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)
2240 if (btrfs_delayed_ref_is_head(node)) {
2241 struct btrfs_delayed_ref_head *head;
2243 * we've hit the end of the chain and we were supposed
2244 * to insert this extent into the tree. But, it got
2245 * deleted before we ever needed to insert it, so all
2246 * we have to do is clean up the accounting
2249 head = btrfs_delayed_node_to_head(node);
2250 trace_run_delayed_ref_head(node, head, node->action);
2252 if (insert_reserved) {
2253 btrfs_pin_extent(root, node->bytenr,
2254 node->num_bytes, 1);
2255 if (head->is_data) {
2256 ret = btrfs_del_csums(trans, root,
2264 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2265 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2266 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2268 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2269 node->type == BTRFS_SHARED_DATA_REF_KEY)
2270 ret = run_delayed_data_ref(trans, root, node, extent_op,
2277 static noinline struct btrfs_delayed_ref_node *
2278 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2280 struct rb_node *node;
2281 struct btrfs_delayed_ref_node *ref;
2282 int action = BTRFS_ADD_DELAYED_REF;
2285 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2286 * this prevents ref count from going down to zero when
2287 * there still are pending delayed ref.
2289 node = rb_prev(&head->node.rb_node);
2293 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2295 if (ref->bytenr != head->node.bytenr)
2297 if (ref->action == action)
2299 node = rb_prev(node);
2301 if (action == BTRFS_ADD_DELAYED_REF) {
2302 action = BTRFS_DROP_DELAYED_REF;
2309 * Returns 0 on success or if called with an already aborted transaction.
2310 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2312 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct list_head *cluster)
2316 struct btrfs_delayed_ref_root *delayed_refs;
2317 struct btrfs_delayed_ref_node *ref;
2318 struct btrfs_delayed_ref_head *locked_ref = NULL;
2319 struct btrfs_delayed_extent_op *extent_op;
2320 struct btrfs_fs_info *fs_info = root->fs_info;
2323 int must_insert_reserved = 0;
2325 delayed_refs = &trans->transaction->delayed_refs;
2328 /* pick a new head ref from the cluster list */
2329 if (list_empty(cluster))
2332 locked_ref = list_entry(cluster->next,
2333 struct btrfs_delayed_ref_head, cluster);
2335 /* grab the lock that says we are going to process
2336 * all the refs for this head */
2337 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2340 * we may have dropped the spin lock to get the head
2341 * mutex lock, and that might have given someone else
2342 * time to free the head. If that's true, it has been
2343 * removed from our list and we can move on.
2345 if (ret == -EAGAIN) {
2353 * We need to try and merge add/drops of the same ref since we
2354 * can run into issues with relocate dropping the implicit ref
2355 * and then it being added back again before the drop can
2356 * finish. If we merged anything we need to re-loop so we can
2359 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2363 * locked_ref is the head node, so we have to go one
2364 * node back for any delayed ref updates
2366 ref = select_delayed_ref(locked_ref);
2368 if (ref && ref->seq &&
2369 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2371 * there are still refs with lower seq numbers in the
2372 * process of being added. Don't run this ref yet.
2374 list_del_init(&locked_ref->cluster);
2375 btrfs_delayed_ref_unlock(locked_ref);
2377 delayed_refs->num_heads_ready++;
2378 spin_unlock(&delayed_refs->lock);
2380 spin_lock(&delayed_refs->lock);
2385 * record the must insert reserved flag before we
2386 * drop the spin lock.
2388 must_insert_reserved = locked_ref->must_insert_reserved;
2389 locked_ref->must_insert_reserved = 0;
2391 extent_op = locked_ref->extent_op;
2392 locked_ref->extent_op = NULL;
2395 /* All delayed refs have been processed, Go ahead
2396 * and send the head node to run_one_delayed_ref,
2397 * so that any accounting fixes can happen
2399 ref = &locked_ref->node;
2401 if (extent_op && must_insert_reserved) {
2402 btrfs_free_delayed_extent_op(extent_op);
2407 spin_unlock(&delayed_refs->lock);
2409 ret = run_delayed_extent_op(trans, root,
2411 btrfs_free_delayed_extent_op(extent_op);
2414 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2415 spin_lock(&delayed_refs->lock);
2416 btrfs_delayed_ref_unlock(locked_ref);
2425 rb_erase(&ref->rb_node, &delayed_refs->root);
2426 delayed_refs->num_entries--;
2427 if (!btrfs_delayed_ref_is_head(ref)) {
2429 * when we play the delayed ref, also correct the
2432 switch (ref->action) {
2433 case BTRFS_ADD_DELAYED_REF:
2434 case BTRFS_ADD_DELAYED_EXTENT:
2435 locked_ref->node.ref_mod -= ref->ref_mod;
2437 case BTRFS_DROP_DELAYED_REF:
2438 locked_ref->node.ref_mod += ref->ref_mod;
2444 list_del_init(&locked_ref->cluster);
2446 spin_unlock(&delayed_refs->lock);
2448 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2449 must_insert_reserved);
2451 btrfs_free_delayed_extent_op(extent_op);
2453 btrfs_delayed_ref_unlock(locked_ref);
2454 btrfs_put_delayed_ref(ref);
2455 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2456 spin_lock(&delayed_refs->lock);
2461 * If this node is a head, that means all the refs in this head
2462 * have been dealt with, and we will pick the next head to deal
2463 * with, so we must unlock the head and drop it from the cluster
2464 * list before we release it.
2466 if (btrfs_delayed_ref_is_head(ref)) {
2467 btrfs_delayed_ref_unlock(locked_ref);
2470 btrfs_put_delayed_ref(ref);
2474 spin_lock(&delayed_refs->lock);
2479 #ifdef SCRAMBLE_DELAYED_REFS
2481 * Normally delayed refs get processed in ascending bytenr order. This
2482 * correlates in most cases to the order added. To expose dependencies on this
2483 * order, we start to process the tree in the middle instead of the beginning
2485 static u64 find_middle(struct rb_root *root)
2487 struct rb_node *n = root->rb_node;
2488 struct btrfs_delayed_ref_node *entry;
2491 u64 first = 0, last = 0;
2495 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2496 first = entry->bytenr;
2500 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501 last = entry->bytenr;
2506 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2507 WARN_ON(!entry->in_tree);
2509 middle = entry->bytenr;
2522 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2523 struct btrfs_fs_info *fs_info)
2525 struct qgroup_update *qgroup_update;
2528 if (list_empty(&trans->qgroup_ref_list) !=
2529 !trans->delayed_ref_elem.seq) {
2530 /* list without seq or seq without list */
2532 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2533 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2534 (u32)(trans->delayed_ref_elem.seq >> 32),
2535 (u32)trans->delayed_ref_elem.seq);
2539 if (!trans->delayed_ref_elem.seq)
2542 while (!list_empty(&trans->qgroup_ref_list)) {
2543 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2544 struct qgroup_update, list);
2545 list_del(&qgroup_update->list);
2547 ret = btrfs_qgroup_account_ref(
2548 trans, fs_info, qgroup_update->node,
2549 qgroup_update->extent_op);
2550 kfree(qgroup_update);
2553 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2558 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2561 int val = atomic_read(&delayed_refs->ref_seq);
2563 if (val < seq || val >= seq + count)
2568 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2572 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2573 sizeof(struct btrfs_extent_inline_ref));
2574 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2575 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2578 * We don't ever fill up leaves all the way so multiply by 2 just to be
2579 * closer to what we're really going to want to ouse.
2581 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2584 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2585 struct btrfs_root *root)
2587 struct btrfs_block_rsv *global_rsv;
2588 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2592 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2593 num_heads = heads_to_leaves(root, num_heads);
2595 num_bytes += (num_heads - 1) * root->leafsize;
2597 global_rsv = &root->fs_info->global_block_rsv;
2600 * If we can't allocate any more chunks lets make sure we have _lots_ of
2601 * wiggle room since running delayed refs can create more delayed refs.
2603 if (global_rsv->space_info->full)
2606 spin_lock(&global_rsv->lock);
2607 if (global_rsv->reserved <= num_bytes)
2609 spin_unlock(&global_rsv->lock);
2614 * this starts processing the delayed reference count updates and
2615 * extent insertions we have queued up so far. count can be
2616 * 0, which means to process everything in the tree at the start
2617 * of the run (but not newly added entries), or it can be some target
2618 * number you'd like to process.
2620 * Returns 0 on success or if called with an aborted transaction
2621 * Returns <0 on error and aborts the transaction
2623 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2624 struct btrfs_root *root, unsigned long count)
2626 struct rb_node *node;
2627 struct btrfs_delayed_ref_root *delayed_refs;
2628 struct btrfs_delayed_ref_node *ref;
2629 struct list_head cluster;
2632 int run_all = count == (unsigned long)-1;
2636 /* We'll clean this up in btrfs_cleanup_transaction */
2640 if (root == root->fs_info->extent_root)
2641 root = root->fs_info->tree_root;
2643 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2645 delayed_refs = &trans->transaction->delayed_refs;
2646 INIT_LIST_HEAD(&cluster);
2648 count = delayed_refs->num_entries * 2;
2652 if (!run_all && !run_most) {
2654 int seq = atomic_read(&delayed_refs->ref_seq);
2657 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2659 DEFINE_WAIT(__wait);
2660 if (delayed_refs->flushing ||
2661 !btrfs_should_throttle_delayed_refs(trans, root))
2664 prepare_to_wait(&delayed_refs->wait, &__wait,
2665 TASK_UNINTERRUPTIBLE);
2667 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2670 finish_wait(&delayed_refs->wait, &__wait);
2672 if (!refs_newer(delayed_refs, seq, 256))
2677 finish_wait(&delayed_refs->wait, &__wait);
2683 atomic_inc(&delayed_refs->procs_running_refs);
2688 spin_lock(&delayed_refs->lock);
2690 #ifdef SCRAMBLE_DELAYED_REFS
2691 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2695 if (!(run_all || run_most) &&
2696 !btrfs_should_throttle_delayed_refs(trans, root))
2700 * go find something we can process in the rbtree. We start at
2701 * the beginning of the tree, and then build a cluster
2702 * of refs to process starting at the first one we are able to
2705 delayed_start = delayed_refs->run_delayed_start;
2706 ret = btrfs_find_ref_cluster(trans, &cluster,
2707 delayed_refs->run_delayed_start);
2711 ret = run_clustered_refs(trans, root, &cluster);
2713 btrfs_release_ref_cluster(&cluster);
2714 spin_unlock(&delayed_refs->lock);
2715 btrfs_abort_transaction(trans, root, ret);
2716 atomic_dec(&delayed_refs->procs_running_refs);
2717 wake_up(&delayed_refs->wait);
2721 atomic_add(ret, &delayed_refs->ref_seq);
2723 count -= min_t(unsigned long, ret, count);
2728 if (delayed_start >= delayed_refs->run_delayed_start) {
2731 * btrfs_find_ref_cluster looped. let's do one
2732 * more cycle. if we don't run any delayed ref
2733 * during that cycle (because we can't because
2734 * all of them are blocked), bail out.
2739 * no runnable refs left, stop trying
2746 /* refs were run, let's reset staleness detection */
2752 if (!list_empty(&trans->new_bgs)) {
2753 spin_unlock(&delayed_refs->lock);
2754 btrfs_create_pending_block_groups(trans, root);
2755 spin_lock(&delayed_refs->lock);
2758 node = rb_first(&delayed_refs->root);
2761 count = (unsigned long)-1;
2764 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2766 if (btrfs_delayed_ref_is_head(ref)) {
2767 struct btrfs_delayed_ref_head *head;
2769 head = btrfs_delayed_node_to_head(ref);
2770 atomic_inc(&ref->refs);
2772 spin_unlock(&delayed_refs->lock);
2774 * Mutex was contended, block until it's
2775 * released and try again
2777 mutex_lock(&head->mutex);
2778 mutex_unlock(&head->mutex);
2780 btrfs_put_delayed_ref(ref);
2784 node = rb_next(node);
2786 spin_unlock(&delayed_refs->lock);
2787 schedule_timeout(1);
2791 atomic_dec(&delayed_refs->procs_running_refs);
2793 if (waitqueue_active(&delayed_refs->wait))
2794 wake_up(&delayed_refs->wait);
2796 spin_unlock(&delayed_refs->lock);
2797 assert_qgroups_uptodate(trans);
2801 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2802 struct btrfs_root *root,
2803 u64 bytenr, u64 num_bytes, u64 flags,
2804 int level, int is_data)
2806 struct btrfs_delayed_extent_op *extent_op;
2809 extent_op = btrfs_alloc_delayed_extent_op();
2813 extent_op->flags_to_set = flags;
2814 extent_op->update_flags = 1;
2815 extent_op->update_key = 0;
2816 extent_op->is_data = is_data ? 1 : 0;
2817 extent_op->level = level;
2819 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2820 num_bytes, extent_op);
2822 btrfs_free_delayed_extent_op(extent_op);
2826 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2827 struct btrfs_root *root,
2828 struct btrfs_path *path,
2829 u64 objectid, u64 offset, u64 bytenr)
2831 struct btrfs_delayed_ref_head *head;
2832 struct btrfs_delayed_ref_node *ref;
2833 struct btrfs_delayed_data_ref *data_ref;
2834 struct btrfs_delayed_ref_root *delayed_refs;
2835 struct rb_node *node;
2839 delayed_refs = &trans->transaction->delayed_refs;
2840 spin_lock(&delayed_refs->lock);
2841 head = btrfs_find_delayed_ref_head(trans, bytenr);
2845 if (!mutex_trylock(&head->mutex)) {
2846 atomic_inc(&head->node.refs);
2847 spin_unlock(&delayed_refs->lock);
2849 btrfs_release_path(path);
2852 * Mutex was contended, block until it's released and let
2855 mutex_lock(&head->mutex);
2856 mutex_unlock(&head->mutex);
2857 btrfs_put_delayed_ref(&head->node);
2861 node = rb_prev(&head->node.rb_node);
2865 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2867 if (ref->bytenr != bytenr)
2871 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2874 data_ref = btrfs_delayed_node_to_data_ref(ref);
2876 node = rb_prev(node);
2880 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2881 if (ref->bytenr == bytenr && ref->seq == seq)
2885 if (data_ref->root != root->root_key.objectid ||
2886 data_ref->objectid != objectid || data_ref->offset != offset)
2891 mutex_unlock(&head->mutex);
2893 spin_unlock(&delayed_refs->lock);
2897 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2898 struct btrfs_root *root,
2899 struct btrfs_path *path,
2900 u64 objectid, u64 offset, u64 bytenr)
2902 struct btrfs_root *extent_root = root->fs_info->extent_root;
2903 struct extent_buffer *leaf;
2904 struct btrfs_extent_data_ref *ref;
2905 struct btrfs_extent_inline_ref *iref;
2906 struct btrfs_extent_item *ei;
2907 struct btrfs_key key;
2911 key.objectid = bytenr;
2912 key.offset = (u64)-1;
2913 key.type = BTRFS_EXTENT_ITEM_KEY;
2915 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2918 BUG_ON(ret == 0); /* Corruption */
2921 if (path->slots[0] == 0)
2925 leaf = path->nodes[0];
2926 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2928 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2932 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2933 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2934 if (item_size < sizeof(*ei)) {
2935 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2939 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2941 if (item_size != sizeof(*ei) +
2942 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2945 if (btrfs_extent_generation(leaf, ei) <=
2946 btrfs_root_last_snapshot(&root->root_item))
2949 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2950 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2951 BTRFS_EXTENT_DATA_REF_KEY)
2954 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2955 if (btrfs_extent_refs(leaf, ei) !=
2956 btrfs_extent_data_ref_count(leaf, ref) ||
2957 btrfs_extent_data_ref_root(leaf, ref) !=
2958 root->root_key.objectid ||
2959 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2960 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2968 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2969 struct btrfs_root *root,
2970 u64 objectid, u64 offset, u64 bytenr)
2972 struct btrfs_path *path;
2976 path = btrfs_alloc_path();
2981 ret = check_committed_ref(trans, root, path, objectid,
2983 if (ret && ret != -ENOENT)
2986 ret2 = check_delayed_ref(trans, root, path, objectid,
2988 } while (ret2 == -EAGAIN);
2990 if (ret2 && ret2 != -ENOENT) {
2995 if (ret != -ENOENT || ret2 != -ENOENT)
2998 btrfs_free_path(path);
2999 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3004 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3005 struct btrfs_root *root,
3006 struct extent_buffer *buf,
3007 int full_backref, int inc, int for_cow)
3014 struct btrfs_key key;
3015 struct btrfs_file_extent_item *fi;
3019 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3020 u64, u64, u64, u64, u64, u64, int);
3022 ref_root = btrfs_header_owner(buf);
3023 nritems = btrfs_header_nritems(buf);
3024 level = btrfs_header_level(buf);
3026 if (!root->ref_cows && level == 0)
3030 process_func = btrfs_inc_extent_ref;
3032 process_func = btrfs_free_extent;
3035 parent = buf->start;
3039 for (i = 0; i < nritems; i++) {
3041 btrfs_item_key_to_cpu(buf, &key, i);
3042 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3044 fi = btrfs_item_ptr(buf, i,
3045 struct btrfs_file_extent_item);
3046 if (btrfs_file_extent_type(buf, fi) ==
3047 BTRFS_FILE_EXTENT_INLINE)
3049 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3053 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3054 key.offset -= btrfs_file_extent_offset(buf, fi);
3055 ret = process_func(trans, root, bytenr, num_bytes,
3056 parent, ref_root, key.objectid,
3057 key.offset, for_cow);
3061 bytenr = btrfs_node_blockptr(buf, i);
3062 num_bytes = btrfs_level_size(root, level - 1);
3063 ret = process_func(trans, root, bytenr, num_bytes,
3064 parent, ref_root, level - 1, 0,
3075 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3076 struct extent_buffer *buf, int full_backref, int for_cow)
3078 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3081 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3082 struct extent_buffer *buf, int full_backref, int for_cow)
3084 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3087 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3088 struct btrfs_root *root,
3089 struct btrfs_path *path,
3090 struct btrfs_block_group_cache *cache)
3093 struct btrfs_root *extent_root = root->fs_info->extent_root;
3095 struct extent_buffer *leaf;
3097 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3100 BUG_ON(ret); /* Corruption */
3102 leaf = path->nodes[0];
3103 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3104 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3105 btrfs_mark_buffer_dirty(leaf);
3106 btrfs_release_path(path);
3109 btrfs_abort_transaction(trans, root, ret);
3116 static struct btrfs_block_group_cache *
3117 next_block_group(struct btrfs_root *root,
3118 struct btrfs_block_group_cache *cache)
3120 struct rb_node *node;
3121 spin_lock(&root->fs_info->block_group_cache_lock);
3122 node = rb_next(&cache->cache_node);
3123 btrfs_put_block_group(cache);
3125 cache = rb_entry(node, struct btrfs_block_group_cache,
3127 btrfs_get_block_group(cache);
3130 spin_unlock(&root->fs_info->block_group_cache_lock);
3134 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3135 struct btrfs_trans_handle *trans,
3136 struct btrfs_path *path)
3138 struct btrfs_root *root = block_group->fs_info->tree_root;
3139 struct inode *inode = NULL;
3141 int dcs = BTRFS_DC_ERROR;
3147 * If this block group is smaller than 100 megs don't bother caching the
3150 if (block_group->key.offset < (100 * 1024 * 1024)) {
3151 spin_lock(&block_group->lock);
3152 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3153 spin_unlock(&block_group->lock);
3158 inode = lookup_free_space_inode(root, block_group, path);
3159 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3160 ret = PTR_ERR(inode);
3161 btrfs_release_path(path);
3165 if (IS_ERR(inode)) {
3169 if (block_group->ro)
3172 ret = create_free_space_inode(root, trans, block_group, path);
3178 /* We've already setup this transaction, go ahead and exit */
3179 if (block_group->cache_generation == trans->transid &&
3180 i_size_read(inode)) {
3181 dcs = BTRFS_DC_SETUP;
3186 * We want to set the generation to 0, that way if anything goes wrong
3187 * from here on out we know not to trust this cache when we load up next
3190 BTRFS_I(inode)->generation = 0;
3191 ret = btrfs_update_inode(trans, root, inode);
3194 if (i_size_read(inode) > 0) {
3195 ret = btrfs_check_trunc_cache_free_space(root,
3196 &root->fs_info->global_block_rsv);
3200 ret = btrfs_truncate_free_space_cache(root, trans, path,
3206 spin_lock(&block_group->lock);
3207 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3208 !btrfs_test_opt(root, SPACE_CACHE)) {
3210 * don't bother trying to write stuff out _if_
3211 * a) we're not cached,
3212 * b) we're with nospace_cache mount option.
3214 dcs = BTRFS_DC_WRITTEN;
3215 spin_unlock(&block_group->lock);
3218 spin_unlock(&block_group->lock);
3221 * Try to preallocate enough space based on how big the block group is.
3222 * Keep in mind this has to include any pinned space which could end up
3223 * taking up quite a bit since it's not folded into the other space
3226 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3231 num_pages *= PAGE_CACHE_SIZE;
3233 ret = btrfs_check_data_free_space(inode, num_pages);
3237 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3238 num_pages, num_pages,
3241 dcs = BTRFS_DC_SETUP;
3242 btrfs_free_reserved_data_space(inode, num_pages);
3247 btrfs_release_path(path);
3249 spin_lock(&block_group->lock);
3250 if (!ret && dcs == BTRFS_DC_SETUP)
3251 block_group->cache_generation = trans->transid;
3252 block_group->disk_cache_state = dcs;
3253 spin_unlock(&block_group->lock);
3258 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root)
3261 struct btrfs_block_group_cache *cache;
3263 struct btrfs_path *path;
3266 path = btrfs_alloc_path();
3272 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3274 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3276 cache = next_block_group(root, cache);
3284 err = cache_save_setup(cache, trans, path);
3285 last = cache->key.objectid + cache->key.offset;
3286 btrfs_put_block_group(cache);
3291 err = btrfs_run_delayed_refs(trans, root,
3293 if (err) /* File system offline */
3297 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3299 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3300 btrfs_put_block_group(cache);
3306 cache = next_block_group(root, cache);
3315 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3316 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3318 last = cache->key.objectid + cache->key.offset;
3320 err = write_one_cache_group(trans, root, path, cache);
3321 if (err) /* File system offline */
3324 btrfs_put_block_group(cache);
3329 * I don't think this is needed since we're just marking our
3330 * preallocated extent as written, but just in case it can't
3334 err = btrfs_run_delayed_refs(trans, root,
3336 if (err) /* File system offline */
3340 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3343 * Really this shouldn't happen, but it could if we
3344 * couldn't write the entire preallocated extent and
3345 * splitting the extent resulted in a new block.
3348 btrfs_put_block_group(cache);
3351 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3353 cache = next_block_group(root, cache);
3362 err = btrfs_write_out_cache(root, trans, cache, path);
3365 * If we didn't have an error then the cache state is still
3366 * NEED_WRITE, so we can set it to WRITTEN.
3368 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3369 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3370 last = cache->key.objectid + cache->key.offset;
3371 btrfs_put_block_group(cache);
3375 btrfs_free_path(path);
3379 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3381 struct btrfs_block_group_cache *block_group;
3384 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3385 if (!block_group || block_group->ro)
3388 btrfs_put_block_group(block_group);
3392 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3393 u64 total_bytes, u64 bytes_used,
3394 struct btrfs_space_info **space_info)
3396 struct btrfs_space_info *found;
3401 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3402 BTRFS_BLOCK_GROUP_RAID10))
3407 found = __find_space_info(info, flags);
3409 spin_lock(&found->lock);
3410 found->total_bytes += total_bytes;
3411 found->disk_total += total_bytes * factor;
3412 found->bytes_used += bytes_used;
3413 found->disk_used += bytes_used * factor;
3415 spin_unlock(&found->lock);
3416 *space_info = found;
3419 found = kzalloc(sizeof(*found), GFP_NOFS);
3423 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3429 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3430 INIT_LIST_HEAD(&found->block_groups[i]);
3431 init_rwsem(&found->groups_sem);
3432 spin_lock_init(&found->lock);
3433 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3434 found->total_bytes = total_bytes;
3435 found->disk_total = total_bytes * factor;
3436 found->bytes_used = bytes_used;
3437 found->disk_used = bytes_used * factor;
3438 found->bytes_pinned = 0;
3439 found->bytes_reserved = 0;
3440 found->bytes_readonly = 0;
3441 found->bytes_may_use = 0;
3443 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3444 found->chunk_alloc = 0;
3446 init_waitqueue_head(&found->wait);
3447 *space_info = found;
3448 list_add_rcu(&found->list, &info->space_info);
3449 if (flags & BTRFS_BLOCK_GROUP_DATA)
3450 info->data_sinfo = found;
3454 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3456 u64 extra_flags = chunk_to_extended(flags) &
3457 BTRFS_EXTENDED_PROFILE_MASK;
3459 write_seqlock(&fs_info->profiles_lock);
3460 if (flags & BTRFS_BLOCK_GROUP_DATA)
3461 fs_info->avail_data_alloc_bits |= extra_flags;
3462 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3463 fs_info->avail_metadata_alloc_bits |= extra_flags;
3464 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3465 fs_info->avail_system_alloc_bits |= extra_flags;
3466 write_sequnlock(&fs_info->profiles_lock);
3470 * returns target flags in extended format or 0 if restripe for this
3471 * chunk_type is not in progress
3473 * should be called with either volume_mutex or balance_lock held
3475 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3477 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3483 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3484 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3485 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3486 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3487 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3488 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3489 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3490 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3491 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3498 * @flags: available profiles in extended format (see ctree.h)
3500 * Returns reduced profile in chunk format. If profile changing is in
3501 * progress (either running or paused) picks the target profile (if it's
3502 * already available), otherwise falls back to plain reducing.
3504 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3507 * we add in the count of missing devices because we want
3508 * to make sure that any RAID levels on a degraded FS
3509 * continue to be honored.
3511 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3512 root->fs_info->fs_devices->missing_devices;
3517 * see if restripe for this chunk_type is in progress, if so
3518 * try to reduce to the target profile
3520 spin_lock(&root->fs_info->balance_lock);
3521 target = get_restripe_target(root->fs_info, flags);
3523 /* pick target profile only if it's already available */
3524 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3525 spin_unlock(&root->fs_info->balance_lock);
3526 return extended_to_chunk(target);
3529 spin_unlock(&root->fs_info->balance_lock);
3531 /* First, mask out the RAID levels which aren't possible */
3532 if (num_devices == 1)
3533 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3534 BTRFS_BLOCK_GROUP_RAID5);
3535 if (num_devices < 3)
3536 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3537 if (num_devices < 4)
3538 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3540 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3541 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3542 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3545 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3546 tmp = BTRFS_BLOCK_GROUP_RAID6;
3547 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3548 tmp = BTRFS_BLOCK_GROUP_RAID5;
3549 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3550 tmp = BTRFS_BLOCK_GROUP_RAID10;
3551 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3552 tmp = BTRFS_BLOCK_GROUP_RAID1;
3553 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3554 tmp = BTRFS_BLOCK_GROUP_RAID0;
3556 return extended_to_chunk(flags | tmp);
3559 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3564 seq = read_seqbegin(&root->fs_info->profiles_lock);
3566 if (flags & BTRFS_BLOCK_GROUP_DATA)
3567 flags |= root->fs_info->avail_data_alloc_bits;
3568 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3569 flags |= root->fs_info->avail_system_alloc_bits;
3570 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3571 flags |= root->fs_info->avail_metadata_alloc_bits;
3572 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3574 return btrfs_reduce_alloc_profile(root, flags);
3577 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3583 flags = BTRFS_BLOCK_GROUP_DATA;
3584 else if (root == root->fs_info->chunk_root)
3585 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3587 flags = BTRFS_BLOCK_GROUP_METADATA;
3589 ret = get_alloc_profile(root, flags);
3594 * This will check the space that the inode allocates from to make sure we have
3595 * enough space for bytes.
3597 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3599 struct btrfs_space_info *data_sinfo;
3600 struct btrfs_root *root = BTRFS_I(inode)->root;
3601 struct btrfs_fs_info *fs_info = root->fs_info;
3603 int ret = 0, committed = 0, alloc_chunk = 1;
3605 /* make sure bytes are sectorsize aligned */
3606 bytes = ALIGN(bytes, root->sectorsize);
3608 if (root == root->fs_info->tree_root ||
3609 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3614 data_sinfo = fs_info->data_sinfo;
3619 /* make sure we have enough space to handle the data first */
3620 spin_lock(&data_sinfo->lock);
3621 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3622 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3623 data_sinfo->bytes_may_use;
3625 if (used + bytes > data_sinfo->total_bytes) {
3626 struct btrfs_trans_handle *trans;
3629 * if we don't have enough free bytes in this space then we need
3630 * to alloc a new chunk.
3632 if (!data_sinfo->full && alloc_chunk) {
3635 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3636 spin_unlock(&data_sinfo->lock);
3638 alloc_target = btrfs_get_alloc_profile(root, 1);
3639 trans = btrfs_join_transaction(root);
3641 return PTR_ERR(trans);
3643 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3645 CHUNK_ALLOC_NO_FORCE);
3646 btrfs_end_transaction(trans, root);
3655 data_sinfo = fs_info->data_sinfo;
3661 * If we don't have enough pinned space to deal with this
3662 * allocation don't bother committing the transaction.
3664 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3667 spin_unlock(&data_sinfo->lock);
3669 /* commit the current transaction and try again */
3672 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3675 trans = btrfs_join_transaction(root);
3677 return PTR_ERR(trans);
3678 ret = btrfs_commit_transaction(trans, root);
3686 data_sinfo->bytes_may_use += bytes;
3687 trace_btrfs_space_reservation(root->fs_info, "space_info",
3688 data_sinfo->flags, bytes, 1);
3689 spin_unlock(&data_sinfo->lock);
3695 * Called if we need to clear a data reservation for this inode.
3697 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3699 struct btrfs_root *root = BTRFS_I(inode)->root;
3700 struct btrfs_space_info *data_sinfo;
3702 /* make sure bytes are sectorsize aligned */
3703 bytes = ALIGN(bytes, root->sectorsize);
3705 data_sinfo = root->fs_info->data_sinfo;
3706 spin_lock(&data_sinfo->lock);
3707 WARN_ON(data_sinfo->bytes_may_use < bytes);
3708 data_sinfo->bytes_may_use -= bytes;
3709 trace_btrfs_space_reservation(root->fs_info, "space_info",
3710 data_sinfo->flags, bytes, 0);
3711 spin_unlock(&data_sinfo->lock);
3714 static void force_metadata_allocation(struct btrfs_fs_info *info)
3716 struct list_head *head = &info->space_info;
3717 struct btrfs_space_info *found;
3720 list_for_each_entry_rcu(found, head, list) {
3721 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3722 found->force_alloc = CHUNK_ALLOC_FORCE;
3727 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3729 return (global->size << 1);
3732 static int should_alloc_chunk(struct btrfs_root *root,
3733 struct btrfs_space_info *sinfo, int force)
3735 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3736 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3737 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3740 if (force == CHUNK_ALLOC_FORCE)
3744 * We need to take into account the global rsv because for all intents
3745 * and purposes it's used space. Don't worry about locking the
3746 * global_rsv, it doesn't change except when the transaction commits.
3748 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3749 num_allocated += calc_global_rsv_need_space(global_rsv);
3752 * in limited mode, we want to have some free space up to
3753 * about 1% of the FS size.
3755 if (force == CHUNK_ALLOC_LIMITED) {
3756 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3757 thresh = max_t(u64, 64 * 1024 * 1024,
3758 div_factor_fine(thresh, 1));
3760 if (num_bytes - num_allocated < thresh)
3764 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3769 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3773 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3774 BTRFS_BLOCK_GROUP_RAID0 |
3775 BTRFS_BLOCK_GROUP_RAID5 |
3776 BTRFS_BLOCK_GROUP_RAID6))
3777 num_dev = root->fs_info->fs_devices->rw_devices;
3778 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3781 num_dev = 1; /* DUP or single */
3783 /* metadata for updaing devices and chunk tree */
3784 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3787 static void check_system_chunk(struct btrfs_trans_handle *trans,
3788 struct btrfs_root *root, u64 type)
3790 struct btrfs_space_info *info;
3794 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3795 spin_lock(&info->lock);
3796 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3797 info->bytes_reserved - info->bytes_readonly;
3798 spin_unlock(&info->lock);
3800 thresh = get_system_chunk_thresh(root, type);
3801 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3802 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3803 left, thresh, type);
3804 dump_space_info(info, 0, 0);
3807 if (left < thresh) {
3810 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3811 btrfs_alloc_chunk(trans, root, flags);
3815 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3816 struct btrfs_root *extent_root, u64 flags, int force)
3818 struct btrfs_space_info *space_info;
3819 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3820 int wait_for_alloc = 0;
3823 /* Don't re-enter if we're already allocating a chunk */
3824 if (trans->allocating_chunk)
3827 space_info = __find_space_info(extent_root->fs_info, flags);
3829 ret = update_space_info(extent_root->fs_info, flags,
3831 BUG_ON(ret); /* -ENOMEM */
3833 BUG_ON(!space_info); /* Logic error */
3836 spin_lock(&space_info->lock);
3837 if (force < space_info->force_alloc)
3838 force = space_info->force_alloc;
3839 if (space_info->full) {
3840 if (should_alloc_chunk(extent_root, space_info, force))
3844 spin_unlock(&space_info->lock);
3848 if (!should_alloc_chunk(extent_root, space_info, force)) {
3849 spin_unlock(&space_info->lock);
3851 } else if (space_info->chunk_alloc) {
3854 space_info->chunk_alloc = 1;
3857 spin_unlock(&space_info->lock);
3859 mutex_lock(&fs_info->chunk_mutex);
3862 * The chunk_mutex is held throughout the entirety of a chunk
3863 * allocation, so once we've acquired the chunk_mutex we know that the
3864 * other guy is done and we need to recheck and see if we should
3867 if (wait_for_alloc) {
3868 mutex_unlock(&fs_info->chunk_mutex);
3873 trans->allocating_chunk = true;
3876 * If we have mixed data/metadata chunks we want to make sure we keep
3877 * allocating mixed chunks instead of individual chunks.
3879 if (btrfs_mixed_space_info(space_info))
3880 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3883 * if we're doing a data chunk, go ahead and make sure that
3884 * we keep a reasonable number of metadata chunks allocated in the
3887 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3888 fs_info->data_chunk_allocations++;
3889 if (!(fs_info->data_chunk_allocations %
3890 fs_info->metadata_ratio))
3891 force_metadata_allocation(fs_info);
3895 * Check if we have enough space in SYSTEM chunk because we may need
3896 * to update devices.
3898 check_system_chunk(trans, extent_root, flags);
3900 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3901 trans->allocating_chunk = false;
3903 spin_lock(&space_info->lock);
3904 if (ret < 0 && ret != -ENOSPC)
3907 space_info->full = 1;
3911 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3913 space_info->chunk_alloc = 0;
3914 spin_unlock(&space_info->lock);
3915 mutex_unlock(&fs_info->chunk_mutex);
3919 static int can_overcommit(struct btrfs_root *root,
3920 struct btrfs_space_info *space_info, u64 bytes,
3921 enum btrfs_reserve_flush_enum flush)
3923 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3924 u64 profile = btrfs_get_alloc_profile(root, 0);
3930 used = space_info->bytes_used + space_info->bytes_reserved +
3931 space_info->bytes_pinned + space_info->bytes_readonly;
3934 * We only want to allow over committing if we have lots of actual space
3935 * free, but if we don't have enough space to handle the global reserve
3936 * space then we could end up having a real enospc problem when trying
3937 * to allocate a chunk or some other such important allocation.
3939 spin_lock(&global_rsv->lock);
3940 space_size = calc_global_rsv_need_space(global_rsv);
3941 spin_unlock(&global_rsv->lock);
3942 if (used + space_size >= space_info->total_bytes)
3945 used += space_info->bytes_may_use;
3947 spin_lock(&root->fs_info->free_chunk_lock);
3948 avail = root->fs_info->free_chunk_space;
3949 spin_unlock(&root->fs_info->free_chunk_lock);
3952 * If we have dup, raid1 or raid10 then only half of the free
3953 * space is actually useable. For raid56, the space info used
3954 * doesn't include the parity drive, so we don't have to
3957 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3958 BTRFS_BLOCK_GROUP_RAID1 |
3959 BTRFS_BLOCK_GROUP_RAID10))
3962 to_add = space_info->total_bytes;
3965 * If we aren't flushing all things, let us overcommit up to
3966 * 1/2th of the space. If we can flush, don't let us overcommit
3967 * too much, let it overcommit up to 1/8 of the space.
3969 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3975 * Limit the overcommit to the amount of free space we could possibly
3976 * allocate for chunks.
3978 to_add = min(avail, to_add);
3980 if (used + bytes < space_info->total_bytes + to_add)
3985 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3986 unsigned long nr_pages)
3988 struct super_block *sb = root->fs_info->sb;
3990 if (down_read_trylock(&sb->s_umount)) {
3991 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3992 up_read(&sb->s_umount);
3995 * We needn't worry the filesystem going from r/w to r/o though
3996 * we don't acquire ->s_umount mutex, because the filesystem
3997 * should guarantee the delalloc inodes list be empty after
3998 * the filesystem is readonly(all dirty pages are written to
4001 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
4002 if (!current->journal_info)
4003 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4008 * shrink metadata reservation for delalloc
4010 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4013 struct btrfs_block_rsv *block_rsv;
4014 struct btrfs_space_info *space_info;
4015 struct btrfs_trans_handle *trans;
4019 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
4021 enum btrfs_reserve_flush_enum flush;
4023 trans = (struct btrfs_trans_handle *)current->journal_info;
4024 block_rsv = &root->fs_info->delalloc_block_rsv;
4025 space_info = block_rsv->space_info;
4028 delalloc_bytes = percpu_counter_sum_positive(
4029 &root->fs_info->delalloc_bytes);
4030 if (delalloc_bytes == 0) {
4033 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4037 while (delalloc_bytes && loops < 3) {
4038 max_reclaim = min(delalloc_bytes, to_reclaim);
4039 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4040 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4042 * We need to wait for the async pages to actually start before
4045 wait_event(root->fs_info->async_submit_wait,
4046 !atomic_read(&root->fs_info->async_delalloc_pages));
4049 flush = BTRFS_RESERVE_FLUSH_ALL;
4051 flush = BTRFS_RESERVE_NO_FLUSH;
4052 spin_lock(&space_info->lock);
4053 if (can_overcommit(root, space_info, orig, flush)) {
4054 spin_unlock(&space_info->lock);
4057 spin_unlock(&space_info->lock);
4060 if (wait_ordered && !trans) {
4061 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4063 time_left = schedule_timeout_killable(1);
4068 delalloc_bytes = percpu_counter_sum_positive(
4069 &root->fs_info->delalloc_bytes);
4074 * maybe_commit_transaction - possibly commit the transaction if its ok to
4075 * @root - the root we're allocating for
4076 * @bytes - the number of bytes we want to reserve
4077 * @force - force the commit
4079 * This will check to make sure that committing the transaction will actually
4080 * get us somewhere and then commit the transaction if it does. Otherwise it
4081 * will return -ENOSPC.
4083 static int may_commit_transaction(struct btrfs_root *root,
4084 struct btrfs_space_info *space_info,
4085 u64 bytes, int force)
4087 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4088 struct btrfs_trans_handle *trans;
4090 trans = (struct btrfs_trans_handle *)current->journal_info;
4097 /* See if there is enough pinned space to make this reservation */
4098 spin_lock(&space_info->lock);
4099 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4101 spin_unlock(&space_info->lock);
4104 spin_unlock(&space_info->lock);
4107 * See if there is some space in the delayed insertion reservation for
4110 if (space_info != delayed_rsv->space_info)
4113 spin_lock(&space_info->lock);
4114 spin_lock(&delayed_rsv->lock);
4115 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4116 bytes - delayed_rsv->size) >= 0) {
4117 spin_unlock(&delayed_rsv->lock);
4118 spin_unlock(&space_info->lock);
4121 spin_unlock(&delayed_rsv->lock);
4122 spin_unlock(&space_info->lock);
4125 trans = btrfs_join_transaction(root);
4129 return btrfs_commit_transaction(trans, root);
4133 FLUSH_DELAYED_ITEMS_NR = 1,
4134 FLUSH_DELAYED_ITEMS = 2,
4136 FLUSH_DELALLOC_WAIT = 4,
4141 static int flush_space(struct btrfs_root *root,
4142 struct btrfs_space_info *space_info, u64 num_bytes,
4143 u64 orig_bytes, int state)
4145 struct btrfs_trans_handle *trans;
4150 case FLUSH_DELAYED_ITEMS_NR:
4151 case FLUSH_DELAYED_ITEMS:
4152 if (state == FLUSH_DELAYED_ITEMS_NR) {
4153 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4155 nr = (int)div64_u64(num_bytes, bytes);
4162 trans = btrfs_join_transaction(root);
4163 if (IS_ERR(trans)) {
4164 ret = PTR_ERR(trans);
4167 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4168 btrfs_end_transaction(trans, root);
4170 case FLUSH_DELALLOC:
4171 case FLUSH_DELALLOC_WAIT:
4172 shrink_delalloc(root, num_bytes, orig_bytes,
4173 state == FLUSH_DELALLOC_WAIT);
4176 trans = btrfs_join_transaction(root);
4177 if (IS_ERR(trans)) {
4178 ret = PTR_ERR(trans);
4181 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4182 btrfs_get_alloc_profile(root, 0),
4183 CHUNK_ALLOC_NO_FORCE);
4184 btrfs_end_transaction(trans, root);
4189 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4199 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4200 * @root - the root we're allocating for
4201 * @block_rsv - the block_rsv we're allocating for
4202 * @orig_bytes - the number of bytes we want
4203 * @flush - whether or not we can flush to make our reservation
4205 * This will reserve orgi_bytes number of bytes from the space info associated
4206 * with the block_rsv. If there is not enough space it will make an attempt to
4207 * flush out space to make room. It will do this by flushing delalloc if
4208 * possible or committing the transaction. If flush is 0 then no attempts to
4209 * regain reservations will be made and this will fail if there is not enough
4212 static int reserve_metadata_bytes(struct btrfs_root *root,
4213 struct btrfs_block_rsv *block_rsv,
4215 enum btrfs_reserve_flush_enum flush)
4217 struct btrfs_space_info *space_info = block_rsv->space_info;
4219 u64 num_bytes = orig_bytes;
4220 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4222 bool flushing = false;
4226 spin_lock(&space_info->lock);
4228 * We only want to wait if somebody other than us is flushing and we
4229 * are actually allowed to flush all things.
4231 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4232 space_info->flush) {
4233 spin_unlock(&space_info->lock);
4235 * If we have a trans handle we can't wait because the flusher
4236 * may have to commit the transaction, which would mean we would
4237 * deadlock since we are waiting for the flusher to finish, but
4238 * hold the current transaction open.
4240 if (current->journal_info)
4242 ret = wait_event_killable(space_info->wait, !space_info->flush);
4243 /* Must have been killed, return */
4247 spin_lock(&space_info->lock);
4251 used = space_info->bytes_used + space_info->bytes_reserved +
4252 space_info->bytes_pinned + space_info->bytes_readonly +
4253 space_info->bytes_may_use;
4256 * The idea here is that we've not already over-reserved the block group
4257 * then we can go ahead and save our reservation first and then start
4258 * flushing if we need to. Otherwise if we've already overcommitted
4259 * lets start flushing stuff first and then come back and try to make
4262 if (used <= space_info->total_bytes) {
4263 if (used + orig_bytes <= space_info->total_bytes) {
4264 space_info->bytes_may_use += orig_bytes;
4265 trace_btrfs_space_reservation(root->fs_info,
4266 "space_info", space_info->flags, orig_bytes, 1);
4270 * Ok set num_bytes to orig_bytes since we aren't
4271 * overocmmitted, this way we only try and reclaim what
4274 num_bytes = orig_bytes;
4278 * Ok we're over committed, set num_bytes to the overcommitted
4279 * amount plus the amount of bytes that we need for this
4282 num_bytes = used - space_info->total_bytes +
4286 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4287 space_info->bytes_may_use += orig_bytes;
4288 trace_btrfs_space_reservation(root->fs_info, "space_info",
4289 space_info->flags, orig_bytes,
4295 * Couldn't make our reservation, save our place so while we're trying
4296 * to reclaim space we can actually use it instead of somebody else
4297 * stealing it from us.
4299 * We make the other tasks wait for the flush only when we can flush
4302 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4304 space_info->flush = 1;
4307 spin_unlock(&space_info->lock);
4309 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4312 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4317 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4318 * would happen. So skip delalloc flush.
4320 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4321 (flush_state == FLUSH_DELALLOC ||
4322 flush_state == FLUSH_DELALLOC_WAIT))
4323 flush_state = ALLOC_CHUNK;
4327 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4328 flush_state < COMMIT_TRANS)
4330 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4331 flush_state <= COMMIT_TRANS)
4335 if (ret == -ENOSPC &&
4336 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4337 struct btrfs_block_rsv *global_rsv =
4338 &root->fs_info->global_block_rsv;
4340 if (block_rsv != global_rsv &&
4341 !block_rsv_use_bytes(global_rsv, orig_bytes))
4345 spin_lock(&space_info->lock);
4346 space_info->flush = 0;
4347 wake_up_all(&space_info->wait);
4348 spin_unlock(&space_info->lock);
4353 static struct btrfs_block_rsv *get_block_rsv(
4354 const struct btrfs_trans_handle *trans,
4355 const struct btrfs_root *root)
4357 struct btrfs_block_rsv *block_rsv = NULL;
4360 block_rsv = trans->block_rsv;
4362 if (root == root->fs_info->csum_root && trans->adding_csums)
4363 block_rsv = trans->block_rsv;
4365 if (root == root->fs_info->uuid_root)
4366 block_rsv = trans->block_rsv;
4369 block_rsv = root->block_rsv;
4372 block_rsv = &root->fs_info->empty_block_rsv;
4377 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4381 spin_lock(&block_rsv->lock);
4382 if (block_rsv->reserved >= num_bytes) {
4383 block_rsv->reserved -= num_bytes;
4384 if (block_rsv->reserved < block_rsv->size)
4385 block_rsv->full = 0;
4388 spin_unlock(&block_rsv->lock);
4392 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4393 u64 num_bytes, int update_size)
4395 spin_lock(&block_rsv->lock);
4396 block_rsv->reserved += num_bytes;
4398 block_rsv->size += num_bytes;
4399 else if (block_rsv->reserved >= block_rsv->size)
4400 block_rsv->full = 1;
4401 spin_unlock(&block_rsv->lock);
4404 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4405 struct btrfs_block_rsv *dest, u64 num_bytes,
4408 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4411 if (global_rsv->space_info != dest->space_info)
4414 spin_lock(&global_rsv->lock);
4415 min_bytes = div_factor(global_rsv->size, min_factor);
4416 if (global_rsv->reserved < min_bytes + num_bytes) {
4417 spin_unlock(&global_rsv->lock);
4420 global_rsv->reserved -= num_bytes;
4421 if (global_rsv->reserved < global_rsv->size)
4422 global_rsv->full = 0;
4423 spin_unlock(&global_rsv->lock);
4425 block_rsv_add_bytes(dest, num_bytes, 1);
4429 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4430 struct btrfs_block_rsv *block_rsv,
4431 struct btrfs_block_rsv *dest, u64 num_bytes)
4433 struct btrfs_space_info *space_info = block_rsv->space_info;
4435 spin_lock(&block_rsv->lock);
4436 if (num_bytes == (u64)-1)
4437 num_bytes = block_rsv->size;
4438 block_rsv->size -= num_bytes;
4439 if (block_rsv->reserved >= block_rsv->size) {
4440 num_bytes = block_rsv->reserved - block_rsv->size;
4441 block_rsv->reserved = block_rsv->size;
4442 block_rsv->full = 1;
4446 spin_unlock(&block_rsv->lock);
4448 if (num_bytes > 0) {
4450 spin_lock(&dest->lock);
4454 bytes_to_add = dest->size - dest->reserved;
4455 bytes_to_add = min(num_bytes, bytes_to_add);
4456 dest->reserved += bytes_to_add;
4457 if (dest->reserved >= dest->size)
4459 num_bytes -= bytes_to_add;
4461 spin_unlock(&dest->lock);
4464 spin_lock(&space_info->lock);
4465 space_info->bytes_may_use -= num_bytes;
4466 trace_btrfs_space_reservation(fs_info, "space_info",
4467 space_info->flags, num_bytes, 0);
4468 space_info->reservation_progress++;
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 sinfo->reservation_progress++;
4670 block_rsv->reserved = block_rsv->size;
4671 block_rsv->full = 1;
4674 spin_unlock(&block_rsv->lock);
4675 spin_unlock(&sinfo->lock);
4678 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4680 struct btrfs_space_info *space_info;
4682 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4683 fs_info->chunk_block_rsv.space_info = space_info;
4685 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4686 fs_info->global_block_rsv.space_info = space_info;
4687 fs_info->delalloc_block_rsv.space_info = space_info;
4688 fs_info->trans_block_rsv.space_info = space_info;
4689 fs_info->empty_block_rsv.space_info = space_info;
4690 fs_info->delayed_block_rsv.space_info = space_info;
4692 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4693 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4694 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4695 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4696 if (fs_info->quota_root)
4697 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4698 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4700 update_global_block_rsv(fs_info);
4703 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4705 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4707 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4708 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4709 WARN_ON(fs_info->trans_block_rsv.size > 0);
4710 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4711 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4712 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4713 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4714 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4717 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4718 struct btrfs_root *root)
4720 if (!trans->block_rsv)
4723 if (!trans->bytes_reserved)
4726 trace_btrfs_space_reservation(root->fs_info, "transaction",
4727 trans->transid, trans->bytes_reserved, 0);
4728 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4729 trans->bytes_reserved = 0;
4732 /* Can only return 0 or -ENOSPC */
4733 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4734 struct inode *inode)
4736 struct btrfs_root *root = BTRFS_I(inode)->root;
4737 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4738 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4741 * We need to hold space in order to delete our orphan item once we've
4742 * added it, so this takes the reservation so we can release it later
4743 * when we are truly done with the orphan item.
4745 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4746 trace_btrfs_space_reservation(root->fs_info, "orphan",
4747 btrfs_ino(inode), num_bytes, 1);
4748 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4751 void btrfs_orphan_release_metadata(struct inode *inode)
4753 struct btrfs_root *root = BTRFS_I(inode)->root;
4754 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4755 trace_btrfs_space_reservation(root->fs_info, "orphan",
4756 btrfs_ino(inode), num_bytes, 0);
4757 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4761 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4762 * root: the root of the parent directory
4763 * rsv: block reservation
4764 * items: the number of items that we need do reservation
4765 * qgroup_reserved: used to return the reserved size in qgroup
4767 * This function is used to reserve the space for snapshot/subvolume
4768 * creation and deletion. Those operations are different with the
4769 * common file/directory operations, they change two fs/file trees
4770 * and root tree, the number of items that the qgroup reserves is
4771 * different with the free space reservation. So we can not use
4772 * the space reseravtion mechanism in start_transaction().
4774 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4775 struct btrfs_block_rsv *rsv,
4777 u64 *qgroup_reserved,
4778 bool use_global_rsv)
4782 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4784 if (root->fs_info->quota_enabled) {
4785 /* One for parent inode, two for dir entries */
4786 num_bytes = 3 * root->leafsize;
4787 ret = btrfs_qgroup_reserve(root, num_bytes);
4794 *qgroup_reserved = num_bytes;
4796 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4797 rsv->space_info = __find_space_info(root->fs_info,
4798 BTRFS_BLOCK_GROUP_METADATA);
4799 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4800 BTRFS_RESERVE_FLUSH_ALL);
4802 if (ret == -ENOSPC && use_global_rsv)
4803 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4806 if (*qgroup_reserved)
4807 btrfs_qgroup_free(root, *qgroup_reserved);
4813 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4814 struct btrfs_block_rsv *rsv,
4815 u64 qgroup_reserved)
4817 btrfs_block_rsv_release(root, rsv, (u64)-1);
4818 if (qgroup_reserved)
4819 btrfs_qgroup_free(root, qgroup_reserved);
4823 * drop_outstanding_extent - drop an outstanding extent
4824 * @inode: the inode we're dropping the extent for
4826 * This is called when we are freeing up an outstanding extent, either called
4827 * after an error or after an extent is written. This will return the number of
4828 * reserved extents that need to be freed. This must be called with
4829 * BTRFS_I(inode)->lock held.
4831 static unsigned drop_outstanding_extent(struct inode *inode)
4833 unsigned drop_inode_space = 0;
4834 unsigned dropped_extents = 0;
4836 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4837 BTRFS_I(inode)->outstanding_extents--;
4839 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4840 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4841 &BTRFS_I(inode)->runtime_flags))
4842 drop_inode_space = 1;
4845 * If we have more or the same amount of outsanding extents than we have
4846 * reserved then we need to leave the reserved extents count alone.
4848 if (BTRFS_I(inode)->outstanding_extents >=
4849 BTRFS_I(inode)->reserved_extents)
4850 return drop_inode_space;
4852 dropped_extents = BTRFS_I(inode)->reserved_extents -
4853 BTRFS_I(inode)->outstanding_extents;
4854 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4855 return dropped_extents + drop_inode_space;
4859 * calc_csum_metadata_size - return the amount of metada space that must be
4860 * reserved/free'd for the given bytes.
4861 * @inode: the inode we're manipulating
4862 * @num_bytes: the number of bytes in question
4863 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4865 * This adjusts the number of csum_bytes in the inode and then returns the
4866 * correct amount of metadata that must either be reserved or freed. We
4867 * calculate how many checksums we can fit into one leaf and then divide the
4868 * number of bytes that will need to be checksumed by this value to figure out
4869 * how many checksums will be required. If we are adding bytes then the number
4870 * may go up and we will return the number of additional bytes that must be
4871 * reserved. If it is going down we will return the number of bytes that must
4874 * This must be called with BTRFS_I(inode)->lock held.
4876 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4879 struct btrfs_root *root = BTRFS_I(inode)->root;
4881 int num_csums_per_leaf;
4885 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4886 BTRFS_I(inode)->csum_bytes == 0)
4889 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4891 BTRFS_I(inode)->csum_bytes += num_bytes;
4893 BTRFS_I(inode)->csum_bytes -= num_bytes;
4894 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4895 num_csums_per_leaf = (int)div64_u64(csum_size,
4896 sizeof(struct btrfs_csum_item) +
4897 sizeof(struct btrfs_disk_key));
4898 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4899 num_csums = num_csums + num_csums_per_leaf - 1;
4900 num_csums = num_csums / num_csums_per_leaf;
4902 old_csums = old_csums + num_csums_per_leaf - 1;
4903 old_csums = old_csums / num_csums_per_leaf;
4905 /* No change, no need to reserve more */
4906 if (old_csums == num_csums)
4910 return btrfs_calc_trans_metadata_size(root,
4911 num_csums - old_csums);
4913 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4916 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4918 struct btrfs_root *root = BTRFS_I(inode)->root;
4919 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4922 unsigned nr_extents = 0;
4923 int extra_reserve = 0;
4924 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4926 bool delalloc_lock = true;
4930 /* If we are a free space inode we need to not flush since we will be in
4931 * the middle of a transaction commit. We also don't need the delalloc
4932 * mutex since we won't race with anybody. We need this mostly to make
4933 * lockdep shut its filthy mouth.
4935 if (btrfs_is_free_space_inode(inode)) {
4936 flush = BTRFS_RESERVE_NO_FLUSH;
4937 delalloc_lock = false;
4940 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4941 btrfs_transaction_in_commit(root->fs_info))
4942 schedule_timeout(1);
4945 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4947 num_bytes = ALIGN(num_bytes, root->sectorsize);
4949 spin_lock(&BTRFS_I(inode)->lock);
4950 BTRFS_I(inode)->outstanding_extents++;
4952 if (BTRFS_I(inode)->outstanding_extents >
4953 BTRFS_I(inode)->reserved_extents)
4954 nr_extents = BTRFS_I(inode)->outstanding_extents -
4955 BTRFS_I(inode)->reserved_extents;
4958 * Add an item to reserve for updating the inode when we complete the
4961 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4962 &BTRFS_I(inode)->runtime_flags)) {
4967 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4968 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4969 csum_bytes = BTRFS_I(inode)->csum_bytes;
4970 spin_unlock(&BTRFS_I(inode)->lock);
4972 if (root->fs_info->quota_enabled) {
4973 ret = btrfs_qgroup_reserve(root, num_bytes +
4974 nr_extents * root->leafsize);
4979 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4980 if (unlikely(ret)) {
4981 if (root->fs_info->quota_enabled)
4982 btrfs_qgroup_free(root, num_bytes +
4983 nr_extents * root->leafsize);
4987 spin_lock(&BTRFS_I(inode)->lock);
4988 if (extra_reserve) {
4989 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4990 &BTRFS_I(inode)->runtime_flags);
4993 BTRFS_I(inode)->reserved_extents += nr_extents;
4994 spin_unlock(&BTRFS_I(inode)->lock);
4997 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5000 trace_btrfs_space_reservation(root->fs_info,"delalloc",
5001 btrfs_ino(inode), to_reserve, 1);
5002 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5007 spin_lock(&BTRFS_I(inode)->lock);
5008 dropped = drop_outstanding_extent(inode);
5010 * If the inodes csum_bytes is the same as the original
5011 * csum_bytes then we know we haven't raced with any free()ers
5012 * so we can just reduce our inodes csum bytes and carry on.
5014 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5015 calc_csum_metadata_size(inode, num_bytes, 0);
5017 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5021 * This is tricky, but first we need to figure out how much we
5022 * free'd from any free-ers that occured during this
5023 * reservation, so we reset ->csum_bytes to the csum_bytes
5024 * before we dropped our lock, and then call the free for the
5025 * number of bytes that were freed while we were trying our
5028 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5029 BTRFS_I(inode)->csum_bytes = csum_bytes;
5030 to_free = calc_csum_metadata_size(inode, bytes, 0);
5034 * Now we need to see how much we would have freed had we not
5035 * been making this reservation and our ->csum_bytes were not
5036 * artificially inflated.
5038 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5039 bytes = csum_bytes - orig_csum_bytes;
5040 bytes = calc_csum_metadata_size(inode, bytes, 0);
5043 * Now reset ->csum_bytes to what it should be. If bytes is
5044 * more than to_free then we would have free'd more space had we
5045 * not had an artificially high ->csum_bytes, so we need to free
5046 * the remainder. If bytes is the same or less then we don't
5047 * need to do anything, the other free-ers did the correct
5050 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5051 if (bytes > to_free)
5052 to_free = bytes - to_free;
5056 spin_unlock(&BTRFS_I(inode)->lock);
5058 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5061 btrfs_block_rsv_release(root, block_rsv, to_free);
5062 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5063 btrfs_ino(inode), to_free, 0);
5066 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5071 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5072 * @inode: the inode to release the reservation for
5073 * @num_bytes: the number of bytes we're releasing
5075 * This will release the metadata reservation for an inode. This can be called
5076 * once we complete IO for a given set of bytes to release their metadata
5079 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5081 struct btrfs_root *root = BTRFS_I(inode)->root;
5085 num_bytes = ALIGN(num_bytes, root->sectorsize);
5086 spin_lock(&BTRFS_I(inode)->lock);
5087 dropped = drop_outstanding_extent(inode);
5090 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5091 spin_unlock(&BTRFS_I(inode)->lock);
5093 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5095 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5096 btrfs_ino(inode), to_free, 0);
5097 if (root->fs_info->quota_enabled) {
5098 btrfs_qgroup_free(root, num_bytes +
5099 dropped * root->leafsize);
5102 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5107 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5108 * @inode: inode we're writing to
5109 * @num_bytes: the number of bytes we want to allocate
5111 * This will do the following things
5113 * o reserve space in the data space info for num_bytes
5114 * o reserve space in the metadata space info based on number of outstanding
5115 * extents and how much csums will be needed
5116 * o add to the inodes ->delalloc_bytes
5117 * o add it to the fs_info's delalloc inodes list.
5119 * This will return 0 for success and -ENOSPC if there is no space left.
5121 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5125 ret = btrfs_check_data_free_space(inode, num_bytes);
5129 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5131 btrfs_free_reserved_data_space(inode, num_bytes);
5139 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5140 * @inode: inode we're releasing space for
5141 * @num_bytes: the number of bytes we want to free up
5143 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5144 * called in the case that we don't need the metadata AND data reservations
5145 * anymore. So if there is an error or we insert an inline extent.
5147 * This function will release the metadata space that was not used and will
5148 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5149 * list if there are no delalloc bytes left.
5151 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5153 btrfs_delalloc_release_metadata(inode, num_bytes);
5154 btrfs_free_reserved_data_space(inode, num_bytes);
5157 static int update_block_group(struct btrfs_root *root,
5158 u64 bytenr, u64 num_bytes, int alloc)
5160 struct btrfs_block_group_cache *cache = NULL;
5161 struct btrfs_fs_info *info = root->fs_info;
5162 u64 total = num_bytes;
5167 /* block accounting for super block */
5168 spin_lock(&info->delalloc_root_lock);
5169 old_val = btrfs_super_bytes_used(info->super_copy);
5171 old_val += num_bytes;
5173 old_val -= num_bytes;
5174 btrfs_set_super_bytes_used(info->super_copy, old_val);
5175 spin_unlock(&info->delalloc_root_lock);
5178 cache = btrfs_lookup_block_group(info, bytenr);
5181 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5182 BTRFS_BLOCK_GROUP_RAID1 |
5183 BTRFS_BLOCK_GROUP_RAID10))
5188 * If this block group has free space cache written out, we
5189 * need to make sure to load it if we are removing space. This
5190 * is because we need the unpinning stage to actually add the
5191 * space back to the block group, otherwise we will leak space.
5193 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5194 cache_block_group(cache, 1);
5196 byte_in_group = bytenr - cache->key.objectid;
5197 WARN_ON(byte_in_group > cache->key.offset);
5199 spin_lock(&cache->space_info->lock);
5200 spin_lock(&cache->lock);
5202 if (btrfs_test_opt(root, SPACE_CACHE) &&
5203 cache->disk_cache_state < BTRFS_DC_CLEAR)
5204 cache->disk_cache_state = BTRFS_DC_CLEAR;
5207 old_val = btrfs_block_group_used(&cache->item);
5208 num_bytes = min(total, cache->key.offset - byte_in_group);
5210 old_val += num_bytes;
5211 btrfs_set_block_group_used(&cache->item, old_val);
5212 cache->reserved -= num_bytes;
5213 cache->space_info->bytes_reserved -= num_bytes;
5214 cache->space_info->bytes_used += num_bytes;
5215 cache->space_info->disk_used += num_bytes * factor;
5216 spin_unlock(&cache->lock);
5217 spin_unlock(&cache->space_info->lock);
5219 old_val -= num_bytes;
5220 btrfs_set_block_group_used(&cache->item, old_val);
5221 cache->pinned += num_bytes;
5222 cache->space_info->bytes_pinned += num_bytes;
5223 cache->space_info->bytes_used -= num_bytes;
5224 cache->space_info->disk_used -= num_bytes * factor;
5225 spin_unlock(&cache->lock);
5226 spin_unlock(&cache->space_info->lock);
5228 set_extent_dirty(info->pinned_extents,
5229 bytenr, bytenr + num_bytes - 1,
5230 GFP_NOFS | __GFP_NOFAIL);
5232 btrfs_put_block_group(cache);
5234 bytenr += num_bytes;
5239 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5241 struct btrfs_block_group_cache *cache;
5244 spin_lock(&root->fs_info->block_group_cache_lock);
5245 bytenr = root->fs_info->first_logical_byte;
5246 spin_unlock(&root->fs_info->block_group_cache_lock);
5248 if (bytenr < (u64)-1)
5251 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5255 bytenr = cache->key.objectid;
5256 btrfs_put_block_group(cache);
5261 static int pin_down_extent(struct btrfs_root *root,
5262 struct btrfs_block_group_cache *cache,
5263 u64 bytenr, u64 num_bytes, int reserved)
5265 spin_lock(&cache->space_info->lock);
5266 spin_lock(&cache->lock);
5267 cache->pinned += num_bytes;
5268 cache->space_info->bytes_pinned += num_bytes;
5270 cache->reserved -= num_bytes;
5271 cache->space_info->bytes_reserved -= num_bytes;
5273 spin_unlock(&cache->lock);
5274 spin_unlock(&cache->space_info->lock);
5276 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5277 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5282 * this function must be called within transaction
5284 int btrfs_pin_extent(struct btrfs_root *root,
5285 u64 bytenr, u64 num_bytes, int reserved)
5287 struct btrfs_block_group_cache *cache;
5289 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5290 BUG_ON(!cache); /* Logic error */
5292 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5294 btrfs_put_block_group(cache);
5299 * this function must be called within transaction
5301 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5302 u64 bytenr, u64 num_bytes)
5304 struct btrfs_block_group_cache *cache;
5307 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5312 * pull in the free space cache (if any) so that our pin
5313 * removes the free space from the cache. We have load_only set
5314 * to one because the slow code to read in the free extents does check
5315 * the pinned extents.
5317 cache_block_group(cache, 1);
5319 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5321 /* remove us from the free space cache (if we're there at all) */
5322 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5323 btrfs_put_block_group(cache);
5327 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5330 struct btrfs_block_group_cache *block_group;
5331 struct btrfs_caching_control *caching_ctl;
5333 block_group = btrfs_lookup_block_group(root->fs_info, start);
5337 cache_block_group(block_group, 0);
5338 caching_ctl = get_caching_control(block_group);
5342 BUG_ON(!block_group_cache_done(block_group));
5343 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5345 mutex_lock(&caching_ctl->mutex);
5347 if (start >= caching_ctl->progress) {
5348 ret = add_excluded_extent(root, start, num_bytes);
5349 } else if (start + num_bytes <= caching_ctl->progress) {
5350 ret = btrfs_remove_free_space(block_group,
5353 num_bytes = caching_ctl->progress - start;
5354 ret = btrfs_remove_free_space(block_group,
5359 num_bytes = (start + num_bytes) -
5360 caching_ctl->progress;
5361 start = caching_ctl->progress;
5362 ret = add_excluded_extent(root, start, num_bytes);
5365 mutex_unlock(&caching_ctl->mutex);
5366 put_caching_control(caching_ctl);
5368 btrfs_put_block_group(block_group);
5372 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5373 struct extent_buffer *eb)
5375 struct btrfs_file_extent_item *item;
5376 struct btrfs_key key;
5380 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5383 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5384 btrfs_item_key_to_cpu(eb, &key, i);
5385 if (key.type != BTRFS_EXTENT_DATA_KEY)
5387 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5388 found_type = btrfs_file_extent_type(eb, item);
5389 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5391 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5393 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5394 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5395 __exclude_logged_extent(log, key.objectid, key.offset);
5402 * btrfs_update_reserved_bytes - update the block_group and space info counters
5403 * @cache: The cache we are manipulating
5404 * @num_bytes: The number of bytes in question
5405 * @reserve: One of the reservation enums
5407 * This is called by the allocator when it reserves space, or by somebody who is
5408 * freeing space that was never actually used on disk. For example if you
5409 * reserve some space for a new leaf in transaction A and before transaction A
5410 * commits you free that leaf, you call this with reserve set to 0 in order to
5411 * clear the reservation.
5413 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5414 * ENOSPC accounting. For data we handle the reservation through clearing the
5415 * delalloc bits in the io_tree. We have to do this since we could end up
5416 * allocating less disk space for the amount of data we have reserved in the
5417 * case of compression.
5419 * If this is a reservation and the block group has become read only we cannot
5420 * make the reservation and return -EAGAIN, otherwise this function always
5423 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5424 u64 num_bytes, int reserve)
5426 struct btrfs_space_info *space_info = cache->space_info;
5429 spin_lock(&space_info->lock);
5430 spin_lock(&cache->lock);
5431 if (reserve != RESERVE_FREE) {
5435 cache->reserved += num_bytes;
5436 space_info->bytes_reserved += num_bytes;
5437 if (reserve == RESERVE_ALLOC) {
5438 trace_btrfs_space_reservation(cache->fs_info,
5439 "space_info", space_info->flags,
5441 space_info->bytes_may_use -= num_bytes;
5446 space_info->bytes_readonly += num_bytes;
5447 cache->reserved -= num_bytes;
5448 space_info->bytes_reserved -= num_bytes;
5449 space_info->reservation_progress++;
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);
5986 add_pinned_bytes(root->fs_info, buf->len,
5987 btrfs_header_level(buf),
5988 root->root_key.objectid);
5991 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5994 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5995 btrfs_put_block_group(cache);
5998 /* Can return -ENOMEM */
5999 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6000 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6001 u64 owner, u64 offset, int for_cow)
6004 struct btrfs_fs_info *fs_info = root->fs_info;
6006 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6009 * tree log blocks never actually go into the extent allocation
6010 * tree, just update pinning info and exit early.
6012 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6013 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6014 /* unlocks the pinned mutex */
6015 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6017 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6018 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6020 parent, root_objectid, (int)owner,
6021 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6023 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6025 parent, root_objectid, owner,
6026 offset, BTRFS_DROP_DELAYED_REF,
6032 static u64 stripe_align(struct btrfs_root *root,
6033 struct btrfs_block_group_cache *cache,
6034 u64 val, u64 num_bytes)
6036 u64 ret = ALIGN(val, root->stripesize);
6041 * when we wait for progress in the block group caching, its because
6042 * our allocation attempt failed at least once. So, we must sleep
6043 * and let some progress happen before we try again.
6045 * This function will sleep at least once waiting for new free space to
6046 * show up, and then it will check the block group free space numbers
6047 * for our min num_bytes. Another option is to have it go ahead
6048 * and look in the rbtree for a free extent of a given size, but this
6051 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6052 * any of the information in this block group.
6054 static noinline void
6055 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6058 struct btrfs_caching_control *caching_ctl;
6060 caching_ctl = get_caching_control(cache);
6064 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6065 (cache->free_space_ctl->free_space >= num_bytes));
6067 put_caching_control(caching_ctl);
6071 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6073 struct btrfs_caching_control *caching_ctl;
6076 caching_ctl = get_caching_control(cache);
6078 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6080 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6081 if (cache->cached == BTRFS_CACHE_ERROR)
6083 put_caching_control(caching_ctl);
6087 int __get_raid_index(u64 flags)
6089 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6090 return BTRFS_RAID_RAID10;
6091 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6092 return BTRFS_RAID_RAID1;
6093 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6094 return BTRFS_RAID_DUP;
6095 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6096 return BTRFS_RAID_RAID0;
6097 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6098 return BTRFS_RAID_RAID5;
6099 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6100 return BTRFS_RAID_RAID6;
6102 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6105 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6107 return __get_raid_index(cache->flags);
6110 enum btrfs_loop_type {
6111 LOOP_CACHING_NOWAIT = 0,
6112 LOOP_CACHING_WAIT = 1,
6113 LOOP_ALLOC_CHUNK = 2,
6114 LOOP_NO_EMPTY_SIZE = 3,
6118 * walks the btree of allocated extents and find a hole of a given size.
6119 * The key ins is changed to record the hole:
6120 * ins->objectid == start position
6121 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6122 * ins->offset == the size of the hole.
6123 * Any available blocks before search_start are skipped.
6125 * If there is no suitable free space, we will record the max size of
6126 * the free space extent currently.
6128 static noinline int find_free_extent(struct btrfs_root *orig_root,
6129 u64 num_bytes, u64 empty_size,
6130 u64 hint_byte, struct btrfs_key *ins,
6134 struct btrfs_root *root = orig_root->fs_info->extent_root;
6135 struct btrfs_free_cluster *last_ptr = NULL;
6136 struct btrfs_block_group_cache *block_group = NULL;
6137 struct btrfs_block_group_cache *used_block_group;
6138 u64 search_start = 0;
6139 u64 max_extent_size = 0;
6140 int empty_cluster = 2 * 1024 * 1024;
6141 struct btrfs_space_info *space_info;
6143 int index = __get_raid_index(flags);
6144 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6145 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6146 bool found_uncached_bg = false;
6147 bool failed_cluster_refill = false;
6148 bool failed_alloc = false;
6149 bool use_cluster = true;
6150 bool have_caching_bg = false;
6152 WARN_ON(num_bytes < root->sectorsize);
6153 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6157 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6159 space_info = __find_space_info(root->fs_info, flags);
6161 btrfs_err(root->fs_info, "No space info for %llu", flags);
6166 * If the space info is for both data and metadata it means we have a
6167 * small filesystem and we can't use the clustering stuff.
6169 if (btrfs_mixed_space_info(space_info))
6170 use_cluster = false;
6172 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6173 last_ptr = &root->fs_info->meta_alloc_cluster;
6174 if (!btrfs_test_opt(root, SSD))
6175 empty_cluster = 64 * 1024;
6178 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6179 btrfs_test_opt(root, SSD)) {
6180 last_ptr = &root->fs_info->data_alloc_cluster;
6184 spin_lock(&last_ptr->lock);
6185 if (last_ptr->block_group)
6186 hint_byte = last_ptr->window_start;
6187 spin_unlock(&last_ptr->lock);
6190 search_start = max(search_start, first_logical_byte(root, 0));
6191 search_start = max(search_start, hint_byte);
6196 if (search_start == hint_byte) {
6197 block_group = btrfs_lookup_block_group(root->fs_info,
6199 used_block_group = block_group;
6201 * we don't want to use the block group if it doesn't match our
6202 * allocation bits, or if its not cached.
6204 * However if we are re-searching with an ideal block group
6205 * picked out then we don't care that the block group is cached.
6207 if (block_group && block_group_bits(block_group, flags) &&
6208 block_group->cached != BTRFS_CACHE_NO) {
6209 down_read(&space_info->groups_sem);
6210 if (list_empty(&block_group->list) ||
6213 * someone is removing this block group,
6214 * we can't jump into the have_block_group
6215 * target because our list pointers are not
6218 btrfs_put_block_group(block_group);
6219 up_read(&space_info->groups_sem);
6221 index = get_block_group_index(block_group);
6222 goto have_block_group;
6224 } else if (block_group) {
6225 btrfs_put_block_group(block_group);
6229 have_caching_bg = false;
6230 down_read(&space_info->groups_sem);
6231 list_for_each_entry(block_group, &space_info->block_groups[index],
6236 used_block_group = block_group;
6237 btrfs_get_block_group(block_group);
6238 search_start = block_group->key.objectid;
6241 * this can happen if we end up cycling through all the
6242 * raid types, but we want to make sure we only allocate
6243 * for the proper type.
6245 if (!block_group_bits(block_group, flags)) {
6246 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6247 BTRFS_BLOCK_GROUP_RAID1 |
6248 BTRFS_BLOCK_GROUP_RAID5 |
6249 BTRFS_BLOCK_GROUP_RAID6 |
6250 BTRFS_BLOCK_GROUP_RAID10;
6253 * if they asked for extra copies and this block group
6254 * doesn't provide them, bail. This does allow us to
6255 * fill raid0 from raid1.
6257 if ((flags & extra) && !(block_group->flags & extra))
6262 cached = block_group_cache_done(block_group);
6263 if (unlikely(!cached)) {
6264 found_uncached_bg = true;
6265 ret = cache_block_group(block_group, 0);
6270 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6272 if (unlikely(block_group->ro))
6276 * Ok we want to try and use the cluster allocator, so
6280 unsigned long aligned_cluster;
6282 * the refill lock keeps out other
6283 * people trying to start a new cluster
6285 spin_lock(&last_ptr->refill_lock);
6286 used_block_group = last_ptr->block_group;
6287 if (used_block_group != block_group &&
6288 (!used_block_group ||
6289 used_block_group->ro ||
6290 !block_group_bits(used_block_group, flags))) {
6291 used_block_group = block_group;
6292 goto refill_cluster;
6295 if (used_block_group != block_group)
6296 btrfs_get_block_group(used_block_group);
6298 offset = btrfs_alloc_from_cluster(used_block_group,
6301 used_block_group->key.objectid,
6304 /* we have a block, we're done */
6305 spin_unlock(&last_ptr->refill_lock);
6306 trace_btrfs_reserve_extent_cluster(root,
6307 block_group, search_start, num_bytes);
6311 WARN_ON(last_ptr->block_group != used_block_group);
6312 if (used_block_group != block_group) {
6313 btrfs_put_block_group(used_block_group);
6314 used_block_group = block_group;
6317 BUG_ON(used_block_group != block_group);
6318 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6319 * set up a new clusters, so lets just skip it
6320 * and let the allocator find whatever block
6321 * it can find. If we reach this point, we
6322 * will have tried the cluster allocator
6323 * plenty of times and not have found
6324 * anything, so we are likely way too
6325 * fragmented for the clustering stuff to find
6328 * However, if the cluster is taken from the
6329 * current block group, release the cluster
6330 * first, so that we stand a better chance of
6331 * succeeding in the unclustered
6333 if (loop >= LOOP_NO_EMPTY_SIZE &&
6334 last_ptr->block_group != block_group) {
6335 spin_unlock(&last_ptr->refill_lock);
6336 goto unclustered_alloc;
6340 * this cluster didn't work out, free it and
6343 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6345 if (loop >= LOOP_NO_EMPTY_SIZE) {
6346 spin_unlock(&last_ptr->refill_lock);
6347 goto unclustered_alloc;
6350 aligned_cluster = max_t(unsigned long,
6351 empty_cluster + empty_size,
6352 block_group->full_stripe_len);
6354 /* allocate a cluster in this block group */
6355 ret = btrfs_find_space_cluster(root, block_group,
6356 last_ptr, search_start,
6361 * now pull our allocation out of this
6364 offset = btrfs_alloc_from_cluster(block_group,
6370 /* we found one, proceed */
6371 spin_unlock(&last_ptr->refill_lock);
6372 trace_btrfs_reserve_extent_cluster(root,
6373 block_group, search_start,
6377 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6378 && !failed_cluster_refill) {
6379 spin_unlock(&last_ptr->refill_lock);
6381 failed_cluster_refill = true;
6382 wait_block_group_cache_progress(block_group,
6383 num_bytes + empty_cluster + empty_size);
6384 goto have_block_group;
6388 * at this point we either didn't find a cluster
6389 * or we weren't able to allocate a block from our
6390 * cluster. Free the cluster we've been trying
6391 * to use, and go to the next block group
6393 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6394 spin_unlock(&last_ptr->refill_lock);
6399 spin_lock(&block_group->free_space_ctl->tree_lock);
6401 block_group->free_space_ctl->free_space <
6402 num_bytes + empty_cluster + empty_size) {
6403 if (block_group->free_space_ctl->free_space >
6406 block_group->free_space_ctl->free_space;
6407 spin_unlock(&block_group->free_space_ctl->tree_lock);
6410 spin_unlock(&block_group->free_space_ctl->tree_lock);
6412 offset = btrfs_find_space_for_alloc(block_group, search_start,
6413 num_bytes, empty_size,
6416 * If we didn't find a chunk, and we haven't failed on this
6417 * block group before, and this block group is in the middle of
6418 * caching and we are ok with waiting, then go ahead and wait
6419 * for progress to be made, and set failed_alloc to true.
6421 * If failed_alloc is true then we've already waited on this
6422 * block group once and should move on to the next block group.
6424 if (!offset && !failed_alloc && !cached &&
6425 loop > LOOP_CACHING_NOWAIT) {
6426 wait_block_group_cache_progress(block_group,
6427 num_bytes + empty_size);
6428 failed_alloc = true;
6429 goto have_block_group;
6430 } else if (!offset) {
6432 have_caching_bg = true;
6436 search_start = stripe_align(root, used_block_group,
6439 /* move on to the next group */
6440 if (search_start + num_bytes >
6441 used_block_group->key.objectid + used_block_group->key.offset) {
6442 btrfs_add_free_space(used_block_group, offset, num_bytes);
6446 if (offset < search_start)
6447 btrfs_add_free_space(used_block_group, offset,
6448 search_start - offset);
6449 BUG_ON(offset > search_start);
6451 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6453 if (ret == -EAGAIN) {
6454 btrfs_add_free_space(used_block_group, offset, num_bytes);
6458 /* we are all good, lets return */
6459 ins->objectid = search_start;
6460 ins->offset = num_bytes;
6462 trace_btrfs_reserve_extent(orig_root, block_group,
6463 search_start, num_bytes);
6464 if (used_block_group != block_group)
6465 btrfs_put_block_group(used_block_group);
6466 btrfs_put_block_group(block_group);
6469 failed_cluster_refill = false;
6470 failed_alloc = false;
6471 BUG_ON(index != get_block_group_index(block_group));
6472 if (used_block_group != block_group)
6473 btrfs_put_block_group(used_block_group);
6474 btrfs_put_block_group(block_group);
6476 up_read(&space_info->groups_sem);
6478 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6481 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6485 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6486 * caching kthreads as we move along
6487 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6488 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6489 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6492 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6495 if (loop == LOOP_ALLOC_CHUNK) {
6496 struct btrfs_trans_handle *trans;
6498 trans = btrfs_join_transaction(root);
6499 if (IS_ERR(trans)) {
6500 ret = PTR_ERR(trans);
6504 ret = do_chunk_alloc(trans, root, flags,
6507 * Do not bail out on ENOSPC since we
6508 * can do more things.
6510 if (ret < 0 && ret != -ENOSPC)
6511 btrfs_abort_transaction(trans,
6515 btrfs_end_transaction(trans, root);
6520 if (loop == LOOP_NO_EMPTY_SIZE) {
6526 } else if (!ins->objectid) {
6528 } else if (ins->objectid) {
6533 ins->offset = max_extent_size;
6537 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6538 int dump_block_groups)
6540 struct btrfs_block_group_cache *cache;
6543 spin_lock(&info->lock);
6544 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6546 info->total_bytes - info->bytes_used - info->bytes_pinned -
6547 info->bytes_reserved - info->bytes_readonly,
6548 (info->full) ? "" : "not ");
6549 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6550 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6551 info->total_bytes, info->bytes_used, info->bytes_pinned,
6552 info->bytes_reserved, info->bytes_may_use,
6553 info->bytes_readonly);
6554 spin_unlock(&info->lock);
6556 if (!dump_block_groups)
6559 down_read(&info->groups_sem);
6561 list_for_each_entry(cache, &info->block_groups[index], list) {
6562 spin_lock(&cache->lock);
6563 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6564 cache->key.objectid, cache->key.offset,
6565 btrfs_block_group_used(&cache->item), cache->pinned,
6566 cache->reserved, cache->ro ? "[readonly]" : "");
6567 btrfs_dump_free_space(cache, bytes);
6568 spin_unlock(&cache->lock);
6570 if (++index < BTRFS_NR_RAID_TYPES)
6572 up_read(&info->groups_sem);
6575 int btrfs_reserve_extent(struct btrfs_root *root,
6576 u64 num_bytes, u64 min_alloc_size,
6577 u64 empty_size, u64 hint_byte,
6578 struct btrfs_key *ins, int is_data)
6580 bool final_tried = false;
6584 flags = btrfs_get_alloc_profile(root, is_data);
6586 WARN_ON(num_bytes < root->sectorsize);
6587 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6590 if (ret == -ENOSPC) {
6591 if (!final_tried && ins->offset) {
6592 num_bytes = min(num_bytes >> 1, ins->offset);
6593 num_bytes = round_down(num_bytes, root->sectorsize);
6594 num_bytes = max(num_bytes, min_alloc_size);
6595 if (num_bytes == min_alloc_size)
6598 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6599 struct btrfs_space_info *sinfo;
6601 sinfo = __find_space_info(root->fs_info, flags);
6602 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6605 dump_space_info(sinfo, num_bytes, 1);
6609 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6614 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6615 u64 start, u64 len, int pin)
6617 struct btrfs_block_group_cache *cache;
6620 cache = btrfs_lookup_block_group(root->fs_info, start);
6622 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6627 if (btrfs_test_opt(root, DISCARD))
6628 ret = btrfs_discard_extent(root, start, len, NULL);
6631 pin_down_extent(root, cache, start, len, 1);
6633 btrfs_add_free_space(cache, start, len);
6634 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6636 btrfs_put_block_group(cache);
6638 trace_btrfs_reserved_extent_free(root, start, len);
6643 int btrfs_free_reserved_extent(struct btrfs_root *root,
6646 return __btrfs_free_reserved_extent(root, start, len, 0);
6649 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6652 return __btrfs_free_reserved_extent(root, start, len, 1);
6655 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6656 struct btrfs_root *root,
6657 u64 parent, u64 root_objectid,
6658 u64 flags, u64 owner, u64 offset,
6659 struct btrfs_key *ins, int ref_mod)
6662 struct btrfs_fs_info *fs_info = root->fs_info;
6663 struct btrfs_extent_item *extent_item;
6664 struct btrfs_extent_inline_ref *iref;
6665 struct btrfs_path *path;
6666 struct extent_buffer *leaf;
6671 type = BTRFS_SHARED_DATA_REF_KEY;
6673 type = BTRFS_EXTENT_DATA_REF_KEY;
6675 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6677 path = btrfs_alloc_path();
6681 path->leave_spinning = 1;
6682 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6685 btrfs_free_path(path);
6689 leaf = path->nodes[0];
6690 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6691 struct btrfs_extent_item);
6692 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6693 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6694 btrfs_set_extent_flags(leaf, extent_item,
6695 flags | BTRFS_EXTENT_FLAG_DATA);
6697 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6698 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6700 struct btrfs_shared_data_ref *ref;
6701 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6702 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6703 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6705 struct btrfs_extent_data_ref *ref;
6706 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6707 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6708 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6709 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6710 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6713 btrfs_mark_buffer_dirty(path->nodes[0]);
6714 btrfs_free_path(path);
6716 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6717 if (ret) { /* -ENOENT, logic error */
6718 btrfs_err(fs_info, "update block group failed for %llu %llu",
6719 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();
6749 path->leave_spinning = 1;
6750 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6753 btrfs_free_path(path);
6757 leaf = path->nodes[0];
6758 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6759 struct btrfs_extent_item);
6760 btrfs_set_extent_refs(leaf, extent_item, 1);
6761 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6762 btrfs_set_extent_flags(leaf, extent_item,
6763 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6765 if (skinny_metadata) {
6766 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6768 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6769 btrfs_set_tree_block_key(leaf, block_info, key);
6770 btrfs_set_tree_block_level(leaf, block_info, level);
6771 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6775 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6776 btrfs_set_extent_inline_ref_type(leaf, iref,
6777 BTRFS_SHARED_BLOCK_REF_KEY);
6778 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6780 btrfs_set_extent_inline_ref_type(leaf, iref,
6781 BTRFS_TREE_BLOCK_REF_KEY);
6782 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6785 btrfs_mark_buffer_dirty(leaf);
6786 btrfs_free_path(path);
6788 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6789 if (ret) { /* -ENOENT, logic error */
6790 btrfs_err(fs_info, "update block group failed for %llu %llu",
6791 ins->objectid, ins->offset);
6797 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6798 struct btrfs_root *root,
6799 u64 root_objectid, u64 owner,
6800 u64 offset, struct btrfs_key *ins)
6804 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6806 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6808 root_objectid, owner, offset,
6809 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6814 * this is used by the tree logging recovery code. It records that
6815 * an extent has been allocated and makes sure to clear the free
6816 * space cache bits as well
6818 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6819 struct btrfs_root *root,
6820 u64 root_objectid, u64 owner, u64 offset,
6821 struct btrfs_key *ins)
6824 struct btrfs_block_group_cache *block_group;
6827 * Mixed block groups will exclude before processing the log so we only
6828 * need to do the exlude dance if this fs isn't mixed.
6830 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6831 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6836 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6840 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6841 RESERVE_ALLOC_NO_ACCOUNT);
6842 BUG_ON(ret); /* logic error */
6843 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6844 0, owner, offset, ins, 1);
6845 btrfs_put_block_group(block_group);
6849 static struct extent_buffer *
6850 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6851 u64 bytenr, u32 blocksize, int level)
6853 struct extent_buffer *buf;
6855 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6857 return ERR_PTR(-ENOMEM);
6858 btrfs_set_header_generation(buf, trans->transid);
6859 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6860 btrfs_tree_lock(buf);
6861 clean_tree_block(trans, root, buf);
6862 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6864 btrfs_set_lock_blocking(buf);
6865 btrfs_set_buffer_uptodate(buf);
6867 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6869 * we allow two log transactions at a time, use different
6870 * EXENT bit to differentiate dirty pages.
6872 if (root->log_transid % 2 == 0)
6873 set_extent_dirty(&root->dirty_log_pages, buf->start,
6874 buf->start + buf->len - 1, GFP_NOFS);
6876 set_extent_new(&root->dirty_log_pages, buf->start,
6877 buf->start + buf->len - 1, GFP_NOFS);
6879 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6880 buf->start + buf->len - 1, GFP_NOFS);
6882 trans->blocks_used++;
6883 /* this returns a buffer locked for blocking */
6887 static struct btrfs_block_rsv *
6888 use_block_rsv(struct btrfs_trans_handle *trans,
6889 struct btrfs_root *root, u32 blocksize)
6891 struct btrfs_block_rsv *block_rsv;
6892 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6894 bool global_updated = false;
6896 block_rsv = get_block_rsv(trans, root);
6898 if (unlikely(block_rsv->size == 0))
6901 ret = block_rsv_use_bytes(block_rsv, blocksize);
6905 if (block_rsv->failfast)
6906 return ERR_PTR(ret);
6908 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6909 global_updated = true;
6910 update_global_block_rsv(root->fs_info);
6914 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6915 static DEFINE_RATELIMIT_STATE(_rs,
6916 DEFAULT_RATELIMIT_INTERVAL * 10,
6917 /*DEFAULT_RATELIMIT_BURST*/ 1);
6918 if (__ratelimit(&_rs))
6920 "btrfs: block rsv returned %d\n", ret);
6923 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6924 BTRFS_RESERVE_NO_FLUSH);
6928 * If we couldn't reserve metadata bytes try and use some from
6929 * the global reserve if its space type is the same as the global
6932 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6933 block_rsv->space_info == global_rsv->space_info) {
6934 ret = block_rsv_use_bytes(global_rsv, blocksize);
6938 return ERR_PTR(ret);
6941 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6942 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6944 block_rsv_add_bytes(block_rsv, blocksize, 0);
6945 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6949 * finds a free extent and does all the dirty work required for allocation
6950 * returns the key for the extent through ins, and a tree buffer for
6951 * the first block of the extent through buf.
6953 * returns the tree buffer or NULL.
6955 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6956 struct btrfs_root *root, u32 blocksize,
6957 u64 parent, u64 root_objectid,
6958 struct btrfs_disk_key *key, int level,
6959 u64 hint, u64 empty_size)
6961 struct btrfs_key ins;
6962 struct btrfs_block_rsv *block_rsv;
6963 struct extent_buffer *buf;
6966 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6969 block_rsv = use_block_rsv(trans, root, blocksize);
6970 if (IS_ERR(block_rsv))
6971 return ERR_CAST(block_rsv);
6973 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6974 empty_size, hint, &ins, 0);
6976 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6977 return ERR_PTR(ret);
6980 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6982 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6984 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6986 parent = ins.objectid;
6987 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6991 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6992 struct btrfs_delayed_extent_op *extent_op;
6993 extent_op = btrfs_alloc_delayed_extent_op();
6994 BUG_ON(!extent_op); /* -ENOMEM */
6996 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6998 memset(&extent_op->key, 0, sizeof(extent_op->key));
6999 extent_op->flags_to_set = flags;
7000 if (skinny_metadata)
7001 extent_op->update_key = 0;
7003 extent_op->update_key = 1;
7004 extent_op->update_flags = 1;
7005 extent_op->is_data = 0;
7006 extent_op->level = level;
7008 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7010 ins.offset, parent, root_objectid,
7011 level, BTRFS_ADD_DELAYED_EXTENT,
7013 BUG_ON(ret); /* -ENOMEM */
7018 struct walk_control {
7019 u64 refs[BTRFS_MAX_LEVEL];
7020 u64 flags[BTRFS_MAX_LEVEL];
7021 struct btrfs_key update_progress;
7032 #define DROP_REFERENCE 1
7033 #define UPDATE_BACKREF 2
7035 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7036 struct btrfs_root *root,
7037 struct walk_control *wc,
7038 struct btrfs_path *path)
7046 struct btrfs_key key;
7047 struct extent_buffer *eb;
7052 if (path->slots[wc->level] < wc->reada_slot) {
7053 wc->reada_count = wc->reada_count * 2 / 3;
7054 wc->reada_count = max(wc->reada_count, 2);
7056 wc->reada_count = wc->reada_count * 3 / 2;
7057 wc->reada_count = min_t(int, wc->reada_count,
7058 BTRFS_NODEPTRS_PER_BLOCK(root));
7061 eb = path->nodes[wc->level];
7062 nritems = btrfs_header_nritems(eb);
7063 blocksize = btrfs_level_size(root, wc->level - 1);
7065 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7066 if (nread >= wc->reada_count)
7070 bytenr = btrfs_node_blockptr(eb, slot);
7071 generation = btrfs_node_ptr_generation(eb, slot);
7073 if (slot == path->slots[wc->level])
7076 if (wc->stage == UPDATE_BACKREF &&
7077 generation <= root->root_key.offset)
7080 /* We don't lock the tree block, it's OK to be racy here */
7081 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7082 wc->level - 1, 1, &refs,
7084 /* We don't care about errors in readahead. */
7089 if (wc->stage == DROP_REFERENCE) {
7093 if (wc->level == 1 &&
7094 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7096 if (!wc->update_ref ||
7097 generation <= root->root_key.offset)
7099 btrfs_node_key_to_cpu(eb, &key, slot);
7100 ret = btrfs_comp_cpu_keys(&key,
7101 &wc->update_progress);
7105 if (wc->level == 1 &&
7106 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7110 ret = readahead_tree_block(root, bytenr, blocksize,
7116 wc->reada_slot = slot;
7120 * helper to process tree block while walking down the tree.
7122 * when wc->stage == UPDATE_BACKREF, this function updates
7123 * back refs for pointers in the block.
7125 * NOTE: return value 1 means we should stop walking down.
7127 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7128 struct btrfs_root *root,
7129 struct btrfs_path *path,
7130 struct walk_control *wc, int lookup_info)
7132 int level = wc->level;
7133 struct extent_buffer *eb = path->nodes[level];
7134 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7137 if (wc->stage == UPDATE_BACKREF &&
7138 btrfs_header_owner(eb) != root->root_key.objectid)
7142 * when reference count of tree block is 1, it won't increase
7143 * again. once full backref flag is set, we never clear it.
7146 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7147 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7148 BUG_ON(!path->locks[level]);
7149 ret = btrfs_lookup_extent_info(trans, root,
7150 eb->start, level, 1,
7153 BUG_ON(ret == -ENOMEM);
7156 BUG_ON(wc->refs[level] == 0);
7159 if (wc->stage == DROP_REFERENCE) {
7160 if (wc->refs[level] > 1)
7163 if (path->locks[level] && !wc->keep_locks) {
7164 btrfs_tree_unlock_rw(eb, path->locks[level]);
7165 path->locks[level] = 0;
7170 /* wc->stage == UPDATE_BACKREF */
7171 if (!(wc->flags[level] & flag)) {
7172 BUG_ON(!path->locks[level]);
7173 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7174 BUG_ON(ret); /* -ENOMEM */
7175 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7176 BUG_ON(ret); /* -ENOMEM */
7177 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7179 btrfs_header_level(eb), 0);
7180 BUG_ON(ret); /* -ENOMEM */
7181 wc->flags[level] |= flag;
7185 * the block is shared by multiple trees, so it's not good to
7186 * keep the tree lock
7188 if (path->locks[level] && level > 0) {
7189 btrfs_tree_unlock_rw(eb, path->locks[level]);
7190 path->locks[level] = 0;
7196 * helper to process tree block pointer.
7198 * when wc->stage == DROP_REFERENCE, this function checks
7199 * reference count of the block pointed to. if the block
7200 * is shared and we need update back refs for the subtree
7201 * rooted at the block, this function changes wc->stage to
7202 * UPDATE_BACKREF. if the block is shared and there is no
7203 * need to update back, this function drops the reference
7206 * NOTE: return value 1 means we should stop walking down.
7208 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7209 struct btrfs_root *root,
7210 struct btrfs_path *path,
7211 struct walk_control *wc, int *lookup_info)
7217 struct btrfs_key key;
7218 struct extent_buffer *next;
7219 int level = wc->level;
7223 generation = btrfs_node_ptr_generation(path->nodes[level],
7224 path->slots[level]);
7226 * if the lower level block was created before the snapshot
7227 * was created, we know there is no need to update back refs
7230 if (wc->stage == UPDATE_BACKREF &&
7231 generation <= root->root_key.offset) {
7236 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7237 blocksize = btrfs_level_size(root, level - 1);
7239 next = btrfs_find_tree_block(root, bytenr, blocksize);
7241 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7244 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7248 btrfs_tree_lock(next);
7249 btrfs_set_lock_blocking(next);
7251 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7252 &wc->refs[level - 1],
7253 &wc->flags[level - 1]);
7255 btrfs_tree_unlock(next);
7259 if (unlikely(wc->refs[level - 1] == 0)) {
7260 btrfs_err(root->fs_info, "Missing references.");
7265 if (wc->stage == DROP_REFERENCE) {
7266 if (wc->refs[level - 1] > 1) {
7268 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7271 if (!wc->update_ref ||
7272 generation <= root->root_key.offset)
7275 btrfs_node_key_to_cpu(path->nodes[level], &key,
7276 path->slots[level]);
7277 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7281 wc->stage = UPDATE_BACKREF;
7282 wc->shared_level = level - 1;
7286 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7290 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7291 btrfs_tree_unlock(next);
7292 free_extent_buffer(next);
7298 if (reada && level == 1)
7299 reada_walk_down(trans, root, wc, path);
7300 next = read_tree_block(root, bytenr, blocksize, generation);
7301 if (!next || !extent_buffer_uptodate(next)) {
7302 free_extent_buffer(next);
7305 btrfs_tree_lock(next);
7306 btrfs_set_lock_blocking(next);
7310 BUG_ON(level != btrfs_header_level(next));
7311 path->nodes[level] = next;
7312 path->slots[level] = 0;
7313 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7319 wc->refs[level - 1] = 0;
7320 wc->flags[level - 1] = 0;
7321 if (wc->stage == DROP_REFERENCE) {
7322 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7323 parent = path->nodes[level]->start;
7325 BUG_ON(root->root_key.objectid !=
7326 btrfs_header_owner(path->nodes[level]));
7330 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7331 root->root_key.objectid, level - 1, 0, 0);
7332 BUG_ON(ret); /* -ENOMEM */
7334 btrfs_tree_unlock(next);
7335 free_extent_buffer(next);
7341 * helper to process tree block while walking up the tree.
7343 * when wc->stage == DROP_REFERENCE, this function drops
7344 * reference count on the block.
7346 * when wc->stage == UPDATE_BACKREF, this function changes
7347 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7348 * to UPDATE_BACKREF previously while processing the block.
7350 * NOTE: return value 1 means we should stop walking up.
7352 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7353 struct btrfs_root *root,
7354 struct btrfs_path *path,
7355 struct walk_control *wc)
7358 int level = wc->level;
7359 struct extent_buffer *eb = path->nodes[level];
7362 if (wc->stage == UPDATE_BACKREF) {
7363 BUG_ON(wc->shared_level < level);
7364 if (level < wc->shared_level)
7367 ret = find_next_key(path, level + 1, &wc->update_progress);
7371 wc->stage = DROP_REFERENCE;
7372 wc->shared_level = -1;
7373 path->slots[level] = 0;
7376 * check reference count again if the block isn't locked.
7377 * we should start walking down the tree again if reference
7380 if (!path->locks[level]) {
7382 btrfs_tree_lock(eb);
7383 btrfs_set_lock_blocking(eb);
7384 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7386 ret = btrfs_lookup_extent_info(trans, root,
7387 eb->start, level, 1,
7391 btrfs_tree_unlock_rw(eb, path->locks[level]);
7392 path->locks[level] = 0;
7395 BUG_ON(wc->refs[level] == 0);
7396 if (wc->refs[level] == 1) {
7397 btrfs_tree_unlock_rw(eb, path->locks[level]);
7398 path->locks[level] = 0;
7404 /* wc->stage == DROP_REFERENCE */
7405 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7407 if (wc->refs[level] == 1) {
7409 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7410 ret = btrfs_dec_ref(trans, root, eb, 1,
7413 ret = btrfs_dec_ref(trans, root, eb, 0,
7415 BUG_ON(ret); /* -ENOMEM */
7417 /* make block locked assertion in clean_tree_block happy */
7418 if (!path->locks[level] &&
7419 btrfs_header_generation(eb) == trans->transid) {
7420 btrfs_tree_lock(eb);
7421 btrfs_set_lock_blocking(eb);
7422 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7424 clean_tree_block(trans, root, eb);
7427 if (eb == root->node) {
7428 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7431 BUG_ON(root->root_key.objectid !=
7432 btrfs_header_owner(eb));
7434 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7435 parent = path->nodes[level + 1]->start;
7437 BUG_ON(root->root_key.objectid !=
7438 btrfs_header_owner(path->nodes[level + 1]));
7441 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7443 wc->refs[level] = 0;
7444 wc->flags[level] = 0;
7448 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7449 struct btrfs_root *root,
7450 struct btrfs_path *path,
7451 struct walk_control *wc)
7453 int level = wc->level;
7454 int lookup_info = 1;
7457 while (level >= 0) {
7458 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7465 if (path->slots[level] >=
7466 btrfs_header_nritems(path->nodes[level]))
7469 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7471 path->slots[level]++;
7480 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7481 struct btrfs_root *root,
7482 struct btrfs_path *path,
7483 struct walk_control *wc, int max_level)
7485 int level = wc->level;
7488 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7489 while (level < max_level && path->nodes[level]) {
7491 if (path->slots[level] + 1 <
7492 btrfs_header_nritems(path->nodes[level])) {
7493 path->slots[level]++;
7496 ret = walk_up_proc(trans, root, path, wc);
7500 if (path->locks[level]) {
7501 btrfs_tree_unlock_rw(path->nodes[level],
7502 path->locks[level]);
7503 path->locks[level] = 0;
7505 free_extent_buffer(path->nodes[level]);
7506 path->nodes[level] = NULL;
7514 * drop a subvolume tree.
7516 * this function traverses the tree freeing any blocks that only
7517 * referenced by the tree.
7519 * when a shared tree block is found. this function decreases its
7520 * reference count by one. if update_ref is true, this function
7521 * also make sure backrefs for the shared block and all lower level
7522 * blocks are properly updated.
7524 * If called with for_reloc == 0, may exit early with -EAGAIN
7526 int btrfs_drop_snapshot(struct btrfs_root *root,
7527 struct btrfs_block_rsv *block_rsv, int update_ref,
7530 struct btrfs_path *path;
7531 struct btrfs_trans_handle *trans;
7532 struct btrfs_root *tree_root = root->fs_info->tree_root;
7533 struct btrfs_root_item *root_item = &root->root_item;
7534 struct walk_control *wc;
7535 struct btrfs_key key;
7539 bool root_dropped = false;
7541 path = btrfs_alloc_path();
7547 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7549 btrfs_free_path(path);
7554 trans = btrfs_start_transaction(tree_root, 0);
7555 if (IS_ERR(trans)) {
7556 err = PTR_ERR(trans);
7561 trans->block_rsv = block_rsv;
7563 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7564 level = btrfs_header_level(root->node);
7565 path->nodes[level] = btrfs_lock_root_node(root);
7566 btrfs_set_lock_blocking(path->nodes[level]);
7567 path->slots[level] = 0;
7568 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7569 memset(&wc->update_progress, 0,
7570 sizeof(wc->update_progress));
7572 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7573 memcpy(&wc->update_progress, &key,
7574 sizeof(wc->update_progress));
7576 level = root_item->drop_level;
7578 path->lowest_level = level;
7579 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7580 path->lowest_level = 0;
7588 * unlock our path, this is safe because only this
7589 * function is allowed to delete this snapshot
7591 btrfs_unlock_up_safe(path, 0);
7593 level = btrfs_header_level(root->node);
7595 btrfs_tree_lock(path->nodes[level]);
7596 btrfs_set_lock_blocking(path->nodes[level]);
7597 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7599 ret = btrfs_lookup_extent_info(trans, root,
7600 path->nodes[level]->start,
7601 level, 1, &wc->refs[level],
7607 BUG_ON(wc->refs[level] == 0);
7609 if (level == root_item->drop_level)
7612 btrfs_tree_unlock(path->nodes[level]);
7613 path->locks[level] = 0;
7614 WARN_ON(wc->refs[level] != 1);
7620 wc->shared_level = -1;
7621 wc->stage = DROP_REFERENCE;
7622 wc->update_ref = update_ref;
7624 wc->for_reloc = for_reloc;
7625 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7629 ret = walk_down_tree(trans, root, path, wc);
7635 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7642 BUG_ON(wc->stage != DROP_REFERENCE);
7646 if (wc->stage == DROP_REFERENCE) {
7648 btrfs_node_key(path->nodes[level],
7649 &root_item->drop_progress,
7650 path->slots[level]);
7651 root_item->drop_level = level;
7654 BUG_ON(wc->level == 0);
7655 if (btrfs_should_end_transaction(trans, tree_root) ||
7656 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7657 ret = btrfs_update_root(trans, tree_root,
7661 btrfs_abort_transaction(trans, tree_root, ret);
7666 btrfs_end_transaction_throttle(trans, tree_root);
7667 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7668 pr_debug("btrfs: drop snapshot early exit\n");
7673 trans = btrfs_start_transaction(tree_root, 0);
7674 if (IS_ERR(trans)) {
7675 err = PTR_ERR(trans);
7679 trans->block_rsv = block_rsv;
7682 btrfs_release_path(path);
7686 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7688 btrfs_abort_transaction(trans, tree_root, ret);
7692 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7693 ret = btrfs_find_root(tree_root, &root->root_key, path,
7696 btrfs_abort_transaction(trans, tree_root, ret);
7699 } else if (ret > 0) {
7700 /* if we fail to delete the orphan item this time
7701 * around, it'll get picked up the next time.
7703 * The most common failure here is just -ENOENT.
7705 btrfs_del_orphan_item(trans, tree_root,
7706 root->root_key.objectid);
7710 if (root->in_radix) {
7711 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7713 free_extent_buffer(root->node);
7714 free_extent_buffer(root->commit_root);
7715 btrfs_put_fs_root(root);
7717 root_dropped = true;
7719 btrfs_end_transaction_throttle(trans, tree_root);
7722 btrfs_free_path(path);
7725 * So if we need to stop dropping the snapshot for whatever reason we
7726 * need to make sure to add it back to the dead root list so that we
7727 * keep trying to do the work later. This also cleans up roots if we
7728 * don't have it in the radix (like when we recover after a power fail
7729 * or unmount) so we don't leak memory.
7731 if (!for_reloc && root_dropped == false)
7732 btrfs_add_dead_root(root);
7734 btrfs_std_error(root->fs_info, err);
7739 * drop subtree rooted at tree block 'node'.
7741 * NOTE: this function will unlock and release tree block 'node'
7742 * only used by relocation code
7744 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7745 struct btrfs_root *root,
7746 struct extent_buffer *node,
7747 struct extent_buffer *parent)
7749 struct btrfs_path *path;
7750 struct walk_control *wc;
7756 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7758 path = btrfs_alloc_path();
7762 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7764 btrfs_free_path(path);
7768 btrfs_assert_tree_locked(parent);
7769 parent_level = btrfs_header_level(parent);
7770 extent_buffer_get(parent);
7771 path->nodes[parent_level] = parent;
7772 path->slots[parent_level] = btrfs_header_nritems(parent);
7774 btrfs_assert_tree_locked(node);
7775 level = btrfs_header_level(node);
7776 path->nodes[level] = node;
7777 path->slots[level] = 0;
7778 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7780 wc->refs[parent_level] = 1;
7781 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7783 wc->shared_level = -1;
7784 wc->stage = DROP_REFERENCE;
7788 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7791 wret = walk_down_tree(trans, root, path, wc);
7797 wret = walk_up_tree(trans, root, path, wc, parent_level);
7805 btrfs_free_path(path);
7809 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7815 * if restripe for this chunk_type is on pick target profile and
7816 * return, otherwise do the usual balance
7818 stripped = get_restripe_target(root->fs_info, flags);
7820 return extended_to_chunk(stripped);
7823 * we add in the count of missing devices because we want
7824 * to make sure that any RAID levels on a degraded FS
7825 * continue to be honored.
7827 num_devices = root->fs_info->fs_devices->rw_devices +
7828 root->fs_info->fs_devices->missing_devices;
7830 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7831 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7832 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7834 if (num_devices == 1) {
7835 stripped |= BTRFS_BLOCK_GROUP_DUP;
7836 stripped = flags & ~stripped;
7838 /* turn raid0 into single device chunks */
7839 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7842 /* turn mirroring into duplication */
7843 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7844 BTRFS_BLOCK_GROUP_RAID10))
7845 return stripped | BTRFS_BLOCK_GROUP_DUP;
7847 /* they already had raid on here, just return */
7848 if (flags & stripped)
7851 stripped |= BTRFS_BLOCK_GROUP_DUP;
7852 stripped = flags & ~stripped;
7854 /* switch duplicated blocks with raid1 */
7855 if (flags & BTRFS_BLOCK_GROUP_DUP)
7856 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7858 /* this is drive concat, leave it alone */
7864 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7866 struct btrfs_space_info *sinfo = cache->space_info;
7868 u64 min_allocable_bytes;
7873 * We need some metadata space and system metadata space for
7874 * allocating chunks in some corner cases until we force to set
7875 * it to be readonly.
7878 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7880 min_allocable_bytes = 1 * 1024 * 1024;
7882 min_allocable_bytes = 0;
7884 spin_lock(&sinfo->lock);
7885 spin_lock(&cache->lock);
7892 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7893 cache->bytes_super - btrfs_block_group_used(&cache->item);
7895 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7896 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7897 min_allocable_bytes <= sinfo->total_bytes) {
7898 sinfo->bytes_readonly += num_bytes;
7903 spin_unlock(&cache->lock);
7904 spin_unlock(&sinfo->lock);
7908 int btrfs_set_block_group_ro(struct btrfs_root *root,
7909 struct btrfs_block_group_cache *cache)
7912 struct btrfs_trans_handle *trans;
7918 trans = btrfs_join_transaction(root);
7920 return PTR_ERR(trans);
7922 alloc_flags = update_block_group_flags(root, cache->flags);
7923 if (alloc_flags != cache->flags) {
7924 ret = do_chunk_alloc(trans, root, alloc_flags,
7930 ret = set_block_group_ro(cache, 0);
7933 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7934 ret = do_chunk_alloc(trans, root, alloc_flags,
7938 ret = set_block_group_ro(cache, 0);
7940 btrfs_end_transaction(trans, root);
7944 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7945 struct btrfs_root *root, u64 type)
7947 u64 alloc_flags = get_alloc_profile(root, type);
7948 return do_chunk_alloc(trans, root, alloc_flags,
7953 * helper to account the unused space of all the readonly block group in the
7954 * list. takes mirrors into account.
7956 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7958 struct btrfs_block_group_cache *block_group;
7962 list_for_each_entry(block_group, groups_list, list) {
7963 spin_lock(&block_group->lock);
7965 if (!block_group->ro) {
7966 spin_unlock(&block_group->lock);
7970 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7971 BTRFS_BLOCK_GROUP_RAID10 |
7972 BTRFS_BLOCK_GROUP_DUP))
7977 free_bytes += (block_group->key.offset -
7978 btrfs_block_group_used(&block_group->item)) *
7981 spin_unlock(&block_group->lock);
7988 * helper to account the unused space of all the readonly block group in the
7989 * space_info. takes mirrors into account.
7991 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7996 spin_lock(&sinfo->lock);
7998 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7999 if (!list_empty(&sinfo->block_groups[i]))
8000 free_bytes += __btrfs_get_ro_block_group_free_space(
8001 &sinfo->block_groups[i]);
8003 spin_unlock(&sinfo->lock);
8008 void btrfs_set_block_group_rw(struct btrfs_root *root,
8009 struct btrfs_block_group_cache *cache)
8011 struct btrfs_space_info *sinfo = cache->space_info;
8016 spin_lock(&sinfo->lock);
8017 spin_lock(&cache->lock);
8018 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8019 cache->bytes_super - btrfs_block_group_used(&cache->item);
8020 sinfo->bytes_readonly -= num_bytes;
8022 spin_unlock(&cache->lock);
8023 spin_unlock(&sinfo->lock);
8027 * checks to see if its even possible to relocate this block group.
8029 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8030 * ok to go ahead and try.
8032 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8034 struct btrfs_block_group_cache *block_group;
8035 struct btrfs_space_info *space_info;
8036 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8037 struct btrfs_device *device;
8038 struct btrfs_trans_handle *trans;
8047 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8049 /* odd, couldn't find the block group, leave it alone */
8053 min_free = btrfs_block_group_used(&block_group->item);
8055 /* no bytes used, we're good */
8059 space_info = block_group->space_info;
8060 spin_lock(&space_info->lock);
8062 full = space_info->full;
8065 * if this is the last block group we have in this space, we can't
8066 * relocate it unless we're able to allocate a new chunk below.
8068 * Otherwise, we need to make sure we have room in the space to handle
8069 * all of the extents from this block group. If we can, we're good
8071 if ((space_info->total_bytes != block_group->key.offset) &&
8072 (space_info->bytes_used + space_info->bytes_reserved +
8073 space_info->bytes_pinned + space_info->bytes_readonly +
8074 min_free < space_info->total_bytes)) {
8075 spin_unlock(&space_info->lock);
8078 spin_unlock(&space_info->lock);
8081 * ok we don't have enough space, but maybe we have free space on our
8082 * devices to allocate new chunks for relocation, so loop through our
8083 * alloc devices and guess if we have enough space. if this block
8084 * group is going to be restriped, run checks against the target
8085 * profile instead of the current one.
8097 target = get_restripe_target(root->fs_info, block_group->flags);
8099 index = __get_raid_index(extended_to_chunk(target));
8102 * this is just a balance, so if we were marked as full
8103 * we know there is no space for a new chunk
8108 index = get_block_group_index(block_group);
8111 if (index == BTRFS_RAID_RAID10) {
8115 } else if (index == BTRFS_RAID_RAID1) {
8117 } else if (index == BTRFS_RAID_DUP) {
8120 } else if (index == BTRFS_RAID_RAID0) {
8121 dev_min = fs_devices->rw_devices;
8122 do_div(min_free, dev_min);
8125 /* We need to do this so that we can look at pending chunks */
8126 trans = btrfs_join_transaction(root);
8127 if (IS_ERR(trans)) {
8128 ret = PTR_ERR(trans);
8132 mutex_lock(&root->fs_info->chunk_mutex);
8133 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8137 * check to make sure we can actually find a chunk with enough
8138 * space to fit our block group in.
8140 if (device->total_bytes > device->bytes_used + min_free &&
8141 !device->is_tgtdev_for_dev_replace) {
8142 ret = find_free_dev_extent(trans, device, min_free,
8147 if (dev_nr >= dev_min)
8153 mutex_unlock(&root->fs_info->chunk_mutex);
8154 btrfs_end_transaction(trans, root);
8156 btrfs_put_block_group(block_group);
8160 static int find_first_block_group(struct btrfs_root *root,
8161 struct btrfs_path *path, struct btrfs_key *key)
8164 struct btrfs_key found_key;
8165 struct extent_buffer *leaf;
8168 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8173 slot = path->slots[0];
8174 leaf = path->nodes[0];
8175 if (slot >= btrfs_header_nritems(leaf)) {
8176 ret = btrfs_next_leaf(root, path);
8183 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8185 if (found_key.objectid >= key->objectid &&
8186 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8196 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8198 struct btrfs_block_group_cache *block_group;
8202 struct inode *inode;
8204 block_group = btrfs_lookup_first_block_group(info, last);
8205 while (block_group) {
8206 spin_lock(&block_group->lock);
8207 if (block_group->iref)
8209 spin_unlock(&block_group->lock);
8210 block_group = next_block_group(info->tree_root,
8220 inode = block_group->inode;
8221 block_group->iref = 0;
8222 block_group->inode = NULL;
8223 spin_unlock(&block_group->lock);
8225 last = block_group->key.objectid + block_group->key.offset;
8226 btrfs_put_block_group(block_group);
8230 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8232 struct btrfs_block_group_cache *block_group;
8233 struct btrfs_space_info *space_info;
8234 struct btrfs_caching_control *caching_ctl;
8237 down_write(&info->extent_commit_sem);
8238 while (!list_empty(&info->caching_block_groups)) {
8239 caching_ctl = list_entry(info->caching_block_groups.next,
8240 struct btrfs_caching_control, list);
8241 list_del(&caching_ctl->list);
8242 put_caching_control(caching_ctl);
8244 up_write(&info->extent_commit_sem);
8246 spin_lock(&info->block_group_cache_lock);
8247 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8248 block_group = rb_entry(n, struct btrfs_block_group_cache,
8250 rb_erase(&block_group->cache_node,
8251 &info->block_group_cache_tree);
8252 spin_unlock(&info->block_group_cache_lock);
8254 down_write(&block_group->space_info->groups_sem);
8255 list_del(&block_group->list);
8256 up_write(&block_group->space_info->groups_sem);
8258 if (block_group->cached == BTRFS_CACHE_STARTED)
8259 wait_block_group_cache_done(block_group);
8262 * We haven't cached this block group, which means we could
8263 * possibly have excluded extents on this block group.
8265 if (block_group->cached == BTRFS_CACHE_NO ||
8266 block_group->cached == BTRFS_CACHE_ERROR)
8267 free_excluded_extents(info->extent_root, block_group);
8269 btrfs_remove_free_space_cache(block_group);
8270 btrfs_put_block_group(block_group);
8272 spin_lock(&info->block_group_cache_lock);
8274 spin_unlock(&info->block_group_cache_lock);
8276 /* now that all the block groups are freed, go through and
8277 * free all the space_info structs. This is only called during
8278 * the final stages of unmount, and so we know nobody is
8279 * using them. We call synchronize_rcu() once before we start,
8280 * just to be on the safe side.
8284 release_global_block_rsv(info);
8286 while(!list_empty(&info->space_info)) {
8287 space_info = list_entry(info->space_info.next,
8288 struct btrfs_space_info,
8290 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8291 if (space_info->bytes_pinned > 0 ||
8292 space_info->bytes_reserved > 0 ||
8293 space_info->bytes_may_use > 0) {
8295 dump_space_info(space_info, 0, 0);
8298 percpu_counter_destroy(&space_info->total_bytes_pinned);
8299 list_del(&space_info->list);
8305 static void __link_block_group(struct btrfs_space_info *space_info,
8306 struct btrfs_block_group_cache *cache)
8308 int index = get_block_group_index(cache);
8310 down_write(&space_info->groups_sem);
8311 list_add_tail(&cache->list, &space_info->block_groups[index]);
8312 up_write(&space_info->groups_sem);
8315 int btrfs_read_block_groups(struct btrfs_root *root)
8317 struct btrfs_path *path;
8319 struct btrfs_block_group_cache *cache;
8320 struct btrfs_fs_info *info = root->fs_info;
8321 struct btrfs_space_info *space_info;
8322 struct btrfs_key key;
8323 struct btrfs_key found_key;
8324 struct extent_buffer *leaf;
8328 root = info->extent_root;
8331 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8332 path = btrfs_alloc_path();
8337 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8338 if (btrfs_test_opt(root, SPACE_CACHE) &&
8339 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8341 if (btrfs_test_opt(root, CLEAR_CACHE))
8345 ret = find_first_block_group(root, path, &key);
8350 leaf = path->nodes[0];
8351 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8352 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8357 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8359 if (!cache->free_space_ctl) {
8365 atomic_set(&cache->count, 1);
8366 spin_lock_init(&cache->lock);
8367 cache->fs_info = info;
8368 INIT_LIST_HEAD(&cache->list);
8369 INIT_LIST_HEAD(&cache->cluster_list);
8373 * When we mount with old space cache, we need to
8374 * set BTRFS_DC_CLEAR and set dirty flag.
8376 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8377 * truncate the old free space cache inode and
8379 * b) Setting 'dirty flag' makes sure that we flush
8380 * the new space cache info onto disk.
8382 cache->disk_cache_state = BTRFS_DC_CLEAR;
8383 if (btrfs_test_opt(root, SPACE_CACHE))
8387 read_extent_buffer(leaf, &cache->item,
8388 btrfs_item_ptr_offset(leaf, path->slots[0]),
8389 sizeof(cache->item));
8390 memcpy(&cache->key, &found_key, sizeof(found_key));
8392 key.objectid = found_key.objectid + found_key.offset;
8393 btrfs_release_path(path);
8394 cache->flags = btrfs_block_group_flags(&cache->item);
8395 cache->sectorsize = root->sectorsize;
8396 cache->full_stripe_len = btrfs_full_stripe_len(root,
8397 &root->fs_info->mapping_tree,
8398 found_key.objectid);
8399 btrfs_init_free_space_ctl(cache);
8402 * We need to exclude the super stripes now so that the space
8403 * info has super bytes accounted for, otherwise we'll think
8404 * we have more space than we actually do.
8406 ret = exclude_super_stripes(root, cache);
8409 * We may have excluded something, so call this just in
8412 free_excluded_extents(root, cache);
8413 kfree(cache->free_space_ctl);
8419 * check for two cases, either we are full, and therefore
8420 * don't need to bother with the caching work since we won't
8421 * find any space, or we are empty, and we can just add all
8422 * the space in and be done with it. This saves us _alot_ of
8423 * time, particularly in the full case.
8425 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8426 cache->last_byte_to_unpin = (u64)-1;
8427 cache->cached = BTRFS_CACHE_FINISHED;
8428 free_excluded_extents(root, cache);
8429 } else if (btrfs_block_group_used(&cache->item) == 0) {
8430 cache->last_byte_to_unpin = (u64)-1;
8431 cache->cached = BTRFS_CACHE_FINISHED;
8432 add_new_free_space(cache, root->fs_info,
8434 found_key.objectid +
8436 free_excluded_extents(root, cache);
8439 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8441 btrfs_remove_free_space_cache(cache);
8442 btrfs_put_block_group(cache);
8446 ret = update_space_info(info, cache->flags, found_key.offset,
8447 btrfs_block_group_used(&cache->item),
8450 btrfs_remove_free_space_cache(cache);
8451 spin_lock(&info->block_group_cache_lock);
8452 rb_erase(&cache->cache_node,
8453 &info->block_group_cache_tree);
8454 spin_unlock(&info->block_group_cache_lock);
8455 btrfs_put_block_group(cache);
8459 cache->space_info = space_info;
8460 spin_lock(&cache->space_info->lock);
8461 cache->space_info->bytes_readonly += cache->bytes_super;
8462 spin_unlock(&cache->space_info->lock);
8464 __link_block_group(space_info, cache);
8466 set_avail_alloc_bits(root->fs_info, cache->flags);
8467 if (btrfs_chunk_readonly(root, cache->key.objectid))
8468 set_block_group_ro(cache, 1);
8471 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8472 if (!(get_alloc_profile(root, space_info->flags) &
8473 (BTRFS_BLOCK_GROUP_RAID10 |
8474 BTRFS_BLOCK_GROUP_RAID1 |
8475 BTRFS_BLOCK_GROUP_RAID5 |
8476 BTRFS_BLOCK_GROUP_RAID6 |
8477 BTRFS_BLOCK_GROUP_DUP)))
8480 * avoid allocating from un-mirrored block group if there are
8481 * mirrored block groups.
8483 list_for_each_entry(cache,
8484 &space_info->block_groups[BTRFS_RAID_RAID0],
8486 set_block_group_ro(cache, 1);
8487 list_for_each_entry(cache,
8488 &space_info->block_groups[BTRFS_RAID_SINGLE],
8490 set_block_group_ro(cache, 1);
8493 init_global_block_rsv(info);
8496 btrfs_free_path(path);
8500 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8501 struct btrfs_root *root)
8503 struct btrfs_block_group_cache *block_group, *tmp;
8504 struct btrfs_root *extent_root = root->fs_info->extent_root;
8505 struct btrfs_block_group_item item;
8506 struct btrfs_key key;
8509 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8511 list_del_init(&block_group->new_bg_list);
8516 spin_lock(&block_group->lock);
8517 memcpy(&item, &block_group->item, sizeof(item));
8518 memcpy(&key, &block_group->key, sizeof(key));
8519 spin_unlock(&block_group->lock);
8521 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8524 btrfs_abort_transaction(trans, extent_root, ret);
8525 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8526 key.objectid, key.offset);
8528 btrfs_abort_transaction(trans, extent_root, ret);
8532 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8533 struct btrfs_root *root, u64 bytes_used,
8534 u64 type, u64 chunk_objectid, u64 chunk_offset,
8538 struct btrfs_root *extent_root;
8539 struct btrfs_block_group_cache *cache;
8541 extent_root = root->fs_info->extent_root;
8543 root->fs_info->last_trans_log_full_commit = trans->transid;
8545 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8548 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8550 if (!cache->free_space_ctl) {
8555 cache->key.objectid = chunk_offset;
8556 cache->key.offset = size;
8557 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8558 cache->sectorsize = root->sectorsize;
8559 cache->fs_info = root->fs_info;
8560 cache->full_stripe_len = btrfs_full_stripe_len(root,
8561 &root->fs_info->mapping_tree,
8564 atomic_set(&cache->count, 1);
8565 spin_lock_init(&cache->lock);
8566 INIT_LIST_HEAD(&cache->list);
8567 INIT_LIST_HEAD(&cache->cluster_list);
8568 INIT_LIST_HEAD(&cache->new_bg_list);
8570 btrfs_init_free_space_ctl(cache);
8572 btrfs_set_block_group_used(&cache->item, bytes_used);
8573 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8574 cache->flags = type;
8575 btrfs_set_block_group_flags(&cache->item, type);
8577 cache->last_byte_to_unpin = (u64)-1;
8578 cache->cached = BTRFS_CACHE_FINISHED;
8579 ret = exclude_super_stripes(root, cache);
8582 * We may have excluded something, so call this just in
8585 free_excluded_extents(root, cache);
8586 kfree(cache->free_space_ctl);
8591 add_new_free_space(cache, root->fs_info, chunk_offset,
8592 chunk_offset + size);
8594 free_excluded_extents(root, cache);
8596 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8598 btrfs_remove_free_space_cache(cache);
8599 btrfs_put_block_group(cache);
8603 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8604 &cache->space_info);
8606 btrfs_remove_free_space_cache(cache);
8607 spin_lock(&root->fs_info->block_group_cache_lock);
8608 rb_erase(&cache->cache_node,
8609 &root->fs_info->block_group_cache_tree);
8610 spin_unlock(&root->fs_info->block_group_cache_lock);
8611 btrfs_put_block_group(cache);
8614 update_global_block_rsv(root->fs_info);
8616 spin_lock(&cache->space_info->lock);
8617 cache->space_info->bytes_readonly += cache->bytes_super;
8618 spin_unlock(&cache->space_info->lock);
8620 __link_block_group(cache->space_info, cache);
8622 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8624 set_avail_alloc_bits(extent_root->fs_info, type);
8629 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8631 u64 extra_flags = chunk_to_extended(flags) &
8632 BTRFS_EXTENDED_PROFILE_MASK;
8634 write_seqlock(&fs_info->profiles_lock);
8635 if (flags & BTRFS_BLOCK_GROUP_DATA)
8636 fs_info->avail_data_alloc_bits &= ~extra_flags;
8637 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8638 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8639 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8640 fs_info->avail_system_alloc_bits &= ~extra_flags;
8641 write_sequnlock(&fs_info->profiles_lock);
8644 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8645 struct btrfs_root *root, u64 group_start)
8647 struct btrfs_path *path;
8648 struct btrfs_block_group_cache *block_group;
8649 struct btrfs_free_cluster *cluster;
8650 struct btrfs_root *tree_root = root->fs_info->tree_root;
8651 struct btrfs_key key;
8652 struct inode *inode;
8657 root = root->fs_info->extent_root;
8659 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8660 BUG_ON(!block_group);
8661 BUG_ON(!block_group->ro);
8664 * Free the reserved super bytes from this block group before
8667 free_excluded_extents(root, block_group);
8669 memcpy(&key, &block_group->key, sizeof(key));
8670 index = get_block_group_index(block_group);
8671 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8672 BTRFS_BLOCK_GROUP_RAID1 |
8673 BTRFS_BLOCK_GROUP_RAID10))
8678 /* make sure this block group isn't part of an allocation cluster */
8679 cluster = &root->fs_info->data_alloc_cluster;
8680 spin_lock(&cluster->refill_lock);
8681 btrfs_return_cluster_to_free_space(block_group, cluster);
8682 spin_unlock(&cluster->refill_lock);
8685 * make sure this block group isn't part of a metadata
8686 * allocation cluster
8688 cluster = &root->fs_info->meta_alloc_cluster;
8689 spin_lock(&cluster->refill_lock);
8690 btrfs_return_cluster_to_free_space(block_group, cluster);
8691 spin_unlock(&cluster->refill_lock);
8693 path = btrfs_alloc_path();
8699 inode = lookup_free_space_inode(tree_root, block_group, path);
8700 if (!IS_ERR(inode)) {
8701 ret = btrfs_orphan_add(trans, inode);
8703 btrfs_add_delayed_iput(inode);
8707 /* One for the block groups ref */
8708 spin_lock(&block_group->lock);
8709 if (block_group->iref) {
8710 block_group->iref = 0;
8711 block_group->inode = NULL;
8712 spin_unlock(&block_group->lock);
8715 spin_unlock(&block_group->lock);
8717 /* One for our lookup ref */
8718 btrfs_add_delayed_iput(inode);
8721 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8722 key.offset = block_group->key.objectid;
8725 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8729 btrfs_release_path(path);
8731 ret = btrfs_del_item(trans, tree_root, path);
8734 btrfs_release_path(path);
8737 spin_lock(&root->fs_info->block_group_cache_lock);
8738 rb_erase(&block_group->cache_node,
8739 &root->fs_info->block_group_cache_tree);
8741 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8742 root->fs_info->first_logical_byte = (u64)-1;
8743 spin_unlock(&root->fs_info->block_group_cache_lock);
8745 down_write(&block_group->space_info->groups_sem);
8747 * we must use list_del_init so people can check to see if they
8748 * are still on the list after taking the semaphore
8750 list_del_init(&block_group->list);
8751 if (list_empty(&block_group->space_info->block_groups[index]))
8752 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8753 up_write(&block_group->space_info->groups_sem);
8755 if (block_group->cached == BTRFS_CACHE_STARTED)
8756 wait_block_group_cache_done(block_group);
8758 btrfs_remove_free_space_cache(block_group);
8760 spin_lock(&block_group->space_info->lock);
8761 block_group->space_info->total_bytes -= block_group->key.offset;
8762 block_group->space_info->bytes_readonly -= block_group->key.offset;
8763 block_group->space_info->disk_total -= block_group->key.offset * factor;
8764 spin_unlock(&block_group->space_info->lock);
8766 memcpy(&key, &block_group->key, sizeof(key));
8768 btrfs_clear_space_info_full(root->fs_info);
8770 btrfs_put_block_group(block_group);
8771 btrfs_put_block_group(block_group);
8773 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8779 ret = btrfs_del_item(trans, root, path);
8781 btrfs_free_path(path);
8785 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8787 struct btrfs_space_info *space_info;
8788 struct btrfs_super_block *disk_super;
8794 disk_super = fs_info->super_copy;
8795 if (!btrfs_super_root(disk_super))
8798 features = btrfs_super_incompat_flags(disk_super);
8799 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8802 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8803 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8808 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8809 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8811 flags = BTRFS_BLOCK_GROUP_METADATA;
8812 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8816 flags = BTRFS_BLOCK_GROUP_DATA;
8817 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8823 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8825 return unpin_extent_range(root, start, end);
8828 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8829 u64 num_bytes, u64 *actual_bytes)
8831 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8834 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8836 struct btrfs_fs_info *fs_info = root->fs_info;
8837 struct btrfs_block_group_cache *cache = NULL;
8842 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8846 * try to trim all FS space, our block group may start from non-zero.
8848 if (range->len == total_bytes)
8849 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8851 cache = btrfs_lookup_block_group(fs_info, range->start);
8854 if (cache->key.objectid >= (range->start + range->len)) {
8855 btrfs_put_block_group(cache);
8859 start = max(range->start, cache->key.objectid);
8860 end = min(range->start + range->len,
8861 cache->key.objectid + cache->key.offset);
8863 if (end - start >= range->minlen) {
8864 if (!block_group_cache_done(cache)) {
8865 ret = cache_block_group(cache, 0);
8867 btrfs_put_block_group(cache);
8870 ret = wait_block_group_cache_done(cache);
8872 btrfs_put_block_group(cache);
8876 ret = btrfs_trim_block_group(cache,
8882 trimmed += group_trimmed;
8884 btrfs_put_block_group(cache);
8889 cache = next_block_group(fs_info->tree_root, cache);
8892 range->len = trimmed;