs can create more delayed refs.
2615 if (global_rsv->space_info->full)
2618 spin_lock(&global_rsv->lock);
2619 if (global_rsv->reserved <= num_bytes)
2621 spin_unlock(&global_rsv->lock);
2626 * this starts processing the delayed reference count updates and
2627 * extent insertions we have queued up so far. count can be
2628 * 0, which means to process everything in the tree at the start
2629 * of the run (but not newly added entries), or it can be some target
2630 * number you'd like to process.
2632 * Returns 0 on success or if called with an aborted transaction
2633 * Returns <0 on error and aborts the transaction
2635 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2636 struct btrfs_root *root, unsigned long count)
2638 struct rb_node *node;
2639 struct btrfs_delayed_ref_root *delayed_refs;
2640 struct btrfs_delayed_ref_node *ref;
2641 struct list_head cluster;
2644 int run_all = count == (unsigned long)-1;
2648 /* We'll clean this up in btrfs_cleanup_transaction */
2652 if (root == root->fs_info->extent_root)
2653 root = root->fs_info->tree_root;
2655 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2657 delayed_refs = &trans->transaction->delayed_refs;
2658 INIT_LIST_HEAD(&cluster);
2660 count = delayed_refs->num_entries * 2;
2664 if (!run_all && !run_most) {
2666 int seq = atomic_read(&delayed_refs->ref_seq);
2669 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2671 DEFINE_WAIT(__wait);
2672 if (delayed_refs->flushing ||
2673 !btrfs_should_throttle_delayed_refs(trans, root))
2676 prepare_to_wait(&delayed_refs->wait, &__wait,
2677 TASK_UNINTERRUPTIBLE);
2679 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2682 finish_wait(&delayed_refs->wait, &__wait);
2684 if (!refs_newer(delayed_refs, seq, 256))
2689 finish_wait(&delayed_refs->wait, &__wait);
2695 atomic_inc(&delayed_refs->procs_running_refs);
2700 spin_lock(&delayed_refs->lock);
2702 #ifdef SCRAMBLE_DELAYED_REFS
2703 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2707 if (!(run_all || run_most) &&
2708 !btrfs_should_throttle_delayed_refs(trans, root))
2712 * go find something we can process in the rbtree. We start at
2713 * the beginning of the tree, and then build a cluster
2714 * of refs to process starting at the first one we are able to
2717 delayed_start = delayed_refs->run_delayed_start;
2718 ret = btrfs_find_ref_cluster(trans, &cluster,
2719 delayed_refs->run_delayed_start);
2723 ret = run_clustered_refs(trans, root, &cluster);
2725 btrfs_release_ref_cluster(&cluster);
2726 spin_unlock(&delayed_refs->lock);
2727 btrfs_abort_transaction(trans, root, ret);
2728 atomic_dec(&delayed_refs->procs_running_refs);
2729 wake_up(&delayed_refs->wait);
2733 atomic_add(ret, &delayed_refs->ref_seq);
2735 count -= min_t(unsigned long, ret, count);
2740 if (delayed_start >= delayed_refs->run_delayed_start) {
2743 * btrfs_find_ref_cluster looped. let's do one
2744 * more cycle. if we don't run any delayed ref
2745 * during that cycle (because we can't because
2746 * all of them are blocked), bail out.
2751 * no runnable refs left, stop trying
2758 /* refs were run, let's reset staleness detection */
2764 if (!list_empty(&trans->new_bgs)) {
2765 spin_unlock(&delayed_refs->lock);
2766 btrfs_create_pending_block_groups(trans, root);
2767 spin_lock(&delayed_refs->lock);
2770 node = rb_first(&delayed_refs->root);
2773 count = (unsigned long)-1;
2776 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2778 if (btrfs_delayed_ref_is_head(ref)) {
2779 struct btrfs_delayed_ref_head *head;
2781 head = btrfs_delayed_node_to_head(ref);
2782 atomic_inc(&ref->refs);
2784 spin_unlock(&delayed_refs->lock);
2786 * Mutex was contended, block until it's
2787 * released and try again
2789 mutex_lock(&head->mutex);
2790 mutex_unlock(&head->mutex);
2792 btrfs_put_delayed_ref(ref);
2796 node = rb_next(node);
2798 spin_unlock(&delayed_refs->lock);
2799 schedule_timeout(1);
2803 atomic_dec(&delayed_refs->procs_running_refs);
2805 if (waitqueue_active(&delayed_refs->wait))
2806 wake_up(&delayed_refs->wait);
2808 spin_unlock(&delayed_refs->lock);
2809 assert_qgroups_uptodate(trans);
2813 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2814 struct btrfs_root *root,
2815 u64 bytenr, u64 num_bytes, u64 flags,
2816 int level, int is_data)
2818 struct btrfs_delayed_extent_op *extent_op;
2821 extent_op = btrfs_alloc_delayed_extent_op();
2825 extent_op->flags_to_set = flags;
2826 extent_op->update_flags = 1;
2827 extent_op->update_key = 0;
2828 extent_op->is_data = is_data ? 1 : 0;
2829 extent_op->level = level;
2831 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2832 num_bytes, extent_op);
2834 btrfs_free_delayed_extent_op(extent_op);
2838 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2839 struct btrfs_root *root,
2840 struct btrfs_path *path,
2841 u64 objectid, u64 offset, u64 bytenr)
2843 struct btrfs_delayed_ref_head *head;
2844 struct btrfs_delayed_ref_node *ref;
2845 struct btrfs_delayed_data_ref *data_ref;
2846 struct btrfs_delayed_ref_root *delayed_refs;
2847 struct rb_node *node;
2851 delayed_refs = &trans->transaction->delayed_refs;
2852 spin_lock(&delayed_refs->lock);
2853 head = btrfs_find_delayed_ref_head(trans, bytenr);
2857 if (!mutex_trylock(&head->mutex)) {
2858 atomic_inc(&head->node.refs);
2859 spin_unlock(&delayed_refs->lock);
2861 btrfs_release_path(path);
2864 * Mutex was contended, block until it's released and let
2867 mutex_lock(&head->mutex);
2868 mutex_unlock(&head->mutex);
2869 btrfs_put_delayed_ref(&head->node);
2873 node = rb_prev(&head->node.rb_node);
2877 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2879 if (ref->bytenr != bytenr)
2883 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2886 data_ref = btrfs_delayed_node_to_data_ref(ref);
2888 node = rb_prev(node);
2892 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2893 if (ref->bytenr == bytenr && ref->seq == seq)
2897 if (data_ref->root != root->root_key.objectid ||
2898 data_ref->objectid != objectid || data_ref->offset != offset)
2903 mutex_unlock(&head->mutex);
2905 spin_unlock(&delayed_refs->lock);
2909 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2910 struct btrfs_root *root,
2911 struct btrfs_path *path,
2912 u64 objectid, u64 offset, u64 bytenr)
2914 struct btrfs_root *extent_root = root->fs_info->extent_root;
2915 struct extent_buffer *leaf;
2916 struct btrfs_extent_data_ref *ref;
2917 struct btrfs_extent_inline_ref *iref;
2918 struct btrfs_extent_item *ei;
2919 struct btrfs_key key;
2923 key.objectid = bytenr;
2924 key.offset = (u64)-1;
2925 key.type = BTRFS_EXTENT_ITEM_KEY;
2927 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2930 BUG_ON(ret == 0); /* Corruption */
2933 if (path->slots[0] == 0)
2937 leaf = path->nodes[0];
2938 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2940 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2944 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2945 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2946 if (item_size < sizeof(*ei)) {
2947 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2951 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2953 if (item_size != sizeof(*ei) +
2954 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2957 if (btrfs_extent_generation(leaf, ei) <=
2958 btrfs_root_last_snapshot(&root->root_item))
2961 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2962 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2963 BTRFS_EXTENT_DATA_REF_KEY)
2966 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2967 if (btrfs_extent_refs(leaf, ei) !=
2968 btrfs_extent_data_ref_count(leaf, ref) ||
2969 btrfs_extent_data_ref_root(leaf, ref) !=
2970 root->root_key.objectid ||
2971 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2972 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2980 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2981 struct btrfs_root *root,
2982 u64 objectid, u64 offset, u64 bytenr)
2984 struct btrfs_path *path;
2988 path = btrfs_alloc_path();
2993 ret = check_committed_ref(trans, root, path, objectid,
2995 if (ret && ret != -ENOENT)
2998 ret2 = check_delayed_ref(trans, root, path, objectid,
3000 } while (ret2 == -EAGAIN);
3002 if (ret2 && ret2 != -ENOENT) {
3007 if (ret != -ENOENT || ret2 != -ENOENT)
3010 btrfs_free_path(path);
3011 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3016 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3017 struct btrfs_root *root,
3018 struct extent_buffer *buf,
3019 int full_backref, int inc, int for_cow)
3026 struct btrfs_key key;
3027 struct btrfs_file_extent_item *fi;
3031 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3032 u64, u64, u64, u64, u64, u64, int);
3034 ref_root = btrfs_header_owner(buf);
3035 nritems = btrfs_header_nritems(buf);
3036 level = btrfs_header_level(buf);
3038 if (!root->ref_cows && level == 0)
3042 process_func = btrfs_inc_extent_ref;
3044 process_func = btrfs_free_extent;
3047 parent = buf->start;
3051 for (i = 0; i < nritems; i++) {
3053 btrfs_item_key_to_cpu(buf, &key, i);
3054 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3056 fi = btrfs_item_ptr(buf, i,
3057 struct btrfs_file_extent_item);
3058 if (btrfs_file_extent_type(buf, fi) ==
3059 BTRFS_FILE_EXTENT_INLINE)
3061 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3065 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3066 key.offset -= btrfs_file_extent_offset(buf, fi);
3067 ret = process_func(trans, root, bytenr, num_bytes,
3068 parent, ref_root, key.objectid,
3069 key.offset, for_cow);
3073 bytenr = btrfs_node_blockptr(buf, i);
3074 num_bytes = btrfs_level_size(root, level - 1);
3075 ret = process_func(trans, root, bytenr, num_bytes,
3076 parent, ref_root, level - 1, 0,
3087 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3088 struct extent_buffer *buf, int full_backref, int for_cow)
3090 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3093 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3094 struct extent_buffer *buf, int full_backref, int for_cow)
3096 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3099 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3100 struct btrfs_root *root,
3101 struct btrfs_path *path,
3102 struct btrfs_block_group_cache *cache)
3105 struct btrfs_root *extent_root = root->fs_info->extent_root;
3107 struct extent_buffer *leaf;
3109 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3112 BUG_ON(ret); /* Corruption */
3114 leaf = path->nodes[0];
3115 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3116 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3117 btrfs_mark_buffer_dirty(leaf);
3118 btrfs_release_path(path);
3121 btrfs_abort_transaction(trans, root, ret);
3128 static struct btrfs_block_group_cache *
3129 next_block_group(struct btrfs_root *root,
3130 struct btrfs_block_group_cache *cache)
3132 struct rb_node *node;
3133 spin_lock(&root->fs_info->block_group_cache_lock);
3134 node = rb_next(&cache->cache_node);
3135 btrfs_put_block_group(cache);
3137 cache = rb_entry(node, struct btrfs_block_group_cache,
3139 btrfs_get_block_group(cache);
3142 spin_unlock(&root->fs_info->block_group_cache_lock);
3146 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3147 struct btrfs_trans_handle *trans,
3148 struct btrfs_path *path)
3150 struct btrfs_root *root = block_group->fs_info->tree_root;
3151 struct inode *inode = NULL;
3153 int dcs = BTRFS_DC_ERROR;
3159 * If this block group is smaller than 100 megs don't bother caching the
3162 if (block_group->key.offset < (100 * 1024 * 1024)) {
3163 spin_lock(&block_group->lock);
3164 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3165 spin_unlock(&block_group->lock);
3170 inode = lookup_free_space_inode(root, block_group, path);
3171 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3172 ret = PTR_ERR(inode);
3173 btrfs_release_path(path);
3177 if (IS_ERR(inode)) {
3181 if (block_group->ro)
3184 ret = create_free_space_inode(root, trans, block_group, path);
3190 /* We've already setup this transaction, go ahead and exit */
3191 if (block_group->cache_generation == trans->transid &&
3192 i_size_read(inode)) {
3193 dcs = BTRFS_DC_SETUP;
3198 * We want to set the generation to 0, that way if anything goes wrong
3199 * from here on out we know not to trust this cache when we load up next
3202 BTRFS_I(inode)->generation = 0;
3203 ret = btrfs_update_inode(trans, root, inode);
3206 if (i_size_read(inode) > 0) {
3207 ret = btrfs_check_trunc_cache_free_space(root,
3208 &root->fs_info->global_block_rsv);
3212 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3217 spin_lock(&block_group->lock);
3218 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3219 !btrfs_test_opt(root, SPACE_CACHE)) {
3221 * don't bother trying to write stuff out _if_
3222 * a) we're not cached,
3223 * b) we're with nospace_cache mount option.
3225 dcs = BTRFS_DC_WRITTEN;
3226 spin_unlock(&block_group->lock);
3229 spin_unlock(&block_group->lock);
3232 * Try to preallocate enough space based on how big the block group is.
3233 * Keep in mind this has to include any pinned space which could end up
3234 * taking up quite a bit since it's not folded into the other space
3237 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3242 num_pages *= PAGE_CACHE_SIZE;
3244 ret = btrfs_check_data_free_space(inode, num_pages);
3248 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3249 num_pages, num_pages,
3252 dcs = BTRFS_DC_SETUP;
3253 btrfs_free_reserved_data_space(inode, num_pages);
3258 btrfs_release_path(path);
3260 spin_lock(&block_group->lock);
3261 if (!ret && dcs == BTRFS_DC_SETUP)
3262 block_group->cache_generation = trans->transid;
3263 block_group->disk_cache_state = dcs;
3264 spin_unlock(&block_group->lock);
3269 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3270 struct btrfs_root *root)
3272 struct btrfs_block_group_cache *cache;
3274 struct btrfs_path *path;
3277 path = btrfs_alloc_path();
3283 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3285 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3287 cache = next_block_group(root, cache);
3295 err = cache_save_setup(cache, trans, path);
3296 last = cache->key.objectid + cache->key.offset;
3297 btrfs_put_block_group(cache);
3302 err = btrfs_run_delayed_refs(trans, root,
3304 if (err) /* File system offline */
3308 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3310 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3311 btrfs_put_block_group(cache);
3317 cache = next_block_group(root, cache);
3326 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3327 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3329 last = cache->key.objectid + cache->key.offset;
3331 err = write_one_cache_group(trans, root, path, cache);
3332 btrfs_put_block_group(cache);
3333 if (err) /* File system offline */
3339 * I don't think this is needed since we're just marking our
3340 * preallocated extent as written, but just in case it can't
3344 err = btrfs_run_delayed_refs(trans, root,
3346 if (err) /* File system offline */
3350 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3353 * Really this shouldn't happen, but it could if we
3354 * couldn't write the entire preallocated extent and
3355 * splitting the extent resulted in a new block.
3358 btrfs_put_block_group(cache);
3361 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3363 cache = next_block_group(root, cache);
3372 err = btrfs_write_out_cache(root, trans, cache, path);
3375 * If we didn't have an error then the cache state is still
3376 * NEED_WRITE, so we can set it to WRITTEN.
3378 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3379 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3380 last = cache->key.objectid + cache->key.offset;
3381 btrfs_put_block_group(cache);
3385 btrfs_free_path(path);
3389 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3391 struct btrfs_block_group_cache *block_group;
3394 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3395 if (!block_group || block_group->ro)
3398 btrfs_put_block_group(block_group);
3402 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3403 u64 total_bytes, u64 bytes_used,
3404 struct btrfs_space_info **space_info)
3406 struct btrfs_space_info *found;
3411 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3412 BTRFS_BLOCK_GROUP_RAID10))
3417 found = __find_space_info(info, flags);
3419 spin_lock(&found->lock);
3420 found->total_bytes += total_bytes;
3421 found->disk_total += total_bytes * factor;
3422 found->bytes_used += bytes_used;
3423 found->disk_used += bytes_used * factor;
3425 spin_unlock(&found->lock);
3426 *space_info = found;
3429 found = kzalloc(sizeof(*found), GFP_NOFS);
3433 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3439 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3440 INIT_LIST_HEAD(&found->block_groups[i]);
3441 init_rwsem(&found->groups_sem);
3442 spin_lock_init(&found->lock);
3443 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3444 found->total_bytes = total_bytes;
3445 found->disk_total = total_bytes * factor;
3446 found->bytes_used = bytes_used;
3447 found->disk_used = bytes_used * factor;
3448 found->bytes_pinned = 0;
3449 found->bytes_reserved = 0;
3450 found->bytes_readonly = 0;
3451 found->bytes_may_use = 0;
3453 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3454 found->chunk_alloc = 0;
3456 init_waitqueue_head(&found->wait);
3457 *space_info = found;
3458 list_add_rcu(&found->list, &info->space_info);
3459 if (flags & BTRFS_BLOCK_GROUP_DATA)
3460 info->data_sinfo = found;
3464 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3466 u64 extra_flags = chunk_to_extended(flags) &
3467 BTRFS_EXTENDED_PROFILE_MASK;
3469 write_seqlock(&fs_info->profiles_lock);
3470 if (flags & BTRFS_BLOCK_GROUP_DATA)
3471 fs_info->avail_data_alloc_bits |= extra_flags;
3472 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3473 fs_info->avail_metadata_alloc_bits |= extra_flags;
3474 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3475 fs_info->avail_system_alloc_bits |= extra_flags;
3476 write_sequnlock(&fs_info->profiles_lock);
3480 * returns target flags in extended format or 0 if restripe for this
3481 * chunk_type is not in progress
3483 * should be called with either volume_mutex or balance_lock held
3485 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3487 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3493 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3494 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3495 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3496 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3497 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3498 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3499 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3500 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3501 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3508 * @flags: available profiles in extended format (see ctree.h)
3510 * Returns reduced profile in chunk format. If profile changing is in
3511 * progress (either running or paused) picks the target profile (if it's
3512 * already available), otherwise falls back to plain reducing.
3514 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3517 * we add in the count of missing devices because we want
3518 * to make sure that any RAID levels on a degraded FS
3519 * continue to be honored.
3521 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3522 root->fs_info->fs_devices->missing_devices;
3527 * see if restripe for this chunk_type is in progress, if so
3528 * try to reduce to the target profile
3530 spin_lock(&root->fs_info->balance_lock);
3531 target = get_restripe_target(root->fs_info, flags);
3533 /* pick target profile only if it's already available */
3534 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3535 spin_unlock(&root->fs_info->balance_lock);
3536 return extended_to_chunk(target);
3539 spin_unlock(&root->fs_info->balance_lock);
3541 /* First, mask out the RAID levels which aren't possible */
3542 if (num_devices == 1)
3543 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3544 BTRFS_BLOCK_GROUP_RAID5);
3545 if (num_devices < 3)
3546 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3547 if (num_devices < 4)
3548 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3550 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3551 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3552 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3555 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3556 tmp = BTRFS_BLOCK_GROUP_RAID6;
3557 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3558 tmp = BTRFS_BLOCK_GROUP_RAID5;
3559 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3560 tmp = BTRFS_BLOCK_GROUP_RAID10;
3561 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3562 tmp = BTRFS_BLOCK_GROUP_RAID1;
3563 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3564 tmp = BTRFS_BLOCK_GROUP_RAID0;
3566 return extended_to_chunk(flags | tmp);
3569 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3574 seq = read_seqbegin(&root->fs_info->profiles_lock);
3576 if (flags & BTRFS_BLOCK_GROUP_DATA)
3577 flags |= root->fs_info->avail_data_alloc_bits;
3578 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3579 flags |= root->fs_info->avail_system_alloc_bits;
3580 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3581 flags |= root->fs_info->avail_metadata_alloc_bits;
3582 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3584 return btrfs_reduce_alloc_profile(root, flags);
3587 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3593 flags = BTRFS_BLOCK_GROUP_DATA;
3594 else if (root == root->fs_info->chunk_root)
3595 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3597 flags = BTRFS_BLOCK_GROUP_METADATA;
3599 ret = get_alloc_profile(root, flags);
3604 * This will check the space that the inode allocates from to make sure we have
3605 * enough space for bytes.
3607 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3609 struct btrfs_space_info *data_sinfo;
3610 struct btrfs_root *root = BTRFS_I(inode)->root;
3611 struct btrfs_fs_info *fs_info = root->fs_info;
3613 int ret = 0, committed = 0, alloc_chunk = 1;
3615 /* make sure bytes are sectorsize aligned */
3616 bytes = ALIGN(bytes, root->sectorsize);
3618 if (root == root->fs_info->tree_root ||
3619 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3624 data_sinfo = fs_info->data_sinfo;
3629 /* make sure we have enough space to handle the data first */
3630 spin_lock(&data_sinfo->lock);
3631 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3632 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3633 data_sinfo->bytes_may_use;
3635 if (used + bytes > data_sinfo->total_bytes) {
3636 struct btrfs_trans_handle *trans;
3639 * if we don't have enough free bytes in this space then we need
3640 * to alloc a new chunk.
3642 if (!data_sinfo->full && alloc_chunk) {
3645 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3646 spin_unlock(&data_sinfo->lock);
3648 alloc_target = btrfs_get_alloc_profile(root, 1);
3649 trans = btrfs_join_transaction(root);
3651 return PTR_ERR(trans);
3653 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3655 CHUNK_ALLOC_NO_FORCE);
3656 btrfs_end_transaction(trans, root);
3665 data_sinfo = fs_info->data_sinfo;
3671 * If we don't have enough pinned space to deal with this
3672 * allocation don't bother committing the transaction.
3674 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3677 spin_unlock(&data_sinfo->lock);
3679 /* commit the current transaction and try again */
3682 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3685 trans = btrfs_join_transaction(root);
3687 return PTR_ERR(trans);
3688 ret = btrfs_commit_transaction(trans, root);
3696 data_sinfo->bytes_may_use += bytes;
3697 trace_btrfs_space_reservation(root->fs_info, "space_info",
3698 data_sinfo->flags, bytes, 1);
3699 spin_unlock(&data_sinfo->lock);
3705 * Called if we need to clear a data reservation for this inode.
3707 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3709 struct btrfs_root *root = BTRFS_I(inode)->root;
3710 struct btrfs_space_info *data_sinfo;
3712 /* make sure bytes are sectorsize aligned */
3713 bytes = ALIGN(bytes, root->sectorsize);
3715 data_sinfo = root->fs_info->data_sinfo;
3716 spin_lock(&data_sinfo->lock);
3717 WARN_ON(data_sinfo->bytes_may_use < bytes);
3718 data_sinfo->bytes_may_use -= bytes;
3719 trace_btrfs_space_reservation(root->fs_info, "space_info",
3720 data_sinfo->flags, bytes, 0);
3721 spin_unlock(&data_sinfo->lock);
3724 static void force_metadata_allocation(struct btrfs_fs_info *info)
3726 struct list_head *head = &info->space_info;
3727 struct btrfs_space_info *found;
3730 list_for_each_entry_rcu(found, head, list) {
3731 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3732 found->force_alloc = CHUNK_ALLOC_FORCE;
3737 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3739 return (global->size << 1);
3742 static int should_alloc_chunk(struct btrfs_root *root,
3743 struct btrfs_space_info *sinfo, int force)
3745 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3746 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3747 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3750 if (force == CHUNK_ALLOC_FORCE)
3754 * We need to take into account the global rsv because for all intents
3755 * and purposes it's used space. Don't worry about locking the
3756 * global_rsv, it doesn't change except when the transaction commits.
3758 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3759 num_allocated += calc_global_rsv_need_space(global_rsv);
3762 * in limited mode, we want to have some free space up to
3763 * about 1% of the FS size.
3765 if (force == CHUNK_ALLOC_LIMITED) {
3766 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3767 thresh = max_t(u64, 64 * 1024 * 1024,
3768 div_factor_fine(thresh, 1));
3770 if (num_bytes - num_allocated < thresh)
3774 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3779 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3783 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3784 BTRFS_BLOCK_GROUP_RAID0 |
3785 BTRFS_BLOCK_GROUP_RAID5 |
3786 BTRFS_BLOCK_GROUP_RAID6))
3787 num_dev = root->fs_info->fs_devices->rw_devices;
3788 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3791 num_dev = 1; /* DUP or single */
3793 /* metadata for updaing devices and chunk tree */
3794 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3797 static void check_system_chunk(struct btrfs_trans_handle *trans,
3798 struct btrfs_root *root, u64 type)
3800 struct btrfs_space_info *info;
3804 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3805 spin_lock(&info->lock);
3806 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3807 info->bytes_reserved - info->bytes_readonly;
3808 spin_unlock(&info->lock);
3810 thresh = get_system_chunk_thresh(root, type);
3811 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3812 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3813 left, thresh, type);
3814 dump_space_info(info, 0, 0);
3817 if (left < thresh) {
3820 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3821 btrfs_alloc_chunk(trans, root, flags);
3825 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3826 struct btrfs_root *extent_root, u64 flags, int force)
3828 struct btrfs_space_info *space_info;
3829 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3830 int wait_for_alloc = 0;
3833 /* Don't re-enter if we're already allocating a chunk */
3834 if (trans->allocating_chunk)
3837 space_info = __find_space_info(extent_root->fs_info, flags);
3839 ret = update_space_info(extent_root->fs_info, flags,
3841 BUG_ON(ret); /* -ENOMEM */
3843 BUG_ON(!space_info); /* Logic error */
3846 spin_lock(&space_info->lock);
3847 if (force < space_info->force_alloc)
3848 force = space_info->force_alloc;
3849 if (space_info->full) {
3850 if (should_alloc_chunk(extent_root, space_info, force))
3854 spin_unlock(&space_info->lock);
3858 if (!should_alloc_chunk(extent_root, space_info, force)) {
3859 spin_unlock(&space_info->lock);
3861 } else if (space_info->chunk_alloc) {
3864 space_info->chunk_alloc = 1;
3867 spin_unlock(&space_info->lock);
3869 mutex_lock(&fs_info->chunk_mutex);
3872 * The chunk_mutex is held throughout the entirety of a chunk
3873 * allocation, so once we've acquired the chunk_mutex we know that the
3874 * other guy is done and we need to recheck and see if we should
3877 if (wait_for_alloc) {
3878 mutex_unlock(&fs_info->chunk_mutex);
3883 trans->allocating_chunk = true;
3886 * If we have mixed data/metadata chunks we want to make sure we keep
3887 * allocating mixed chunks instead of individual chunks.
3889 if (btrfs_mixed_space_info(space_info))
3890 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3893 * if we're doing a data chunk, go ahead and make sure that
3894 * we keep a reasonable number of metadata chunks allocated in the
3897 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3898 fs_info->data_chunk_allocations++;
3899 if (!(fs_info->data_chunk_allocations %
3900 fs_info->metadata_ratio))
3901 force_metadata_allocation(fs_info);
3905 * Check if we have enough space in SYSTEM chunk because we may need
3906 * to update devices.
3908 check_system_chunk(trans, extent_root, flags);
3910 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3911 trans->allocating_chunk = false;
3913 spin_lock(&space_info->lock);
3914 if (ret < 0 && ret != -ENOSPC)
3917 space_info->full = 1;
3921 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3923 space_info->chunk_alloc = 0;
3924 spin_unlock(&space_info->lock);
3925 mutex_unlock(&fs_info->chunk_mutex);
3929 static int can_overcommit(struct btrfs_root *root,
3930 struct btrfs_space_info *space_info, u64 bytes,
3931 enum btrfs_reserve_flush_enum flush)
3933 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3934 u64 profile = btrfs_get_alloc_profile(root, 0);
3939 used = space_info->bytes_used + space_info->bytes_reserved +
3940 space_info->bytes_pinned + space_info->bytes_readonly;
3943 * We only want to allow over committing if we have lots of actual space
3944 * free, but if we don't have enough space to handle the global reserve
3945 * space then we could end up having a real enospc problem when trying
3946 * to allocate a chunk or some other such important allocation.
3948 spin_lock(&global_rsv->lock);
3949 space_size = calc_global_rsv_need_space(global_rsv);
3950 spin_unlock(&global_rsv->lock);
3951 if (used + space_size >= space_info->total_bytes)
3954 used += space_info->bytes_may_use;
3956 spin_lock(&root->fs_info->free_chunk_lock);
3957 avail = root->fs_info->free_chunk_space;
3958 spin_unlock(&root->fs_info->free_chunk_lock);
3961 * If we have dup, raid1 or raid10 then only half of the free
3962 * space is actually useable. For raid56, the space info used
3963 * doesn't include the parity drive, so we don't have to
3966 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3967 BTRFS_BLOCK_GROUP_RAID1 |
3968 BTRFS_BLOCK_GROUP_RAID10))
3972 * If we aren't flushing all things, let us overcommit up to
3973 * 1/2th of the space. If we can flush, don't let us overcommit
3974 * too much, let it overcommit up to 1/8 of the space.
3976 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3981 if (used + bytes < space_info->total_bytes + avail)
3986 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3987 unsigned long nr_pages)
3989 struct super_block *sb = root->fs_info->sb;
3991 if (down_read_trylock(&sb->s_umount)) {
3992 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3993 up_read(&sb->s_umount);
3996 * We needn't worry the filesystem going from r/w to r/o though
3997 * we don't acquire ->s_umount mutex, because the filesystem
3998 * should guarantee the delalloc inodes list be empty after
3999 * the filesystem is readonly(all dirty pages are written to
4002 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
4003 if (!current->journal_info)
4004 btrfs_wait_all_ordered_extents(root->fs_info);
4009 * shrink metadata reservation for delalloc
4011 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4014 struct btrfs_block_rsv *block_rsv;
4015 struct btrfs_space_info *space_info;
4016 struct btrfs_trans_handle *trans;
4020 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
4022 enum btrfs_reserve_flush_enum flush;
4024 trans = (struct btrfs_trans_handle *)current->journal_info;
4025 block_rsv = &root->fs_info->delalloc_block_rsv;
4026 space_info = block_rsv->space_info;
4029 delalloc_bytes = percpu_counter_sum_positive(
4030 &root->fs_info->delalloc_bytes);
4031 if (delalloc_bytes == 0) {
4034 btrfs_wait_all_ordered_extents(root->fs_info);
4038 while (delalloc_bytes && loops < 3) {
4039 max_reclaim = min(delalloc_bytes, to_reclaim);
4040 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4041 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4043 * We need to wait for the async pages to actually start before
4046 wait_event(root->fs_info->async_submit_wait,
4047 !atomic_read(&root->fs_info->async_delalloc_pages));
4050 flush = BTRFS_RESERVE_FLUSH_ALL;
4052 flush = BTRFS_RESERVE_NO_FLUSH;
4053 spin_lock(&space_info->lock);
4054 if (can_overcommit(root, space_info, orig, flush)) {
4055 spin_unlock(&space_info->lock);
4058 spin_unlock(&space_info->lock);
4061 if (wait_ordered && !trans) {
4062 btrfs_wait_all_ordered_extents(root->fs_info);
4064 time_left = schedule_timeout_killable(1);
4069 delalloc_bytes = percpu_counter_sum_positive(
4070 &root->fs_info->delalloc_bytes);
4075 * maybe_commit_transaction - possibly commit the transaction if its ok to
4076 * @root - the root we're allocating for
4077 * @bytes - the number of bytes we want to reserve
4078 * @force - force the commit
4080 * This will check to make sure that committing the transaction will actually
4081 * get us somewhere and then commit the transaction if it does. Otherwise it
4082 * will return -ENOSPC.
4084 static int may_commit_transaction(struct btrfs_root *root,
4085 struct btrfs_space_info *space_info,
4086 u64 bytes, int force)
4088 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4089 struct btrfs_trans_handle *trans;
4091 trans = (struct btrfs_trans_handle *)current->journal_info;
4098 /* See if there is enough pinned space to make this reservation */
4099 spin_lock(&space_info->lock);
4100 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4102 spin_unlock(&space_info->lock);
4105 spin_unlock(&space_info->lock);
4108 * See if there is some space in the delayed insertion reservation for
4111 if (space_info != delayed_rsv->space_info)
4114 spin_lock(&space_info->lock);
4115 spin_lock(&delayed_rsv->lock);
4116 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4117 bytes - delayed_rsv->size) >= 0) {
4118 spin_unlock(&delayed_rsv->lock);
4119 spin_unlock(&space_info->lock);
4122 spin_unlock(&delayed_rsv->lock);
4123 spin_unlock(&space_info->lock);
4126 trans = btrfs_join_transaction(root);
4130 return btrfs_commit_transaction(trans, root);
4134 FLUSH_DELAYED_ITEMS_NR = 1,
4135 FLUSH_DELAYED_ITEMS = 2,
4137 FLUSH_DELALLOC_WAIT = 4,
4142 static int flush_space(struct btrfs_root *root,
4143 struct btrfs_space_info *space_info, u64 num_bytes,
4144 u64 orig_bytes, int state)
4146 struct btrfs_trans_handle *trans;
4151 case FLUSH_DELAYED_ITEMS_NR:
4152 case FLUSH_DELAYED_ITEMS:
4153 if (state == FLUSH_DELAYED_ITEMS_NR) {
4154 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4156 nr = (int)div64_u64(num_bytes, bytes);
4163 trans = btrfs_join_transaction(root);
4164 if (IS_ERR(trans)) {
4165 ret = PTR_ERR(trans);
4168 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4169 btrfs_end_transaction(trans, root);
4171 case FLUSH_DELALLOC:
4172 case FLUSH_DELALLOC_WAIT:
4173 shrink_delalloc(root, num_bytes, orig_bytes,
4174 state == FLUSH_DELALLOC_WAIT);
4177 trans = btrfs_join_transaction(root);
4178 if (IS_ERR(trans)) {
4179 ret = PTR_ERR(trans);
4182 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4183 btrfs_get_alloc_profile(root, 0),
4184 CHUNK_ALLOC_NO_FORCE);
4185 btrfs_end_transaction(trans, root);
4190 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4200 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4201 * @root - the root we're allocating for
4202 * @block_rsv - the block_rsv we're allocating for
4203 * @orig_bytes - the number of bytes we want
4204 * @flush - whether or not we can flush to make our reservation
4206 * This will reserve orgi_bytes number of bytes from the space info associated
4207 * with the block_rsv. If there is not enough space it will make an attempt to
4208 * flush out space to make room. It will do this by flushing delalloc if
4209 * possible or committing the transaction. If flush is 0 then no attempts to
4210 * regain reservations will be made and this will fail if there is not enough
4213 static int reserve_metadata_bytes(struct btrfs_root *root,
4214 struct btrfs_block_rsv *block_rsv,
4216 enum btrfs_reserve_flush_enum flush)
4218 struct btrfs_space_info *space_info = block_rsv->space_info;
4220 u64 num_bytes = orig_bytes;
4221 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4223 bool flushing = false;
4227 spin_lock(&space_info->lock);
4229 * We only want to wait if somebody other than us is flushing and we
4230 * are actually allowed to flush all things.
4232 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4233 space_info->flush) {
4234 spin_unlock(&space_info->lock);
4236 * If we have a trans handle we can't wait because the flusher
4237 * may have to commit the transaction, which would mean we would
4238 * deadlock since we are waiting for the flusher to finish, but
4239 * hold the current transaction open.
4241 if (current->journal_info)
4243 ret = wait_event_killable(space_info->wait, !space_info->flush);
4244 /* Must have been killed, return */
4248 spin_lock(&space_info->lock);
4252 used = space_info->bytes_used + space_info->bytes_reserved +
4253 space_info->bytes_pinned + space_info->bytes_readonly +
4254 space_info->bytes_may_use;
4257 * The idea here is that we've not already over-reserved the block group
4258 * then we can go ahead and save our reservation first and then start
4259 * flushing if we need to. Otherwise if we've already overcommitted
4260 * lets start flushing stuff first and then come back and try to make
4263 if (used <= space_info->total_bytes) {
4264 if (used + orig_bytes <= space_info->total_bytes) {
4265 space_info->bytes_may_use += orig_bytes;
4266 trace_btrfs_space_reservation(root->fs_info,
4267 "space_info", space_info->flags, orig_bytes, 1);
4271 * Ok set num_bytes to orig_bytes since we aren't
4272 * overocmmitted, this way we only try and reclaim what
4275 num_bytes = orig_bytes;
4279 * Ok we're over committed, set num_bytes to the overcommitted
4280 * amount plus the amount of bytes that we need for this
4283 num_bytes = used - space_info->total_bytes +
4287 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4288 space_info->bytes_may_use += orig_bytes;
4289 trace_btrfs_space_reservation(root->fs_info, "space_info",
4290 space_info->flags, orig_bytes,
4296 * Couldn't make our reservation, save our place so while we're trying
4297 * to reclaim space we can actually use it instead of somebody else
4298 * stealing it from us.
4300 * We make the other tasks wait for the flush only when we can flush
4303 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4305 space_info->flush = 1;
4308 spin_unlock(&space_info->lock);
4310 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4313 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4318 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4319 * would happen. So skip delalloc flush.
4321 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4322 (flush_state == FLUSH_DELALLOC ||
4323 flush_state == FLUSH_DELALLOC_WAIT))
4324 flush_state = ALLOC_CHUNK;
4328 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4329 flush_state < COMMIT_TRANS)
4331 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4332 flush_state <= COMMIT_TRANS)
4336 if (ret == -ENOSPC &&
4337 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4338 struct btrfs_block_rsv *global_rsv =
4339 &root->fs_info->global_block_rsv;
4341 if (block_rsv != global_rsv &&
4342 !block_rsv_use_bytes(global_rsv, orig_bytes))
4346 spin_lock(&space_info->lock);
4347 space_info->flush = 0;
4348 wake_up_all(&space_info->wait);
4349 spin_unlock(&space_info->lock);
4354 static struct btrfs_block_rsv *get_block_rsv(
4355 const struct btrfs_trans_handle *trans,
4356 const struct btrfs_root *root)
4358 struct btrfs_block_rsv *block_rsv = NULL;
4361 block_rsv = trans->block_rsv;
4363 if (root == root->fs_info->csum_root && trans->adding_csums)
4364 block_rsv = trans->block_rsv;
4366 if (root == root->fs_info->uuid_root)
4367 block_rsv = trans->block_rsv;
4370 block_rsv = root->block_rsv;
4373 block_rsv = &root->fs_info->empty_block_rsv;
4378 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4382 spin_lock(&block_rsv->lock);
4383 if (block_rsv->reserved >= num_bytes) {
4384 block_rsv->reserved -= num_bytes;
4385 if (block_rsv->reserved < block_rsv->size)
4386 block_rsv->full = 0;
4389 spin_unlock(&block_rsv->lock);
4393 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4394 u64 num_bytes, int update_size)
4396 spin_lock(&block_rsv->lock);
4397 block_rsv->reserved += num_bytes;
4399 block_rsv->size += num_bytes;
4400 else if (block_rsv->reserved >= block_rsv->size)
4401 block_rsv->full = 1;
4402 spin_unlock(&block_rsv->lock);
4405 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4406 struct btrfs_block_rsv *dest, u64 num_bytes,
4409 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4412 if (global_rsv->space_info != dest->space_info)
4415 spin_lock(&global_rsv->lock);
4416 min_bytes = div_factor(global_rsv->size, min_factor);
4417 if (global_rsv->reserved < min_bytes + num_bytes) {
4418 spin_unlock(&global_rsv->lock);
4421 global_rsv->reserved -= num_bytes;
4422 if (global_rsv->reserved < global_rsv->size)
4423 global_rsv->full = 0;
4424 spin_unlock(&global_rsv->lock);
4426 block_rsv_add_bytes(dest, num_bytes, 1);
4430 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4431 struct btrfs_block_rsv *block_rsv,
4432 struct btrfs_block_rsv *dest, u64 num_bytes)
4434 struct btrfs_space_info *space_info = block_rsv->space_info;
4436 spin_lock(&block_rsv->lock);
4437 if (num_bytes == (u64)-1)
4438 num_bytes = block_rsv->size;
4439 block_rsv->size -= num_bytes;
4440 if (block_rsv->reserved >= block_rsv->size) {
4441 num_bytes = block_rsv->reserved - block_rsv->size;
4442 block_rsv->reserved = block_rsv->size;
4443 block_rsv->full = 1;
4447 spin_unlock(&block_rsv->lock);
4449 if (num_bytes > 0) {
4451 spin_lock(&dest->lock);
4455 bytes_to_add = dest->size - dest->reserved;
4456 bytes_to_add = min(num_bytes, bytes_to_add);
4457 dest->reserved += bytes_to_add;
4458 if (dest->reserved >= dest->size)
4460 num_bytes -= bytes_to_add;
4462 spin_unlock(&dest->lock);
4465 spin_lock(&space_info->lock);
4466 space_info->bytes_may_use -= num_bytes;
4467 trace_btrfs_space_reservation(fs_info, "space_info",
4468 space_info->flags, num_bytes, 0);
4469 spin_unlock(&space_info->lock);
4474 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4475 struct btrfs_block_rsv *dst, u64 num_bytes)
4479 ret = block_rsv_use_bytes(src, num_bytes);
4483 block_rsv_add_bytes(dst, num_bytes, 1);
4487 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4489 memset(rsv, 0, sizeof(*rsv));
4490 spin_lock_init(&rsv->lock);
4494 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4495 unsigned short type)
4497 struct btrfs_block_rsv *block_rsv;
4498 struct btrfs_fs_info *fs_info = root->fs_info;
4500 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4504 btrfs_init_block_rsv(block_rsv, type);
4505 block_rsv->space_info = __find_space_info(fs_info,
4506 BTRFS_BLOCK_GROUP_METADATA);
4510 void btrfs_free_block_rsv(struct btrfs_root *root,
4511 struct btrfs_block_rsv *rsv)
4515 btrfs_block_rsv_release(root, rsv, (u64)-1);
4519 int btrfs_block_rsv_add(struct btrfs_root *root,
4520 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4521 enum btrfs_reserve_flush_enum flush)
4528 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4530 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4537 int btrfs_block_rsv_check(struct btrfs_root *root,
4538 struct btrfs_block_rsv *block_rsv, int min_factor)
4546 spin_lock(&block_rsv->lock);
4547 num_bytes = div_factor(block_rsv->size, min_factor);
4548 if (block_rsv->reserved >= num_bytes)
4550 spin_unlock(&block_rsv->lock);
4555 int btrfs_block_rsv_refill(struct btrfs_root *root,
4556 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4557 enum btrfs_reserve_flush_enum flush)
4565 spin_lock(&block_rsv->lock);
4566 num_bytes = min_reserved;
4567 if (block_rsv->reserved >= num_bytes)
4570 num_bytes -= block_rsv->reserved;
4571 spin_unlock(&block_rsv->lock);
4576 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4578 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4585 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4586 struct btrfs_block_rsv *dst_rsv,
4589 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4592 void btrfs_block_rsv_release(struct btrfs_root *root,
4593 struct btrfs_block_rsv *block_rsv,
4596 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4597 if (global_rsv->full || global_rsv == block_rsv ||
4598 block_rsv->space_info != global_rsv->space_info)
4600 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4605 * helper to calculate size of global block reservation.
4606 * the desired value is sum of space used by extent tree,
4607 * checksum tree and root tree
4609 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4611 struct btrfs_space_info *sinfo;
4615 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4617 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4618 spin_lock(&sinfo->lock);
4619 data_used = sinfo->bytes_used;
4620 spin_unlock(&sinfo->lock);
4622 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4623 spin_lock(&sinfo->lock);
4624 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4626 meta_used = sinfo->bytes_used;
4627 spin_unlock(&sinfo->lock);
4629 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4631 num_bytes += div64_u64(data_used + meta_used, 50);
4633 if (num_bytes * 3 > meta_used)
4634 num_bytes = div64_u64(meta_used, 3);
4636 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4639 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4641 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4642 struct btrfs_space_info *sinfo = block_rsv->space_info;
4645 num_bytes = calc_global_metadata_size(fs_info);
4647 spin_lock(&sinfo->lock);
4648 spin_lock(&block_rsv->lock);
4650 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4652 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4653 sinfo->bytes_reserved + sinfo->bytes_readonly +
4654 sinfo->bytes_may_use;
4656 if (sinfo->total_bytes > num_bytes) {
4657 num_bytes = sinfo->total_bytes - num_bytes;
4658 block_rsv->reserved += num_bytes;
4659 sinfo->bytes_may_use += num_bytes;
4660 trace_btrfs_space_reservation(fs_info, "space_info",
4661 sinfo->flags, num_bytes, 1);
4664 if (block_rsv->reserved >= block_rsv->size) {
4665 num_bytes = block_rsv->reserved - block_rsv->size;
4666 sinfo->bytes_may_use -= num_bytes;
4667 trace_btrfs_space_reservation(fs_info, "space_info",
4668 sinfo->flags, num_bytes, 0);
4669 block_rsv->reserved = block_rsv->size;
4670 block_rsv->full = 1;
4673 spin_unlock(&block_rsv->lock);
4674 spin_unlock(&sinfo->lock);
4677 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4679 struct btrfs_space_info *space_info;
4681 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4682 fs_info->chunk_block_rsv.space_info = space_info;
4684 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4685 fs_info->global_block_rsv.space_info = space_info;
4686 fs_info->delalloc_block_rsv.space_info = space_info;
4687 fs_info->trans_block_rsv.space_info = space_info;
4688 fs_info->empty_block_rsv.space_info = space_info;
4689 fs_info->delayed_block_rsv.space_info = space_info;
4691 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4692 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4693 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4694 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4695 if (fs_info->quota_root)
4696 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4697 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4699 update_global_block_rsv(fs_info);
4702 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4704 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4706 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4707 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4708 WARN_ON(fs_info->trans_block_rsv.size > 0);
4709 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4710 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4711 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4712 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4713 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4716 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4717 struct btrfs_root *root)
4719 if (!trans->block_rsv)
4722 if (!trans->bytes_reserved)
4725 trace_btrfs_space_reservation(root->fs_info, "transaction",
4726 trans->transid, trans->bytes_reserved, 0);
4727 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4728 trans->bytes_reserved = 0;
4731 /* Can only return 0 or -ENOSPC */
4732 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4733 struct inode *inode)
4735 struct btrfs_root *root = BTRFS_I(inode)->root;
4736 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4737 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4740 * We need to hold space in order to delete our orphan item once we've
4741 * added it, so this takes the reservation so we can release it later
4742 * when we are truly done with the orphan item.
4744 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4745 trace_btrfs_space_reservation(root->fs_info, "orphan",
4746 btrfs_ino(inode), num_bytes, 1);
4747 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4750 void btrfs_orphan_release_metadata(struct inode *inode)
4752 struct btrfs_root *root = BTRFS_I(inode)->root;
4753 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4754 trace_btrfs_space_reservation(root->fs_info, "orphan",
4755 btrfs_ino(inode), num_bytes, 0);
4756 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4760 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4761 * root: the root of the parent directory
4762 * rsv: block reservation
4763 * items: the number of items that we need do reservation
4764 * qgroup_reserved: used to return the reserved size in qgroup
4766 * This function is used to reserve the space for snapshot/subvolume
4767 * creation and deletion. Those operations are different with the
4768 * common file/directory operations, they change two fs/file trees
4769 * and root tree, the number of items that the qgroup reserves is
4770 * different with the free space reservation. So we can not use
4771 * the space reseravtion mechanism in start_transaction().
4773 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4774 struct btrfs_block_rsv *rsv,
4776 u64 *qgroup_reserved,
4777 bool use_global_rsv)
4781 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4783 if (root->fs_info->quota_enabled) {
4784 /* One for parent inode, two for dir entries */
4785 num_bytes = 3 * root->leafsize;
4786 ret = btrfs_qgroup_reserve(root, num_bytes);
4793 *qgroup_reserved = num_bytes;
4795 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4796 rsv->space_info = __find_space_info(root->fs_info,
4797 BTRFS_BLOCK_GROUP_METADATA);
4798 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4799 BTRFS_RESERVE_FLUSH_ALL);
4801 if (ret == -ENOSPC && use_global_rsv)
4802 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4805 if (*qgroup_reserved)
4806 btrfs_qgroup_free(root, *qgroup_reserved);
4812 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4813 struct btrfs_block_rsv *rsv,
4814 u64 qgroup_reserved)
4816 btrfs_block_rsv_release(root, rsv, (u64)-1);
4817 if (qgroup_reserved)
4818 btrfs_qgroup_free(root, qgroup_reserved);
4822 * drop_outstanding_extent - drop an outstanding extent
4823 * @inode: the inode we're dropping the extent for
4825 * This is called when we are freeing up an outstanding extent, either called
4826 * after an error or after an extent is written. This will return the number of
4827 * reserved extents that need to be freed. This must be called with
4828 * BTRFS_I(inode)->lock held.
4830 static unsigned drop_outstanding_extent(struct inode *inode)
4832 unsigned drop_inode_space = 0;
4833 unsigned dropped_extents = 0;
4835 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4836 BTRFS_I(inode)->outstanding_extents--;
4838 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4839 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4840 &BTRFS_I(inode)->runtime_flags))
4841 drop_inode_space = 1;
4844 * If we have more or the same amount of outsanding extents than we have
4845 * reserved then we need to leave the reserved extents count alone.
4847 if (BTRFS_I(inode)->outstanding_extents >=
4848 BTRFS_I(inode)->reserved_extents)
4849 return drop_inode_space;
4851 dropped_extents = BTRFS_I(inode)->reserved_extents -
4852 BTRFS_I(inode)->outstanding_extents;
4853 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4854 return dropped_extents + drop_inode_space;
4858 * calc_csum_metadata_size - return the amount of metada space that must be
4859 * reserved/free'd for the given bytes.
4860 * @inode: the inode we're manipulating
4861 * @num_bytes: the number of bytes in question
4862 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4864 * This adjusts the number of csum_bytes in the inode and then returns the
4865 * correct amount of metadata that must either be reserved or freed. We
4866 * calculate how many checksums we can fit into one leaf and then divide the
4867 * number of bytes that will need to be checksumed by this value to figure out
4868 * how many checksums will be required. If we are adding bytes then the number
4869 * may go up and we will return the number of additional bytes that must be
4870 * reserved. If it is going down we will return the number of bytes that must
4873 * This must be called with BTRFS_I(inode)->lock held.
4875 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4878 struct btrfs_root *root = BTRFS_I(inode)->root;
4880 int num_csums_per_leaf;
4884 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4885 BTRFS_I(inode)->csum_bytes == 0)
4888 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4890 BTRFS_I(inode)->csum_bytes += num_bytes;
4892 BTRFS_I(inode)->csum_bytes -= num_bytes;
4893 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4894 num_csums_per_leaf = (int)div64_u64(csum_size,
4895 sizeof(struct btrfs_csum_item) +
4896 sizeof(struct btrfs_disk_key));
4897 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4898 num_csums = num_csums + num_csums_per_leaf - 1;
4899 num_csums = num_csums / num_csums_per_leaf;
4901 old_csums = old_csums + num_csums_per_leaf - 1;
4902 old_csums = old_csums / num_csums_per_leaf;
4904 /* No change, no need to reserve more */
4905 if (old_csums == num_csums)
4909 return btrfs_calc_trans_metadata_size(root,
4910 num_csums - old_csums);
4912 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4915 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4917 struct btrfs_root *root = BTRFS_I(inode)->root;
4918 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4921 unsigned nr_extents = 0;
4922 int extra_reserve = 0;
4923 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4925 bool delalloc_lock = true;
4929 /* If we are a free space inode we need to not flush since we will be in
4930 * the middle of a transaction commit. We also don't need the delalloc
4931 * mutex since we won't race with anybody. We need this mostly to make
4932 * lockdep shut its filthy mouth.
4934 if (btrfs_is_free_space_inode(inode)) {
4935 flush = BTRFS_RESERVE_NO_FLUSH;
4936 delalloc_lock = false;
4939 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4940 btrfs_transaction_in_commit(root->fs_info))
4941 schedule_timeout(1);
4944 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4946 num_bytes = ALIGN(num_bytes, root->sectorsize);
4948 spin_lock(&BTRFS_I(inode)->lock);
4949 BTRFS_I(inode)->outstanding_extents++;
4951 if (BTRFS_I(inode)->outstanding_extents >
4952 BTRFS_I(inode)->reserved_extents)
4953 nr_extents = BTRFS_I(inode)->outstanding_extents -
4954 BTRFS_I(inode)->reserved_extents;
4957 * Add an item to reserve for updating the inode when we complete the
4960 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4961 &BTRFS_I(inode)->runtime_flags)) {
4966 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4967 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4968 csum_bytes = BTRFS_I(inode)->csum_bytes;
4969 spin_unlock(&BTRFS_I(inode)->lock);
4971 if (root->fs_info->quota_enabled) {
4972 ret = btrfs_qgroup_reserve(root, num_bytes +
4973 nr_extents * root->leafsize);
4978 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4979 if (unlikely(ret)) {
4980 if (root->fs_info->quota_enabled)
4981 btrfs_qgroup_free(root, num_bytes +
4982 nr_extents * root->leafsize);
4986 spin_lock(&BTRFS_I(inode)->lock);
4987 if (extra_reserve) {
4988 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4989 &BTRFS_I(inode)->runtime_flags);
4992 BTRFS_I(inode)->reserved_extents += nr_extents;
4993 spin_unlock(&BTRFS_I(inode)->lock);
4996 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4999 trace_btrfs_space_reservation(root->fs_info,"delalloc",
5000 btrfs_ino(inode), to_reserve, 1);
5001 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5006 spin_lock(&BTRFS_I(inode)->lock);
5007 dropped = drop_outstanding_extent(inode);
5009 * If the inodes csum_bytes is the same as the original
5010 * csum_bytes then we know we haven't raced with any free()ers
5011 * so we can just reduce our inodes csum bytes and carry on.
5013 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5014 calc_csum_metadata_size(inode, num_bytes, 0);
5016 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5020 * This is tricky, but first we need to figure out how much we
5021 * free'd from any free-ers that occured during this
5022 * reservation, so we reset ->csum_bytes to the csum_bytes
5023 * before we dropped our lock, and then call the free for the
5024 * number of bytes that were freed while we were trying our
5027 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5028 BTRFS_I(inode)->csum_bytes = csum_bytes;
5029 to_free = calc_csum_metadata_size(inode, bytes, 0);
5033 * Now we need to see how much we would have freed had we not
5034 * been making this reservation and our ->csum_bytes were not
5035 * artificially inflated.
5037 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5038 bytes = csum_bytes - orig_csum_bytes;
5039 bytes = calc_csum_metadata_size(inode, bytes, 0);
5042 * Now reset ->csum_bytes to what it should be. If bytes is
5043 * more than to_free then we would have free'd more space had we
5044 * not had an artificially high ->csum_bytes, so we need to free
5045 * the remainder. If bytes is the same or less then we don't
5046 * need to do anything, the other free-ers did the correct
5049 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5050 if (bytes > to_free)
5051 to_free = bytes - to_free;
5055 spin_unlock(&BTRFS_I(inode)->lock);
5057 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5060 btrfs_block_rsv_release(root, block_rsv, to_free);
5061 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5062 btrfs_ino(inode), to_free, 0);
5065 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5070 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5071 * @inode: the inode to release the reservation for
5072 * @num_bytes: the number of bytes we're releasing
5074 * This will release the metadata reservation for an inode. This can be called
5075 * once we complete IO for a given set of bytes to release their metadata
5078 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5080 struct btrfs_root *root = BTRFS_I(inode)->root;
5084 num_bytes = ALIGN(num_bytes, root->sectorsize);
5085 spin_lock(&BTRFS_I(inode)->lock);
5086 dropped = drop_outstanding_extent(inode);
5089 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5090 spin_unlock(&BTRFS_I(inode)->lock);
5092 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5094 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5095 btrfs_ino(inode), to_free, 0);
5096 if (root->fs_info->quota_enabled) {
5097 btrfs_qgroup_free(root, num_bytes +
5098 dropped * root->leafsize);
5101 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5106 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5107 * @inode: inode we're writing to
5108 * @num_bytes: the number of bytes we want to allocate
5110 * This will do the following things
5112 * o reserve space in the data space info for num_bytes
5113 * o reserve space in the metadata space info based on number of outstanding
5114 * extents and how much csums will be needed
5115 * o add to the inodes ->delalloc_bytes
5116 * o add it to the fs_info's delalloc inodes list.
5118 * This will return 0 for success and -ENOSPC if there is no space left.
5120 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5124 ret = btrfs_check_data_free_space(inode, num_bytes);
5128 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5130 btrfs_free_reserved_data_space(inode, num_bytes);
5138 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5139 * @inode: inode we're releasing space for
5140 * @num_bytes: the number of bytes we want to free up
5142 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5143 * called in the case that we don't need the metadata AND data reservations
5144 * anymore. So if there is an error or we insert an inline extent.
5146 * This function will release the metadata space that was not used and will
5147 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5148 * list if there are no delalloc bytes left.
5150 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5152 btrfs_delalloc_release_metadata(inode, num_bytes);
5153 btrfs_free_reserved_data_space(inode, num_bytes);
5156 static int update_block_group(struct btrfs_root *root,
5157 u64 bytenr, u64 num_bytes, int alloc)
5159 struct btrfs_block_group_cache *cache = NULL;
5160 struct btrfs_fs_info *info = root->fs_info;
5161 u64 total = num_bytes;
5166 /* block accounting for super block */
5167 spin_lock(&info->delalloc_root_lock);
5168 old_val = btrfs_super_bytes_used(info->super_copy);
5170 old_val += num_bytes;
5172 old_val -= num_bytes;
5173 btrfs_set_super_bytes_used(info->super_copy, old_val);
5174 spin_unlock(&info->delalloc_root_lock);
5177 cache = btrfs_lookup_block_group(info, bytenr);
5180 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5181 BTRFS_BLOCK_GROUP_RAID1 |
5182 BTRFS_BLOCK_GROUP_RAID10))
5187 * If this block group has free space cache written out, we
5188 * need to make sure to load it if we are removing space. This
5189 * is because we need the unpinning stage to actually add the
5190 * space back to the block group, otherwise we will leak space.
5192 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5193 cache_block_group(cache, 1);
5195 byte_in_group = bytenr - cache->key.objectid;
5196 WARN_ON(byte_in_group > cache->key.offset);
5198 spin_lock(&cache->space_info->lock);
5199 spin_lock(&cache->lock);
5201 if (btrfs_test_opt(root, SPACE_CACHE) &&
5202 cache->disk_cache_state < BTRFS_DC_CLEAR)
5203 cache->disk_cache_state = BTRFS_DC_CLEAR;
5206 old_val = btrfs_block_group_used(&cache->item);
5207 num_bytes = min(total, cache->key.offset - byte_in_group);
5209 old_val += num_bytes;
5210 btrfs_set_block_group_used(&cache->item, old_val);
5211 cache->reserved -= num_bytes;
5212 cache->space_info->bytes_reserved -= num_bytes;
5213 cache->space_info->bytes_used += num_bytes;
5214 cache->space_info->disk_used += num_bytes * factor;
5215 spin_unlock(&cache->lock);
5216 spin_unlock(&cache->space_info->lock);
5218 old_val -= num_bytes;
5219 btrfs_set_block_group_used(&cache->item, old_val);
5220 cache->pinned += num_bytes;
5221 cache->space_info->bytes_pinned += num_bytes;
5222 cache->space_info->bytes_used -= num_bytes;
5223 cache->space_info->disk_used -= num_bytes * factor;
5224 spin_unlock(&cache->lock);
5225 spin_unlock(&cache->space_info->lock);
5227 set_extent_dirty(info->pinned_extents,
5228 bytenr, bytenr + num_bytes - 1,
5229 GFP_NOFS | __GFP_NOFAIL);
5231 btrfs_put_block_group(cache);
5233 bytenr += num_bytes;
5238 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5240 struct btrfs_block_group_cache *cache;
5243 spin_lock(&root->fs_info->block_group_cache_lock);
5244 bytenr = root->fs_info->first_logical_byte;
5245 spin_unlock(&root->fs_info->block_group_cache_lock);
5247 if (bytenr < (u64)-1)
5250 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5254 bytenr = cache->key.objectid;
5255 btrfs_put_block_group(cache);
5260 static int pin_down_extent(struct btrfs_root *root,
5261 struct btrfs_block_group_cache *cache,
5262 u64 bytenr, u64 num_bytes, int reserved)
5264 spin_lock(&cache->space_info->lock);
5265 spin_lock(&cache->lock);
5266 cache->pinned += num_bytes;
5267 cache->space_info->bytes_pinned += num_bytes;
5269 cache->reserved -= num_bytes;
5270 cache->space_info->bytes_reserved -= num_bytes;
5272 spin_unlock(&cache->lock);
5273 spin_unlock(&cache->space_info->lock);
5275 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5276 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5278 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5283 * this function must be called within transaction
5285 int btrfs_pin_extent(struct btrfs_root *root,
5286 u64 bytenr, u64 num_bytes, int reserved)
5288 struct btrfs_block_group_cache *cache;
5290 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5291 BUG_ON(!cache); /* Logic error */
5293 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5295 btrfs_put_block_group(cache);
5300 * this function must be called within transaction
5302 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5303 u64 bytenr, u64 num_bytes)
5305 struct btrfs_block_group_cache *cache;
5308 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5313 * pull in the free space cache (if any) so that our pin
5314 * removes the free space from the cache. We have load_only set
5315 * to one because the slow code to read in the free extents does check
5316 * the pinned extents.
5318 cache_block_group(cache, 1);
5320 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5322 /* remove us from the free space cache (if we're there at all) */
5323 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5324 btrfs_put_block_group(cache);
5328 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5331 struct btrfs_block_group_cache *block_group;
5332 struct btrfs_caching_control *caching_ctl;
5334 block_group = btrfs_lookup_block_group(root->fs_info, start);
5338 cache_block_group(block_group, 0);
5339 caching_ctl = get_caching_control(block_group);
5343 BUG_ON(!block_group_cache_done(block_group));
5344 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5346 mutex_lock(&caching_ctl->mutex);
5348 if (start >= caching_ctl->progress) {
5349 ret = add_excluded_extent(root, start, num_bytes);
5350 } else if (start + num_bytes <= caching_ctl->progress) {
5351 ret = btrfs_remove_free_space(block_group,
5354 num_bytes = caching_ctl->progress - start;
5355 ret = btrfs_remove_free_space(block_group,
5360 num_bytes = (start + num_bytes) -
5361 caching_ctl->progress;
5362 start = caching_ctl->progress;
5363 ret = add_excluded_extent(root, start, num_bytes);
5366 mutex_unlock(&caching_ctl->mutex);
5367 put_caching_control(caching_ctl);
5369 btrfs_put_block_group(block_group);
5373 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5374 struct extent_buffer *eb)
5376 struct btrfs_file_extent_item *item;
5377 struct btrfs_key key;
5381 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5384 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5385 btrfs_item_key_to_cpu(eb, &key, i);
5386 if (key.type != BTRFS_EXTENT_DATA_KEY)
5388 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5389 found_type = btrfs_file_extent_type(eb, item);
5390 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5392 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5394 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5395 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5396 __exclude_logged_extent(log, key.objectid, key.offset);
5403 * btrfs_update_reserved_bytes - update the block_group and space info counters
5404 * @cache: The cache we are manipulating
5405 * @num_bytes: The number of bytes in question
5406 * @reserve: One of the reservation enums
5408 * This is called by the allocator when it reserves space, or by somebody who is
5409 * freeing space that was never actually used on disk. For example if you
5410 * reserve some space for a new leaf in transaction A and before transaction A
5411 * commits you free that leaf, you call this with reserve set to 0 in order to
5412 * clear the reservation.
5414 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5415 * ENOSPC accounting. For data we handle the reservation through clearing the
5416 * delalloc bits in the io_tree. We have to do this since we could end up
5417 * allocating less disk space for the amount of data we have reserved in the
5418 * case of compression.
5420 * If this is a reservation and the block group has become read only we cannot
5421 * make the reservation and return -EAGAIN, otherwise this function always
5424 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5425 u64 num_bytes, int reserve)
5427 struct btrfs_space_info *space_info = cache->space_info;
5430 spin_lock(&space_info->lock);
5431 spin_lock(&cache->lock);
5432 if (reserve != RESERVE_FREE) {
5436 cache->reserved += num_bytes;
5437 space_info->bytes_reserved += num_bytes;
5438 if (reserve == RESERVE_ALLOC) {
5439 trace_btrfs_space_reservation(cache->fs_info,
5440 "space_info", space_info->flags,
5442 space_info->bytes_may_use -= num_bytes;
5447 space_info->bytes_readonly += num_bytes;
5448 cache->reserved -= num_bytes;
5449 space_info->bytes_reserved -= num_bytes;
5451 spin_unlock(&cache->lock);
5452 spin_unlock(&space_info->lock);
5456 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5457 struct btrfs_root *root)
5459 struct btrfs_fs_info *fs_info = root->fs_info;
5460 struct btrfs_caching_control *next;
5461 struct btrfs_caching_control *caching_ctl;
5462 struct btrfs_block_group_cache *cache;
5463 struct btrfs_space_info *space_info;
5465 down_write(&fs_info->extent_commit_sem);
5467 list_for_each_entry_safe(caching_ctl, next,
5468 &fs_info->caching_block_groups, list) {
5469 cache = caching_ctl->block_group;
5470 if (block_group_cache_done(cache)) {
5471 cache->last_byte_to_unpin = (u64)-1;
5472 list_del_init(&caching_ctl->list);
5473 put_caching_control(caching_ctl);
5475 cache->last_byte_to_unpin = caching_ctl->progress;
5479 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5480 fs_info->pinned_extents = &fs_info->freed_extents[1];
5482 fs_info->pinned_extents = &fs_info->freed_extents[0];
5484 up_write(&fs_info->extent_commit_sem);
5486 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5487 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5489 update_global_block_rsv(fs_info);
5492 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5494 struct btrfs_fs_info *fs_info = root->fs_info;
5495 struct btrfs_block_group_cache *cache = NULL;
5496 struct btrfs_space_info *space_info;
5497 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5501 while (start <= end) {
5504 start >= cache->key.objectid + cache->key.offset) {
5506 btrfs_put_block_group(cache);
5507 cache = btrfs_lookup_block_group(fs_info, start);
5508 BUG_ON(!cache); /* Logic error */
5511 len = cache->key.objectid + cache->key.offset - start;
5512 len = min(len, end + 1 - start);
5514 if (start < cache->last_byte_to_unpin) {
5515 len = min(len, cache->last_byte_to_unpin - start);
5516 btrfs_add_free_space(cache, start, len);
5520 space_info = cache->space_info;
5522 spin_lock(&space_info->lock);
5523 spin_lock(&cache->lock);
5524 cache->pinned -= len;
5525 space_info->bytes_pinned -= len;
5527 space_info->bytes_readonly += len;
5530 spin_unlock(&cache->lock);
5531 if (!readonly && global_rsv->space_info == space_info) {
5532 spin_lock(&global_rsv->lock);
5533 if (!global_rsv->full) {
5534 len = min(len, global_rsv->size -
5535 global_rsv->reserved);
5536 global_rsv->reserved += len;
5537 space_info->bytes_may_use += len;
5538 if (global_rsv->reserved >= global_rsv->size)
5539 global_rsv->full = 1;
5541 spin_unlock(&global_rsv->lock);
5543 spin_unlock(&space_info->lock);
5547 btrfs_put_block_group(cache);
5551 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5552 struct btrfs_root *root)
5554 struct btrfs_fs_info *fs_info = root->fs_info;
5555 struct extent_io_tree *unpin;
5563 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5564 unpin = &fs_info->freed_extents[1];
5566 unpin = &fs_info->freed_extents[0];
5569 ret = find_first_extent_bit(unpin, 0, &start, &end,
5570 EXTENT_DIRTY, NULL);
5574 if (btrfs_test_opt(root, DISCARD))
5575 ret = btrfs_discard_extent(root, start,
5576 end + 1 - start, NULL);
5578 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5579 unpin_extent_range(root, start, end);
5586 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5587 u64 owner, u64 root_objectid)
5589 struct btrfs_space_info *space_info;
5592 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5593 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5594 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5596 flags = BTRFS_BLOCK_GROUP_METADATA;
5598 flags = BTRFS_BLOCK_GROUP_DATA;
5601 space_info = __find_space_info(fs_info, flags);
5602 BUG_ON(!space_info); /* Logic bug */
5603 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5607 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5608 struct btrfs_root *root,
5609 u64 bytenr, u64 num_bytes, u64 parent,
5610 u64 root_objectid, u64 owner_objectid,
5611 u64 owner_offset, int refs_to_drop,
5612 struct btrfs_delayed_extent_op *extent_op)
5614 struct btrfs_key key;
5615 struct btrfs_path *path;
5616 struct btrfs_fs_info *info = root->fs_info;
5617 struct btrfs_root *extent_root = info->extent_root;
5618 struct extent_buffer *leaf;
5619 struct btrfs_extent_item *ei;
5620 struct btrfs_extent_inline_ref *iref;
5623 int extent_slot = 0;
5624 int found_extent = 0;
5628 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5631 path = btrfs_alloc_path();
5636 path->leave_spinning = 1;
5638 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5639 BUG_ON(!is_data && refs_to_drop != 1);
5642 skinny_metadata = 0;
5644 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5645 bytenr, num_bytes, parent,
5646 root_objectid, owner_objectid,
5649 extent_slot = path->slots[0];
5650 while (extent_slot >= 0) {
5651 btrfs_item_key_to_cpu(path->nodes[0], &key,
5653 if (key.objectid != bytenr)
5655 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5656 key.offset == num_bytes) {
5660 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5661 key.offset == owner_objectid) {
5665 if (path->slots[0] - extent_slot > 5)
5669 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5670 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5671 if (found_extent && item_size < sizeof(*ei))
5674 if (!found_extent) {
5676 ret = remove_extent_backref(trans, extent_root, path,
5680 btrfs_abort_transaction(trans, extent_root, ret);
5683 btrfs_release_path(path);
5684 path->leave_spinning = 1;
5686 key.objectid = bytenr;
5687 key.type = BTRFS_EXTENT_ITEM_KEY;
5688 key.offset = num_bytes;
5690 if (!is_data && skinny_metadata) {
5691 key.type = BTRFS_METADATA_ITEM_KEY;
5692 key.offset = owner_objectid;
5695 ret = btrfs_search_slot(trans, extent_root,
5697 if (ret > 0 && skinny_metadata && path->slots[0]) {
5699 * Couldn't find our skinny metadata item,
5700 * see if we have ye olde extent item.
5703 btrfs_item_key_to_cpu(path->nodes[0], &key,
5705 if (key.objectid == bytenr &&
5706 key.type == BTRFS_EXTENT_ITEM_KEY &&
5707 key.offset == num_bytes)
5711 if (ret > 0 && skinny_metadata) {
5712 skinny_metadata = false;
5713 key.type = BTRFS_EXTENT_ITEM_KEY;
5714 key.offset = num_bytes;
5715 btrfs_release_path(path);
5716 ret = btrfs_search_slot(trans, extent_root,
5721 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5724 btrfs_print_leaf(extent_root,
5728 btrfs_abort_transaction(trans, extent_root, ret);
5731 extent_slot = path->slots[0];
5733 } else if (ret == -ENOENT) {
5734 btrfs_print_leaf(extent_root, path->nodes[0]);
5737 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5738 bytenr, parent, root_objectid, owner_objectid,
5741 btrfs_abort_transaction(trans, extent_root, ret);
5745 leaf = path->nodes[0];
5746 item_size = btrfs_item_size_nr(leaf, extent_slot);
5747 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5748 if (item_size < sizeof(*ei)) {
5749 BUG_ON(found_extent || extent_slot != path->slots[0]);
5750 ret = convert_extent_item_v0(trans, extent_root, path,
5753 btrfs_abort_transaction(trans, extent_root, ret);
5757 btrfs_release_path(path);
5758 path->leave_spinning = 1;
5760 key.objectid = bytenr;
5761 key.type = BTRFS_EXTENT_ITEM_KEY;
5762 key.offset = num_bytes;
5764 ret = btrfs_search_slot(trans, extent_root, &key, path,
5767 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5769 btrfs_print_leaf(extent_root, path->nodes[0]);
5772 btrfs_abort_transaction(trans, extent_root, ret);
5776 extent_slot = path->slots[0];
5777 leaf = path->nodes[0];
5778 item_size = btrfs_item_size_nr(leaf, extent_slot);
5781 BUG_ON(item_size < sizeof(*ei));
5782 ei = btrfs_item_ptr(leaf, extent_slot,
5783 struct btrfs_extent_item);
5784 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5785 key.type == BTRFS_EXTENT_ITEM_KEY) {
5786 struct btrfs_tree_block_info *bi;
5787 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5788 bi = (struct btrfs_tree_block_info *)(ei + 1);
5789 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5792 refs = btrfs_extent_refs(leaf, ei);
5793 if (refs < refs_to_drop) {
5794 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5795 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5797 btrfs_abort_transaction(trans, extent_root, ret);
5800 refs -= refs_to_drop;
5804 __run_delayed_extent_op(extent_op, leaf, ei);
5806 * In the case of inline back ref, reference count will
5807 * be updated by remove_extent_backref
5810 BUG_ON(!found_extent);
5812 btrfs_set_extent_refs(leaf, ei, refs);
5813 btrfs_mark_buffer_dirty(leaf);
5816 ret = remove_extent_backref(trans, extent_root, path,
5820 btrfs_abort_transaction(trans, extent_root, ret);
5824 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5828 BUG_ON(is_data && refs_to_drop !=
5829 extent_data_ref_count(root, path, iref));
5831 BUG_ON(path->slots[0] != extent_slot);
5833 BUG_ON(path->slots[0] != extent_slot + 1);
5834 path->slots[0] = extent_slot;
5839 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5842 btrfs_abort_transaction(trans, extent_root, ret);
5845 btrfs_release_path(path);
5848 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5850 btrfs_abort_transaction(trans, extent_root, ret);
5855 ret = update_block_group(root, bytenr, num_bytes, 0);
5857 btrfs_abort_transaction(trans, extent_root, ret);
5862 btrfs_free_path(path);
5867 * when we free an block, it is possible (and likely) that we free the last
5868 * delayed ref for that extent as well. This searches the delayed ref tree for
5869 * a given extent, and if there are no other delayed refs to be processed, it
5870 * removes it from the tree.
5872 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5873 struct btrfs_root *root, u64 bytenr)
5875 struct btrfs_delayed_ref_head *head;
5876 struct btrfs_delayed_ref_root *delayed_refs;
5877 struct btrfs_delayed_ref_node *ref;
5878 struct rb_node *node;
5881 delayed_refs = &trans->transaction->delayed_refs;
5882 spin_lock(&delayed_refs->lock);
5883 head = btrfs_find_delayed_ref_head(trans, bytenr);
5887 node = rb_prev(&head->node.rb_node);
5891 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5893 /* there are still entries for this ref, we can't drop it */
5894 if (ref->bytenr == bytenr)
5897 if (head->extent_op) {
5898 if (!head->must_insert_reserved)
5900 btrfs_free_delayed_extent_op(head->extent_op);
5901 head->extent_op = NULL;
5905 * waiting for the lock here would deadlock. If someone else has it
5906 * locked they are already in the process of dropping it anyway
5908 if (!mutex_trylock(&head->mutex))
5912 * at this point we have a head with no other entries. Go
5913 * ahead and process it.
5915 head->node.in_tree = 0;
5916 rb_erase(&head->node.rb_node, &delayed_refs->root);
5918 delayed_refs->num_entries--;
5921 * we don't take a ref on the node because we're removing it from the
5922 * tree, so we just steal the ref the tree was holding.
5924 delayed_refs->num_heads--;
5925 if (list_empty(&head->cluster))
5926 delayed_refs->num_heads_ready--;
5928 list_del_init(&head->cluster);
5929 spin_unlock(&delayed_refs->lock);
5931 BUG_ON(head->extent_op);
5932 if (head->must_insert_reserved)
5935 mutex_unlock(&head->mutex);
5936 btrfs_put_delayed_ref(&head->node);
5939 spin_unlock(&delayed_refs->lock);
5943 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5944 struct btrfs_root *root,
5945 struct extent_buffer *buf,
5946 u64 parent, int last_ref)
5948 struct btrfs_block_group_cache *cache = NULL;
5952 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5953 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5954 buf->start, buf->len,
5955 parent, root->root_key.objectid,
5956 btrfs_header_level(buf),
5957 BTRFS_DROP_DELAYED_REF, NULL, 0);
5958 BUG_ON(ret); /* -ENOMEM */
5964 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5966 if (btrfs_header_generation(buf) == trans->transid) {
5967 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5968 ret = check_ref_cleanup(trans, root, buf->start);
5973 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5974 pin_down_extent(root, cache, buf->start, buf->len, 1);
5978 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5980 btrfs_add_free_space(cache, buf->start, buf->len);
5981 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5982 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5987 add_pinned_bytes(root->fs_info, buf->len,
5988 btrfs_header_level(buf),
5989 root->root_key.objectid);
5992 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5995 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5996 btrfs_put_block_group(cache);
5999 /* Can return -ENOMEM */
6000 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6001 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6002 u64 owner, u64 offset, int for_cow)
6005 struct btrfs_fs_info *fs_info = root->fs_info;
6007 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6010 * tree log blocks never actually go into the extent allocation
6011 * tree, just update pinning info and exit early.
6013 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6014 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6015 /* unlocks the pinned mutex */
6016 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6018 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6019 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6021 parent, root_objectid, (int)owner,
6022 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6024 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6026 parent, root_objectid, owner,
6027 offset, BTRFS_DROP_DELAYED_REF,
6033 static u64 stripe_align(struct btrfs_root *root,
6034 struct btrfs_block_group_cache *cache,
6035 u64 val, u64 num_bytes)
6037 u64 ret = ALIGN(val, root->stripesize);
6042 * when we wait for progress in the block group caching, its because
6043 * our allocation attempt failed at least once. So, we must sleep
6044 * and let some progress happen before we try again.
6046 * This function will sleep at least once waiting for new free space to
6047 * show up, and then it will check the block group free space numbers
6048 * for our min num_bytes. Another option is to have it go ahead
6049 * and look in the rbtree for a free extent of a given size, but this
6052 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6053 * any of the information in this block group.
6055 static noinline void
6056 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6059 struct btrfs_caching_control *caching_ctl;
6061 caching_ctl = get_caching_control(cache);
6065 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6066 (cache->free_space_ctl->free_space >= num_bytes));
6068 put_caching_control(caching_ctl);
6072 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6074 struct btrfs_caching_control *caching_ctl;
6077 caching_ctl = get_caching_control(cache);
6079 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6081 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6082 if (cache->cached == BTRFS_CACHE_ERROR)
6084 put_caching_control(caching_ctl);
6088 int __get_raid_index(u64 flags)
6090 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6091 return BTRFS_RAID_RAID10;
6092 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6093 return BTRFS_RAID_RAID1;
6094 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6095 return BTRFS_RAID_DUP;
6096 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6097 return BTRFS_RAID_RAID0;
6098 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6099 return BTRFS_RAID_RAID5;
6100 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6101 return BTRFS_RAID_RAID6;
6103 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6106 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6108 return __get_raid_index(cache->flags);
6111 enum btrfs_loop_type {
6112 LOOP_CACHING_NOWAIT = 0,
6113 LOOP_CACHING_WAIT = 1,
6114 LOOP_ALLOC_CHUNK = 2,
6115 LOOP_NO_EMPTY_SIZE = 3,
6119 * walks the btree of allocated extents and find a hole of a given size.
6120 * The key ins is changed to record the hole:
6121 * ins->objectid == start position
6122 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6123 * ins->offset == the size of the hole.
6124 * Any available blocks before search_start are skipped.
6126 * If there is no suitable free space, we will record the max size of
6127 * the free space extent currently.
6129 static noinline int find_free_extent(struct btrfs_root *orig_root,
6130 u64 num_bytes, u64 empty_size,
6131 u64 hint_byte, struct btrfs_key *ins,
6135 struct btrfs_root *root = orig_root->fs_info->extent_root;
6136 struct btrfs_free_cluster *last_ptr = NULL;
6137 struct btrfs_block_group_cache *block_group = NULL;
6138 struct btrfs_block_group_cache *used_block_group;
6139 u64 search_start = 0;
6140 u64 max_extent_size = 0;
6141 int empty_cluster = 2 * 1024 * 1024;
6142 struct btrfs_space_info *space_info;
6144 int index = __get_raid_index(flags);
6145 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6146 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6147 bool found_uncached_bg = false;
6148 bool failed_cluster_refill = false;
6149 bool failed_alloc = false;
6150 bool use_cluster = true;
6151 bool have_caching_bg = false;
6153 WARN_ON(num_bytes < root->sectorsize);
6154 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6158 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6160 space_info = __find_space_info(root->fs_info, flags);
6162 btrfs_err(root->fs_info, "No space info for %llu", flags);
6167 * If the space info is for both data and metadata it means we have a
6168 * small filesystem and we can't use the clustering stuff.
6170 if (btrfs_mixed_space_info(space_info))
6171 use_cluster = false;
6173 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6174 last_ptr = &root->fs_info->meta_alloc_cluster;
6175 if (!btrfs_test_opt(root, SSD))
6176 empty_cluster = 64 * 1024;
6179 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6180 btrfs_test_opt(root, SSD)) {
6181 last_ptr = &root->fs_info->data_alloc_cluster;
6185 spin_lock(&last_ptr->lock);
6186 if (last_ptr->block_group)
6187 hint_byte = last_ptr->window_start;
6188 spin_unlock(&last_ptr->lock);
6191 search_start = max(search_start, first_logical_byte(root, 0));
6192 search_start = max(search_start, hint_byte);
6197 if (search_start == hint_byte) {
6198 block_group = btrfs_lookup_block_group(root->fs_info,
6200 used_block_group = block_group;
6202 * we don't want to use the block group if it doesn't match our
6203 * allocation bits, or if its not cached.
6205 * However if we are re-searching with an ideal block group
6206 * picked out then we don't care that the block group is cached.
6208 if (block_group && block_group_bits(block_group, flags) &&
6209 block_group->cached != BTRFS_CACHE_NO) {
6210 down_read(&space_info->groups_sem);
6211 if (list_empty(&block_group->list) ||
6214 * someone is removing this block group,
6215 * we can't jump into the have_block_group
6216 * target because our list pointers are not
6219 btrfs_put_block_group(block_group);
6220 up_read(&space_info->groups_sem);
6222 index = get_block_group_index(block_group);
6223 goto have_block_group;
6225 } else if (block_group) {
6226 btrfs_put_block_group(block_group);
6230 have_caching_bg = false;
6231 down_read(&space_info->groups_sem);
6232 list_for_each_entry(block_group, &space_info->block_groups[index],
6237 used_block_group = block_group;
6238 btrfs_get_block_group(block_group);
6239 search_start = block_group->key.objectid;
6242 * this can happen if we end up cycling through all the
6243 * raid types, but we want to make sure we only allocate
6244 * for the proper type.
6246 if (!block_group_bits(block_group, flags)) {
6247 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6248 BTRFS_BLOCK_GROUP_RAID1 |
6249 BTRFS_BLOCK_GROUP_RAID5 |
6250 BTRFS_BLOCK_GROUP_RAID6 |
6251 BTRFS_BLOCK_GROUP_RAID10;
6254 * if they asked for extra copies and this block group
6255 * doesn't provide them, bail. This does allow us to
6256 * fill raid0 from raid1.
6258 if ((flags & extra) && !(block_group->flags & extra))
6263 cached = block_group_cache_done(block_group);
6264 if (unlikely(!cached)) {
6265 found_uncached_bg = true;
6266 ret = cache_block_group(block_group, 0);
6271 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6273 if (unlikely(block_group->ro))
6277 * Ok we want to try and use the cluster allocator, so
6281 unsigned long aligned_cluster;
6283 * the refill lock keeps out other
6284 * people trying to start a new cluster
6286 spin_lock(&last_ptr->refill_lock);
6287 used_block_group = last_ptr->block_group;
6288 if (used_block_group != block_group &&
6289 (!used_block_group ||
6290 used_block_group->ro ||
6291 !block_group_bits(used_block_group, flags))) {
6292 used_block_group = block_group;
6293 goto refill_cluster;
6296 if (used_block_group != block_group)
6297 btrfs_get_block_group(used_block_group);
6299 offset = btrfs_alloc_from_cluster(used_block_group,
6302 used_block_group->key.objectid,
6305 /* we have a block, we're done */
6306 spin_unlock(&last_ptr->refill_lock);
6307 trace_btrfs_reserve_extent_cluster(root,
6308 block_group, search_start, num_bytes);
6312 WARN_ON(last_ptr->block_group != used_block_group);
6313 if (used_block_group != block_group) {
6314 btrfs_put_block_group(used_block_group);
6315 used_block_group = block_group;
6318 BUG_ON(used_block_group != block_group);
6319 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6320 * set up a new clusters, so lets just skip it
6321 * and let the allocator find whatever block
6322 * it can find. If we reach this point, we
6323 * will have tried the cluster allocator
6324 * plenty of times and not have found
6325 * anything, so we are likely way too
6326 * fragmented for the clustering stuff to find
6329 * However, if the cluster is taken from the
6330 * current block group, release the cluster
6331 * first, so that we stand a better chance of
6332 * succeeding in the unclustered
6334 if (loop >= LOOP_NO_EMPTY_SIZE &&
6335 last_ptr->block_group != block_group) {
6336 spin_unlock(&last_ptr->refill_lock);
6337 goto unclustered_alloc;
6341 * this cluster didn't work out, free it and
6344 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6346 if (loop >= LOOP_NO_EMPTY_SIZE) {
6347 spin_unlock(&last_ptr->refill_lock);
6348 goto unclustered_alloc;
6351 aligned_cluster = max_t(unsigned long,
6352 empty_cluster + empty_size,
6353 block_group->full_stripe_len);
6355 /* allocate a cluster in this block group */
6356 ret = btrfs_find_space_cluster(root, block_group,
6357 last_ptr, search_start,
6362 * now pull our allocation out of this
6365 offset = btrfs_alloc_from_cluster(block_group,
6371 /* we found one, proceed */
6372 spin_unlock(&last_ptr->refill_lock);
6373 trace_btrfs_reserve_extent_cluster(root,
6374 block_group, search_start,
6378 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6379 && !failed_cluster_refill) {
6380 spin_unlock(&last_ptr->refill_lock);
6382 failed_cluster_refill = true;
6383 wait_block_group_cache_progress(block_group,
6384 num_bytes + empty_cluster + empty_size);
6385 goto have_block_group;
6389 * at this point we either didn't find a cluster
6390 * or we weren't able to allocate a block from our
6391 * cluster. Free the cluster we've been trying
6392 * to use, and go to the next block group
6394 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6395 spin_unlock(&last_ptr->refill_lock);
6400 spin_lock(&block_group->free_space_ctl->tree_lock);
6402 block_group->free_space_ctl->free_space <
6403 num_bytes + empty_cluster + empty_size) {
6404 if (block_group->free_space_ctl->free_space >
6407 block_group->free_space_ctl->free_space;
6408 spin_unlock(&block_group->free_space_ctl->tree_lock);
6411 spin_unlock(&block_group->free_space_ctl->tree_lock);
6413 offset = btrfs_find_space_for_alloc(block_group, search_start,
6414 num_bytes, empty_size,
6417 * If we didn't find a chunk, and we haven't failed on this
6418 * block group before, and this block group is in the middle of
6419 * caching and we are ok with waiting, then go ahead and wait
6420 * for progress to be made, and set failed_alloc to true.
6422 * If failed_alloc is true then we've already waited on this
6423 * block group once and should move on to the next block group.
6425 if (!offset && !failed_alloc && !cached &&
6426 loop > LOOP_CACHING_NOWAIT) {
6427 wait_block_group_cache_progress(block_group,
6428 num_bytes + empty_size);
6429 failed_alloc = true;
6430 goto have_block_group;
6431 } else if (!offset) {
6433 have_caching_bg = true;
6437 search_start = stripe_align(root, used_block_group,
6440 /* move on to the next group */
6441 if (search_start + num_bytes >
6442 used_block_group->key.objectid + used_block_group->key.offset) {
6443 btrfs_add_free_space(used_block_group, offset, num_bytes);
6447 if (offset < search_start)
6448 btrfs_add_free_space(used_block_group, offset,
6449 search_start - offset);
6450 BUG_ON(offset > search_start);
6452 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6454 if (ret == -EAGAIN) {
6455 btrfs_add_free_space(used_block_group, offset, num_bytes);
6459 /* we are all good, lets return */
6460 ins->objectid = search_start;
6461 ins->offset = num_bytes;
6463 trace_btrfs_reserve_extent(orig_root, block_group,
6464 search_start, num_bytes);
6465 if (used_block_group != block_group)
6466 btrfs_put_block_group(used_block_group);
6467 btrfs_put_block_group(block_group);
6470 failed_cluster_refill = false;
6471 failed_alloc = false;
6472 BUG_ON(index != get_block_group_index(block_group));
6473 if (used_block_group != block_group)
6474 btrfs_put_block_group(used_block_group);
6475 btrfs_put_block_group(block_group);
6477 up_read(&space_info->groups_sem);
6479 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6482 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6486 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6487 * caching kthreads as we move along
6488 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6489 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6490 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6493 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6496 if (loop == LOOP_ALLOC_CHUNK) {
6497 struct btrfs_trans_handle *trans;
6499 trans = btrfs_join_transaction(root);
6500 if (IS_ERR(trans)) {
6501 ret = PTR_ERR(trans);
6505 ret = do_chunk_alloc(trans, root, flags,
6508 * Do not bail out on ENOSPC since we
6509 * can do more things.
6511 if (ret < 0 && ret != -ENOSPC)
6512 btrfs_abort_transaction(trans,
6516 btrfs_end_transaction(trans, root);
6521 if (loop == LOOP_NO_EMPTY_SIZE) {
6527 } else if (!ins->objectid) {
6529 } else if (ins->objectid) {
6534 ins->offset = max_extent_size;
6538 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6539 int dump_block_groups)
6541 struct btrfs_block_group_cache *cache;
6544 spin_lock(&info->lock);
6545 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6547 info->total_bytes - info->bytes_used - info->bytes_pinned -
6548 info->bytes_reserved - info->bytes_readonly,
6549 (info->full) ? "" : "not ");
6550 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6551 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6552 info->total_bytes, info->bytes_used, info->bytes_pinned,
6553 info->bytes_reserved, info->bytes_may_use,
6554 info->bytes_readonly);
6555 spin_unlock(&info->lock);
6557 if (!dump_block_groups)
6560 down_read(&info->groups_sem);
6562 list_for_each_entry(cache, &info->block_groups[index], list) {
6563 spin_lock(&cache->lock);
6564 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6565 cache->key.objectid, cache->key.offset,
6566 btrfs_block_group_used(&cache->item), cache->pinned,
6567 cache->reserved, cache->ro ? "[readonly]" : "");
6568 btrfs_dump_free_space(cache, bytes);
6569 spin_unlock(&cache->lock);
6571 if (++index < BTRFS_NR_RAID_TYPES)
6573 up_read(&info->groups_sem);
6576 int btrfs_reserve_extent(struct btrfs_root *root,
6577 u64 num_bytes, u64 min_alloc_size,
6578 u64 empty_size, u64 hint_byte,
6579 struct btrfs_key *ins, int is_data)
6581 bool final_tried = false;
6585 flags = btrfs_get_alloc_profile(root, is_data);
6587 WARN_ON(num_bytes < root->sectorsize);
6588 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6591 if (ret == -ENOSPC) {
6592 if (!final_tried && ins->offset) {
6593 num_bytes = min(num_bytes >> 1, ins->offset);
6594 num_bytes = round_down(num_bytes, root->sectorsize);
6595 num_bytes = max(num_bytes, min_alloc_size);
6596 if (num_bytes == min_alloc_size)
6599 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6600 struct btrfs_space_info *sinfo;
6602 sinfo = __find_space_info(root->fs_info, flags);
6603 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6606 dump_space_info(sinfo, num_bytes, 1);
6613 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6614 u64 start, u64 len, int pin)
6616 struct btrfs_block_group_cache *cache;
6619 cache = btrfs_lookup_block_group(root->fs_info, start);
6621 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6626 if (btrfs_test_opt(root, DISCARD))
6627 ret = btrfs_discard_extent(root, start, len, NULL);
6630 pin_down_extent(root, cache, start, len, 1);
6632 btrfs_add_free_space(cache, start, len);
6633 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6635 btrfs_put_block_group(cache);
6637 trace_btrfs_reserved_extent_free(root, start, len);
6642 int btrfs_free_reserved_extent(struct btrfs_root *root,
6645 return __btrfs_free_reserved_extent(root, start, len, 0);
6648 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6651 return __btrfs_free_reserved_extent(root, start, len, 1);
6654 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6655 struct btrfs_root *root,
6656 u64 parent, u64 root_objectid,
6657 u64 flags, u64 owner, u64 offset,
6658 struct btrfs_key *ins, int ref_mod)
6661 struct btrfs_fs_info *fs_info = root->fs_info;
6662 struct btrfs_extent_item *extent_item;
6663 struct btrfs_extent_inline_ref *iref;
6664 struct btrfs_path *path;
6665 struct extent_buffer *leaf;
6670 type = BTRFS_SHARED_DATA_REF_KEY;
6672 type = BTRFS_EXTENT_DATA_REF_KEY;
6674 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6676 path = btrfs_alloc_path();
6680 path->leave_spinning = 1;
6681 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6684 btrfs_free_path(path);
6688 leaf = path->nodes[0];
6689 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6690 struct btrfs_extent_item);
6691 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6692 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6693 btrfs_set_extent_flags(leaf, extent_item,
6694 flags | BTRFS_EXTENT_FLAG_DATA);
6696 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6697 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6699 struct btrfs_shared_data_ref *ref;
6700 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6701 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6702 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6704 struct btrfs_extent_data_ref *ref;
6705 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6706 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6707 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6708 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6709 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6712 btrfs_mark_buffer_dirty(path->nodes[0]);
6713 btrfs_free_path(path);
6715 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6716 if (ret) { /* -ENOENT, logic error */
6717 btrfs_err(fs_info, "update block group failed for %llu %llu",
6718 ins->objectid, ins->offset);
6721 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6725 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6726 struct btrfs_root *root,
6727 u64 parent, u64 root_objectid,
6728 u64 flags, struct btrfs_disk_key *key,
6729 int level, struct btrfs_key *ins)
6732 struct btrfs_fs_info *fs_info = root->fs_info;
6733 struct btrfs_extent_item *extent_item;
6734 struct btrfs_tree_block_info *block_info;
6735 struct btrfs_extent_inline_ref *iref;
6736 struct btrfs_path *path;
6737 struct extent_buffer *leaf;
6738 u32 size = sizeof(*extent_item) + sizeof(*iref);
6739 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6742 if (!skinny_metadata)
6743 size += sizeof(*block_info);
6745 path = btrfs_alloc_path();
6747 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6752 path->leave_spinning = 1;
6753 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6756 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6758 btrfs_free_path(path);
6762 leaf = path->nodes[0];
6763 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6764 struct btrfs_extent_item);
6765 btrfs_set_extent_refs(leaf, extent_item, 1);
6766 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6767 btrfs_set_extent_flags(leaf, extent_item,
6768 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6770 if (skinny_metadata) {
6771 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6773 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6774 btrfs_set_tree_block_key(leaf, block_info, key);
6775 btrfs_set_tree_block_level(leaf, block_info, level);
6776 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6780 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6781 btrfs_set_extent_inline_ref_type(leaf, iref,
6782 BTRFS_SHARED_BLOCK_REF_KEY);
6783 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6785 btrfs_set_extent_inline_ref_type(leaf, iref,
6786 BTRFS_TREE_BLOCK_REF_KEY);
6787 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6790 btrfs_mark_buffer_dirty(leaf);
6791 btrfs_free_path(path);
6793 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6794 if (ret) { /* -ENOENT, logic error */
6795 btrfs_err(fs_info, "update block group failed for %llu %llu",
6796 ins->objectid, ins->offset);
6800 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6804 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6805 struct btrfs_root *root,
6806 u64 root_objectid, u64 owner,
6807 u64 offset, struct btrfs_key *ins)
6811 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6813 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6815 root_objectid, owner, offset,
6816 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6821 * this is used by the tree logging recovery code. It records that
6822 * an extent has been allocated and makes sure to clear the free
6823 * space cache bits as well
6825 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6826 struct btrfs_root *root,
6827 u64 root_objectid, u64 owner, u64 offset,
6828 struct btrfs_key *ins)
6831 struct btrfs_block_group_cache *block_group;
6834 * Mixed block groups will exclude before processing the log so we only
6835 * need to do the exlude dance if this fs isn't mixed.
6837 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6838 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6843 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6847 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6848 RESERVE_ALLOC_NO_ACCOUNT);
6849 BUG_ON(ret); /* logic error */
6850 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6851 0, owner, offset, ins, 1);
6852 btrfs_put_block_group(block_group);
6856 static struct extent_buffer *
6857 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6858 u64 bytenr, u32 blocksize, int level)
6860 struct extent_buffer *buf;
6862 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6864 return ERR_PTR(-ENOMEM);
6865 btrfs_set_header_generation(buf, trans->transid);
6866 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6867 btrfs_tree_lock(buf);
6868 clean_tree_block(trans, root, buf);
6869 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6871 btrfs_set_lock_blocking(buf);
6872 btrfs_set_buffer_uptodate(buf);
6874 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6876 * we allow two log transactions at a time, use different
6877 * EXENT bit to differentiate dirty pages.
6879 if (root->log_transid % 2 == 0)
6880 set_extent_dirty(&root->dirty_log_pages, buf->start,
6881 buf->start + buf->len - 1, GFP_NOFS);
6883 set_extent_new(&root->dirty_log_pages, buf->start,
6884 buf->start + buf->len - 1, GFP_NOFS);
6886 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6887 buf->start + buf->len - 1, GFP_NOFS);
6889 trans->blocks_used++;
6890 /* this returns a buffer locked for blocking */
6894 static struct btrfs_block_rsv *
6895 use_block_rsv(struct btrfs_trans_handle *trans,
6896 struct btrfs_root *root, u32 blocksize)
6898 struct btrfs_block_rsv *block_rsv;
6899 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6901 bool global_updated = false;
6903 block_rsv = get_block_rsv(trans, root);
6905 if (unlikely(block_rsv->size == 0))
6908 ret = block_rsv_use_bytes(block_rsv, blocksize);
6912 if (block_rsv->failfast)
6913 return ERR_PTR(ret);
6915 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6916 global_updated = true;
6917 update_global_block_rsv(root->fs_info);
6921 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6922 static DEFINE_RATELIMIT_STATE(_rs,
6923 DEFAULT_RATELIMIT_INTERVAL * 10,
6924 /*DEFAULT_RATELIMIT_BURST*/ 1);
6925 if (__ratelimit(&_rs))
6927 "btrfs: block rsv returned %d\n", ret);
6930 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6931 BTRFS_RESERVE_NO_FLUSH);
6935 * If we couldn't reserve metadata bytes try and use some from
6936 * the global reserve if its space type is the same as the global
6939 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6940 block_rsv->space_info == global_rsv->space_info) {
6941 ret = block_rsv_use_bytes(global_rsv, blocksize);
6945 return ERR_PTR(ret);
6948 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6949 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6951 block_rsv_add_bytes(block_rsv, blocksize, 0);
6952 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6956 * finds a free extent and does all the dirty work required for allocation
6957 * returns the key for the extent through ins, and a tree buffer for
6958 * the first block of the extent through buf.
6960 * returns the tree buffer or NULL.
6962 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6963 struct btrfs_root *root, u32 blocksize,
6964 u64 parent, u64 root_objectid,
6965 struct btrfs_disk_key *key, int level,
6966 u64 hint, u64 empty_size)
6968 struct btrfs_key ins;
6969 struct btrfs_block_rsv *block_rsv;
6970 struct extent_buffer *buf;
6973 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6976 block_rsv = use_block_rsv(trans, root, blocksize);
6977 if (IS_ERR(block_rsv))
6978 return ERR_CAST(block_rsv);
6980 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6981 empty_size, hint, &ins, 0);
6983 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6984 return ERR_PTR(ret);
6987 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6989 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6991 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6993 parent = ins.objectid;
6994 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6998 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6999 struct btrfs_delayed_extent_op *extent_op;
7000 extent_op = btrfs_alloc_delayed_extent_op();
7001 BUG_ON(!extent_op); /* -ENOMEM */
7003 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7005 memset(&extent_op->key, 0, sizeof(extent_op->key));
7006 extent_op->flags_to_set = flags;
7007 if (skinny_metadata)
7008 extent_op->update_key = 0;
7010 extent_op->update_key = 1;
7011 extent_op->update_flags = 1;
7012 extent_op->is_data = 0;
7013 extent_op->level = level;
7015 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7017 ins.offset, parent, root_objectid,
7018 level, BTRFS_ADD_DELAYED_EXTENT,
7020 BUG_ON(ret); /* -ENOMEM */
7025 struct walk_control {
7026 u64 refs[BTRFS_MAX_LEVEL];
7027 u64 flags[BTRFS_MAX_LEVEL];
7028 struct btrfs_key update_progress;
7039 #define DROP_REFERENCE 1
7040 #define UPDATE_BACKREF 2
7042 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7043 struct btrfs_root *root,
7044 struct walk_control *wc,
7045 struct btrfs_path *path)
7053 struct btrfs_key key;
7054 struct extent_buffer *eb;
7059 if (path->slots[wc->level] < wc->reada_slot) {
7060 wc->reada_count = wc->reada_count * 2 / 3;
7061 wc->reada_count = max(wc->reada_count, 2);
7063 wc->reada_count = wc->reada_count * 3 / 2;
7064 wc->reada_count = min_t(int, wc->reada_count,
7065 BTRFS_NODEPTRS_PER_BLOCK(root));
7068 eb = path->nodes[wc->level];
7069 nritems = btrfs_header_nritems(eb);
7070 blocksize = btrfs_level_size(root, wc->level - 1);
7072 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7073 if (nread >= wc->reada_count)
7077 bytenr = btrfs_node_blockptr(eb, slot);
7078 generation = btrfs_node_ptr_generation(eb, slot);
7080 if (slot == path->slots[wc->level])
7083 if (wc->stage == UPDATE_BACKREF &&
7084 generation <= root->root_key.offset)
7087 /* We don't lock the tree block, it's OK to be racy here */
7088 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7089 wc->level - 1, 1, &refs,
7091 /* We don't care about errors in readahead. */
7096 if (wc->stage == DROP_REFERENCE) {
7100 if (wc->level == 1 &&
7101 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7103 if (!wc->update_ref ||
7104 generation <= root->root_key.offset)
7106 btrfs_node_key_to_cpu(eb, &key, slot);
7107 ret = btrfs_comp_cpu_keys(&key,
7108 &wc->update_progress);
7112 if (wc->level == 1 &&
7113 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7117 ret = readahead_tree_block(root, bytenr, blocksize,
7123 wc->reada_slot = slot;
7127 * helper to process tree block while walking down the tree.
7129 * when wc->stage == UPDATE_BACKREF, this function updates
7130 * back refs for pointers in the block.
7132 * NOTE: return value 1 means we should stop walking down.
7134 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7135 struct btrfs_root *root,
7136 struct btrfs_path *path,
7137 struct walk_control *wc, int lookup_info)
7139 int level = wc->level;
7140 struct extent_buffer *eb = path->nodes[level];
7141 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7144 if (wc->stage == UPDATE_BACKREF &&
7145 btrfs_header_owner(eb) != root->root_key.objectid)
7149 * when reference count of tree block is 1, it won't increase
7150 * again. once full backref flag is set, we never clear it.
7153 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7154 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7155 BUG_ON(!path->locks[level]);
7156 ret = btrfs_lookup_extent_info(trans, root,
7157 eb->start, level, 1,
7160 BUG_ON(ret == -ENOMEM);
7163 BUG_ON(wc->refs[level] == 0);
7166 if (wc->stage == DROP_REFERENCE) {
7167 if (wc->refs[level] > 1)
7170 if (path->locks[level] && !wc->keep_locks) {
7171 btrfs_tree_unlock_rw(eb, path->locks[level]);
7172 path->locks[level] = 0;
7177 /* wc->stage == UPDATE_BACKREF */
7178 if (!(wc->flags[level] & flag)) {
7179 BUG_ON(!path->locks[level]);
7180 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7181 BUG_ON(ret); /* -ENOMEM */
7182 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7183 BUG_ON(ret); /* -ENOMEM */
7184 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7186 btrfs_header_level(eb), 0);
7187 BUG_ON(ret); /* -ENOMEM */
7188 wc->flags[level] |= flag;
7192 * the block is shared by multiple trees, so it's not good to
7193 * keep the tree lock
7195 if (path->locks[level] && level > 0) {
7196 btrfs_tree_unlock_rw(eb, path->locks[level]);
7197 path->locks[level] = 0;
7203 * helper to process tree block pointer.
7205 * when wc->stage == DROP_REFERENCE, this function checks
7206 * reference count of the block pointed to. if the block
7207 * is shared and we need update back refs for the subtree
7208 * rooted at the block, this function changes wc->stage to
7209 * UPDATE_BACKREF. if the block is shared and there is no
7210 * need to update back, this function drops the reference
7213 * NOTE: return value 1 means we should stop walking down.
7215 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7216 struct btrfs_root *root,
7217 struct btrfs_path *path,
7218 struct walk_control *wc, int *lookup_info)
7224 struct btrfs_key key;
7225 struct extent_buffer *next;
7226 int level = wc->level;
7230 generation = btrfs_node_ptr_generation(path->nodes[level],
7231 path->slots[level]);
7233 * if the lower level block was created before the snapshot
7234 * was created, we know there is no need to update back refs
7237 if (wc->stage == UPDATE_BACKREF &&
7238 generation <= root->root_key.offset) {
7243 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7244 blocksize = btrfs_level_size(root, level - 1);
7246 next = btrfs_find_tree_block(root, bytenr, blocksize);
7248 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7251 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7255 btrfs_tree_lock(next);
7256 btrfs_set_lock_blocking(next);
7258 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7259 &wc->refs[level - 1],
7260 &wc->flags[level - 1]);
7262 btrfs_tree_unlock(next);
7266 if (unlikely(wc->refs[level - 1] == 0)) {
7267 btrfs_err(root->fs_info, "Missing references.");
7272 if (wc->stage == DROP_REFERENCE) {
7273 if (wc->refs[level - 1] > 1) {
7275 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7278 if (!wc->update_ref ||
7279 generation <= root->root_key.offset)
7282 btrfs_node_key_to_cpu(path->nodes[level], &key,
7283 path->slots[level]);
7284 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7288 wc->stage = UPDATE_BACKREF;
7289 wc->shared_level = level - 1;
7293 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7297 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7298 btrfs_tree_unlock(next);
7299 free_extent_buffer(next);
7305 if (reada && level == 1)
7306 reada_walk_down(trans, root, wc, path);
7307 next = read_tree_block(root, bytenr, blocksize, generation);
7308 if (!next || !extent_buffer_uptodate(next)) {
7309 free_extent_buffer(next);
7312 btrfs_tree_lock(next);
7313 btrfs_set_lock_blocking(next);
7317 BUG_ON(level != btrfs_header_level(next));
7318 path->nodes[level] = next;
7319 path->slots[level] = 0;
7320 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7326 wc->refs[level - 1] = 0;
7327 wc->flags[level - 1] = 0;
7328 if (wc->stage == DROP_REFERENCE) {
7329 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7330 parent = path->nodes[level]->start;
7332 BUG_ON(root->root_key.objectid !=
7333 btrfs_header_owner(path->nodes[level]));
7337 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7338 root->root_key.objectid, level - 1, 0, 0);
7339 BUG_ON(ret); /* -ENOMEM */
7341 btrfs_tree_unlock(next);
7342 free_extent_buffer(next);
7348 * helper to process tree block while walking up the tree.
7350 * when wc->stage == DROP_REFERENCE, this function drops
7351 * reference count on the block.
7353 * when wc->stage == UPDATE_BACKREF, this function changes
7354 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7355 * to UPDATE_BACKREF previously while processing the block.
7357 * NOTE: return value 1 means we should stop walking up.
7359 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7360 struct btrfs_root *root,
7361 struct btrfs_path *path,
7362 struct walk_control *wc)
7365 int level = wc->level;
7366 struct extent_buffer *eb = path->nodes[level];
7369 if (wc->stage == UPDATE_BACKREF) {
7370 BUG_ON(wc->shared_level < level);
7371 if (level < wc->shared_level)
7374 ret = find_next_key(path, level + 1, &wc->update_progress);
7378 wc->stage = DROP_REFERENCE;
7379 wc->shared_level = -1;
7380 path->slots[level] = 0;
7383 * check reference count again if the block isn't locked.
7384 * we should start walking down the tree again if reference
7387 if (!path->locks[level]) {
7389 btrfs_tree_lock(eb);
7390 btrfs_set_lock_blocking(eb);
7391 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7393 ret = btrfs_lookup_extent_info(trans, root,
7394 eb->start, level, 1,
7398 btrfs_tree_unlock_rw(eb, path->locks[level]);
7399 path->locks[level] = 0;
7402 BUG_ON(wc->refs[level] == 0);
7403 if (wc->refs[level] == 1) {
7404 btrfs_tree_unlock_rw(eb, path->locks[level]);
7405 path->locks[level] = 0;
7411 /* wc->stage == DROP_REFERENCE */
7412 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7414 if (wc->refs[level] == 1) {
7416 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7417 ret = btrfs_dec_ref(trans, root, eb, 1,
7420 ret = btrfs_dec_ref(trans, root, eb, 0,
7422 BUG_ON(ret); /* -ENOMEM */
7424 /* make block locked assertion in clean_tree_block happy */
7425 if (!path->locks[level] &&
7426 btrfs_header_generation(eb) == trans->transid) {
7427 btrfs_tree_lock(eb);
7428 btrfs_set_lock_blocking(eb);
7429 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7431 clean_tree_block(trans, root, eb);
7434 if (eb == root->node) {
7435 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7438 BUG_ON(root->root_key.objectid !=
7439 btrfs_header_owner(eb));
7441 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7442 parent = path->nodes[level + 1]->start;
7444 BUG_ON(root->root_key.objectid !=
7445 btrfs_header_owner(path->nodes[level + 1]));
7448 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7450 wc->refs[level] = 0;
7451 wc->flags[level] = 0;
7455 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7456 struct btrfs_root *root,
7457 struct btrfs_path *path,
7458 struct walk_control *wc)
7460 int level = wc->level;
7461 int lookup_info = 1;
7464 while (level >= 0) {
7465 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7472 if (path->slots[level] >=
7473 btrfs_header_nritems(path->nodes[level]))
7476 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7478 path->slots[level]++;
7487 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7488 struct btrfs_root *root,
7489 struct btrfs_path *path,
7490 struct walk_control *wc, int max_level)
7492 int level = wc->level;
7495 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7496 while (level < max_level && path->nodes[level]) {
7498 if (path->slots[level] + 1 <
7499 btrfs_header_nritems(path->nodes[level])) {
7500 path->slots[level]++;
7503 ret = walk_up_proc(trans, root, path, wc);
7507 if (path->locks[level]) {
7508 btrfs_tree_unlock_rw(path->nodes[level],
7509 path->locks[level]);
7510 path->locks[level] = 0;
7512 free_extent_buffer(path->nodes[level]);
7513 path->nodes[level] = NULL;
7521 * drop a subvolume tree.
7523 * this function traverses the tree freeing any blocks that only
7524 * referenced by the tree.
7526 * when a shared tree block is found. this function decreases its
7527 * reference count by one. if update_ref is true, this function
7528 * also make sure backrefs for the shared block and all lower level
7529 * blocks are properly updated.
7531 * If called with for_reloc == 0, may exit early with -EAGAIN
7533 int btrfs_drop_snapshot(struct btrfs_root *root,
7534 struct btrfs_block_rsv *block_rsv, int update_ref,
7537 struct btrfs_path *path;
7538 struct btrfs_trans_handle *trans;
7539 struct btrfs_root *tree_root = root->fs_info->tree_root;
7540 struct btrfs_root_item *root_item = &root->root_item;
7541 struct walk_control *wc;
7542 struct btrfs_key key;
7546 bool root_dropped = false;
7548 path = btrfs_alloc_path();
7554 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7556 btrfs_free_path(path);
7561 trans = btrfs_start_transaction(tree_root, 0);
7562 if (IS_ERR(trans)) {
7563 err = PTR_ERR(trans);
7568 trans->block_rsv = block_rsv;
7570 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7571 level = btrfs_header_level(root->node);
7572 path->nodes[level] = btrfs_lock_root_node(root);
7573 btrfs_set_lock_blocking(path->nodes[level]);
7574 path->slots[level] = 0;
7575 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7576 memset(&wc->update_progress, 0,
7577 sizeof(wc->update_progress));
7579 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7580 memcpy(&wc->update_progress, &key,
7581 sizeof(wc->update_progress));
7583 level = root_item->drop_level;
7585 path->lowest_level = level;
7586 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7587 path->lowest_level = 0;
7595 * unlock our path, this is safe because only this
7596 * function is allowed to delete this snapshot
7598 btrfs_unlock_up_safe(path, 0);
7600 level = btrfs_header_level(root->node);
7602 btrfs_tree_lock(path->nodes[level]);
7603 btrfs_set_lock_blocking(path->nodes[level]);
7604 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7606 ret = btrfs_lookup_extent_info(trans, root,
7607 path->nodes[level]->start,
7608 level, 1, &wc->refs[level],
7614 BUG_ON(wc->refs[level] == 0);
7616 if (level == root_item->drop_level)
7619 btrfs_tree_unlock(path->nodes[level]);
7620 path->locks[level] = 0;
7621 WARN_ON(wc->refs[level] != 1);
7627 wc->shared_level = -1;
7628 wc->stage = DROP_REFERENCE;
7629 wc->update_ref = update_ref;
7631 wc->for_reloc = for_reloc;
7632 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7636 ret = walk_down_tree(trans, root, path, wc);
7642 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7649 BUG_ON(wc->stage != DROP_REFERENCE);
7653 if (wc->stage == DROP_REFERENCE) {
7655 btrfs_node_key(path->nodes[level],
7656 &root_item->drop_progress,
7657 path->slots[level]);
7658 root_item->drop_level = level;
7661 BUG_ON(wc->level == 0);
7662 if (btrfs_should_end_transaction(trans, tree_root) ||
7663 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7664 ret = btrfs_update_root(trans, tree_root,
7668 btrfs_abort_transaction(trans, tree_root, ret);
7673 btrfs_end_transaction_throttle(trans, tree_root);
7674 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7675 pr_debug("btrfs: drop snapshot early exit\n");
7680 trans = btrfs_start_transaction(tree_root, 0);
7681 if (IS_ERR(trans)) {
7682 err = PTR_ERR(trans);
7686 trans->block_rsv = block_rsv;
7689 btrfs_release_path(path);
7693 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7695 btrfs_abort_transaction(trans, tree_root, ret);
7699 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7700 ret = btrfs_find_root(tree_root, &root->root_key, path,
7703 btrfs_abort_transaction(trans, tree_root, ret);
7706 } else if (ret > 0) {
7707 /* if we fail to delete the orphan item this time
7708 * around, it'll get picked up the next time.
7710 * The most common failure here is just -ENOENT.
7712 btrfs_del_orphan_item(trans, tree_root,
7713 root->root_key.objectid);
7717 if (root->in_radix) {
7718 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7720 free_extent_buffer(root->node);
7721 free_extent_buffer(root->commit_root);
7722 btrfs_put_fs_root(root);
7724 root_dropped = true;
7726 btrfs_end_transaction_throttle(trans, tree_root);
7729 btrfs_free_path(path);
7732 * So if we need to stop dropping the snapshot for whatever reason we
7733 * need to make sure to add it back to the dead root list so that we
7734 * keep trying to do the work later. This also cleans up roots if we
7735 * don't have it in the radix (like when we recover after a power fail
7736 * or unmount) so we don't leak memory.
7738 if (!for_reloc && root_dropped == false)
7739 btrfs_add_dead_root(root);
7741 btrfs_std_error(root->fs_info, err);
7746 * drop subtree rooted at tree block 'node'.
7748 * NOTE: this function will unlock and release tree block 'node'
7749 * only used by relocation code
7751 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7752 struct btrfs_root *root,
7753 struct extent_buffer *node,
7754 struct extent_buffer *parent)
7756 struct btrfs_path *path;
7757 struct walk_control *wc;
7763 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7765 path = btrfs_alloc_path();
7769 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7771 btrfs_free_path(path);
7775 btrfs_assert_tree_locked(parent);
7776 parent_level = btrfs_header_level(parent);
7777 extent_buffer_get(parent);
7778 path->nodes[parent_level] = parent;
7779 path->slots[parent_level] = btrfs_header_nritems(parent);
7781 btrfs_assert_tree_locked(node);
7782 level = btrfs_header_level(node);
7783 path->nodes[level] = node;
7784 path->slots[level] = 0;
7785 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7787 wc->refs[parent_level] = 1;
7788 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7790 wc->shared_level = -1;
7791 wc->stage = DROP_REFERENCE;
7795 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7798 wret = walk_down_tree(trans, root, path, wc);
7804 wret = walk_up_tree(trans, root, path, wc, parent_level);
7812 btrfs_free_path(path);
7816 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7822 * if restripe for this chunk_type is on pick target profile and
7823 * return, otherwise do the usual balance
7825 stripped = get_restripe_target(root->fs_info, flags);
7827 return extended_to_chunk(stripped);
7830 * we add in the count of missing devices because we want
7831 * to make sure that any RAID levels on a degraded FS
7832 * continue to be honored.
7834 num_devices = root->fs_info->fs_devices->rw_devices +
7835 root->fs_info->fs_devices->missing_devices;
7837 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7838 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7839 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7841 if (num_devices == 1) {
7842 stripped |= BTRFS_BLOCK_GROUP_DUP;
7843 stripped = flags & ~stripped;
7845 /* turn raid0 into single device chunks */
7846 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7849 /* turn mirroring into duplication */
7850 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7851 BTRFS_BLOCK_GROUP_RAID10))
7852 return stripped | BTRFS_BLOCK_GROUP_DUP;
7854 /* they already had raid on here, just return */
7855 if (flags & stripped)
7858 stripped |= BTRFS_BLOCK_GROUP_DUP;
7859 stripped = flags & ~stripped;
7861 /* switch duplicated blocks with raid1 */
7862 if (flags & BTRFS_BLOCK_GROUP_DUP)
7863 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7865 /* this is drive concat, leave it alone */
7871 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7873 struct btrfs_space_info *sinfo = cache->space_info;
7875 u64 min_allocable_bytes;
7880 * We need some metadata space and system metadata space for
7881 * allocating chunks in some corner cases until we force to set
7882 * it to be readonly.
7885 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7887 min_allocable_bytes = 1 * 1024 * 1024;
7889 min_allocable_bytes = 0;
7891 spin_lock(&sinfo->lock);
7892 spin_lock(&cache->lock);
7899 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7900 cache->bytes_super - btrfs_block_group_used(&cache->item);
7902 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7903 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7904 min_allocable_bytes <= sinfo->total_bytes) {
7905 sinfo->bytes_readonly += num_bytes;
7910 spin_unlock(&cache->lock);
7911 spin_unlock(&sinfo->lock);
7915 int btrfs_set_block_group_ro(struct btrfs_root *root,
7916 struct btrfs_block_group_cache *cache)
7919 struct btrfs_trans_handle *trans;
7925 trans = btrfs_join_transaction(root);
7927 return PTR_ERR(trans);
7929 alloc_flags = update_block_group_flags(root, cache->flags);
7930 if (alloc_flags != cache->flags) {
7931 ret = do_chunk_alloc(trans, root, alloc_flags,
7937 ret = set_block_group_ro(cache, 0);
7940 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7941 ret = do_chunk_alloc(trans, root, alloc_flags,
7945 ret = set_block_group_ro(cache, 0);
7947 btrfs_end_transaction(trans, root);
7951 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7952 struct btrfs_root *root, u64 type)
7954 u64 alloc_flags = get_alloc_profile(root, type);
7955 return do_chunk_alloc(trans, root, alloc_flags,
7960 * helper to account the unused space of all the readonly block group in the
7961 * list. takes mirrors into account.
7963 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7965 struct btrfs_block_group_cache *block_group;
7969 list_for_each_entry(block_group, groups_list, list) {
7970 spin_lock(&block_group->lock);
7972 if (!block_group->ro) {
7973 spin_unlock(&block_group->lock);
7977 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7978 BTRFS_BLOCK_GROUP_RAID10 |
7979 BTRFS_BLOCK_GROUP_DUP))
7984 free_bytes += (block_group->key.offset -
7985 btrfs_block_group_used(&block_group->item)) *
7988 spin_unlock(&block_group->lock);
7995 * helper to account the unused space of all the readonly block group in the
7996 * space_info. takes mirrors into account.
7998 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8003 spin_lock(&sinfo->lock);
8005 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8006 if (!list_empty(&sinfo->block_groups[i]))
8007 free_bytes += __btrfs_get_ro_block_group_free_space(
8008 &sinfo->block_groups[i]);
8010 spin_unlock(&sinfo->lock);
8015 void btrfs_set_block_group_rw(struct btrfs_root *root,
8016 struct btrfs_block_group_cache *cache)
8018 struct btrfs_space_info *sinfo = cache->space_info;
8023 spin_lock(&sinfo->lock);
8024 spin_lock(&cache->lock);
8025 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8026 cache->bytes_super - btrfs_block_group_used(&cache->item);
8027 sinfo->bytes_readonly -= num_bytes;
8029 spin_unlock(&cache->lock);
8030 spin_unlock(&sinfo->lock);
8034 * checks to see if its even possible to relocate this block group.
8036 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8037 * ok to go ahead and try.
8039 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8041 struct btrfs_block_group_cache *block_group;
8042 struct btrfs_space_info *space_info;
8043 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8044 struct btrfs_device *device;
8045 struct btrfs_trans_handle *trans;
8054 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8056 /* odd, couldn't find the block group, leave it alone */
8060 min_free = btrfs_block_group_used(&block_group->item);
8062 /* no bytes used, we're good */
8066 space_info = block_group->space_info;
8067 spin_lock(&space_info->lock);
8069 full = space_info->full;
8072 * if this is the last block group we have in this space, we can't
8073 * relocate it unless we're able to allocate a new chunk below.
8075 * Otherwise, we need to make sure we have room in the space to handle
8076 * all of the extents from this block group. If we can, we're good
8078 if ((space_info->total_bytes != block_group->key.offset) &&
8079 (space_info->bytes_used + space_info->bytes_reserved +
8080 space_info->bytes_pinned + space_info->bytes_readonly +
8081 min_free < space_info->total_bytes)) {
8082 spin_unlock(&space_info->lock);
8085 spin_unlock(&space_info->lock);
8088 * ok we don't have enough space, but maybe we have free space on our
8089 * devices to allocate new chunks for relocation, so loop through our
8090 * alloc devices and guess if we have enough space. if this block
8091 * group is going to be restriped, run checks against the target
8092 * profile instead of the current one.
8104 target = get_restripe_target(root->fs_info, block_group->flags);
8106 index = __get_raid_index(extended_to_chunk(target));
8109 * this is just a balance, so if we were marked as full
8110 * we know there is no space for a new chunk
8115 index = get_block_group_index(block_group);
8118 if (index == BTRFS_RAID_RAID10) {
8122 } else if (index == BTRFS_RAID_RAID1) {
8124 } else if (index == BTRFS_RAID_DUP) {
8127 } else if (index == BTRFS_RAID_RAID0) {
8128 dev_min = fs_devices->rw_devices;
8129 do_div(min_free, dev_min);
8132 /* We need to do this so that we can look at pending chunks */
8133 trans = btrfs_join_transaction(root);
8134 if (IS_ERR(trans)) {
8135 ret = PTR_ERR(trans);
8139 mutex_lock(&root->fs_info->chunk_mutex);
8140 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8144 * check to make sure we can actually find a chunk with enough
8145 * space to fit our block group in.
8147 if (device->total_bytes > device->bytes_used + min_free &&
8148 !device->is_tgtdev_for_dev_replace) {
8149 ret = find_free_dev_extent(trans, device, min_free,
8154 if (dev_nr >= dev_min)
8160 mutex_unlock(&root->fs_info->chunk_mutex);
8161 btrfs_end_transaction(trans, root);
8163 btrfs_put_block_group(block_group);
8167 static int find_first_block_group(struct btrfs_root *root,
8168 struct btrfs_path *path, struct btrfs_key *key)
8171 struct btrfs_key found_key;
8172 struct extent_buffer *leaf;
8175 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8180 slot = path->slots[0];
8181 leaf = path->nodes[0];
8182 if (slot >= btrfs_header_nritems(leaf)) {
8183 ret = btrfs_next_leaf(root, path);
8190 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8192 if (found_key.objectid >= key->objectid &&
8193 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8203 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8205 struct btrfs_block_group_cache *block_group;
8209 struct inode *inode;
8211 block_group = btrfs_lookup_first_block_group(info, last);
8212 while (block_group) {
8213 spin_lock(&block_group->lock);
8214 if (block_group->iref)
8216 spin_unlock(&block_group->lock);
8217 block_group = next_block_group(info->tree_root,
8227 inode = block_group->inode;
8228 block_group->iref = 0;
8229 block_group->inode = NULL;
8230 spin_unlock(&block_group->lock);
8232 last = block_group->key.objectid + block_group->key.offset;
8233 btrfs_put_block_group(block_group);
8237 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8239 struct btrfs_block_group_cache *block_group;
8240 struct btrfs_space_info *space_info;
8241 struct btrfs_caching_control *caching_ctl;
8244 down_write(&info->extent_commit_sem);
8245 while (!list_empty(&info->caching_block_groups)) {
8246 caching_ctl = list_entry(info->caching_block_groups.next,
8247 struct btrfs_caching_control, list);
8248 list_del(&caching_ctl->list);
8249 put_caching_control(caching_ctl);
8251 up_write(&info->extent_commit_sem);
8253 spin_lock(&info->block_group_cache_lock);
8254 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8255 block_group = rb_entry(n, struct btrfs_block_group_cache,
8257 rb_erase(&block_group->cache_node,
8258 &info->block_group_cache_tree);
8259 spin_unlock(&info->block_group_cache_lock);
8261 down_write(&block_group->space_info->groups_sem);
8262 list_del(&block_group->list);
8263 up_write(&block_group->space_info->groups_sem);
8265 if (block_group->cached == BTRFS_CACHE_STARTED)
8266 wait_block_group_cache_done(block_group);
8269 * We haven't cached this block group, which means we could
8270 * possibly have excluded extents on this block group.
8272 if (block_group->cached == BTRFS_CACHE_NO ||
8273 block_group->cached == BTRFS_CACHE_ERROR)
8274 free_excluded_extents(info->extent_root, block_group);
8276 btrfs_remove_free_space_cache(block_group);
8277 btrfs_put_block_group(block_group);
8279 spin_lock(&info->block_group_cache_lock);
8281 spin_unlock(&info->block_group_cache_lock);
8283 /* now that all the block groups are freed, go through and
8284 * free all the space_info structs. This is only called during
8285 * the final stages of unmount, and so we know nobody is
8286 * using them. We call synchronize_rcu() once before we start,
8287 * just to be on the safe side.
8291 release_global_block_rsv(info);
8293 while(!list_empty(&info->space_info)) {
8294 space_info = list_entry(info->space_info.next,
8295 struct btrfs_space_info,
8297 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8298 if (space_info->bytes_pinned > 0 ||
8299 space_info->bytes_reserved > 0 ||
8300 space_info->bytes_may_use > 0) {
8302 dump_space_info(space_info, 0, 0);
8305 percpu_counter_destroy(&space_info->total_bytes_pinned);
8306 list_del(&space_info->list);
8312 static void __link_block_group(struct btrfs_space_info *space_info,
8313 struct btrfs_block_group_cache *cache)
8315 int index = get_block_group_index(cache);
8317 down_write(&space_info->groups_sem);
8318 list_add_tail(&cache->list, &space_info->block_groups[index]);
8319 up_write(&space_info->groups_sem);
8322 int btrfs_read_block_groups(struct btrfs_root *root)
8324 struct btrfs_path *path;
8326 struct btrfs_block_group_cache *cache;
8327 struct btrfs_fs_info *info = root->fs_info;
8328 struct btrfs_space_info *space_info;
8329 struct btrfs_key key;
8330 struct btrfs_key found_key;
8331 struct extent_buffer *leaf;
8335 root = info->extent_root;
8338 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8339 path = btrfs_alloc_path();
8344 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8345 if (btrfs_test_opt(root, SPACE_CACHE) &&
8346 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8348 if (btrfs_test_opt(root, CLEAR_CACHE))
8352 ret = find_first_block_group(root, path, &key);
8357 leaf = path->nodes[0];
8358 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8359 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8364 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8366 if (!cache->free_space_ctl) {
8372 atomic_set(&cache->count, 1);
8373 spin_lock_init(&cache->lock);
8374 cache->fs_info = info;
8375 INIT_LIST_HEAD(&cache->list);
8376 INIT_LIST_HEAD(&cache->cluster_list);
8380 * When we mount with old space cache, we need to
8381 * set BTRFS_DC_CLEAR and set dirty flag.
8383 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8384 * truncate the old free space cache inode and
8386 * b) Setting 'dirty flag' makes sure that we flush
8387 * the new space cache info onto disk.
8389 cache->disk_cache_state = BTRFS_DC_CLEAR;
8390 if (btrfs_test_opt(root, SPACE_CACHE))
8394 read_extent_buffer(leaf, &cache->item,
8395 btrfs_item_ptr_offset(leaf, path->slots[0]),
8396 sizeof(cache->item));
8397 memcpy(&cache->key, &found_key, sizeof(found_key));
8399 key.objectid = found_key.objectid + found_key.offset;
8400 btrfs_release_path(path);
8401 cache->flags = btrfs_block_group_flags(&cache->item);
8402 cache->sectorsize = root->sectorsize;
8403 cache->full_stripe_len = btrfs_full_stripe_len(root,
8404 &root->fs_info->mapping_tree,
8405 found_key.objectid);
8406 btrfs_init_free_space_ctl(cache);
8409 * We need to exclude the super stripes now so that the space
8410 * info has super bytes accounted for, otherwise we'll think
8411 * we have more space than we actually do.
8413 ret = exclude_super_stripes(root, cache);
8416 * We may have excluded something, so call this just in
8419 free_excluded_extents(root, cache);
8420 kfree(cache->free_space_ctl);
8426 * check for two cases, either we are full, and therefore
8427 * don't need to bother with the caching work since we won't
8428 * find any space, or we are empty, and we can just add all
8429 * the space in and be done with it. This saves us _alot_ of
8430 * time, particularly in the full case.
8432 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8433 cache->last_byte_to_unpin = (u64)-1;
8434 cache->cached = BTRFS_CACHE_FINISHED;
8435 free_excluded_extents(root, cache);
8436 } else if (btrfs_block_group_used(&cache->item) == 0) {
8437 cache->last_byte_to_unpin = (u64)-1;
8438 cache->cached = BTRFS_CACHE_FINISHED;
8439 add_new_free_space(cache, root->fs_info,
8441 found_key.objectid +
8443 free_excluded_extents(root, cache);
8446 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8448 btrfs_remove_free_space_cache(cache);
8449 btrfs_put_block_group(cache);
8453 ret = update_space_info(info, cache->flags, found_key.offset,
8454 btrfs_block_group_used(&cache->item),
8457 btrfs_remove_free_space_cache(cache);
8458 spin_lock(&info->block_group_cache_lock);
8459 rb_erase(&cache->cache_node,
8460 &info->block_group_cache_tree);
8461 spin_unlock(&info->block_group_cache_lock);
8462 btrfs_put_block_group(cache);
8466 cache->space_info = space_info;
8467 spin_lock(&cache->space_info->lock);
8468 cache->space_info->bytes_readonly += cache->bytes_super;
8469 spin_unlock(&cache->space_info->lock);
8471 __link_block_group(space_info, cache);
8473 set_avail_alloc_bits(root->fs_info, cache->flags);
8474 if (btrfs_chunk_readonly(root, cache->key.objectid))
8475 set_block_group_ro(cache, 1);
8478 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8479 if (!(get_alloc_profile(root, space_info->flags) &
8480 (BTRFS_BLOCK_GROUP_RAID10 |
8481 BTRFS_BLOCK_GROUP_RAID1 |
8482 BTRFS_BLOCK_GROUP_RAID5 |
8483 BTRFS_BLOCK_GROUP_RAID6 |
8484 BTRFS_BLOCK_GROUP_DUP)))
8487 * avoid allocating from un-mirrored block group if there are
8488 * mirrored block groups.
8490 list_for_each_entry(cache,
8491 &space_info->block_groups[BTRFS_RAID_RAID0],
8493 set_block_group_ro(cache, 1);
8494 list_for_each_entry(cache,
8495 &space_info->block_groups[BTRFS_RAID_SINGLE],
8497 set_block_group_ro(cache, 1);
8500 init_global_block_rsv(info);
8503 btrfs_free_path(path);
8507 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8508 struct btrfs_root *root)
8510 struct btrfs_block_group_cache *block_group, *tmp;
8511 struct btrfs_root *extent_root = root->fs_info->extent_root;
8512 struct btrfs_block_group_item item;
8513 struct btrfs_key key;
8516 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8518 list_del_init(&block_group->new_bg_list);
8523 spin_lock(&block_group->lock);
8524 memcpy(&item, &block_group->item, sizeof(item));
8525 memcpy(&key, &block_group->key, sizeof(key));
8526 spin_unlock(&block_group->lock);
8528 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8531 btrfs_abort_transaction(trans, extent_root, ret);
8532 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8533 key.objectid, key.offset);
8535 btrfs_abort_transaction(trans, extent_root, ret);
8539 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8540 struct btrfs_root *root, u64 bytes_used,
8541 u64 type, u64 chunk_objectid, u64 chunk_offset,
8545 struct btrfs_root *extent_root;
8546 struct btrfs_block_group_cache *cache;
8548 extent_root = root->fs_info->extent_root;
8550 root->fs_info->last_trans_log_full_commit = trans->transid;
8552 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8555 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8557 if (!cache->free_space_ctl) {
8562 cache->key.objectid = chunk_offset;
8563 cache->key.offset = size;
8564 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8565 cache->sectorsize = root->sectorsize;
8566 cache->fs_info = root->fs_info;
8567 cache->full_stripe_len = btrfs_full_stripe_len(root,
8568 &root->fs_info->mapping_tree,
8571 atomic_set(&cache->count, 1);
8572 spin_lock_init(&cache->lock);
8573 INIT_LIST_HEAD(&cache->list);
8574 INIT_LIST_HEAD(&cache->cluster_list);
8575 INIT_LIST_HEAD(&cache->new_bg_list);
8577 btrfs_init_free_space_ctl(cache);
8579 btrfs_set_block_group_used(&cache->item, bytes_used);
8580 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8581 cache->flags = type;
8582 btrfs_set_block_group_flags(&cache->item, type);
8584 cache->last_byte_to_unpin = (u64)-1;
8585 cache->cached = BTRFS_CACHE_FINISHED;
8586 ret = exclude_super_stripes(root, cache);
8589 * We may have excluded something, so call this just in
8592 free_excluded_extents(root, cache);
8593 kfree(cache->free_space_ctl);
8598 add_new_free_space(cache, root->fs_info, chunk_offset,
8599 chunk_offset + size);
8601 free_excluded_extents(root, cache);
8603 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8605 btrfs_remove_free_space_cache(cache);
8606 btrfs_put_block_group(cache);
8610 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8611 &cache->space_info);
8613 btrfs_remove_free_space_cache(cache);
8614 spin_lock(&root->fs_info->block_group_cache_lock);
8615 rb_erase(&cache->cache_node,
8616 &root->fs_info->block_group_cache_tree);
8617 spin_unlock(&root->fs_info->block_group_cache_lock);
8618 btrfs_put_block_group(cache);
8621 update_global_block_rsv(root->fs_info);
8623 spin_lock(&cache->space_info->lock);
8624 cache->space_info->bytes_readonly += cache->bytes_super;
8625 spin_unlock(&cache->space_info->lock);
8627 __link_block_group(cache->space_info, cache);
8629 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8631 set_avail_alloc_bits(extent_root->fs_info, type);
8636 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8638 u64 extra_flags = chunk_to_extended(flags) &
8639 BTRFS_EXTENDED_PROFILE_MASK;
8641 write_seqlock(&fs_info->profiles_lock);
8642 if (flags & BTRFS_BLOCK_GROUP_DATA)
8643 fs_info->avail_data_alloc_bits &= ~extra_flags;
8644 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8645 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8646 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8647 fs_info->avail_system_alloc_bits &= ~extra_flags;
8648 write_sequnlock(&fs_info->profiles_lock);
8651 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8652 struct btrfs_root *root, u64 group_start)
8654 struct btrfs_path *path;
8655 struct btrfs_block_group_cache *block_group;
8656 struct btrfs_free_cluster *cluster;
8657 struct btrfs_root *tree_root = root->fs_info->tree_root;
8658 struct btrfs_key key;
8659 struct inode *inode;
8664 root = root->fs_info->extent_root;
8666 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8667 BUG_ON(!block_group);
8668 BUG_ON(!block_group->ro);
8671 * Free the reserved super bytes from this block group before
8674 free_excluded_extents(root, block_group);
8676 memcpy(&key, &block_group->key, sizeof(key));
8677 index = get_block_group_index(block_group);
8678 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8679 BTRFS_BLOCK_GROUP_RAID1 |
8680 BTRFS_BLOCK_GROUP_RAID10))
8685 /* make sure this block group isn't part of an allocation cluster */
8686 cluster = &root->fs_info->data_alloc_cluster;
8687 spin_lock(&cluster->refill_lock);
8688 btrfs_return_cluster_to_free_space(block_group, cluster);
8689 spin_unlock(&cluster->refill_lock);
8692 * make sure this block group isn't part of a metadata
8693 * allocation cluster
8695 cluster = &root->fs_info->meta_alloc_cluster;
8696 spin_lock(&cluster->refill_lock);
8697 btrfs_return_cluster_to_free_space(block_group, cluster);
8698 spin_unlock(&cluster->refill_lock);
8700 path = btrfs_alloc_path();
8706 inode = lookup_free_space_inode(tree_root, block_group, path);
8707 if (!IS_ERR(inode)) {
8708 ret = btrfs_orphan_add(trans, inode);
8710 btrfs_add_delayed_iput(inode);
8714 /* One for the block groups ref */
8715 spin_lock(&block_group->lock);
8716 if (block_group->iref) {
8717 block_group->iref = 0;
8718 block_group->inode = NULL;
8719 spin_unlock(&block_group->lock);
8722 spin_unlock(&block_group->lock);
8724 /* One for our lookup ref */
8725 btrfs_add_delayed_iput(inode);
8728 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8729 key.offset = block_group->key.objectid;
8732 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8736 btrfs_release_path(path);
8738 ret = btrfs_del_item(trans, tree_root, path);
8741 btrfs_release_path(path);
8744 spin_lock(&root->fs_info->block_group_cache_lock);
8745 rb_erase(&block_group->cache_node,
8746 &root->fs_info->block_group_cache_tree);
8748 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8749 root->fs_info->first_logical_byte = (u64)-1;
8750 spin_unlock(&root->fs_info->block_group_cache_lock);
8752 down_write(&block_group->space_info->groups_sem);
8754 * we must use list_del_init so people can check to see if they
8755 * are still on the list after taking the semaphore
8757 list_del_init(&block_group->list);
8758 if (list_empty(&block_group->space_info->block_groups[index]))
8759 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8760 up_write(&block_group->space_info->groups_sem);
8762 if (block_group->cached == BTRFS_CACHE_STARTED)
8763 wait_block_group_cache_done(block_group);
8765 btrfs_remove_free_space_cache(block_group);
8767 spin_lock(&block_group->space_info->lock);
8768 block_group->space_info->total_bytes -= block_group->key.offset;
8769 block_group->space_info->bytes_readonly -= block_group->key.offset;
8770 block_group->space_info->disk_total -= block_group->key.offset * factor;
8771 spin_unlock(&block_group->space_info->lock);
8773 memcpy(&key, &block_group->key, sizeof(key));
8775 btrfs_clear_space_info_full(root->fs_info);
8777 btrfs_put_block_group(block_group);
8778 btrfs_put_block_group(block_group);
8780 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8786 ret = btrfs_del_item(trans, root, path);
8788 btrfs_free_path(path);
8792 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8794 struct btrfs_space_info *space_info;
8795 struct btrfs_super_block *disk_super;
8801 disk_super = fs_info->super_copy;
8802 if (!btrfs_super_root(disk_super))
8805 features = btrfs_super_incompat_flags(disk_super);
8806 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8809 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8810 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8815 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8816 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8818 flags = BTRFS_BLOCK_GROUP_METADATA;
8819 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8823 flags = BTRFS_BLOCK_GROUP_DATA;
8824 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8830 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8832 return unpin_extent_range(root, start, end);
8835 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8836 u64 num_bytes, u64 *actual_bytes)
8838 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8841 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8843 struct btrfs_fs_info *fs_info = root->fs_info;
8844 struct btrfs_block_group_cache *cache = NULL;
8849 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8853 * try to trim all FS space, our block group may start from non-zero.
8855 if (range->len == total_bytes)
8856 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8858 cache = btrfs_lookup_block_group(fs_info, range->start);
8861 if (cache->key.objectid >= (range->start + range->len)) {
8862 btrfs_put_block_group(cache);
8866 start = max(range->start, cache->key.objectid);
8867 end = min(range->start + range->len,
8868 cache->key.objectid + cache->key.offset);
8870 if (end - start >= range->minlen) {
8871 if (!block_group_cache_done(cache)) {
8872 ret = cache_block_group(cache, 0);
8874 btrfs_put_block_group(cache);
8877 ret = wait_block_group_cache_done(cache);
8879 btrfs_put_block_group(cache);
8883 ret = btrfs_trim_block_group(cache,
8889 trimmed += group_trimmed;
8891 btrfs_put_block_group(cache);
8896 cache = next_block_group(fs_info->tree_root, cache);
8899 range->len = trimmed;