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 spin_unlock(&delayed_refs->lock);
2446 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2447 must_insert_reserved);
2449 btrfs_free_delayed_extent_op(extent_op);
2451 btrfs_delayed_ref_unlock(locked_ref);
2452 btrfs_put_delayed_ref(ref);
2453 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2454 spin_lock(&delayed_refs->lock);
2459 * If this node is a head, that means all the refs in this head
2460 * have been dealt with, and we will pick the next head to deal
2461 * with, so we must unlock the head and drop it from the cluster
2462 * list before we release it.
2464 if (btrfs_delayed_ref_is_head(ref)) {
2465 list_del_init(&locked_ref->cluster);
2466 btrfs_delayed_ref_unlock(locked_ref);
2469 btrfs_put_delayed_ref(ref);
2473 spin_lock(&delayed_refs->lock);
2478 #ifdef SCRAMBLE_DELAYED_REFS
2480 * Normally delayed refs get processed in ascending bytenr order. This
2481 * correlates in most cases to the order added. To expose dependencies on this
2482 * order, we start to process the tree in the middle instead of the beginning
2484 static u64 find_middle(struct rb_root *root)
2486 struct rb_node *n = root->rb_node;
2487 struct btrfs_delayed_ref_node *entry;
2490 u64 first = 0, last = 0;
2494 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2495 first = entry->bytenr;
2499 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2500 last = entry->bytenr;
2505 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2506 WARN_ON(!entry->in_tree);
2508 middle = entry->bytenr;
2521 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2522 struct btrfs_fs_info *fs_info)
2524 struct qgroup_update *qgroup_update;
2527 if (list_empty(&trans->qgroup_ref_list) !=
2528 !trans->delayed_ref_elem.seq) {
2529 /* list without seq or seq without list */
2531 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2532 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2533 (u32)(trans->delayed_ref_elem.seq >> 32),
2534 (u32)trans->delayed_ref_elem.seq);
2538 if (!trans->delayed_ref_elem.seq)
2541 while (!list_empty(&trans->qgroup_ref_list)) {
2542 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2543 struct qgroup_update, list);
2544 list_del(&qgroup_update->list);
2546 ret = btrfs_qgroup_account_ref(
2547 trans, fs_info, qgroup_update->node,
2548 qgroup_update->extent_op);
2549 kfree(qgroup_update);
2552 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2557 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2560 int val = atomic_read(&delayed_refs->ref_seq);
2562 if (val < seq || val >= seq + count)
2567 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2571 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2572 sizeof(struct btrfs_extent_inline_ref));
2573 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2574 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2577 * We don't ever fill up leaves all the way so multiply by 2 just to be
2578 * closer to what we're really going to want to ouse.
2580 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2583 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2584 struct btrfs_root *root)
2586 struct btrfs_block_rsv *global_rsv;
2587 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2591 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2592 num_heads = heads_to_leaves(root, num_heads);
2594 num_bytes += (num_heads - 1) * root->leafsize;
2596 global_rsv = &root->fs_info->global_block_rsv;
2599 * If we can't allocate any more chunks lets make sure we have _lots_ of
2600 * wiggle room since running delayed refs can create more delayed refs.
2602 if (global_rsv->space_info->full)
2605 spin_lock(&global_rsv->lock);
2606 if (global_rsv->reserved <= num_bytes)
2608 spin_unlock(&global_rsv->lock);
2613 * this starts processing the delayed reference count updates and
2614 * extent insertions we have queued up so far. count can be
2615 * 0, which means to process everything in the tree at the start
2616 * of the run (but not newly added entries), or it can be some target
2617 * number you'd like to process.
2619 * Returns 0 on success or if called with an aborted transaction
2620 * Returns <0 on error and aborts the transaction
2622 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2623 struct btrfs_root *root, unsigned long count)
2625 struct rb_node *node;
2626 struct btrfs_delayed_ref_root *delayed_refs;
2627 struct btrfs_delayed_ref_node *ref;
2628 struct list_head cluster;
2631 int run_all = count == (unsigned long)-1;
2635 /* We'll clean this up in btrfs_cleanup_transaction */
2639 if (root == root->fs_info->extent_root)
2640 root = root->fs_info->tree_root;
2642 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2644 delayed_refs = &trans->transaction->delayed_refs;
2645 INIT_LIST_HEAD(&cluster);
2647 count = delayed_refs->num_entries * 2;
2651 if (!run_all && !run_most) {
2653 int seq = atomic_read(&delayed_refs->ref_seq);
2656 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2658 DEFINE_WAIT(__wait);
2659 if (delayed_refs->flushing ||
2660 !btrfs_should_throttle_delayed_refs(trans, root))
2663 prepare_to_wait(&delayed_refs->wait, &__wait,
2664 TASK_UNINTERRUPTIBLE);
2666 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2669 finish_wait(&delayed_refs->wait, &__wait);
2671 if (!refs_newer(delayed_refs, seq, 256))
2676 finish_wait(&delayed_refs->wait, &__wait);
2682 atomic_inc(&delayed_refs->procs_running_refs);
2687 spin_lock(&delayed_refs->lock);
2689 #ifdef SCRAMBLE_DELAYED_REFS
2690 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2694 if (!(run_all || run_most) &&
2695 !btrfs_should_throttle_delayed_refs(trans, root))
2699 * go find something we can process in the rbtree. We start at
2700 * the beginning of the tree, and then build a cluster
2701 * of refs to process starting at the first one we are able to
2704 delayed_start = delayed_refs->run_delayed_start;
2705 ret = btrfs_find_ref_cluster(trans, &cluster,
2706 delayed_refs->run_delayed_start);
2710 ret = run_clustered_refs(trans, root, &cluster);
2712 btrfs_release_ref_cluster(&cluster);
2713 spin_unlock(&delayed_refs->lock);
2714 btrfs_abort_transaction(trans, root, ret);
2715 atomic_dec(&delayed_refs->procs_running_refs);
2716 wake_up(&delayed_refs->wait);
2720 atomic_add(ret, &delayed_refs->ref_seq);
2722 count -= min_t(unsigned long, ret, count);
2727 if (delayed_start >= delayed_refs->run_delayed_start) {
2730 * btrfs_find_ref_cluster looped. let's do one
2731 * more cycle. if we don't run any delayed ref
2732 * during that cycle (because we can't because
2733 * all of them are blocked), bail out.
2738 * no runnable refs left, stop trying
2745 /* refs were run, let's reset staleness detection */
2751 if (!list_empty(&trans->new_bgs)) {
2752 spin_unlock(&delayed_refs->lock);
2753 btrfs_create_pending_block_groups(trans, root);
2754 spin_lock(&delayed_refs->lock);
2757 node = rb_first(&delayed_refs->root);
2760 count = (unsigned long)-1;
2763 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2765 if (btrfs_delayed_ref_is_head(ref)) {
2766 struct btrfs_delayed_ref_head *head;
2768 head = btrfs_delayed_node_to_head(ref);
2769 atomic_inc(&ref->refs);
2771 spin_unlock(&delayed_refs->lock);
2773 * Mutex was contended, block until it's
2774 * released and try again
2776 mutex_lock(&head->mutex);
2777 mutex_unlock(&head->mutex);
2779 btrfs_put_delayed_ref(ref);
2783 node = rb_next(node);
2785 spin_unlock(&delayed_refs->lock);
2786 schedule_timeout(1);
2790 atomic_dec(&delayed_refs->procs_running_refs);
2792 if (waitqueue_active(&delayed_refs->wait))
2793 wake_up(&delayed_refs->wait);
2795 spin_unlock(&delayed_refs->lock);
2796 assert_qgroups_uptodate(trans);
2800 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2801 struct btrfs_root *root,
2802 u64 bytenr, u64 num_bytes, u64 flags,
2803 int level, int is_data)
2805 struct btrfs_delayed_extent_op *extent_op;
2808 extent_op = btrfs_alloc_delayed_extent_op();
2812 extent_op->flags_to_set = flags;
2813 extent_op->update_flags = 1;
2814 extent_op->update_key = 0;
2815 extent_op->is_data = is_data ? 1 : 0;
2816 extent_op->level = level;
2818 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2819 num_bytes, extent_op);
2821 btrfs_free_delayed_extent_op(extent_op);
2825 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2826 struct btrfs_root *root,
2827 struct btrfs_path *path,
2828 u64 objectid, u64 offset, u64 bytenr)
2830 struct btrfs_delayed_ref_head *head;
2831 struct btrfs_delayed_ref_node *ref;
2832 struct btrfs_delayed_data_ref *data_ref;
2833 struct btrfs_delayed_ref_root *delayed_refs;
2834 struct rb_node *node;
2838 delayed_refs = &trans->transaction->delayed_refs;
2839 spin_lock(&delayed_refs->lock);
2840 head = btrfs_find_delayed_ref_head(trans, bytenr);
2844 if (!mutex_trylock(&head->mutex)) {
2845 atomic_inc(&head->node.refs);
2846 spin_unlock(&delayed_refs->lock);
2848 btrfs_release_path(path);
2851 * Mutex was contended, block until it's released and let
2854 mutex_lock(&head->mutex);
2855 mutex_unlock(&head->mutex);
2856 btrfs_put_delayed_ref(&head->node);
2860 node = rb_prev(&head->node.rb_node);
2864 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2866 if (ref->bytenr != bytenr)
2870 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2873 data_ref = btrfs_delayed_node_to_data_ref(ref);
2875 node = rb_prev(node);
2879 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2880 if (ref->bytenr == bytenr && ref->seq == seq)
2884 if (data_ref->root != root->root_key.objectid ||
2885 data_ref->objectid != objectid || data_ref->offset != offset)
2890 mutex_unlock(&head->mutex);
2892 spin_unlock(&delayed_refs->lock);
2896 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2897 struct btrfs_root *root,
2898 struct btrfs_path *path,
2899 u64 objectid, u64 offset, u64 bytenr)
2901 struct btrfs_root *extent_root = root->fs_info->extent_root;
2902 struct extent_buffer *leaf;
2903 struct btrfs_extent_data_ref *ref;
2904 struct btrfs_extent_inline_ref *iref;
2905 struct btrfs_extent_item *ei;
2906 struct btrfs_key key;
2910 key.objectid = bytenr;
2911 key.offset = (u64)-1;
2912 key.type = BTRFS_EXTENT_ITEM_KEY;
2914 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2917 BUG_ON(ret == 0); /* Corruption */
2920 if (path->slots[0] == 0)
2924 leaf = path->nodes[0];
2925 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2927 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2931 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2932 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2933 if (item_size < sizeof(*ei)) {
2934 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2938 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2940 if (item_size != sizeof(*ei) +
2941 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2944 if (btrfs_extent_generation(leaf, ei) <=
2945 btrfs_root_last_snapshot(&root->root_item))
2948 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2949 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2950 BTRFS_EXTENT_DATA_REF_KEY)
2953 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2954 if (btrfs_extent_refs(leaf, ei) !=
2955 btrfs_extent_data_ref_count(leaf, ref) ||
2956 btrfs_extent_data_ref_root(leaf, ref) !=
2957 root->root_key.objectid ||
2958 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2959 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2967 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2968 struct btrfs_root *root,
2969 u64 objectid, u64 offset, u64 bytenr)
2971 struct btrfs_path *path;
2975 path = btrfs_alloc_path();
2980 ret = check_committed_ref(trans, root, path, objectid,
2982 if (ret && ret != -ENOENT)
2985 ret2 = check_delayed_ref(trans, root, path, objectid,
2987 } while (ret2 == -EAGAIN);
2989 if (ret2 && ret2 != -ENOENT) {
2994 if (ret != -ENOENT || ret2 != -ENOENT)
2997 btrfs_free_path(path);
2998 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3003 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3004 struct btrfs_root *root,
3005 struct extent_buffer *buf,
3006 int full_backref, int inc, int for_cow)
3013 struct btrfs_key key;
3014 struct btrfs_file_extent_item *fi;
3018 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3019 u64, u64, u64, u64, u64, u64, int);
3021 ref_root = btrfs_header_owner(buf);
3022 nritems = btrfs_header_nritems(buf);
3023 level = btrfs_header_level(buf);
3025 if (!root->ref_cows && level == 0)
3029 process_func = btrfs_inc_extent_ref;
3031 process_func = btrfs_free_extent;
3034 parent = buf->start;
3038 for (i = 0; i < nritems; i++) {
3040 btrfs_item_key_to_cpu(buf, &key, i);
3041 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3043 fi = btrfs_item_ptr(buf, i,
3044 struct btrfs_file_extent_item);
3045 if (btrfs_file_extent_type(buf, fi) ==
3046 BTRFS_FILE_EXTENT_INLINE)
3048 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3052 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3053 key.offset -= btrfs_file_extent_offset(buf, fi);
3054 ret = process_func(trans, root, bytenr, num_bytes,
3055 parent, ref_root, key.objectid,
3056 key.offset, for_cow);
3060 bytenr = btrfs_node_blockptr(buf, i);
3061 num_bytes = btrfs_level_size(root, level - 1);
3062 ret = process_func(trans, root, bytenr, num_bytes,
3063 parent, ref_root, level - 1, 0,
3074 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3075 struct extent_buffer *buf, int full_backref, int for_cow)
3077 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3080 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3081 struct extent_buffer *buf, int full_backref, int for_cow)
3083 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3086 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3087 struct btrfs_root *root,
3088 struct btrfs_path *path,
3089 struct btrfs_block_group_cache *cache)
3092 struct btrfs_root *extent_root = root->fs_info->extent_root;
3094 struct extent_buffer *leaf;
3096 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3099 BUG_ON(ret); /* Corruption */
3101 leaf = path->nodes[0];
3102 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3103 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3104 btrfs_mark_buffer_dirty(leaf);
3105 btrfs_release_path(path);
3108 btrfs_abort_transaction(trans, root, ret);
3115 static struct btrfs_block_group_cache *
3116 next_block_group(struct btrfs_root *root,
3117 struct btrfs_block_group_cache *cache)
3119 struct rb_node *node;
3120 spin_lock(&root->fs_info->block_group_cache_lock);
3121 node = rb_next(&cache->cache_node);
3122 btrfs_put_block_group(cache);
3124 cache = rb_entry(node, struct btrfs_block_group_cache,
3126 btrfs_get_block_group(cache);
3129 spin_unlock(&root->fs_info->block_group_cache_lock);
3133 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3134 struct btrfs_trans_handle *trans,
3135 struct btrfs_path *path)
3137 struct btrfs_root *root = block_group->fs_info->tree_root;
3138 struct inode *inode = NULL;
3140 int dcs = BTRFS_DC_ERROR;
3146 * If this block group is smaller than 100 megs don't bother caching the
3149 if (block_group->key.offset < (100 * 1024 * 1024)) {
3150 spin_lock(&block_group->lock);
3151 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3152 spin_unlock(&block_group->lock);
3157 inode = lookup_free_space_inode(root, block_group, path);
3158 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3159 ret = PTR_ERR(inode);
3160 btrfs_release_path(path);
3164 if (IS_ERR(inode)) {
3168 if (block_group->ro)
3171 ret = create_free_space_inode(root, trans, block_group, path);
3177 /* We've already setup this transaction, go ahead and exit */
3178 if (block_group->cache_generation == trans->transid &&
3179 i_size_read(inode)) {
3180 dcs = BTRFS_DC_SETUP;
3185 * We want to set the generation to 0, that way if anything goes wrong
3186 * from here on out we know not to trust this cache when we load up next
3189 BTRFS_I(inode)->generation = 0;
3190 ret = btrfs_update_inode(trans, root, inode);
3193 if (i_size_read(inode) > 0) {
3194 ret = btrfs_check_trunc_cache_free_space(root,
3195 &root->fs_info->global_block_rsv);
3199 ret = btrfs_truncate_free_space_cache(root, trans, path,
3205 spin_lock(&block_group->lock);
3206 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3207 !btrfs_test_opt(root, SPACE_CACHE)) {
3209 * don't bother trying to write stuff out _if_
3210 * a) we're not cached,
3211 * b) we're with nospace_cache mount option.
3213 dcs = BTRFS_DC_WRITTEN;
3214 spin_unlock(&block_group->lock);
3217 spin_unlock(&block_group->lock);
3220 * Try to preallocate enough space based on how big the block group is.
3221 * Keep in mind this has to include any pinned space which could end up
3222 * taking up quite a bit since it's not folded into the other space
3225 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3230 num_pages *= PAGE_CACHE_SIZE;
3232 ret = btrfs_check_data_free_space(inode, num_pages);
3236 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3237 num_pages, num_pages,
3240 dcs = BTRFS_DC_SETUP;
3241 btrfs_free_reserved_data_space(inode, num_pages);
3246 btrfs_release_path(path);
3248 spin_lock(&block_group->lock);
3249 if (!ret && dcs == BTRFS_DC_SETUP)
3250 block_group->cache_generation = trans->transid;
3251 block_group->disk_cache_state = dcs;
3252 spin_unlock(&block_group->lock);
3257 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3258 struct btrfs_root *root)
3260 struct btrfs_block_group_cache *cache;
3262 struct btrfs_path *path;
3265 path = btrfs_alloc_path();
3271 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3273 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3275 cache = next_block_group(root, cache);
3283 err = cache_save_setup(cache, trans, path);
3284 last = cache->key.objectid + cache->key.offset;
3285 btrfs_put_block_group(cache);
3290 err = btrfs_run_delayed_refs(trans, root,
3292 if (err) /* File system offline */
3296 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3298 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3299 btrfs_put_block_group(cache);
3305 cache = next_block_group(root, cache);
3314 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3315 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3317 last = cache->key.objectid + cache->key.offset;
3319 err = write_one_cache_group(trans, root, path, cache);
3320 if (err) /* File system offline */
3323 btrfs_put_block_group(cache);
3328 * I don't think this is needed since we're just marking our
3329 * preallocated extent as written, but just in case it can't
3333 err = btrfs_run_delayed_refs(trans, root,
3335 if (err) /* File system offline */
3339 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3342 * Really this shouldn't happen, but it could if we
3343 * couldn't write the entire preallocated extent and
3344 * splitting the extent resulted in a new block.
3347 btrfs_put_block_group(cache);
3350 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3352 cache = next_block_group(root, cache);
3361 err = btrfs_write_out_cache(root, trans, cache, path);
3364 * If we didn't have an error then the cache state is still
3365 * NEED_WRITE, so we can set it to WRITTEN.
3367 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3368 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3369 last = cache->key.objectid + cache->key.offset;
3370 btrfs_put_block_group(cache);
3374 btrfs_free_path(path);
3378 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3380 struct btrfs_block_group_cache *block_group;
3383 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3384 if (!block_group || block_group->ro)
3387 btrfs_put_block_group(block_group);
3391 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3392 u64 total_bytes, u64 bytes_used,
3393 struct btrfs_space_info **space_info)
3395 struct btrfs_space_info *found;
3400 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3401 BTRFS_BLOCK_GROUP_RAID10))
3406 found = __find_space_info(info, flags);
3408 spin_lock(&found->lock);
3409 found->total_bytes += total_bytes;
3410 found->disk_total += total_bytes * factor;
3411 found->bytes_used += bytes_used;
3412 found->disk_used += bytes_used * factor;
3414 spin_unlock(&found->lock);
3415 *space_info = found;
3418 found = kzalloc(sizeof(*found), GFP_NOFS);
3422 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3428 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3429 INIT_LIST_HEAD(&found->block_groups[i]);
3430 init_rwsem(&found->groups_sem);
3431 spin_lock_init(&found->lock);
3432 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3433 found->total_bytes = total_bytes;
3434 found->disk_total = total_bytes * factor;
3435 found->bytes_used = bytes_used;
3436 found->disk_used = bytes_used * factor;
3437 found->bytes_pinned = 0;
3438 found->bytes_reserved = 0;
3439 found->bytes_readonly = 0;
3440 found->bytes_may_use = 0;
3442 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3443 found->chunk_alloc = 0;
3445 init_waitqueue_head(&found->wait);
3446 *space_info = found;
3447 list_add_rcu(&found->list, &info->space_info);
3448 if (flags & BTRFS_BLOCK_GROUP_DATA)
3449 info->data_sinfo = found;
3453 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3455 u64 extra_flags = chunk_to_extended(flags) &
3456 BTRFS_EXTENDED_PROFILE_MASK;
3458 write_seqlock(&fs_info->profiles_lock);
3459 if (flags & BTRFS_BLOCK_GROUP_DATA)
3460 fs_info->avail_data_alloc_bits |= extra_flags;
3461 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3462 fs_info->avail_metadata_alloc_bits |= extra_flags;
3463 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3464 fs_info->avail_system_alloc_bits |= extra_flags;
3465 write_sequnlock(&fs_info->profiles_lock);
3469 * returns target flags in extended format or 0 if restripe for this
3470 * chunk_type is not in progress
3472 * should be called with either volume_mutex or balance_lock held
3474 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3476 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3482 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3483 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3484 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3485 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3486 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3487 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3488 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3489 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3490 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3497 * @flags: available profiles in extended format (see ctree.h)
3499 * Returns reduced profile in chunk format. If profile changing is in
3500 * progress (either running or paused) picks the target profile (if it's
3501 * already available), otherwise falls back to plain reducing.
3503 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3506 * we add in the count of missing devices because we want
3507 * to make sure that any RAID levels on a degraded FS
3508 * continue to be honored.
3510 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3511 root->fs_info->fs_devices->missing_devices;
3516 * see if restripe for this chunk_type is in progress, if so
3517 * try to reduce to the target profile
3519 spin_lock(&root->fs_info->balance_lock);
3520 target = get_restripe_target(root->fs_info, flags);
3522 /* pick target profile only if it's already available */
3523 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3524 spin_unlock(&root->fs_info->balance_lock);
3525 return extended_to_chunk(target);
3528 spin_unlock(&root->fs_info->balance_lock);
3530 /* First, mask out the RAID levels which aren't possible */
3531 if (num_devices == 1)
3532 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3533 BTRFS_BLOCK_GROUP_RAID5);
3534 if (num_devices < 3)
3535 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3536 if (num_devices < 4)
3537 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3539 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3540 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3541 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3544 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3545 tmp = BTRFS_BLOCK_GROUP_RAID6;
3546 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3547 tmp = BTRFS_BLOCK_GROUP_RAID5;
3548 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3549 tmp = BTRFS_BLOCK_GROUP_RAID10;
3550 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3551 tmp = BTRFS_BLOCK_GROUP_RAID1;
3552 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3553 tmp = BTRFS_BLOCK_GROUP_RAID0;
3555 return extended_to_chunk(flags | tmp);
3558 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3563 seq = read_seqbegin(&root->fs_info->profiles_lock);
3565 if (flags & BTRFS_BLOCK_GROUP_DATA)
3566 flags |= root->fs_info->avail_data_alloc_bits;
3567 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3568 flags |= root->fs_info->avail_system_alloc_bits;
3569 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3570 flags |= root->fs_info->avail_metadata_alloc_bits;
3571 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3573 return btrfs_reduce_alloc_profile(root, flags);
3576 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3582 flags = BTRFS_BLOCK_GROUP_DATA;
3583 else if (root == root->fs_info->chunk_root)
3584 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3586 flags = BTRFS_BLOCK_GROUP_METADATA;
3588 ret = get_alloc_profile(root, flags);
3593 * This will check the space that the inode allocates from to make sure we have
3594 * enough space for bytes.
3596 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3598 struct btrfs_space_info *data_sinfo;
3599 struct btrfs_root *root = BTRFS_I(inode)->root;
3600 struct btrfs_fs_info *fs_info = root->fs_info;
3602 int ret = 0, committed = 0, alloc_chunk = 1;
3604 /* make sure bytes are sectorsize aligned */
3605 bytes = ALIGN(bytes, root->sectorsize);
3607 if (root == root->fs_info->tree_root ||
3608 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3613 data_sinfo = fs_info->data_sinfo;
3618 /* make sure we have enough space to handle the data first */
3619 spin_lock(&data_sinfo->lock);
3620 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3621 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3622 data_sinfo->bytes_may_use;
3624 if (used + bytes > data_sinfo->total_bytes) {
3625 struct btrfs_trans_handle *trans;
3628 * if we don't have enough free bytes in this space then we need
3629 * to alloc a new chunk.
3631 if (!data_sinfo->full && alloc_chunk) {
3634 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3635 spin_unlock(&data_sinfo->lock);
3637 alloc_target = btrfs_get_alloc_profile(root, 1);
3638 trans = btrfs_join_transaction(root);
3640 return PTR_ERR(trans);
3642 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3644 CHUNK_ALLOC_NO_FORCE);
3645 btrfs_end_transaction(trans, root);
3654 data_sinfo = fs_info->data_sinfo;
3660 * If we don't have enough pinned space to deal with this
3661 * allocation don't bother committing the transaction.
3663 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3666 spin_unlock(&data_sinfo->lock);
3668 /* commit the current transaction and try again */
3671 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3674 trans = btrfs_join_transaction(root);
3676 return PTR_ERR(trans);
3677 ret = btrfs_commit_transaction(trans, root);
3685 data_sinfo->bytes_may_use += bytes;
3686 trace_btrfs_space_reservation(root->fs_info, "space_info",
3687 data_sinfo->flags, bytes, 1);
3688 spin_unlock(&data_sinfo->lock);
3694 * Called if we need to clear a data reservation for this inode.
3696 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3698 struct btrfs_root *root = BTRFS_I(inode)->root;
3699 struct btrfs_space_info *data_sinfo;
3701 /* make sure bytes are sectorsize aligned */
3702 bytes = ALIGN(bytes, root->sectorsize);
3704 data_sinfo = root->fs_info->data_sinfo;
3705 spin_lock(&data_sinfo->lock);
3706 WARN_ON(data_sinfo->bytes_may_use < bytes);
3707 data_sinfo->bytes_may_use -= bytes;
3708 trace_btrfs_space_reservation(root->fs_info, "space_info",
3709 data_sinfo->flags, bytes, 0);
3710 spin_unlock(&data_sinfo->lock);
3713 static void force_metadata_allocation(struct btrfs_fs_info *info)
3715 struct list_head *head = &info->space_info;
3716 struct btrfs_space_info *found;
3719 list_for_each_entry_rcu(found, head, list) {
3720 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3721 found->force_alloc = CHUNK_ALLOC_FORCE;
3726 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3728 return (global->size << 1);
3731 static int should_alloc_chunk(struct btrfs_root *root,
3732 struct btrfs_space_info *sinfo, int force)
3734 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3735 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3736 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3739 if (force == CHUNK_ALLOC_FORCE)
3743 * We need to take into account the global rsv because for all intents
3744 * and purposes it's used space. Don't worry about locking the
3745 * global_rsv, it doesn't change except when the transaction commits.
3747 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3748 num_allocated += calc_global_rsv_need_space(global_rsv);
3751 * in limited mode, we want to have some free space up to
3752 * about 1% of the FS size.
3754 if (force == CHUNK_ALLOC_LIMITED) {
3755 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3756 thresh = max_t(u64, 64 * 1024 * 1024,
3757 div_factor_fine(thresh, 1));
3759 if (num_bytes - num_allocated < thresh)
3763 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3768 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3772 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3773 BTRFS_BLOCK_GROUP_RAID0 |
3774 BTRFS_BLOCK_GROUP_RAID5 |
3775 BTRFS_BLOCK_GROUP_RAID6))
3776 num_dev = root->fs_info->fs_devices->rw_devices;
3777 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3780 num_dev = 1; /* DUP or single */
3782 /* metadata for updaing devices and chunk tree */
3783 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3786 static void check_system_chunk(struct btrfs_trans_handle *trans,
3787 struct btrfs_root *root, u64 type)
3789 struct btrfs_space_info *info;
3793 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3794 spin_lock(&info->lock);
3795 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3796 info->bytes_reserved - info->bytes_readonly;
3797 spin_unlock(&info->lock);
3799 thresh = get_system_chunk_thresh(root, type);
3800 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3801 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3802 left, thresh, type);
3803 dump_space_info(info, 0, 0);
3806 if (left < thresh) {
3809 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3810 btrfs_alloc_chunk(trans, root, flags);
3814 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3815 struct btrfs_root *extent_root, u64 flags, int force)
3817 struct btrfs_space_info *space_info;
3818 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3819 int wait_for_alloc = 0;
3822 /* Don't re-enter if we're already allocating a chunk */
3823 if (trans->allocating_chunk)
3826 space_info = __find_space_info(extent_root->fs_info, flags);
3828 ret = update_space_info(extent_root->fs_info, flags,
3830 BUG_ON(ret); /* -ENOMEM */
3832 BUG_ON(!space_info); /* Logic error */
3835 spin_lock(&space_info->lock);
3836 if (force < space_info->force_alloc)
3837 force = space_info->force_alloc;
3838 if (space_info->full) {
3839 if (should_alloc_chunk(extent_root, space_info, force))
3843 spin_unlock(&space_info->lock);
3847 if (!should_alloc_chunk(extent_root, space_info, force)) {
3848 spin_unlock(&space_info->lock);
3850 } else if (space_info->chunk_alloc) {
3853 space_info->chunk_alloc = 1;
3856 spin_unlock(&space_info->lock);
3858 mutex_lock(&fs_info->chunk_mutex);
3861 * The chunk_mutex is held throughout the entirety of a chunk
3862 * allocation, so once we've acquired the chunk_mutex we know that the
3863 * other guy is done and we need to recheck and see if we should
3866 if (wait_for_alloc) {
3867 mutex_unlock(&fs_info->chunk_mutex);
3872 trans->allocating_chunk = true;
3875 * If we have mixed data/metadata chunks we want to make sure we keep
3876 * allocating mixed chunks instead of individual chunks.
3878 if (btrfs_mixed_space_info(space_info))
3879 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3882 * if we're doing a data chunk, go ahead and make sure that
3883 * we keep a reasonable number of metadata chunks allocated in the
3886 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3887 fs_info->data_chunk_allocations++;
3888 if (!(fs_info->data_chunk_allocations %
3889 fs_info->metadata_ratio))
3890 force_metadata_allocation(fs_info);
3894 * Check if we have enough space in SYSTEM chunk because we may need
3895 * to update devices.
3897 check_system_chunk(trans, extent_root, flags);
3899 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3900 trans->allocating_chunk = false;
3902 spin_lock(&space_info->lock);
3903 if (ret < 0 && ret != -ENOSPC)
3906 space_info->full = 1;
3910 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3912 space_info->chunk_alloc = 0;
3913 spin_unlock(&space_info->lock);
3914 mutex_unlock(&fs_info->chunk_mutex);
3918 static int can_overcommit(struct btrfs_root *root,
3919 struct btrfs_space_info *space_info, u64 bytes,
3920 enum btrfs_reserve_flush_enum flush)
3922 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3923 u64 profile = btrfs_get_alloc_profile(root, 0);
3929 used = space_info->bytes_used + space_info->bytes_reserved +
3930 space_info->bytes_pinned + space_info->bytes_readonly;
3933 * We only want to allow over committing if we have lots of actual space
3934 * free, but if we don't have enough space to handle the global reserve
3935 * space then we could end up having a real enospc problem when trying
3936 * to allocate a chunk or some other such important allocation.
3938 spin_lock(&global_rsv->lock);
3939 space_size = calc_global_rsv_need_space(global_rsv);
3940 spin_unlock(&global_rsv->lock);
3941 if (used + space_size >= space_info->total_bytes)
3944 used += space_info->bytes_may_use;
3946 spin_lock(&root->fs_info->free_chunk_lock);
3947 avail = root->fs_info->free_chunk_space;
3948 spin_unlock(&root->fs_info->free_chunk_lock);
3951 * If we have dup, raid1 or raid10 then only half of the free
3952 * space is actually useable. For raid56, the space info used
3953 * doesn't include the parity drive, so we don't have to
3956 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3957 BTRFS_BLOCK_GROUP_RAID1 |
3958 BTRFS_BLOCK_GROUP_RAID10))
3961 to_add = space_info->total_bytes;
3964 * If we aren't flushing all things, let us overcommit up to
3965 * 1/2th of the space. If we can flush, don't let us overcommit
3966 * too much, let it overcommit up to 1/8 of the space.
3968 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3974 * Limit the overcommit to the amount of free space we could possibly
3975 * allocate for chunks.
3977 to_add = min(avail, to_add);
3979 if (used + bytes < space_info->total_bytes + to_add)
3984 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3985 unsigned long nr_pages)
3987 struct super_block *sb = root->fs_info->sb;
3989 if (down_read_trylock(&sb->s_umount)) {
3990 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3991 up_read(&sb->s_umount);
3994 * We needn't worry the filesystem going from r/w to r/o though
3995 * we don't acquire ->s_umount mutex, because the filesystem
3996 * should guarantee the delalloc inodes list be empty after
3997 * the filesystem is readonly(all dirty pages are written to
4000 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
4001 if (!current->journal_info)
4002 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4007 * shrink metadata reservation for delalloc
4009 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4012 struct btrfs_block_rsv *block_rsv;
4013 struct btrfs_space_info *space_info;
4014 struct btrfs_trans_handle *trans;
4018 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
4020 enum btrfs_reserve_flush_enum flush;
4022 trans = (struct btrfs_trans_handle *)current->journal_info;
4023 block_rsv = &root->fs_info->delalloc_block_rsv;
4024 space_info = block_rsv->space_info;
4027 delalloc_bytes = percpu_counter_sum_positive(
4028 &root->fs_info->delalloc_bytes);
4029 if (delalloc_bytes == 0) {
4032 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4036 while (delalloc_bytes && loops < 3) {
4037 max_reclaim = min(delalloc_bytes, to_reclaim);
4038 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4039 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4041 * We need to wait for the async pages to actually start before
4044 wait_event(root->fs_info->async_submit_wait,
4045 !atomic_read(&root->fs_info->async_delalloc_pages));
4048 flush = BTRFS_RESERVE_FLUSH_ALL;
4050 flush = BTRFS_RESERVE_NO_FLUSH;
4051 spin_lock(&space_info->lock);
4052 if (can_overcommit(root, space_info, orig, flush)) {
4053 spin_unlock(&space_info->lock);
4056 spin_unlock(&space_info->lock);
4059 if (wait_ordered && !trans) {
4060 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4062 time_left = schedule_timeout_killable(1);
4067 delalloc_bytes = percpu_counter_sum_positive(
4068 &root->fs_info->delalloc_bytes);
4073 * maybe_commit_transaction - possibly commit the transaction if its ok to
4074 * @root - the root we're allocating for
4075 * @bytes - the number of bytes we want to reserve
4076 * @force - force the commit
4078 * This will check to make sure that committing the transaction will actually
4079 * get us somewhere and then commit the transaction if it does. Otherwise it
4080 * will return -ENOSPC.
4082 static int may_commit_transaction(struct btrfs_root *root,
4083 struct btrfs_space_info *space_info,
4084 u64 bytes, int force)
4086 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4087 struct btrfs_trans_handle *trans;
4089 trans = (struct btrfs_trans_handle *)current->journal_info;
4096 /* See if there is enough pinned space to make this reservation */
4097 spin_lock(&space_info->lock);
4098 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4100 spin_unlock(&space_info->lock);
4103 spin_unlock(&space_info->lock);
4106 * See if there is some space in the delayed insertion reservation for
4109 if (space_info != delayed_rsv->space_info)
4112 spin_lock(&space_info->lock);
4113 spin_lock(&delayed_rsv->lock);
4114 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4115 bytes - delayed_rsv->size) >= 0) {
4116 spin_unlock(&delayed_rsv->lock);
4117 spin_unlock(&space_info->lock);
4120 spin_unlock(&delayed_rsv->lock);
4121 spin_unlock(&space_info->lock);
4124 trans = btrfs_join_transaction(root);
4128 return btrfs_commit_transaction(trans, root);
4132 FLUSH_DELAYED_ITEMS_NR = 1,
4133 FLUSH_DELAYED_ITEMS = 2,
4135 FLUSH_DELALLOC_WAIT = 4,
4140 static int flush_space(struct btrfs_root *root,
4141 struct btrfs_space_info *space_info, u64 num_bytes,
4142 u64 orig_bytes, int state)
4144 struct btrfs_trans_handle *trans;
4149 case FLUSH_DELAYED_ITEMS_NR:
4150 case FLUSH_DELAYED_ITEMS:
4151 if (state == FLUSH_DELAYED_ITEMS_NR) {
4152 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4154 nr = (int)div64_u64(num_bytes, bytes);
4161 trans = btrfs_join_transaction(root);
4162 if (IS_ERR(trans)) {
4163 ret = PTR_ERR(trans);
4166 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4167 btrfs_end_transaction(trans, root);
4169 case FLUSH_DELALLOC:
4170 case FLUSH_DELALLOC_WAIT:
4171 shrink_delalloc(root, num_bytes, orig_bytes,
4172 state == FLUSH_DELALLOC_WAIT);
4175 trans = btrfs_join_transaction(root);
4176 if (IS_ERR(trans)) {
4177 ret = PTR_ERR(trans);
4180 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4181 btrfs_get_alloc_profile(root, 0),
4182 CHUNK_ALLOC_NO_FORCE);
4183 btrfs_end_transaction(trans, root);
4188 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4198 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4199 * @root - the root we're allocating for
4200 * @block_rsv - the block_rsv we're allocating for
4201 * @orig_bytes - the number of bytes we want
4202 * @flush - whether or not we can flush to make our reservation
4204 * This will reserve orgi_bytes number of bytes from the space info associated
4205 * with the block_rsv. If there is not enough space it will make an attempt to
4206 * flush out space to make room. It will do this by flushing delalloc if
4207 * possible or committing the transaction. If flush is 0 then no attempts to
4208 * regain reservations will be made and this will fail if there is not enough
4211 static int reserve_metadata_bytes(struct btrfs_root *root,
4212 struct btrfs_block_rsv *block_rsv,
4214 enum btrfs_reserve_flush_enum flush)
4216 struct btrfs_space_info *space_info = block_rsv->space_info;
4218 u64 num_bytes = orig_bytes;
4219 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4221 bool flushing = false;
4225 spin_lock(&space_info->lock);
4227 * We only want to wait if somebody other than us is flushing and we
4228 * are actually allowed to flush all things.
4230 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4231 space_info->flush) {
4232 spin_unlock(&space_info->lock);
4234 * If we have a trans handle we can't wait because the flusher
4235 * may have to commit the transaction, which would mean we would
4236 * deadlock since we are waiting for the flusher to finish, but
4237 * hold the current transaction open.
4239 if (current->journal_info)
4241 ret = wait_event_killable(space_info->wait, !space_info->flush);
4242 /* Must have been killed, return */
4246 spin_lock(&space_info->lock);
4250 used = space_info->bytes_used + space_info->bytes_reserved +
4251 space_info->bytes_pinned + space_info->bytes_readonly +
4252 space_info->bytes_may_use;
4255 * The idea here is that we've not already over-reserved the block group
4256 * then we can go ahead and save our reservation first and then start
4257 * flushing if we need to. Otherwise if we've already overcommitted
4258 * lets start flushing stuff first and then come back and try to make
4261 if (used <= space_info->total_bytes) {
4262 if (used + orig_bytes <= space_info->total_bytes) {
4263 space_info->bytes_may_use += orig_bytes;
4264 trace_btrfs_space_reservation(root->fs_info,
4265 "space_info", space_info->flags, orig_bytes, 1);
4269 * Ok set num_bytes to orig_bytes since we aren't
4270 * overocmmitted, this way we only try and reclaim what
4273 num_bytes = orig_bytes;
4277 * Ok we're over committed, set num_bytes to the overcommitted
4278 * amount plus the amount of bytes that we need for this
4281 num_bytes = used - space_info->total_bytes +
4285 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4286 space_info->bytes_may_use += orig_bytes;
4287 trace_btrfs_space_reservation(root->fs_info, "space_info",
4288 space_info->flags, orig_bytes,
4294 * Couldn't make our reservation, save our place so while we're trying
4295 * to reclaim space we can actually use it instead of somebody else
4296 * stealing it from us.
4298 * We make the other tasks wait for the flush only when we can flush
4301 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4303 space_info->flush = 1;
4306 spin_unlock(&space_info->lock);
4308 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4311 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4316 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4317 * would happen. So skip delalloc flush.
4319 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4320 (flush_state == FLUSH_DELALLOC ||
4321 flush_state == FLUSH_DELALLOC_WAIT))
4322 flush_state = ALLOC_CHUNK;
4326 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4327 flush_state < COMMIT_TRANS)
4329 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4330 flush_state <= COMMIT_TRANS)
4334 if (ret == -ENOSPC &&
4335 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4336 struct btrfs_block_rsv *global_rsv =
4337 &root->fs_info->global_block_rsv;
4339 if (block_rsv != global_rsv &&
4340 !block_rsv_use_bytes(global_rsv, orig_bytes))
4344 spin_lock(&space_info->lock);
4345 space_info->flush = 0;
4346 wake_up_all(&space_info->wait);
4347 spin_unlock(&space_info->lock);
4352 static struct btrfs_block_rsv *get_block_rsv(
4353 const struct btrfs_trans_handle *trans,
4354 const struct btrfs_root *root)
4356 struct btrfs_block_rsv *block_rsv = NULL;
4359 block_rsv = trans->block_rsv;
4361 if (root == root->fs_info->csum_root && trans->adding_csums)
4362 block_rsv = trans->block_rsv;
4365 block_rsv = root->block_rsv;
4368 block_rsv = &root->fs_info->empty_block_rsv;
4373 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4377 spin_lock(&block_rsv->lock);
4378 if (block_rsv->reserved >= num_bytes) {
4379 block_rsv->reserved -= num_bytes;
4380 if (block_rsv->reserved < block_rsv->size)
4381 block_rsv->full = 0;
4384 spin_unlock(&block_rsv->lock);
4388 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4389 u64 num_bytes, int update_size)
4391 spin_lock(&block_rsv->lock);
4392 block_rsv->reserved += num_bytes;
4394 block_rsv->size += num_bytes;
4395 else if (block_rsv->reserved >= block_rsv->size)
4396 block_rsv->full = 1;
4397 spin_unlock(&block_rsv->lock);
4400 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4401 struct btrfs_block_rsv *dest, u64 num_bytes,
4404 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4407 if (global_rsv->space_info != dest->space_info)
4410 spin_lock(&global_rsv->lock);
4411 min_bytes = div_factor(global_rsv->size, min_factor);
4412 if (global_rsv->reserved < min_bytes + num_bytes) {
4413 spin_unlock(&global_rsv->lock);
4416 global_rsv->reserved -= num_bytes;
4417 if (global_rsv->reserved < global_rsv->size)
4418 global_rsv->full = 0;
4419 spin_unlock(&global_rsv->lock);
4421 block_rsv_add_bytes(dest, num_bytes, 1);
4425 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4426 struct btrfs_block_rsv *block_rsv,
4427 struct btrfs_block_rsv *dest, u64 num_bytes)
4429 struct btrfs_space_info *space_info = block_rsv->space_info;
4431 spin_lock(&block_rsv->lock);
4432 if (num_bytes == (u64)-1)
4433 num_bytes = block_rsv->size;
4434 block_rsv->size -= num_bytes;
4435 if (block_rsv->reserved >= block_rsv->size) {
4436 num_bytes = block_rsv->reserved - block_rsv->size;
4437 block_rsv->reserved = block_rsv->size;
4438 block_rsv->full = 1;
4442 spin_unlock(&block_rsv->lock);
4444 if (num_bytes > 0) {
4446 spin_lock(&dest->lock);
4450 bytes_to_add = dest->size - dest->reserved;
4451 bytes_to_add = min(num_bytes, bytes_to_add);
4452 dest->reserved += bytes_to_add;
4453 if (dest->reserved >= dest->size)
4455 num_bytes -= bytes_to_add;
4457 spin_unlock(&dest->lock);
4460 spin_lock(&space_info->lock);
4461 space_info->bytes_may_use -= num_bytes;
4462 trace_btrfs_space_reservation(fs_info, "space_info",
4463 space_info->flags, num_bytes, 0);
4464 space_info->reservation_progress++;
4465 spin_unlock(&space_info->lock);
4470 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4471 struct btrfs_block_rsv *dst, u64 num_bytes)
4475 ret = block_rsv_use_bytes(src, num_bytes);
4479 block_rsv_add_bytes(dst, num_bytes, 1);
4483 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4485 memset(rsv, 0, sizeof(*rsv));
4486 spin_lock_init(&rsv->lock);
4490 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4491 unsigned short type)
4493 struct btrfs_block_rsv *block_rsv;
4494 struct btrfs_fs_info *fs_info = root->fs_info;
4496 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4500 btrfs_init_block_rsv(block_rsv, type);
4501 block_rsv->space_info = __find_space_info(fs_info,
4502 BTRFS_BLOCK_GROUP_METADATA);
4506 void btrfs_free_block_rsv(struct btrfs_root *root,
4507 struct btrfs_block_rsv *rsv)
4511 btrfs_block_rsv_release(root, rsv, (u64)-1);
4515 int btrfs_block_rsv_add(struct btrfs_root *root,
4516 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4517 enum btrfs_reserve_flush_enum flush)
4524 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4526 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4533 int btrfs_block_rsv_check(struct btrfs_root *root,
4534 struct btrfs_block_rsv *block_rsv, int min_factor)
4542 spin_lock(&block_rsv->lock);
4543 num_bytes = div_factor(block_rsv->size, min_factor);
4544 if (block_rsv->reserved >= num_bytes)
4546 spin_unlock(&block_rsv->lock);
4551 int btrfs_block_rsv_refill(struct btrfs_root *root,
4552 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4553 enum btrfs_reserve_flush_enum flush)
4561 spin_lock(&block_rsv->lock);
4562 num_bytes = min_reserved;
4563 if (block_rsv->reserved >= num_bytes)
4566 num_bytes -= block_rsv->reserved;
4567 spin_unlock(&block_rsv->lock);
4572 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4574 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4581 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4582 struct btrfs_block_rsv *dst_rsv,
4585 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4588 void btrfs_block_rsv_release(struct btrfs_root *root,
4589 struct btrfs_block_rsv *block_rsv,
4592 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4593 if (global_rsv->full || global_rsv == block_rsv ||
4594 block_rsv->space_info != global_rsv->space_info)
4596 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4601 * helper to calculate size of global block reservation.
4602 * the desired value is sum of space used by extent tree,
4603 * checksum tree and root tree
4605 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4607 struct btrfs_space_info *sinfo;
4611 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4613 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4614 spin_lock(&sinfo->lock);
4615 data_used = sinfo->bytes_used;
4616 spin_unlock(&sinfo->lock);
4618 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4619 spin_lock(&sinfo->lock);
4620 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4622 meta_used = sinfo->bytes_used;
4623 spin_unlock(&sinfo->lock);
4625 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4627 num_bytes += div64_u64(data_used + meta_used, 50);
4629 if (num_bytes * 3 > meta_used)
4630 num_bytes = div64_u64(meta_used, 3);
4632 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4635 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4637 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4638 struct btrfs_space_info *sinfo = block_rsv->space_info;
4641 num_bytes = calc_global_metadata_size(fs_info);
4643 spin_lock(&sinfo->lock);
4644 spin_lock(&block_rsv->lock);
4646 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4648 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4649 sinfo->bytes_reserved + sinfo->bytes_readonly +
4650 sinfo->bytes_may_use;
4652 if (sinfo->total_bytes > num_bytes) {
4653 num_bytes = sinfo->total_bytes - num_bytes;
4654 block_rsv->reserved += num_bytes;
4655 sinfo->bytes_may_use += num_bytes;
4656 trace_btrfs_space_reservation(fs_info, "space_info",
4657 sinfo->flags, num_bytes, 1);
4660 if (block_rsv->reserved >= block_rsv->size) {
4661 num_bytes = block_rsv->reserved - block_rsv->size;
4662 sinfo->bytes_may_use -= num_bytes;
4663 trace_btrfs_space_reservation(fs_info, "space_info",
4664 sinfo->flags, num_bytes, 0);
4665 sinfo->reservation_progress++;
4666 block_rsv->reserved = block_rsv->size;
4667 block_rsv->full = 1;
4670 spin_unlock(&block_rsv->lock);
4671 spin_unlock(&sinfo->lock);
4674 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4676 struct btrfs_space_info *space_info;
4678 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4679 fs_info->chunk_block_rsv.space_info = space_info;
4681 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4682 fs_info->global_block_rsv.space_info = space_info;
4683 fs_info->delalloc_block_rsv.space_info = space_info;
4684 fs_info->trans_block_rsv.space_info = space_info;
4685 fs_info->empty_block_rsv.space_info = space_info;
4686 fs_info->delayed_block_rsv.space_info = space_info;
4688 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4689 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4690 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4691 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4692 if (fs_info->quota_root)
4693 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4694 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4696 update_global_block_rsv(fs_info);
4699 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4701 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4703 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4704 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4705 WARN_ON(fs_info->trans_block_rsv.size > 0);
4706 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4707 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4708 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4709 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4710 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4713 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4714 struct btrfs_root *root)
4716 if (!trans->block_rsv)
4719 if (!trans->bytes_reserved)
4722 trace_btrfs_space_reservation(root->fs_info, "transaction",
4723 trans->transid, trans->bytes_reserved, 0);
4724 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4725 trans->bytes_reserved = 0;
4728 /* Can only return 0 or -ENOSPC */
4729 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4730 struct inode *inode)
4732 struct btrfs_root *root = BTRFS_I(inode)->root;
4733 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4734 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4737 * We need to hold space in order to delete our orphan item once we've
4738 * added it, so this takes the reservation so we can release it later
4739 * when we are truly done with the orphan item.
4741 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4742 trace_btrfs_space_reservation(root->fs_info, "orphan",
4743 btrfs_ino(inode), num_bytes, 1);
4744 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4747 void btrfs_orphan_release_metadata(struct inode *inode)
4749 struct btrfs_root *root = BTRFS_I(inode)->root;
4750 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4751 trace_btrfs_space_reservation(root->fs_info, "orphan",
4752 btrfs_ino(inode), num_bytes, 0);
4753 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4757 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4758 * root: the root of the parent directory
4759 * rsv: block reservation
4760 * items: the number of items that we need do reservation
4761 * qgroup_reserved: used to return the reserved size in qgroup
4763 * This function is used to reserve the space for snapshot/subvolume
4764 * creation and deletion. Those operations are different with the
4765 * common file/directory operations, they change two fs/file trees
4766 * and root tree, the number of items that the qgroup reserves is
4767 * different with the free space reservation. So we can not use
4768 * the space reseravtion mechanism in start_transaction().
4770 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4771 struct btrfs_block_rsv *rsv,
4773 u64 *qgroup_reserved,
4774 bool use_global_rsv)
4778 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4780 if (root->fs_info->quota_enabled) {
4781 /* One for parent inode, two for dir entries */
4782 num_bytes = 3 * root->leafsize;
4783 ret = btrfs_qgroup_reserve(root, num_bytes);
4790 *qgroup_reserved = num_bytes;
4792 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4793 rsv->space_info = __find_space_info(root->fs_info,
4794 BTRFS_BLOCK_GROUP_METADATA);
4795 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4796 BTRFS_RESERVE_FLUSH_ALL);
4798 if (ret == -ENOSPC && use_global_rsv)
4799 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4802 if (*qgroup_reserved)
4803 btrfs_qgroup_free(root, *qgroup_reserved);
4809 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4810 struct btrfs_block_rsv *rsv,
4811 u64 qgroup_reserved)
4813 btrfs_block_rsv_release(root, rsv, (u64)-1);
4814 if (qgroup_reserved)
4815 btrfs_qgroup_free(root, qgroup_reserved);
4819 * drop_outstanding_extent - drop an outstanding extent
4820 * @inode: the inode we're dropping the extent for
4822 * This is called when we are freeing up an outstanding extent, either called
4823 * after an error or after an extent is written. This will return the number of
4824 * reserved extents that need to be freed. This must be called with
4825 * BTRFS_I(inode)->lock held.
4827 static unsigned drop_outstanding_extent(struct inode *inode)
4829 unsigned drop_inode_space = 0;
4830 unsigned dropped_extents = 0;
4832 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4833 BTRFS_I(inode)->outstanding_extents--;
4835 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4836 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4837 &BTRFS_I(inode)->runtime_flags))
4838 drop_inode_space = 1;
4841 * If we have more or the same amount of outsanding extents than we have
4842 * reserved then we need to leave the reserved extents count alone.
4844 if (BTRFS_I(inode)->outstanding_extents >=
4845 BTRFS_I(inode)->reserved_extents)
4846 return drop_inode_space;
4848 dropped_extents = BTRFS_I(inode)->reserved_extents -
4849 BTRFS_I(inode)->outstanding_extents;
4850 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4851 return dropped_extents + drop_inode_space;
4855 * calc_csum_metadata_size - return the amount of metada space that must be
4856 * reserved/free'd for the given bytes.
4857 * @inode: the inode we're manipulating
4858 * @num_bytes: the number of bytes in question
4859 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4861 * This adjusts the number of csum_bytes in the inode and then returns the
4862 * correct amount of metadata that must either be reserved or freed. We
4863 * calculate how many checksums we can fit into one leaf and then divide the
4864 * number of bytes that will need to be checksumed by this value to figure out
4865 * how many checksums will be required. If we are adding bytes then the number
4866 * may go up and we will return the number of additional bytes that must be
4867 * reserved. If it is going down we will return the number of bytes that must
4870 * This must be called with BTRFS_I(inode)->lock held.
4872 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4875 struct btrfs_root *root = BTRFS_I(inode)->root;
4877 int num_csums_per_leaf;
4881 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4882 BTRFS_I(inode)->csum_bytes == 0)
4885 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4887 BTRFS_I(inode)->csum_bytes += num_bytes;
4889 BTRFS_I(inode)->csum_bytes -= num_bytes;
4890 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4891 num_csums_per_leaf = (int)div64_u64(csum_size,
4892 sizeof(struct btrfs_csum_item) +
4893 sizeof(struct btrfs_disk_key));
4894 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4895 num_csums = num_csums + num_csums_per_leaf - 1;
4896 num_csums = num_csums / num_csums_per_leaf;
4898 old_csums = old_csums + num_csums_per_leaf - 1;
4899 old_csums = old_csums / num_csums_per_leaf;
4901 /* No change, no need to reserve more */
4902 if (old_csums == num_csums)
4906 return btrfs_calc_trans_metadata_size(root,
4907 num_csums - old_csums);
4909 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4912 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4914 struct btrfs_root *root = BTRFS_I(inode)->root;
4915 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4918 unsigned nr_extents = 0;
4919 int extra_reserve = 0;
4920 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4922 bool delalloc_lock = true;
4926 /* If we are a free space inode we need to not flush since we will be in
4927 * the middle of a transaction commit. We also don't need the delalloc
4928 * mutex since we won't race with anybody. We need this mostly to make
4929 * lockdep shut its filthy mouth.
4931 if (btrfs_is_free_space_inode(inode)) {
4932 flush = BTRFS_RESERVE_NO_FLUSH;
4933 delalloc_lock = false;
4936 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4937 btrfs_transaction_in_commit(root->fs_info))
4938 schedule_timeout(1);
4941 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4943 num_bytes = ALIGN(num_bytes, root->sectorsize);
4945 spin_lock(&BTRFS_I(inode)->lock);
4946 BTRFS_I(inode)->outstanding_extents++;
4948 if (BTRFS_I(inode)->outstanding_extents >
4949 BTRFS_I(inode)->reserved_extents)
4950 nr_extents = BTRFS_I(inode)->outstanding_extents -
4951 BTRFS_I(inode)->reserved_extents;
4954 * Add an item to reserve for updating the inode when we complete the
4957 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4958 &BTRFS_I(inode)->runtime_flags)) {
4963 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4964 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4965 csum_bytes = BTRFS_I(inode)->csum_bytes;
4966 spin_unlock(&BTRFS_I(inode)->lock);
4968 if (root->fs_info->quota_enabled) {
4969 ret = btrfs_qgroup_reserve(root, num_bytes +
4970 nr_extents * root->leafsize);
4975 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4976 if (unlikely(ret)) {
4977 if (root->fs_info->quota_enabled)
4978 btrfs_qgroup_free(root, num_bytes +
4979 nr_extents * root->leafsize);
4983 spin_lock(&BTRFS_I(inode)->lock);
4984 if (extra_reserve) {
4985 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4986 &BTRFS_I(inode)->runtime_flags);
4989 BTRFS_I(inode)->reserved_extents += nr_extents;
4990 spin_unlock(&BTRFS_I(inode)->lock);
4993 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4996 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4997 btrfs_ino(inode), to_reserve, 1);
4998 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5003 spin_lock(&BTRFS_I(inode)->lock);
5004 dropped = drop_outstanding_extent(inode);
5006 * If the inodes csum_bytes is the same as the original
5007 * csum_bytes then we know we haven't raced with any free()ers
5008 * so we can just reduce our inodes csum bytes and carry on.
5010 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5011 calc_csum_metadata_size(inode, num_bytes, 0);
5013 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5017 * This is tricky, but first we need to figure out how much we
5018 * free'd from any free-ers that occured during this
5019 * reservation, so we reset ->csum_bytes to the csum_bytes
5020 * before we dropped our lock, and then call the free for the
5021 * number of bytes that were freed while we were trying our
5024 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5025 BTRFS_I(inode)->csum_bytes = csum_bytes;
5026 to_free = calc_csum_metadata_size(inode, bytes, 0);
5030 * Now we need to see how much we would have freed had we not
5031 * been making this reservation and our ->csum_bytes were not
5032 * artificially inflated.
5034 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5035 bytes = csum_bytes - orig_csum_bytes;
5036 bytes = calc_csum_metadata_size(inode, bytes, 0);
5039 * Now reset ->csum_bytes to what it should be. If bytes is
5040 * more than to_free then we would have free'd more space had we
5041 * not had an artificially high ->csum_bytes, so we need to free
5042 * the remainder. If bytes is the same or less then we don't
5043 * need to do anything, the other free-ers did the correct
5046 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5047 if (bytes > to_free)
5048 to_free = bytes - to_free;
5052 spin_unlock(&BTRFS_I(inode)->lock);
5054 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5057 btrfs_block_rsv_release(root, block_rsv, to_free);
5058 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5059 btrfs_ino(inode), to_free, 0);
5062 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5067 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5068 * @inode: the inode to release the reservation for
5069 * @num_bytes: the number of bytes we're releasing
5071 * This will release the metadata reservation for an inode. This can be called
5072 * once we complete IO for a given set of bytes to release their metadata
5075 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5077 struct btrfs_root *root = BTRFS_I(inode)->root;
5081 num_bytes = ALIGN(num_bytes, root->sectorsize);
5082 spin_lock(&BTRFS_I(inode)->lock);
5083 dropped = drop_outstanding_extent(inode);
5086 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5087 spin_unlock(&BTRFS_I(inode)->lock);
5089 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5091 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5092 btrfs_ino(inode), to_free, 0);
5093 if (root->fs_info->quota_enabled) {
5094 btrfs_qgroup_free(root, num_bytes +
5095 dropped * root->leafsize);
5098 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5103 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5104 * @inode: inode we're writing to
5105 * @num_bytes: the number of bytes we want to allocate
5107 * This will do the following things
5109 * o reserve space in the data space info for num_bytes
5110 * o reserve space in the metadata space info based on number of outstanding
5111 * extents and how much csums will be needed
5112 * o add to the inodes ->delalloc_bytes
5113 * o add it to the fs_info's delalloc inodes list.
5115 * This will return 0 for success and -ENOSPC if there is no space left.
5117 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5121 ret = btrfs_check_data_free_space(inode, num_bytes);
5125 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5127 btrfs_free_reserved_data_space(inode, num_bytes);
5135 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5136 * @inode: inode we're releasing space for
5137 * @num_bytes: the number of bytes we want to free up
5139 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5140 * called in the case that we don't need the metadata AND data reservations
5141 * anymore. So if there is an error or we insert an inline extent.
5143 * This function will release the metadata space that was not used and will
5144 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5145 * list if there are no delalloc bytes left.
5147 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5149 btrfs_delalloc_release_metadata(inode, num_bytes);
5150 btrfs_free_reserved_data_space(inode, num_bytes);
5153 static int update_block_group(struct btrfs_root *root,
5154 u64 bytenr, u64 num_bytes, int alloc)
5156 struct btrfs_block_group_cache *cache = NULL;
5157 struct btrfs_fs_info *info = root->fs_info;
5158 u64 total = num_bytes;
5163 /* block accounting for super block */
5164 spin_lock(&info->delalloc_root_lock);
5165 old_val = btrfs_super_bytes_used(info->super_copy);
5167 old_val += num_bytes;
5169 old_val -= num_bytes;
5170 btrfs_set_super_bytes_used(info->super_copy, old_val);
5171 spin_unlock(&info->delalloc_root_lock);
5174 cache = btrfs_lookup_block_group(info, bytenr);
5177 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5178 BTRFS_BLOCK_GROUP_RAID1 |
5179 BTRFS_BLOCK_GROUP_RAID10))
5184 * If this block group has free space cache written out, we
5185 * need to make sure to load it if we are removing space. This
5186 * is because we need the unpinning stage to actually add the
5187 * space back to the block group, otherwise we will leak space.
5189 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5190 cache_block_group(cache, 1);
5192 byte_in_group = bytenr - cache->key.objectid;
5193 WARN_ON(byte_in_group > cache->key.offset);
5195 spin_lock(&cache->space_info->lock);
5196 spin_lock(&cache->lock);
5198 if (btrfs_test_opt(root, SPACE_CACHE) &&
5199 cache->disk_cache_state < BTRFS_DC_CLEAR)
5200 cache->disk_cache_state = BTRFS_DC_CLEAR;
5203 old_val = btrfs_block_group_used(&cache->item);
5204 num_bytes = min(total, cache->key.offset - byte_in_group);
5206 old_val += num_bytes;
5207 btrfs_set_block_group_used(&cache->item, old_val);
5208 cache->reserved -= num_bytes;
5209 cache->space_info->bytes_reserved -= num_bytes;
5210 cache->space_info->bytes_used += num_bytes;
5211 cache->space_info->disk_used += num_bytes * factor;
5212 spin_unlock(&cache->lock);
5213 spin_unlock(&cache->space_info->lock);
5215 old_val -= num_bytes;
5216 btrfs_set_block_group_used(&cache->item, old_val);
5217 cache->pinned += num_bytes;
5218 cache->space_info->bytes_pinned += num_bytes;
5219 cache->space_info->bytes_used -= num_bytes;
5220 cache->space_info->disk_used -= num_bytes * factor;
5221 spin_unlock(&cache->lock);
5222 spin_unlock(&cache->space_info->lock);
5224 set_extent_dirty(info->pinned_extents,
5225 bytenr, bytenr + num_bytes - 1,
5226 GFP_NOFS | __GFP_NOFAIL);
5228 btrfs_put_block_group(cache);
5230 bytenr += num_bytes;
5235 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5237 struct btrfs_block_group_cache *cache;
5240 spin_lock(&root->fs_info->block_group_cache_lock);
5241 bytenr = root->fs_info->first_logical_byte;
5242 spin_unlock(&root->fs_info->block_group_cache_lock);
5244 if (bytenr < (u64)-1)
5247 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5251 bytenr = cache->key.objectid;
5252 btrfs_put_block_group(cache);
5257 static int pin_down_extent(struct btrfs_root *root,
5258 struct btrfs_block_group_cache *cache,
5259 u64 bytenr, u64 num_bytes, int reserved)
5261 spin_lock(&cache->space_info->lock);
5262 spin_lock(&cache->lock);
5263 cache->pinned += num_bytes;
5264 cache->space_info->bytes_pinned += num_bytes;
5266 cache->reserved -= num_bytes;
5267 cache->space_info->bytes_reserved -= num_bytes;
5269 spin_unlock(&cache->lock);
5270 spin_unlock(&cache->space_info->lock);
5272 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5273 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5278 * this function must be called within transaction
5280 int btrfs_pin_extent(struct btrfs_root *root,
5281 u64 bytenr, u64 num_bytes, int reserved)
5283 struct btrfs_block_group_cache *cache;
5285 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5286 BUG_ON(!cache); /* Logic error */
5288 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5290 btrfs_put_block_group(cache);
5295 * this function must be called within transaction
5297 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5298 u64 bytenr, u64 num_bytes)
5300 struct btrfs_block_group_cache *cache;
5303 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5308 * pull in the free space cache (if any) so that our pin
5309 * removes the free space from the cache. We have load_only set
5310 * to one because the slow code to read in the free extents does check
5311 * the pinned extents.
5313 cache_block_group(cache, 1);
5315 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5317 /* remove us from the free space cache (if we're there at all) */
5318 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5319 btrfs_put_block_group(cache);
5323 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5326 struct btrfs_block_group_cache *block_group;
5327 struct btrfs_caching_control *caching_ctl;
5329 block_group = btrfs_lookup_block_group(root->fs_info, start);
5333 cache_block_group(block_group, 0);
5334 caching_ctl = get_caching_control(block_group);
5338 BUG_ON(!block_group_cache_done(block_group));
5339 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5341 mutex_lock(&caching_ctl->mutex);
5343 if (start >= caching_ctl->progress) {
5344 ret = add_excluded_extent(root, start, num_bytes);
5345 } else if (start + num_bytes <= caching_ctl->progress) {
5346 ret = btrfs_remove_free_space(block_group,
5349 num_bytes = caching_ctl->progress - start;
5350 ret = btrfs_remove_free_space(block_group,
5355 num_bytes = (start + num_bytes) -
5356 caching_ctl->progress;
5357 start = caching_ctl->progress;
5358 ret = add_excluded_extent(root, start, num_bytes);
5361 mutex_unlock(&caching_ctl->mutex);
5362 put_caching_control(caching_ctl);
5364 btrfs_put_block_group(block_group);
5368 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5369 struct extent_buffer *eb)
5371 struct btrfs_file_extent_item *item;
5372 struct btrfs_key key;
5376 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5379 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5380 btrfs_item_key_to_cpu(eb, &key, i);
5381 if (key.type != BTRFS_EXTENT_DATA_KEY)
5383 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5384 found_type = btrfs_file_extent_type(eb, item);
5385 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5387 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5389 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5390 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5391 __exclude_logged_extent(log, key.objectid, key.offset);
5398 * btrfs_update_reserved_bytes - update the block_group and space info counters
5399 * @cache: The cache we are manipulating
5400 * @num_bytes: The number of bytes in question
5401 * @reserve: One of the reservation enums
5403 * This is called by the allocator when it reserves space, or by somebody who is
5404 * freeing space that was never actually used on disk. For example if you
5405 * reserve some space for a new leaf in transaction A and before transaction A
5406 * commits you free that leaf, you call this with reserve set to 0 in order to
5407 * clear the reservation.
5409 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5410 * ENOSPC accounting. For data we handle the reservation through clearing the
5411 * delalloc bits in the io_tree. We have to do this since we could end up
5412 * allocating less disk space for the amount of data we have reserved in the
5413 * case of compression.
5415 * If this is a reservation and the block group has become read only we cannot
5416 * make the reservation and return -EAGAIN, otherwise this function always
5419 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5420 u64 num_bytes, int reserve)
5422 struct btrfs_space_info *space_info = cache->space_info;
5425 spin_lock(&space_info->lock);
5426 spin_lock(&cache->lock);
5427 if (reserve != RESERVE_FREE) {
5431 cache->reserved += num_bytes;
5432 space_info->bytes_reserved += num_bytes;
5433 if (reserve == RESERVE_ALLOC) {
5434 trace_btrfs_space_reservation(cache->fs_info,
5435 "space_info", space_info->flags,
5437 space_info->bytes_may_use -= num_bytes;
5442 space_info->bytes_readonly += num_bytes;
5443 cache->reserved -= num_bytes;
5444 space_info->bytes_reserved -= num_bytes;
5445 space_info->reservation_progress++;
5447 spin_unlock(&cache->lock);
5448 spin_unlock(&space_info->lock);
5452 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5453 struct btrfs_root *root)
5455 struct btrfs_fs_info *fs_info = root->fs_info;
5456 struct btrfs_caching_control *next;
5457 struct btrfs_caching_control *caching_ctl;
5458 struct btrfs_block_group_cache *cache;
5459 struct btrfs_space_info *space_info;
5461 down_write(&fs_info->extent_commit_sem);
5463 list_for_each_entry_safe(caching_ctl, next,
5464 &fs_info->caching_block_groups, list) {
5465 cache = caching_ctl->block_group;
5466 if (block_group_cache_done(cache)) {
5467 cache->last_byte_to_unpin = (u64)-1;
5468 list_del_init(&caching_ctl->list);
5469 put_caching_control(caching_ctl);
5471 cache->last_byte_to_unpin = caching_ctl->progress;
5475 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5476 fs_info->pinned_extents = &fs_info->freed_extents[1];
5478 fs_info->pinned_extents = &fs_info->freed_extents[0];
5480 up_write(&fs_info->extent_commit_sem);
5482 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5483 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5485 update_global_block_rsv(fs_info);
5488 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5490 struct btrfs_fs_info *fs_info = root->fs_info;
5491 struct btrfs_block_group_cache *cache = NULL;
5492 struct btrfs_space_info *space_info;
5493 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5497 while (start <= end) {
5500 start >= cache->key.objectid + cache->key.offset) {
5502 btrfs_put_block_group(cache);
5503 cache = btrfs_lookup_block_group(fs_info, start);
5504 BUG_ON(!cache); /* Logic error */
5507 len = cache->key.objectid + cache->key.offset - start;
5508 len = min(len, end + 1 - start);
5510 if (start < cache->last_byte_to_unpin) {
5511 len = min(len, cache->last_byte_to_unpin - start);
5512 btrfs_add_free_space(cache, start, len);
5516 space_info = cache->space_info;
5518 spin_lock(&space_info->lock);
5519 spin_lock(&cache->lock);
5520 cache->pinned -= len;
5521 space_info->bytes_pinned -= len;
5523 space_info->bytes_readonly += len;
5526 spin_unlock(&cache->lock);
5527 if (!readonly && global_rsv->space_info == space_info) {
5528 spin_lock(&global_rsv->lock);
5529 if (!global_rsv->full) {
5530 len = min(len, global_rsv->size -
5531 global_rsv->reserved);
5532 global_rsv->reserved += len;
5533 space_info->bytes_may_use += len;
5534 if (global_rsv->reserved >= global_rsv->size)
5535 global_rsv->full = 1;
5537 spin_unlock(&global_rsv->lock);
5539 spin_unlock(&space_info->lock);
5543 btrfs_put_block_group(cache);
5547 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5548 struct btrfs_root *root)
5550 struct btrfs_fs_info *fs_info = root->fs_info;
5551 struct extent_io_tree *unpin;
5559 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5560 unpin = &fs_info->freed_extents[1];
5562 unpin = &fs_info->freed_extents[0];
5565 ret = find_first_extent_bit(unpin, 0, &start, &end,
5566 EXTENT_DIRTY, NULL);
5570 if (btrfs_test_opt(root, DISCARD))
5571 ret = btrfs_discard_extent(root, start,
5572 end + 1 - start, NULL);
5574 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5575 unpin_extent_range(root, start, end);
5582 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5583 u64 owner, u64 root_objectid)
5585 struct btrfs_space_info *space_info;
5588 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5589 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5590 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5592 flags = BTRFS_BLOCK_GROUP_METADATA;
5594 flags = BTRFS_BLOCK_GROUP_DATA;
5597 space_info = __find_space_info(fs_info, flags);
5598 BUG_ON(!space_info); /* Logic bug */
5599 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5603 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5604 struct btrfs_root *root,
5605 u64 bytenr, u64 num_bytes, u64 parent,
5606 u64 root_objectid, u64 owner_objectid,
5607 u64 owner_offset, int refs_to_drop,
5608 struct btrfs_delayed_extent_op *extent_op)
5610 struct btrfs_key key;
5611 struct btrfs_path *path;
5612 struct btrfs_fs_info *info = root->fs_info;
5613 struct btrfs_root *extent_root = info->extent_root;
5614 struct extent_buffer *leaf;
5615 struct btrfs_extent_item *ei;
5616 struct btrfs_extent_inline_ref *iref;
5619 int extent_slot = 0;
5620 int found_extent = 0;
5624 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5627 path = btrfs_alloc_path();
5632 path->leave_spinning = 1;
5634 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5635 BUG_ON(!is_data && refs_to_drop != 1);
5638 skinny_metadata = 0;
5640 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5641 bytenr, num_bytes, parent,
5642 root_objectid, owner_objectid,
5645 extent_slot = path->slots[0];
5646 while (extent_slot >= 0) {
5647 btrfs_item_key_to_cpu(path->nodes[0], &key,
5649 if (key.objectid != bytenr)
5651 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5652 key.offset == num_bytes) {
5656 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5657 key.offset == owner_objectid) {
5661 if (path->slots[0] - extent_slot > 5)
5665 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5666 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5667 if (found_extent && item_size < sizeof(*ei))
5670 if (!found_extent) {
5672 ret = remove_extent_backref(trans, extent_root, path,
5676 btrfs_abort_transaction(trans, extent_root, ret);
5679 btrfs_release_path(path);
5680 path->leave_spinning = 1;
5682 key.objectid = bytenr;
5683 key.type = BTRFS_EXTENT_ITEM_KEY;
5684 key.offset = num_bytes;
5686 if (!is_data && skinny_metadata) {
5687 key.type = BTRFS_METADATA_ITEM_KEY;
5688 key.offset = owner_objectid;
5691 ret = btrfs_search_slot(trans, extent_root,
5693 if (ret > 0 && skinny_metadata && path->slots[0]) {
5695 * Couldn't find our skinny metadata item,
5696 * see if we have ye olde extent item.
5699 btrfs_item_key_to_cpu(path->nodes[0], &key,
5701 if (key.objectid == bytenr &&
5702 key.type == BTRFS_EXTENT_ITEM_KEY &&
5703 key.offset == num_bytes)
5707 if (ret > 0 && skinny_metadata) {
5708 skinny_metadata = false;
5709 key.type = BTRFS_EXTENT_ITEM_KEY;
5710 key.offset = num_bytes;
5711 btrfs_release_path(path);
5712 ret = btrfs_search_slot(trans, extent_root,
5717 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5718 ret, (unsigned long long)bytenr);
5720 btrfs_print_leaf(extent_root,
5724 btrfs_abort_transaction(trans, extent_root, ret);
5727 extent_slot = path->slots[0];
5729 } else if (ret == -ENOENT) {
5730 btrfs_print_leaf(extent_root, path->nodes[0]);
5733 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5734 (unsigned long long)bytenr,
5735 (unsigned long long)parent,
5736 (unsigned long long)root_objectid,
5737 (unsigned long long)owner_objectid,
5738 (unsigned long long)owner_offset);
5740 btrfs_abort_transaction(trans, extent_root, ret);
5744 leaf = path->nodes[0];
5745 item_size = btrfs_item_size_nr(leaf, extent_slot);
5746 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5747 if (item_size < sizeof(*ei)) {
5748 BUG_ON(found_extent || extent_slot != path->slots[0]);
5749 ret = convert_extent_item_v0(trans, extent_root, path,
5752 btrfs_abort_transaction(trans, extent_root, ret);
5756 btrfs_release_path(path);
5757 path->leave_spinning = 1;
5759 key.objectid = bytenr;
5760 key.type = BTRFS_EXTENT_ITEM_KEY;
5761 key.offset = num_bytes;
5763 ret = btrfs_search_slot(trans, extent_root, &key, path,
5766 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5767 ret, (unsigned long long)bytenr);
5768 btrfs_print_leaf(extent_root, path->nodes[0]);
5771 btrfs_abort_transaction(trans, extent_root, ret);
5775 extent_slot = path->slots[0];
5776 leaf = path->nodes[0];
5777 item_size = btrfs_item_size_nr(leaf, extent_slot);
5780 BUG_ON(item_size < sizeof(*ei));
5781 ei = btrfs_item_ptr(leaf, extent_slot,
5782 struct btrfs_extent_item);
5783 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5784 key.type == BTRFS_EXTENT_ITEM_KEY) {
5785 struct btrfs_tree_block_info *bi;
5786 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5787 bi = (struct btrfs_tree_block_info *)(ei + 1);
5788 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5791 refs = btrfs_extent_refs(leaf, ei);
5792 if (refs < refs_to_drop) {
5793 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5794 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5796 btrfs_abort_transaction(trans, extent_root, ret);
5799 refs -= refs_to_drop;
5803 __run_delayed_extent_op(extent_op, leaf, ei);
5805 * In the case of inline back ref, reference count will
5806 * be updated by remove_extent_backref
5809 BUG_ON(!found_extent);
5811 btrfs_set_extent_refs(leaf, ei, refs);
5812 btrfs_mark_buffer_dirty(leaf);
5815 ret = remove_extent_backref(trans, extent_root, path,
5819 btrfs_abort_transaction(trans, extent_root, ret);
5823 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5827 BUG_ON(is_data && refs_to_drop !=
5828 extent_data_ref_count(root, path, iref));
5830 BUG_ON(path->slots[0] != extent_slot);
5832 BUG_ON(path->slots[0] != extent_slot + 1);
5833 path->slots[0] = extent_slot;
5838 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5841 btrfs_abort_transaction(trans, extent_root, ret);
5844 btrfs_release_path(path);
5847 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5849 btrfs_abort_transaction(trans, extent_root, ret);
5854 ret = update_block_group(root, bytenr, num_bytes, 0);
5856 btrfs_abort_transaction(trans, extent_root, ret);
5861 btrfs_free_path(path);
5866 * when we free an block, it is possible (and likely) that we free the last
5867 * delayed ref for that extent as well. This searches the delayed ref tree for
5868 * a given extent, and if there are no other delayed refs to be processed, it
5869 * removes it from the tree.
5871 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5872 struct btrfs_root *root, u64 bytenr)
5874 struct btrfs_delayed_ref_head *head;
5875 struct btrfs_delayed_ref_root *delayed_refs;
5876 struct btrfs_delayed_ref_node *ref;
5877 struct rb_node *node;
5880 delayed_refs = &trans->transaction->delayed_refs;
5881 spin_lock(&delayed_refs->lock);
5882 head = btrfs_find_delayed_ref_head(trans, bytenr);
5886 node = rb_prev(&head->node.rb_node);
5890 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5892 /* there are still entries for this ref, we can't drop it */
5893 if (ref->bytenr == bytenr)
5896 if (head->extent_op) {
5897 if (!head->must_insert_reserved)
5899 btrfs_free_delayed_extent_op(head->extent_op);
5900 head->extent_op = NULL;
5904 * waiting for the lock here would deadlock. If someone else has it
5905 * locked they are already in the process of dropping it anyway
5907 if (!mutex_trylock(&head->mutex))
5911 * at this point we have a head with no other entries. Go
5912 * ahead and process it.
5914 head->node.in_tree = 0;
5915 rb_erase(&head->node.rb_node, &delayed_refs->root);
5917 delayed_refs->num_entries--;
5920 * we don't take a ref on the node because we're removing it from the
5921 * tree, so we just steal the ref the tree was holding.
5923 delayed_refs->num_heads--;
5924 if (list_empty(&head->cluster))
5925 delayed_refs->num_heads_ready--;
5927 list_del_init(&head->cluster);
5928 spin_unlock(&delayed_refs->lock);
5930 BUG_ON(head->extent_op);
5931 if (head->must_insert_reserved)
5934 mutex_unlock(&head->mutex);
5935 btrfs_put_delayed_ref(&head->node);
5938 spin_unlock(&delayed_refs->lock);
5942 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5943 struct btrfs_root *root,
5944 struct extent_buffer *buf,
5945 u64 parent, int last_ref)
5947 struct btrfs_block_group_cache *cache = NULL;
5951 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5952 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5953 buf->start, buf->len,
5954 parent, root->root_key.objectid,
5955 btrfs_header_level(buf),
5956 BTRFS_DROP_DELAYED_REF, NULL, 0);
5957 BUG_ON(ret); /* -ENOMEM */
5963 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5965 if (btrfs_header_generation(buf) == trans->transid) {
5966 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5967 ret = check_ref_cleanup(trans, root, buf->start);
5972 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5973 pin_down_extent(root, cache, buf->start, buf->len, 1);
5977 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5979 btrfs_add_free_space(cache, buf->start, buf->len);
5980 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5985 add_pinned_bytes(root->fs_info, buf->len,
5986 btrfs_header_level(buf),
5987 root->root_key.objectid);
5990 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5993 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5994 btrfs_put_block_group(cache);
5997 /* Can return -ENOMEM */
5998 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5999 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6000 u64 owner, u64 offset, int for_cow)
6003 struct btrfs_fs_info *fs_info = root->fs_info;
6005 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6008 * tree log blocks never actually go into the extent allocation
6009 * tree, just update pinning info and exit early.
6011 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6012 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6013 /* unlocks the pinned mutex */
6014 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6016 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6017 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6019 parent, root_objectid, (int)owner,
6020 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6022 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6024 parent, root_objectid, owner,
6025 offset, BTRFS_DROP_DELAYED_REF,
6031 static u64 stripe_align(struct btrfs_root *root,
6032 struct btrfs_block_group_cache *cache,
6033 u64 val, u64 num_bytes)
6035 u64 ret = ALIGN(val, root->stripesize);
6040 * when we wait for progress in the block group caching, its because
6041 * our allocation attempt failed at least once. So, we must sleep
6042 * and let some progress happen before we try again.
6044 * This function will sleep at least once waiting for new free space to
6045 * show up, and then it will check the block group free space numbers
6046 * for our min num_bytes. Another option is to have it go ahead
6047 * and look in the rbtree for a free extent of a given size, but this
6050 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6051 * any of the information in this block group.
6053 static noinline void
6054 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6057 struct btrfs_caching_control *caching_ctl;
6059 caching_ctl = get_caching_control(cache);
6063 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6064 (cache->free_space_ctl->free_space >= num_bytes));
6066 put_caching_control(caching_ctl);
6070 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6072 struct btrfs_caching_control *caching_ctl;
6075 caching_ctl = get_caching_control(cache);
6077 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6079 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6080 if (cache->cached == BTRFS_CACHE_ERROR)
6082 put_caching_control(caching_ctl);
6086 int __get_raid_index(u64 flags)
6088 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6089 return BTRFS_RAID_RAID10;
6090 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6091 return BTRFS_RAID_RAID1;
6092 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6093 return BTRFS_RAID_DUP;
6094 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6095 return BTRFS_RAID_RAID0;
6096 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6097 return BTRFS_RAID_RAID5;
6098 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6099 return BTRFS_RAID_RAID6;
6101 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6104 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6106 return __get_raid_index(cache->flags);
6109 enum btrfs_loop_type {
6110 LOOP_CACHING_NOWAIT = 0,
6111 LOOP_CACHING_WAIT = 1,
6112 LOOP_ALLOC_CHUNK = 2,
6113 LOOP_NO_EMPTY_SIZE = 3,
6117 * walks the btree of allocated extents and find a hole of a given size.
6118 * The key ins is changed to record the hole:
6119 * ins->objectid == block start
6120 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6121 * ins->offset == number of blocks
6122 * Any available blocks before search_start are skipped.
6124 static noinline int find_free_extent(struct btrfs_root *orig_root,
6125 u64 num_bytes, u64 empty_size,
6126 u64 hint_byte, struct btrfs_key *ins,
6130 struct btrfs_root *root = orig_root->fs_info->extent_root;
6131 struct btrfs_free_cluster *last_ptr = NULL;
6132 struct btrfs_block_group_cache *block_group = NULL;
6133 struct btrfs_block_group_cache *used_block_group;
6134 u64 search_start = 0;
6135 int empty_cluster = 2 * 1024 * 1024;
6136 struct btrfs_space_info *space_info;
6138 int index = __get_raid_index(flags);
6139 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6140 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6141 bool found_uncached_bg = false;
6142 bool failed_cluster_refill = false;
6143 bool failed_alloc = false;
6144 bool use_cluster = true;
6145 bool have_caching_bg = false;
6147 WARN_ON(num_bytes < root->sectorsize);
6148 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6152 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6154 space_info = __find_space_info(root->fs_info, flags);
6156 btrfs_err(root->fs_info, "No space info for %llu", flags);
6161 * If the space info is for both data and metadata it means we have a
6162 * small filesystem and we can't use the clustering stuff.
6164 if (btrfs_mixed_space_info(space_info))
6165 use_cluster = false;
6167 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6168 last_ptr = &root->fs_info->meta_alloc_cluster;
6169 if (!btrfs_test_opt(root, SSD))
6170 empty_cluster = 64 * 1024;
6173 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6174 btrfs_test_opt(root, SSD)) {
6175 last_ptr = &root->fs_info->data_alloc_cluster;
6179 spin_lock(&last_ptr->lock);
6180 if (last_ptr->block_group)
6181 hint_byte = last_ptr->window_start;
6182 spin_unlock(&last_ptr->lock);
6185 search_start = max(search_start, first_logical_byte(root, 0));
6186 search_start = max(search_start, hint_byte);
6191 if (search_start == hint_byte) {
6192 block_group = btrfs_lookup_block_group(root->fs_info,
6194 used_block_group = block_group;
6196 * we don't want to use the block group if it doesn't match our
6197 * allocation bits, or if its not cached.
6199 * However if we are re-searching with an ideal block group
6200 * picked out then we don't care that the block group is cached.
6202 if (block_group && block_group_bits(block_group, flags) &&
6203 block_group->cached != BTRFS_CACHE_NO) {
6204 down_read(&space_info->groups_sem);
6205 if (list_empty(&block_group->list) ||
6208 * someone is removing this block group,
6209 * we can't jump into the have_block_group
6210 * target because our list pointers are not
6213 btrfs_put_block_group(block_group);
6214 up_read(&space_info->groups_sem);
6216 index = get_block_group_index(block_group);
6217 goto have_block_group;
6219 } else if (block_group) {
6220 btrfs_put_block_group(block_group);
6224 have_caching_bg = false;
6225 down_read(&space_info->groups_sem);
6226 list_for_each_entry(block_group, &space_info->block_groups[index],
6231 used_block_group = block_group;
6232 btrfs_get_block_group(block_group);
6233 search_start = block_group->key.objectid;
6236 * this can happen if we end up cycling through all the
6237 * raid types, but we want to make sure we only allocate
6238 * for the proper type.
6240 if (!block_group_bits(block_group, flags)) {
6241 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6242 BTRFS_BLOCK_GROUP_RAID1 |
6243 BTRFS_BLOCK_GROUP_RAID5 |
6244 BTRFS_BLOCK_GROUP_RAID6 |
6245 BTRFS_BLOCK_GROUP_RAID10;
6248 * if they asked for extra copies and this block group
6249 * doesn't provide them, bail. This does allow us to
6250 * fill raid0 from raid1.
6252 if ((flags & extra) && !(block_group->flags & extra))
6257 cached = block_group_cache_done(block_group);
6258 if (unlikely(!cached)) {
6259 found_uncached_bg = true;
6260 ret = cache_block_group(block_group, 0);
6265 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6267 if (unlikely(block_group->ro))
6271 * Ok we want to try and use the cluster allocator, so
6275 unsigned long aligned_cluster;
6277 * the refill lock keeps out other
6278 * people trying to start a new cluster
6280 spin_lock(&last_ptr->refill_lock);
6281 used_block_group = last_ptr->block_group;
6282 if (used_block_group != block_group &&
6283 (!used_block_group ||
6284 used_block_group->ro ||
6285 !block_group_bits(used_block_group, flags))) {
6286 used_block_group = block_group;
6287 goto refill_cluster;
6290 if (used_block_group != block_group)
6291 btrfs_get_block_group(used_block_group);
6293 offset = btrfs_alloc_from_cluster(used_block_group,
6294 last_ptr, num_bytes, used_block_group->key.objectid);
6296 /* we have a block, we're done */
6297 spin_unlock(&last_ptr->refill_lock);
6298 trace_btrfs_reserve_extent_cluster(root,
6299 block_group, search_start, num_bytes);
6303 WARN_ON(last_ptr->block_group != used_block_group);
6304 if (used_block_group != block_group) {
6305 btrfs_put_block_group(used_block_group);
6306 used_block_group = block_group;
6309 BUG_ON(used_block_group != block_group);
6310 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6311 * set up a new clusters, so lets just skip it
6312 * and let the allocator find whatever block
6313 * it can find. If we reach this point, we
6314 * will have tried the cluster allocator
6315 * plenty of times and not have found
6316 * anything, so we are likely way too
6317 * fragmented for the clustering stuff to find
6320 * However, if the cluster is taken from the
6321 * current block group, release the cluster
6322 * first, so that we stand a better chance of
6323 * succeeding in the unclustered
6325 if (loop >= LOOP_NO_EMPTY_SIZE &&
6326 last_ptr->block_group != block_group) {
6327 spin_unlock(&last_ptr->refill_lock);
6328 goto unclustered_alloc;
6332 * this cluster didn't work out, free it and
6335 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6337 if (loop >= LOOP_NO_EMPTY_SIZE) {
6338 spin_unlock(&last_ptr->refill_lock);
6339 goto unclustered_alloc;
6342 aligned_cluster = max_t(unsigned long,
6343 empty_cluster + empty_size,
6344 block_group->full_stripe_len);
6346 /* allocate a cluster in this block group */
6347 ret = btrfs_find_space_cluster(root, block_group,
6348 last_ptr, search_start,
6353 * now pull our allocation out of this
6356 offset = btrfs_alloc_from_cluster(block_group,
6357 last_ptr, num_bytes,
6360 /* we found one, proceed */
6361 spin_unlock(&last_ptr->refill_lock);
6362 trace_btrfs_reserve_extent_cluster(root,
6363 block_group, search_start,
6367 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6368 && !failed_cluster_refill) {
6369 spin_unlock(&last_ptr->refill_lock);
6371 failed_cluster_refill = true;
6372 wait_block_group_cache_progress(block_group,
6373 num_bytes + empty_cluster + empty_size);
6374 goto have_block_group;
6378 * at this point we either didn't find a cluster
6379 * or we weren't able to allocate a block from our
6380 * cluster. Free the cluster we've been trying
6381 * to use, and go to the next block group
6383 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6384 spin_unlock(&last_ptr->refill_lock);
6389 spin_lock(&block_group->free_space_ctl->tree_lock);
6391 block_group->free_space_ctl->free_space <
6392 num_bytes + empty_cluster + empty_size) {
6393 spin_unlock(&block_group->free_space_ctl->tree_lock);
6396 spin_unlock(&block_group->free_space_ctl->tree_lock);
6398 offset = btrfs_find_space_for_alloc(block_group, search_start,
6399 num_bytes, empty_size);
6401 * If we didn't find a chunk, and we haven't failed on this
6402 * block group before, and this block group is in the middle of
6403 * caching and we are ok with waiting, then go ahead and wait
6404 * for progress to be made, and set failed_alloc to true.
6406 * If failed_alloc is true then we've already waited on this
6407 * block group once and should move on to the next block group.
6409 if (!offset && !failed_alloc && !cached &&
6410 loop > LOOP_CACHING_NOWAIT) {
6411 wait_block_group_cache_progress(block_group,
6412 num_bytes + empty_size);
6413 failed_alloc = true;
6414 goto have_block_group;
6415 } else if (!offset) {
6417 have_caching_bg = true;
6421 search_start = stripe_align(root, used_block_group,
6424 /* move on to the next group */
6425 if (search_start + num_bytes >
6426 used_block_group->key.objectid + used_block_group->key.offset) {
6427 btrfs_add_free_space(used_block_group, offset, num_bytes);
6431 if (offset < search_start)
6432 btrfs_add_free_space(used_block_group, offset,
6433 search_start - offset);
6434 BUG_ON(offset > search_start);
6436 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6438 if (ret == -EAGAIN) {
6439 btrfs_add_free_space(used_block_group, offset, num_bytes);
6443 /* we are all good, lets return */
6444 ins->objectid = search_start;
6445 ins->offset = num_bytes;
6447 trace_btrfs_reserve_extent(orig_root, block_group,
6448 search_start, num_bytes);
6449 if (used_block_group != block_group)
6450 btrfs_put_block_group(used_block_group);
6451 btrfs_put_block_group(block_group);
6454 failed_cluster_refill = false;
6455 failed_alloc = false;
6456 BUG_ON(index != get_block_group_index(block_group));
6457 if (used_block_group != block_group)
6458 btrfs_put_block_group(used_block_group);
6459 btrfs_put_block_group(block_group);
6461 up_read(&space_info->groups_sem);
6463 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6466 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6470 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6471 * caching kthreads as we move along
6472 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6473 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6474 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6477 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6480 if (loop == LOOP_ALLOC_CHUNK) {
6481 struct btrfs_trans_handle *trans;
6483 trans = btrfs_join_transaction(root);
6484 if (IS_ERR(trans)) {
6485 ret = PTR_ERR(trans);
6489 ret = do_chunk_alloc(trans, root, flags,
6492 * Do not bail out on ENOSPC since we
6493 * can do more things.
6495 if (ret < 0 && ret != -ENOSPC)
6496 btrfs_abort_transaction(trans,
6500 btrfs_end_transaction(trans, root);
6505 if (loop == LOOP_NO_EMPTY_SIZE) {
6511 } else if (!ins->objectid) {
6513 } else if (ins->objectid) {
6521 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6522 int dump_block_groups)
6524 struct btrfs_block_group_cache *cache;
6527 spin_lock(&info->lock);
6528 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6529 (unsigned long long)info->flags,
6530 (unsigned long long)(info->total_bytes - info->bytes_used -
6531 info->bytes_pinned - info->bytes_reserved -
6532 info->bytes_readonly),
6533 (info->full) ? "" : "not ");
6534 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6535 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6536 (unsigned long long)info->total_bytes,
6537 (unsigned long long)info->bytes_used,
6538 (unsigned long long)info->bytes_pinned,
6539 (unsigned long long)info->bytes_reserved,
6540 (unsigned long long)info->bytes_may_use,
6541 (unsigned long long)info->bytes_readonly);
6542 spin_unlock(&info->lock);
6544 if (!dump_block_groups)
6547 down_read(&info->groups_sem);
6549 list_for_each_entry(cache, &info->block_groups[index], list) {
6550 spin_lock(&cache->lock);
6551 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6552 (unsigned long long)cache->key.objectid,
6553 (unsigned long long)cache->key.offset,
6554 (unsigned long long)btrfs_block_group_used(&cache->item),
6555 (unsigned long long)cache->pinned,
6556 (unsigned long long)cache->reserved,
6557 cache->ro ? "[readonly]" : "");
6558 btrfs_dump_free_space(cache, bytes);
6559 spin_unlock(&cache->lock);
6561 if (++index < BTRFS_NR_RAID_TYPES)
6563 up_read(&info->groups_sem);
6566 int btrfs_reserve_extent(struct btrfs_root *root,
6567 u64 num_bytes, u64 min_alloc_size,
6568 u64 empty_size, u64 hint_byte,
6569 struct btrfs_key *ins, int is_data)
6571 bool final_tried = false;
6575 flags = btrfs_get_alloc_profile(root, is_data);
6577 WARN_ON(num_bytes < root->sectorsize);
6578 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6581 if (ret == -ENOSPC) {
6583 num_bytes = num_bytes >> 1;
6584 num_bytes = round_down(num_bytes, root->sectorsize);
6585 num_bytes = max(num_bytes, min_alloc_size);
6586 if (num_bytes == min_alloc_size)
6589 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6590 struct btrfs_space_info *sinfo;
6592 sinfo = __find_space_info(root->fs_info, flags);
6593 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6594 (unsigned long long)flags,
6595 (unsigned long long)num_bytes);
6597 dump_space_info(sinfo, num_bytes, 1);
6601 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6606 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6607 u64 start, u64 len, int pin)
6609 struct btrfs_block_group_cache *cache;
6612 cache = btrfs_lookup_block_group(root->fs_info, start);
6614 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6615 (unsigned long long)start);
6619 if (btrfs_test_opt(root, DISCARD))
6620 ret = btrfs_discard_extent(root, start, len, NULL);
6623 pin_down_extent(root, cache, start, len, 1);
6625 btrfs_add_free_space(cache, start, len);
6626 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6628 btrfs_put_block_group(cache);
6630 trace_btrfs_reserved_extent_free(root, start, len);
6635 int btrfs_free_reserved_extent(struct btrfs_root *root,
6638 return __btrfs_free_reserved_extent(root, start, len, 0);
6641 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6644 return __btrfs_free_reserved_extent(root, start, len, 1);
6647 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6648 struct btrfs_root *root,
6649 u64 parent, u64 root_objectid,
6650 u64 flags, u64 owner, u64 offset,
6651 struct btrfs_key *ins, int ref_mod)
6654 struct btrfs_fs_info *fs_info = root->fs_info;
6655 struct btrfs_extent_item *extent_item;
6656 struct btrfs_extent_inline_ref *iref;
6657 struct btrfs_path *path;
6658 struct extent_buffer *leaf;
6663 type = BTRFS_SHARED_DATA_REF_KEY;
6665 type = BTRFS_EXTENT_DATA_REF_KEY;
6667 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6669 path = btrfs_alloc_path();
6673 path->leave_spinning = 1;
6674 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6677 btrfs_free_path(path);
6681 leaf = path->nodes[0];
6682 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6683 struct btrfs_extent_item);
6684 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6685 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6686 btrfs_set_extent_flags(leaf, extent_item,
6687 flags | BTRFS_EXTENT_FLAG_DATA);
6689 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6690 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6692 struct btrfs_shared_data_ref *ref;
6693 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6694 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6695 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6697 struct btrfs_extent_data_ref *ref;
6698 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6699 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6700 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6701 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6702 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6705 btrfs_mark_buffer_dirty(path->nodes[0]);
6706 btrfs_free_path(path);
6708 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6709 if (ret) { /* -ENOENT, logic error */
6710 btrfs_err(fs_info, "update block group failed for %llu %llu",
6711 (unsigned long long)ins->objectid,
6712 (unsigned long long)ins->offset);
6718 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6719 struct btrfs_root *root,
6720 u64 parent, u64 root_objectid,
6721 u64 flags, struct btrfs_disk_key *key,
6722 int level, struct btrfs_key *ins)
6725 struct btrfs_fs_info *fs_info = root->fs_info;
6726 struct btrfs_extent_item *extent_item;
6727 struct btrfs_tree_block_info *block_info;
6728 struct btrfs_extent_inline_ref *iref;
6729 struct btrfs_path *path;
6730 struct extent_buffer *leaf;
6731 u32 size = sizeof(*extent_item) + sizeof(*iref);
6732 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6735 if (!skinny_metadata)
6736 size += sizeof(*block_info);
6738 path = btrfs_alloc_path();
6742 path->leave_spinning = 1;
6743 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6746 btrfs_free_path(path);
6750 leaf = path->nodes[0];
6751 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6752 struct btrfs_extent_item);
6753 btrfs_set_extent_refs(leaf, extent_item, 1);
6754 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6755 btrfs_set_extent_flags(leaf, extent_item,
6756 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6758 if (skinny_metadata) {
6759 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6761 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6762 btrfs_set_tree_block_key(leaf, block_info, key);
6763 btrfs_set_tree_block_level(leaf, block_info, level);
6764 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6768 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6769 btrfs_set_extent_inline_ref_type(leaf, iref,
6770 BTRFS_SHARED_BLOCK_REF_KEY);
6771 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6773 btrfs_set_extent_inline_ref_type(leaf, iref,
6774 BTRFS_TREE_BLOCK_REF_KEY);
6775 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6778 btrfs_mark_buffer_dirty(leaf);
6779 btrfs_free_path(path);
6781 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6782 if (ret) { /* -ENOENT, logic error */
6783 btrfs_err(fs_info, "update block group failed for %llu %llu",
6784 (unsigned long long)ins->objectid,
6785 (unsigned long long)ins->offset);
6791 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6792 struct btrfs_root *root,
6793 u64 root_objectid, u64 owner,
6794 u64 offset, struct btrfs_key *ins)
6798 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6800 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6802 root_objectid, owner, offset,
6803 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6808 * this is used by the tree logging recovery code. It records that
6809 * an extent has been allocated and makes sure to clear the free
6810 * space cache bits as well
6812 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6813 struct btrfs_root *root,
6814 u64 root_objectid, u64 owner, u64 offset,
6815 struct btrfs_key *ins)
6818 struct btrfs_block_group_cache *block_group;
6821 * Mixed block groups will exclude before processing the log so we only
6822 * need to do the exlude dance if this fs isn't mixed.
6824 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6825 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6830 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6834 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6835 RESERVE_ALLOC_NO_ACCOUNT);
6836 BUG_ON(ret); /* logic error */
6837 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6838 0, owner, offset, ins, 1);
6839 btrfs_put_block_group(block_group);
6843 static struct extent_buffer *
6844 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6845 u64 bytenr, u32 blocksize, int level)
6847 struct extent_buffer *buf;
6849 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6851 return ERR_PTR(-ENOMEM);
6852 btrfs_set_header_generation(buf, trans->transid);
6853 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6854 btrfs_tree_lock(buf);
6855 clean_tree_block(trans, root, buf);
6856 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6858 btrfs_set_lock_blocking(buf);
6859 btrfs_set_buffer_uptodate(buf);
6861 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6863 * we allow two log transactions at a time, use different
6864 * EXENT bit to differentiate dirty pages.
6866 if (root->log_transid % 2 == 0)
6867 set_extent_dirty(&root->dirty_log_pages, buf->start,
6868 buf->start + buf->len - 1, GFP_NOFS);
6870 set_extent_new(&root->dirty_log_pages, buf->start,
6871 buf->start + buf->len - 1, GFP_NOFS);
6873 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6874 buf->start + buf->len - 1, GFP_NOFS);
6876 trans->blocks_used++;
6877 /* this returns a buffer locked for blocking */
6881 static struct btrfs_block_rsv *
6882 use_block_rsv(struct btrfs_trans_handle *trans,
6883 struct btrfs_root *root, u32 blocksize)
6885 struct btrfs_block_rsv *block_rsv;
6886 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6888 bool global_updated = false;
6890 block_rsv = get_block_rsv(trans, root);
6892 if (unlikely(block_rsv->size == 0))
6895 ret = block_rsv_use_bytes(block_rsv, blocksize);
6899 if (block_rsv->failfast)
6900 return ERR_PTR(ret);
6902 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6903 global_updated = true;
6904 update_global_block_rsv(root->fs_info);
6908 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6909 static DEFINE_RATELIMIT_STATE(_rs,
6910 DEFAULT_RATELIMIT_INTERVAL * 10,
6911 /*DEFAULT_RATELIMIT_BURST*/ 1);
6912 if (__ratelimit(&_rs))
6914 "btrfs: block rsv returned %d\n", ret);
6917 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6918 BTRFS_RESERVE_NO_FLUSH);
6922 * If we couldn't reserve metadata bytes try and use some from
6923 * the global reserve if its space type is the same as the global
6926 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6927 block_rsv->space_info == global_rsv->space_info) {
6928 ret = block_rsv_use_bytes(global_rsv, blocksize);
6932 return ERR_PTR(ret);
6935 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6936 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6938 block_rsv_add_bytes(block_rsv, blocksize, 0);
6939 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6943 * finds a free extent and does all the dirty work required for allocation
6944 * returns the key for the extent through ins, and a tree buffer for
6945 * the first block of the extent through buf.
6947 * returns the tree buffer or NULL.
6949 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6950 struct btrfs_root *root, u32 blocksize,
6951 u64 parent, u64 root_objectid,
6952 struct btrfs_disk_key *key, int level,
6953 u64 hint, u64 empty_size)
6955 struct btrfs_key ins;
6956 struct btrfs_block_rsv *block_rsv;
6957 struct extent_buffer *buf;
6960 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6963 block_rsv = use_block_rsv(trans, root, blocksize);
6964 if (IS_ERR(block_rsv))
6965 return ERR_CAST(block_rsv);
6967 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6968 empty_size, hint, &ins, 0);
6970 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6971 return ERR_PTR(ret);
6974 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6976 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6978 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6980 parent = ins.objectid;
6981 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6985 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6986 struct btrfs_delayed_extent_op *extent_op;
6987 extent_op = btrfs_alloc_delayed_extent_op();
6988 BUG_ON(!extent_op); /* -ENOMEM */
6990 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6992 memset(&extent_op->key, 0, sizeof(extent_op->key));
6993 extent_op->flags_to_set = flags;
6994 if (skinny_metadata)
6995 extent_op->update_key = 0;
6997 extent_op->update_key = 1;
6998 extent_op->update_flags = 1;
6999 extent_op->is_data = 0;
7000 extent_op->level = level;
7002 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7004 ins.offset, parent, root_objectid,
7005 level, BTRFS_ADD_DELAYED_EXTENT,
7007 BUG_ON(ret); /* -ENOMEM */
7012 struct walk_control {
7013 u64 refs[BTRFS_MAX_LEVEL];
7014 u64 flags[BTRFS_MAX_LEVEL];
7015 struct btrfs_key update_progress;
7026 #define DROP_REFERENCE 1
7027 #define UPDATE_BACKREF 2
7029 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7030 struct btrfs_root *root,
7031 struct walk_control *wc,
7032 struct btrfs_path *path)
7040 struct btrfs_key key;
7041 struct extent_buffer *eb;
7046 if (path->slots[wc->level] < wc->reada_slot) {
7047 wc->reada_count = wc->reada_count * 2 / 3;
7048 wc->reada_count = max(wc->reada_count, 2);
7050 wc->reada_count = wc->reada_count * 3 / 2;
7051 wc->reada_count = min_t(int, wc->reada_count,
7052 BTRFS_NODEPTRS_PER_BLOCK(root));
7055 eb = path->nodes[wc->level];
7056 nritems = btrfs_header_nritems(eb);
7057 blocksize = btrfs_level_size(root, wc->level - 1);
7059 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7060 if (nread >= wc->reada_count)
7064 bytenr = btrfs_node_blockptr(eb, slot);
7065 generation = btrfs_node_ptr_generation(eb, slot);
7067 if (slot == path->slots[wc->level])
7070 if (wc->stage == UPDATE_BACKREF &&
7071 generation <= root->root_key.offset)
7074 /* We don't lock the tree block, it's OK to be racy here */
7075 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7076 wc->level - 1, 1, &refs,
7078 /* We don't care about errors in readahead. */
7083 if (wc->stage == DROP_REFERENCE) {
7087 if (wc->level == 1 &&
7088 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7090 if (!wc->update_ref ||
7091 generation <= root->root_key.offset)
7093 btrfs_node_key_to_cpu(eb, &key, slot);
7094 ret = btrfs_comp_cpu_keys(&key,
7095 &wc->update_progress);
7099 if (wc->level == 1 &&
7100 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7104 ret = readahead_tree_block(root, bytenr, blocksize,
7110 wc->reada_slot = slot;
7114 * helper to process tree block while walking down the tree.
7116 * when wc->stage == UPDATE_BACKREF, this function updates
7117 * back refs for pointers in the block.
7119 * NOTE: return value 1 means we should stop walking down.
7121 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7122 struct btrfs_root *root,
7123 struct btrfs_path *path,
7124 struct walk_control *wc, int lookup_info)
7126 int level = wc->level;
7127 struct extent_buffer *eb = path->nodes[level];
7128 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7131 if (wc->stage == UPDATE_BACKREF &&
7132 btrfs_header_owner(eb) != root->root_key.objectid)
7136 * when reference count of tree block is 1, it won't increase
7137 * again. once full backref flag is set, we never clear it.
7140 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7141 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7142 BUG_ON(!path->locks[level]);
7143 ret = btrfs_lookup_extent_info(trans, root,
7144 eb->start, level, 1,
7147 BUG_ON(ret == -ENOMEM);
7150 BUG_ON(wc->refs[level] == 0);
7153 if (wc->stage == DROP_REFERENCE) {
7154 if (wc->refs[level] > 1)
7157 if (path->locks[level] && !wc->keep_locks) {
7158 btrfs_tree_unlock_rw(eb, path->locks[level]);
7159 path->locks[level] = 0;
7164 /* wc->stage == UPDATE_BACKREF */
7165 if (!(wc->flags[level] & flag)) {
7166 BUG_ON(!path->locks[level]);
7167 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7168 BUG_ON(ret); /* -ENOMEM */
7169 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7170 BUG_ON(ret); /* -ENOMEM */
7171 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7173 btrfs_header_level(eb), 0);
7174 BUG_ON(ret); /* -ENOMEM */
7175 wc->flags[level] |= flag;
7179 * the block is shared by multiple trees, so it's not good to
7180 * keep the tree lock
7182 if (path->locks[level] && level > 0) {
7183 btrfs_tree_unlock_rw(eb, path->locks[level]);
7184 path->locks[level] = 0;
7190 * helper to process tree block pointer.
7192 * when wc->stage == DROP_REFERENCE, this function checks
7193 * reference count of the block pointed to. if the block
7194 * is shared and we need update back refs for the subtree
7195 * rooted at the block, this function changes wc->stage to
7196 * UPDATE_BACKREF. if the block is shared and there is no
7197 * need to update back, this function drops the reference
7200 * NOTE: return value 1 means we should stop walking down.
7202 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7203 struct btrfs_root *root,
7204 struct btrfs_path *path,
7205 struct walk_control *wc, int *lookup_info)
7211 struct btrfs_key key;
7212 struct extent_buffer *next;
7213 int level = wc->level;
7217 generation = btrfs_node_ptr_generation(path->nodes[level],
7218 path->slots[level]);
7220 * if the lower level block was created before the snapshot
7221 * was created, we know there is no need to update back refs
7224 if (wc->stage == UPDATE_BACKREF &&
7225 generation <= root->root_key.offset) {
7230 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7231 blocksize = btrfs_level_size(root, level - 1);
7233 next = btrfs_find_tree_block(root, bytenr, blocksize);
7235 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7238 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7242 btrfs_tree_lock(next);
7243 btrfs_set_lock_blocking(next);
7245 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7246 &wc->refs[level - 1],
7247 &wc->flags[level - 1]);
7249 btrfs_tree_unlock(next);
7253 if (unlikely(wc->refs[level - 1] == 0)) {
7254 btrfs_err(root->fs_info, "Missing references.");
7259 if (wc->stage == DROP_REFERENCE) {
7260 if (wc->refs[level - 1] > 1) {
7262 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7265 if (!wc->update_ref ||
7266 generation <= root->root_key.offset)
7269 btrfs_node_key_to_cpu(path->nodes[level], &key,
7270 path->slots[level]);
7271 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7275 wc->stage = UPDATE_BACKREF;
7276 wc->shared_level = level - 1;
7280 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7284 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7285 btrfs_tree_unlock(next);
7286 free_extent_buffer(next);
7292 if (reada && level == 1)
7293 reada_walk_down(trans, root, wc, path);
7294 next = read_tree_block(root, bytenr, blocksize, generation);
7295 if (!next || !extent_buffer_uptodate(next)) {
7296 free_extent_buffer(next);
7299 btrfs_tree_lock(next);
7300 btrfs_set_lock_blocking(next);
7304 BUG_ON(level != btrfs_header_level(next));
7305 path->nodes[level] = next;
7306 path->slots[level] = 0;
7307 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7313 wc->refs[level - 1] = 0;
7314 wc->flags[level - 1] = 0;
7315 if (wc->stage == DROP_REFERENCE) {
7316 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7317 parent = path->nodes[level]->start;
7319 BUG_ON(root->root_key.objectid !=
7320 btrfs_header_owner(path->nodes[level]));
7324 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7325 root->root_key.objectid, level - 1, 0, 0);
7326 BUG_ON(ret); /* -ENOMEM */
7328 btrfs_tree_unlock(next);
7329 free_extent_buffer(next);
7335 * helper to process tree block while walking up the tree.
7337 * when wc->stage == DROP_REFERENCE, this function drops
7338 * reference count on the block.
7340 * when wc->stage == UPDATE_BACKREF, this function changes
7341 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7342 * to UPDATE_BACKREF previously while processing the block.
7344 * NOTE: return value 1 means we should stop walking up.
7346 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7347 struct btrfs_root *root,
7348 struct btrfs_path *path,
7349 struct walk_control *wc)
7352 int level = wc->level;
7353 struct extent_buffer *eb = path->nodes[level];
7356 if (wc->stage == UPDATE_BACKREF) {
7357 BUG_ON(wc->shared_level < level);
7358 if (level < wc->shared_level)
7361 ret = find_next_key(path, level + 1, &wc->update_progress);
7365 wc->stage = DROP_REFERENCE;
7366 wc->shared_level = -1;
7367 path->slots[level] = 0;
7370 * check reference count again if the block isn't locked.
7371 * we should start walking down the tree again if reference
7374 if (!path->locks[level]) {
7376 btrfs_tree_lock(eb);
7377 btrfs_set_lock_blocking(eb);
7378 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7380 ret = btrfs_lookup_extent_info(trans, root,
7381 eb->start, level, 1,
7385 btrfs_tree_unlock_rw(eb, path->locks[level]);
7386 path->locks[level] = 0;
7389 BUG_ON(wc->refs[level] == 0);
7390 if (wc->refs[level] == 1) {
7391 btrfs_tree_unlock_rw(eb, path->locks[level]);
7392 path->locks[level] = 0;
7398 /* wc->stage == DROP_REFERENCE */
7399 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7401 if (wc->refs[level] == 1) {
7403 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7404 ret = btrfs_dec_ref(trans, root, eb, 1,
7407 ret = btrfs_dec_ref(trans, root, eb, 0,
7409 BUG_ON(ret); /* -ENOMEM */
7411 /* make block locked assertion in clean_tree_block happy */
7412 if (!path->locks[level] &&
7413 btrfs_header_generation(eb) == trans->transid) {
7414 btrfs_tree_lock(eb);
7415 btrfs_set_lock_blocking(eb);
7416 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7418 clean_tree_block(trans, root, eb);
7421 if (eb == root->node) {
7422 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7425 BUG_ON(root->root_key.objectid !=
7426 btrfs_header_owner(eb));
7428 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7429 parent = path->nodes[level + 1]->start;
7431 BUG_ON(root->root_key.objectid !=
7432 btrfs_header_owner(path->nodes[level + 1]));
7435 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7437 wc->refs[level] = 0;
7438 wc->flags[level] = 0;
7442 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7443 struct btrfs_root *root,
7444 struct btrfs_path *path,
7445 struct walk_control *wc)
7447 int level = wc->level;
7448 int lookup_info = 1;
7451 while (level >= 0) {
7452 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7459 if (path->slots[level] >=
7460 btrfs_header_nritems(path->nodes[level]))
7463 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7465 path->slots[level]++;
7474 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7475 struct btrfs_root *root,
7476 struct btrfs_path *path,
7477 struct walk_control *wc, int max_level)
7479 int level = wc->level;
7482 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7483 while (level < max_level && path->nodes[level]) {
7485 if (path->slots[level] + 1 <
7486 btrfs_header_nritems(path->nodes[level])) {
7487 path->slots[level]++;
7490 ret = walk_up_proc(trans, root, path, wc);
7494 if (path->locks[level]) {
7495 btrfs_tree_unlock_rw(path->nodes[level],
7496 path->locks[level]);
7497 path->locks[level] = 0;
7499 free_extent_buffer(path->nodes[level]);
7500 path->nodes[level] = NULL;
7508 * drop a subvolume tree.
7510 * this function traverses the tree freeing any blocks that only
7511 * referenced by the tree.
7513 * when a shared tree block is found. this function decreases its
7514 * reference count by one. if update_ref is true, this function
7515 * also make sure backrefs for the shared block and all lower level
7516 * blocks are properly updated.
7518 * If called with for_reloc == 0, may exit early with -EAGAIN
7520 int btrfs_drop_snapshot(struct btrfs_root *root,
7521 struct btrfs_block_rsv *block_rsv, int update_ref,
7524 struct btrfs_path *path;
7525 struct btrfs_trans_handle *trans;
7526 struct btrfs_root *tree_root = root->fs_info->tree_root;
7527 struct btrfs_root_item *root_item = &root->root_item;
7528 struct walk_control *wc;
7529 struct btrfs_key key;
7533 bool root_dropped = false;
7535 path = btrfs_alloc_path();
7541 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7543 btrfs_free_path(path);
7548 trans = btrfs_start_transaction(tree_root, 0);
7549 if (IS_ERR(trans)) {
7550 err = PTR_ERR(trans);
7555 trans->block_rsv = block_rsv;
7557 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7558 level = btrfs_header_level(root->node);
7559 path->nodes[level] = btrfs_lock_root_node(root);
7560 btrfs_set_lock_blocking(path->nodes[level]);
7561 path->slots[level] = 0;
7562 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7563 memset(&wc->update_progress, 0,
7564 sizeof(wc->update_progress));
7566 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7567 memcpy(&wc->update_progress, &key,
7568 sizeof(wc->update_progress));
7570 level = root_item->drop_level;
7572 path->lowest_level = level;
7573 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7574 path->lowest_level = 0;
7582 * unlock our path, this is safe because only this
7583 * function is allowed to delete this snapshot
7585 btrfs_unlock_up_safe(path, 0);
7587 level = btrfs_header_level(root->node);
7589 btrfs_tree_lock(path->nodes[level]);
7590 btrfs_set_lock_blocking(path->nodes[level]);
7591 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7593 ret = btrfs_lookup_extent_info(trans, root,
7594 path->nodes[level]->start,
7595 level, 1, &wc->refs[level],
7601 BUG_ON(wc->refs[level] == 0);
7603 if (level == root_item->drop_level)
7606 btrfs_tree_unlock(path->nodes[level]);
7607 path->locks[level] = 0;
7608 WARN_ON(wc->refs[level] != 1);
7614 wc->shared_level = -1;
7615 wc->stage = DROP_REFERENCE;
7616 wc->update_ref = update_ref;
7618 wc->for_reloc = for_reloc;
7619 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7623 ret = walk_down_tree(trans, root, path, wc);
7629 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7636 BUG_ON(wc->stage != DROP_REFERENCE);
7640 if (wc->stage == DROP_REFERENCE) {
7642 btrfs_node_key(path->nodes[level],
7643 &root_item->drop_progress,
7644 path->slots[level]);
7645 root_item->drop_level = level;
7648 BUG_ON(wc->level == 0);
7649 if (btrfs_should_end_transaction(trans, tree_root) ||
7650 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7651 ret = btrfs_update_root(trans, tree_root,
7655 btrfs_abort_transaction(trans, tree_root, ret);
7660 btrfs_end_transaction_throttle(trans, tree_root);
7661 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7662 pr_debug("btrfs: drop snapshot early exit\n");
7667 trans = btrfs_start_transaction(tree_root, 0);
7668 if (IS_ERR(trans)) {
7669 err = PTR_ERR(trans);
7673 trans->block_rsv = block_rsv;
7676 btrfs_release_path(path);
7680 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7682 btrfs_abort_transaction(trans, tree_root, ret);
7686 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7687 ret = btrfs_find_root(tree_root, &root->root_key, path,
7690 btrfs_abort_transaction(trans, tree_root, ret);
7693 } else if (ret > 0) {
7694 /* if we fail to delete the orphan item this time
7695 * around, it'll get picked up the next time.
7697 * The most common failure here is just -ENOENT.
7699 btrfs_del_orphan_item(trans, tree_root,
7700 root->root_key.objectid);
7704 if (root->in_radix) {
7705 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7707 free_extent_buffer(root->node);
7708 free_extent_buffer(root->commit_root);
7709 btrfs_put_fs_root(root);
7711 root_dropped = true;
7713 btrfs_end_transaction_throttle(trans, tree_root);
7716 btrfs_free_path(path);
7719 * So if we need to stop dropping the snapshot for whatever reason we
7720 * need to make sure to add it back to the dead root list so that we
7721 * keep trying to do the work later. This also cleans up roots if we
7722 * don't have it in the radix (like when we recover after a power fail
7723 * or unmount) so we don't leak memory.
7725 if (!for_reloc && root_dropped == false)
7726 btrfs_add_dead_root(root);
7728 btrfs_std_error(root->fs_info, err);
7733 * drop subtree rooted at tree block 'node'.
7735 * NOTE: this function will unlock and release tree block 'node'
7736 * only used by relocation code
7738 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7739 struct btrfs_root *root,
7740 struct extent_buffer *node,
7741 struct extent_buffer *parent)
7743 struct btrfs_path *path;
7744 struct walk_control *wc;
7750 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7752 path = btrfs_alloc_path();
7756 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7758 btrfs_free_path(path);
7762 btrfs_assert_tree_locked(parent);
7763 parent_level = btrfs_header_level(parent);
7764 extent_buffer_get(parent);
7765 path->nodes[parent_level] = parent;
7766 path->slots[parent_level] = btrfs_header_nritems(parent);
7768 btrfs_assert_tree_locked(node);
7769 level = btrfs_header_level(node);
7770 path->nodes[level] = node;
7771 path->slots[level] = 0;
7772 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7774 wc->refs[parent_level] = 1;
7775 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7777 wc->shared_level = -1;
7778 wc->stage = DROP_REFERENCE;
7782 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7785 wret = walk_down_tree(trans, root, path, wc);
7791 wret = walk_up_tree(trans, root, path, wc, parent_level);
7799 btrfs_free_path(path);
7803 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7809 * if restripe for this chunk_type is on pick target profile and
7810 * return, otherwise do the usual balance
7812 stripped = get_restripe_target(root->fs_info, flags);
7814 return extended_to_chunk(stripped);
7817 * we add in the count of missing devices because we want
7818 * to make sure that any RAID levels on a degraded FS
7819 * continue to be honored.
7821 num_devices = root->fs_info->fs_devices->rw_devices +
7822 root->fs_info->fs_devices->missing_devices;
7824 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7825 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7826 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7828 if (num_devices == 1) {
7829 stripped |= BTRFS_BLOCK_GROUP_DUP;
7830 stripped = flags & ~stripped;
7832 /* turn raid0 into single device chunks */
7833 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7836 /* turn mirroring into duplication */
7837 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7838 BTRFS_BLOCK_GROUP_RAID10))
7839 return stripped | BTRFS_BLOCK_GROUP_DUP;
7841 /* they already had raid on here, just return */
7842 if (flags & stripped)
7845 stripped |= BTRFS_BLOCK_GROUP_DUP;
7846 stripped = flags & ~stripped;
7848 /* switch duplicated blocks with raid1 */
7849 if (flags & BTRFS_BLOCK_GROUP_DUP)
7850 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7852 /* this is drive concat, leave it alone */
7858 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7860 struct btrfs_space_info *sinfo = cache->space_info;
7862 u64 min_allocable_bytes;
7867 * We need some metadata space and system metadata space for
7868 * allocating chunks in some corner cases until we force to set
7869 * it to be readonly.
7872 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7874 min_allocable_bytes = 1 * 1024 * 1024;
7876 min_allocable_bytes = 0;
7878 spin_lock(&sinfo->lock);
7879 spin_lock(&cache->lock);
7886 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7887 cache->bytes_super - btrfs_block_group_used(&cache->item);
7889 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7890 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7891 min_allocable_bytes <= sinfo->total_bytes) {
7892 sinfo->bytes_readonly += num_bytes;
7897 spin_unlock(&cache->lock);
7898 spin_unlock(&sinfo->lock);
7902 int btrfs_set_block_group_ro(struct btrfs_root *root,
7903 struct btrfs_block_group_cache *cache)
7906 struct btrfs_trans_handle *trans;
7912 trans = btrfs_join_transaction(root);
7914 return PTR_ERR(trans);
7916 alloc_flags = update_block_group_flags(root, cache->flags);
7917 if (alloc_flags != cache->flags) {
7918 ret = do_chunk_alloc(trans, root, alloc_flags,
7924 ret = set_block_group_ro(cache, 0);
7927 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7928 ret = do_chunk_alloc(trans, root, alloc_flags,
7932 ret = set_block_group_ro(cache, 0);
7934 btrfs_end_transaction(trans, root);
7938 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7939 struct btrfs_root *root, u64 type)
7941 u64 alloc_flags = get_alloc_profile(root, type);
7942 return do_chunk_alloc(trans, root, alloc_flags,
7947 * helper to account the unused space of all the readonly block group in the
7948 * list. takes mirrors into account.
7950 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7952 struct btrfs_block_group_cache *block_group;
7956 list_for_each_entry(block_group, groups_list, list) {
7957 spin_lock(&block_group->lock);
7959 if (!block_group->ro) {
7960 spin_unlock(&block_group->lock);
7964 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7965 BTRFS_BLOCK_GROUP_RAID10 |
7966 BTRFS_BLOCK_GROUP_DUP))
7971 free_bytes += (block_group->key.offset -
7972 btrfs_block_group_used(&block_group->item)) *
7975 spin_unlock(&block_group->lock);
7982 * helper to account the unused space of all the readonly block group in the
7983 * space_info. takes mirrors into account.
7985 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7990 spin_lock(&sinfo->lock);
7992 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7993 if (!list_empty(&sinfo->block_groups[i]))
7994 free_bytes += __btrfs_get_ro_block_group_free_space(
7995 &sinfo->block_groups[i]);
7997 spin_unlock(&sinfo->lock);
8002 void btrfs_set_block_group_rw(struct btrfs_root *root,
8003 struct btrfs_block_group_cache *cache)
8005 struct btrfs_space_info *sinfo = cache->space_info;
8010 spin_lock(&sinfo->lock);
8011 spin_lock(&cache->lock);
8012 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8013 cache->bytes_super - btrfs_block_group_used(&cache->item);
8014 sinfo->bytes_readonly -= num_bytes;
8016 spin_unlock(&cache->lock);
8017 spin_unlock(&sinfo->lock);
8021 * checks to see if its even possible to relocate this block group.
8023 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8024 * ok to go ahead and try.
8026 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8028 struct btrfs_block_group_cache *block_group;
8029 struct btrfs_space_info *space_info;
8030 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8031 struct btrfs_device *device;
8032 struct btrfs_trans_handle *trans;
8041 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8043 /* odd, couldn't find the block group, leave it alone */
8047 min_free = btrfs_block_group_used(&block_group->item);
8049 /* no bytes used, we're good */
8053 space_info = block_group->space_info;
8054 spin_lock(&space_info->lock);
8056 full = space_info->full;
8059 * if this is the last block group we have in this space, we can't
8060 * relocate it unless we're able to allocate a new chunk below.
8062 * Otherwise, we need to make sure we have room in the space to handle
8063 * all of the extents from this block group. If we can, we're good
8065 if ((space_info->total_bytes != block_group->key.offset) &&
8066 (space_info->bytes_used + space_info->bytes_reserved +
8067 space_info->bytes_pinned + space_info->bytes_readonly +
8068 min_free < space_info->total_bytes)) {
8069 spin_unlock(&space_info->lock);
8072 spin_unlock(&space_info->lock);
8075 * ok we don't have enough space, but maybe we have free space on our
8076 * devices to allocate new chunks for relocation, so loop through our
8077 * alloc devices and guess if we have enough space. if this block
8078 * group is going to be restriped, run checks against the target
8079 * profile instead of the current one.
8091 target = get_restripe_target(root->fs_info, block_group->flags);
8093 index = __get_raid_index(extended_to_chunk(target));
8096 * this is just a balance, so if we were marked as full
8097 * we know there is no space for a new chunk
8102 index = get_block_group_index(block_group);
8105 if (index == BTRFS_RAID_RAID10) {
8109 } else if (index == BTRFS_RAID_RAID1) {
8111 } else if (index == BTRFS_RAID_DUP) {
8114 } else if (index == BTRFS_RAID_RAID0) {
8115 dev_min = fs_devices->rw_devices;
8116 do_div(min_free, dev_min);
8119 /* We need to do this so that we can look at pending chunks */
8120 trans = btrfs_join_transaction(root);
8121 if (IS_ERR(trans)) {
8122 ret = PTR_ERR(trans);
8126 mutex_lock(&root->fs_info->chunk_mutex);
8127 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8131 * check to make sure we can actually find a chunk with enough
8132 * space to fit our block group in.
8134 if (device->total_bytes > device->bytes_used + min_free &&
8135 !device->is_tgtdev_for_dev_replace) {
8136 ret = find_free_dev_extent(trans, device, min_free,
8141 if (dev_nr >= dev_min)
8147 mutex_unlock(&root->fs_info->chunk_mutex);
8148 btrfs_end_transaction(trans, root);
8150 btrfs_put_block_group(block_group);
8154 static int find_first_block_group(struct btrfs_root *root,
8155 struct btrfs_path *path, struct btrfs_key *key)
8158 struct btrfs_key found_key;
8159 struct extent_buffer *leaf;
8162 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8167 slot = path->slots[0];
8168 leaf = path->nodes[0];
8169 if (slot >= btrfs_header_nritems(leaf)) {
8170 ret = btrfs_next_leaf(root, path);
8177 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8179 if (found_key.objectid >= key->objectid &&
8180 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8190 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8192 struct btrfs_block_group_cache *block_group;
8196 struct inode *inode;
8198 block_group = btrfs_lookup_first_block_group(info, last);
8199 while (block_group) {
8200 spin_lock(&block_group->lock);
8201 if (block_group->iref)
8203 spin_unlock(&block_group->lock);
8204 block_group = next_block_group(info->tree_root,
8214 inode = block_group->inode;
8215 block_group->iref = 0;
8216 block_group->inode = NULL;
8217 spin_unlock(&block_group->lock);
8219 last = block_group->key.objectid + block_group->key.offset;
8220 btrfs_put_block_group(block_group);
8224 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8226 struct btrfs_block_group_cache *block_group;
8227 struct btrfs_space_info *space_info;
8228 struct btrfs_caching_control *caching_ctl;
8231 down_write(&info->extent_commit_sem);
8232 while (!list_empty(&info->caching_block_groups)) {
8233 caching_ctl = list_entry(info->caching_block_groups.next,
8234 struct btrfs_caching_control, list);
8235 list_del(&caching_ctl->list);
8236 put_caching_control(caching_ctl);
8238 up_write(&info->extent_commit_sem);
8240 spin_lock(&info->block_group_cache_lock);
8241 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8242 block_group = rb_entry(n, struct btrfs_block_group_cache,
8244 rb_erase(&block_group->cache_node,
8245 &info->block_group_cache_tree);
8246 spin_unlock(&info->block_group_cache_lock);
8248 down_write(&block_group->space_info->groups_sem);
8249 list_del(&block_group->list);
8250 up_write(&block_group->space_info->groups_sem);
8252 if (block_group->cached == BTRFS_CACHE_STARTED)
8253 wait_block_group_cache_done(block_group);
8256 * We haven't cached this block group, which means we could
8257 * possibly have excluded extents on this block group.
8259 if (block_group->cached == BTRFS_CACHE_NO ||
8260 block_group->cached == BTRFS_CACHE_ERROR)
8261 free_excluded_extents(info->extent_root, block_group);
8263 btrfs_remove_free_space_cache(block_group);
8264 btrfs_put_block_group(block_group);
8266 spin_lock(&info->block_group_cache_lock);
8268 spin_unlock(&info->block_group_cache_lock);
8270 /* now that all the block groups are freed, go through and
8271 * free all the space_info structs. This is only called during
8272 * the final stages of unmount, and so we know nobody is
8273 * using them. We call synchronize_rcu() once before we start,
8274 * just to be on the safe side.
8278 release_global_block_rsv(info);
8280 while(!list_empty(&info->space_info)) {
8281 space_info = list_entry(info->space_info.next,
8282 struct btrfs_space_info,
8284 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8285 if (space_info->bytes_pinned > 0 ||
8286 space_info->bytes_reserved > 0 ||
8287 space_info->bytes_may_use > 0) {
8289 dump_space_info(space_info, 0, 0);
8292 percpu_counter_destroy(&space_info->total_bytes_pinned);
8293 list_del(&space_info->list);
8299 static void __link_block_group(struct btrfs_space_info *space_info,
8300 struct btrfs_block_group_cache *cache)
8302 int index = get_block_group_index(cache);
8304 down_write(&space_info->groups_sem);
8305 list_add_tail(&cache->list, &space_info->block_groups[index]);
8306 up_write(&space_info->groups_sem);
8309 int btrfs_read_block_groups(struct btrfs_root *root)
8311 struct btrfs_path *path;
8313 struct btrfs_block_group_cache *cache;
8314 struct btrfs_fs_info *info = root->fs_info;
8315 struct btrfs_space_info *space_info;
8316 struct btrfs_key key;
8317 struct btrfs_key found_key;
8318 struct extent_buffer *leaf;
8322 root = info->extent_root;
8325 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8326 path = btrfs_alloc_path();
8331 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8332 if (btrfs_test_opt(root, SPACE_CACHE) &&
8333 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8335 if (btrfs_test_opt(root, CLEAR_CACHE))
8339 ret = find_first_block_group(root, path, &key);
8344 leaf = path->nodes[0];
8345 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8346 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8351 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8353 if (!cache->free_space_ctl) {
8359 atomic_set(&cache->count, 1);
8360 spin_lock_init(&cache->lock);
8361 cache->fs_info = info;
8362 INIT_LIST_HEAD(&cache->list);
8363 INIT_LIST_HEAD(&cache->cluster_list);
8367 * When we mount with old space cache, we need to
8368 * set BTRFS_DC_CLEAR and set dirty flag.
8370 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8371 * truncate the old free space cache inode and
8373 * b) Setting 'dirty flag' makes sure that we flush
8374 * the new space cache info onto disk.
8376 cache->disk_cache_state = BTRFS_DC_CLEAR;
8377 if (btrfs_test_opt(root, SPACE_CACHE))
8381 read_extent_buffer(leaf, &cache->item,
8382 btrfs_item_ptr_offset(leaf, path->slots[0]),
8383 sizeof(cache->item));
8384 memcpy(&cache->key, &found_key, sizeof(found_key));
8386 key.objectid = found_key.objectid + found_key.offset;
8387 btrfs_release_path(path);
8388 cache->flags = btrfs_block_group_flags(&cache->item);
8389 cache->sectorsize = root->sectorsize;
8390 cache->full_stripe_len = btrfs_full_stripe_len(root,
8391 &root->fs_info->mapping_tree,
8392 found_key.objectid);
8393 btrfs_init_free_space_ctl(cache);
8396 * We need to exclude the super stripes now so that the space
8397 * info has super bytes accounted for, otherwise we'll think
8398 * we have more space than we actually do.
8400 ret = exclude_super_stripes(root, cache);
8403 * We may have excluded something, so call this just in
8406 free_excluded_extents(root, cache);
8407 kfree(cache->free_space_ctl);
8413 * check for two cases, either we are full, and therefore
8414 * don't need to bother with the caching work since we won't
8415 * find any space, or we are empty, and we can just add all
8416 * the space in and be done with it. This saves us _alot_ of
8417 * time, particularly in the full case.
8419 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8420 cache->last_byte_to_unpin = (u64)-1;
8421 cache->cached = BTRFS_CACHE_FINISHED;
8422 free_excluded_extents(root, cache);
8423 } else if (btrfs_block_group_used(&cache->item) == 0) {
8424 cache->last_byte_to_unpin = (u64)-1;
8425 cache->cached = BTRFS_CACHE_FINISHED;
8426 add_new_free_space(cache, root->fs_info,
8428 found_key.objectid +
8430 free_excluded_extents(root, cache);
8433 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8435 btrfs_remove_free_space_cache(cache);
8436 btrfs_put_block_group(cache);
8440 ret = update_space_info(info, cache->flags, found_key.offset,
8441 btrfs_block_group_used(&cache->item),
8444 btrfs_remove_free_space_cache(cache);
8445 spin_lock(&info->block_group_cache_lock);
8446 rb_erase(&cache->cache_node,
8447 &info->block_group_cache_tree);
8448 spin_unlock(&info->block_group_cache_lock);
8449 btrfs_put_block_group(cache);
8453 cache->space_info = space_info;
8454 spin_lock(&cache->space_info->lock);
8455 cache->space_info->bytes_readonly += cache->bytes_super;
8456 spin_unlock(&cache->space_info->lock);
8458 __link_block_group(space_info, cache);
8460 set_avail_alloc_bits(root->fs_info, cache->flags);
8461 if (btrfs_chunk_readonly(root, cache->key.objectid))
8462 set_block_group_ro(cache, 1);
8465 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8466 if (!(get_alloc_profile(root, space_info->flags) &
8467 (BTRFS_BLOCK_GROUP_RAID10 |
8468 BTRFS_BLOCK_GROUP_RAID1 |
8469 BTRFS_BLOCK_GROUP_RAID5 |
8470 BTRFS_BLOCK_GROUP_RAID6 |
8471 BTRFS_BLOCK_GROUP_DUP)))
8474 * avoid allocating from un-mirrored block group if there are
8475 * mirrored block groups.
8477 list_for_each_entry(cache,
8478 &space_info->block_groups[BTRFS_RAID_RAID0],
8480 set_block_group_ro(cache, 1);
8481 list_for_each_entry(cache,
8482 &space_info->block_groups[BTRFS_RAID_SINGLE],
8484 set_block_group_ro(cache, 1);
8487 init_global_block_rsv(info);
8490 btrfs_free_path(path);
8494 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8495 struct btrfs_root *root)
8497 struct btrfs_block_group_cache *block_group, *tmp;
8498 struct btrfs_root *extent_root = root->fs_info->extent_root;
8499 struct btrfs_block_group_item item;
8500 struct btrfs_key key;
8503 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8505 list_del_init(&block_group->new_bg_list);
8510 spin_lock(&block_group->lock);
8511 memcpy(&item, &block_group->item, sizeof(item));
8512 memcpy(&key, &block_group->key, sizeof(key));
8513 spin_unlock(&block_group->lock);
8515 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8518 btrfs_abort_transaction(trans, extent_root, ret);
8519 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8520 key.objectid, key.offset);
8522 btrfs_abort_transaction(trans, extent_root, ret);
8526 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8527 struct btrfs_root *root, u64 bytes_used,
8528 u64 type, u64 chunk_objectid, u64 chunk_offset,
8532 struct btrfs_root *extent_root;
8533 struct btrfs_block_group_cache *cache;
8535 extent_root = root->fs_info->extent_root;
8537 root->fs_info->last_trans_log_full_commit = trans->transid;
8539 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8542 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8544 if (!cache->free_space_ctl) {
8549 cache->key.objectid = chunk_offset;
8550 cache->key.offset = size;
8551 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8552 cache->sectorsize = root->sectorsize;
8553 cache->fs_info = root->fs_info;
8554 cache->full_stripe_len = btrfs_full_stripe_len(root,
8555 &root->fs_info->mapping_tree,
8558 atomic_set(&cache->count, 1);
8559 spin_lock_init(&cache->lock);
8560 INIT_LIST_HEAD(&cache->list);
8561 INIT_LIST_HEAD(&cache->cluster_list);
8562 INIT_LIST_HEAD(&cache->new_bg_list);
8564 btrfs_init_free_space_ctl(cache);
8566 btrfs_set_block_group_used(&cache->item, bytes_used);
8567 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8568 cache->flags = type;
8569 btrfs_set_block_group_flags(&cache->item, type);
8571 cache->last_byte_to_unpin = (u64)-1;
8572 cache->cached = BTRFS_CACHE_FINISHED;
8573 ret = exclude_super_stripes(root, cache);
8576 * We may have excluded something, so call this just in
8579 free_excluded_extents(root, cache);
8580 kfree(cache->free_space_ctl);
8585 add_new_free_space(cache, root->fs_info, chunk_offset,
8586 chunk_offset + size);
8588 free_excluded_extents(root, cache);
8590 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8592 btrfs_remove_free_space_cache(cache);
8593 btrfs_put_block_group(cache);
8597 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8598 &cache->space_info);
8600 btrfs_remove_free_space_cache(cache);
8601 spin_lock(&root->fs_info->block_group_cache_lock);
8602 rb_erase(&cache->cache_node,
8603 &root->fs_info->block_group_cache_tree);
8604 spin_unlock(&root->fs_info->block_group_cache_lock);
8605 btrfs_put_block_group(cache);
8608 update_global_block_rsv(root->fs_info);
8610 spin_lock(&cache->space_info->lock);
8611 cache->space_info->bytes_readonly += cache->bytes_super;
8612 spin_unlock(&cache->space_info->lock);
8614 __link_block_group(cache->space_info, cache);
8616 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8618 set_avail_alloc_bits(extent_root->fs_info, type);
8623 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8625 u64 extra_flags = chunk_to_extended(flags) &
8626 BTRFS_EXTENDED_PROFILE_MASK;
8628 write_seqlock(&fs_info->profiles_lock);
8629 if (flags & BTRFS_BLOCK_GROUP_DATA)
8630 fs_info->avail_data_alloc_bits &= ~extra_flags;
8631 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8632 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8633 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8634 fs_info->avail_system_alloc_bits &= ~extra_flags;
8635 write_sequnlock(&fs_info->profiles_lock);
8638 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8639 struct btrfs_root *root, u64 group_start)
8641 struct btrfs_path *path;
8642 struct btrfs_block_group_cache *block_group;
8643 struct btrfs_free_cluster *cluster;
8644 struct btrfs_root *tree_root = root->fs_info->tree_root;
8645 struct btrfs_key key;
8646 struct inode *inode;
8651 root = root->fs_info->extent_root;
8653 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8654 BUG_ON(!block_group);
8655 BUG_ON(!block_group->ro);
8658 * Free the reserved super bytes from this block group before
8661 free_excluded_extents(root, block_group);
8663 memcpy(&key, &block_group->key, sizeof(key));
8664 index = get_block_group_index(block_group);
8665 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8666 BTRFS_BLOCK_GROUP_RAID1 |
8667 BTRFS_BLOCK_GROUP_RAID10))
8672 /* make sure this block group isn't part of an allocation cluster */
8673 cluster = &root->fs_info->data_alloc_cluster;
8674 spin_lock(&cluster->refill_lock);
8675 btrfs_return_cluster_to_free_space(block_group, cluster);
8676 spin_unlock(&cluster->refill_lock);
8679 * make sure this block group isn't part of a metadata
8680 * allocation cluster
8682 cluster = &root->fs_info->meta_alloc_cluster;
8683 spin_lock(&cluster->refill_lock);
8684 btrfs_return_cluster_to_free_space(block_group, cluster);
8685 spin_unlock(&cluster->refill_lock);
8687 path = btrfs_alloc_path();
8693 inode = lookup_free_space_inode(tree_root, block_group, path);
8694 if (!IS_ERR(inode)) {
8695 ret = btrfs_orphan_add(trans, inode);
8697 btrfs_add_delayed_iput(inode);
8701 /* One for the block groups ref */
8702 spin_lock(&block_group->lock);
8703 if (block_group->iref) {
8704 block_group->iref = 0;
8705 block_group->inode = NULL;
8706 spin_unlock(&block_group->lock);
8709 spin_unlock(&block_group->lock);
8711 /* One for our lookup ref */
8712 btrfs_add_delayed_iput(inode);
8715 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8716 key.offset = block_group->key.objectid;
8719 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8723 btrfs_release_path(path);
8725 ret = btrfs_del_item(trans, tree_root, path);
8728 btrfs_release_path(path);
8731 spin_lock(&root->fs_info->block_group_cache_lock);
8732 rb_erase(&block_group->cache_node,
8733 &root->fs_info->block_group_cache_tree);
8735 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8736 root->fs_info->first_logical_byte = (u64)-1;
8737 spin_unlock(&root->fs_info->block_group_cache_lock);
8739 down_write(&block_group->space_info->groups_sem);
8741 * we must use list_del_init so people can check to see if they
8742 * are still on the list after taking the semaphore
8744 list_del_init(&block_group->list);
8745 if (list_empty(&block_group->space_info->block_groups[index]))
8746 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8747 up_write(&block_group->space_info->groups_sem);
8749 if (block_group->cached == BTRFS_CACHE_STARTED)
8750 wait_block_group_cache_done(block_group);
8752 btrfs_remove_free_space_cache(block_group);
8754 spin_lock(&block_group->space_info->lock);
8755 block_group->space_info->total_bytes -= block_group->key.offset;
8756 block_group->space_info->bytes_readonly -= block_group->key.offset;
8757 block_group->space_info->disk_total -= block_group->key.offset * factor;
8758 spin_unlock(&block_group->space_info->lock);
8760 memcpy(&key, &block_group->key, sizeof(key));
8762 btrfs_clear_space_info_full(root->fs_info);
8764 btrfs_put_block_group(block_group);
8765 btrfs_put_block_group(block_group);
8767 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8773 ret = btrfs_del_item(trans, root, path);
8775 btrfs_free_path(path);
8779 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8781 struct btrfs_space_info *space_info;
8782 struct btrfs_super_block *disk_super;
8788 disk_super = fs_info->super_copy;
8789 if (!btrfs_super_root(disk_super))
8792 features = btrfs_super_incompat_flags(disk_super);
8793 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8796 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8797 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8802 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8803 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8805 flags = BTRFS_BLOCK_GROUP_METADATA;
8806 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8810 flags = BTRFS_BLOCK_GROUP_DATA;
8811 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8817 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8819 return unpin_extent_range(root, start, end);
8822 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8823 u64 num_bytes, u64 *actual_bytes)
8825 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8828 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8830 struct btrfs_fs_info *fs_info = root->fs_info;
8831 struct btrfs_block_group_cache *cache = NULL;
8836 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8840 * try to trim all FS space, our block group may start from non-zero.
8842 if (range->len == total_bytes)
8843 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8845 cache = btrfs_lookup_block_group(fs_info, range->start);
8848 if (cache->key.objectid >= (range->start + range->len)) {
8849 btrfs_put_block_group(cache);
8853 start = max(range->start, cache->key.objectid);
8854 end = min(range->start + range->len,
8855 cache->key.objectid + cache->key.offset);
8857 if (end - start >= range->minlen) {
8858 if (!block_group_cache_done(cache)) {
8859 ret = cache_block_group(cache, 0);
8861 btrfs_put_block_group(cache);
8864 ret = wait_block_group_cache_done(cache);
8866 btrfs_put_block_group(cache);
8870 ret = btrfs_trim_block_group(cache,
8876 trimmed += group_trimmed;
8878 btrfs_put_block_group(cache);
8883 cache = next_block_group(fs_info->tree_root, cache);
8886 range->len = trimmed;