2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
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 path->skip_locking = 1;
773 path->search_commit_root = 1;
777 key.objectid = bytenr;
780 key.type = BTRFS_METADATA_ITEM_KEY;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
785 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
790 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
791 if (path->slots[0]) {
793 btrfs_item_key_to_cpu(path->nodes[0], &key,
795 if (key.objectid == bytenr &&
796 key.type == BTRFS_EXTENT_ITEM_KEY &&
797 key.offset == root->leafsize)
801 key.objectid = bytenr;
802 key.type = BTRFS_EXTENT_ITEM_KEY;
803 key.offset = root->leafsize;
804 btrfs_release_path(path);
810 leaf = path->nodes[0];
811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
812 if (item_size >= sizeof(*ei)) {
813 ei = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_extent_item);
815 num_refs = btrfs_extent_refs(leaf, ei);
816 extent_flags = btrfs_extent_flags(leaf, ei);
818 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
819 struct btrfs_extent_item_v0 *ei0;
820 BUG_ON(item_size != sizeof(*ei0));
821 ei0 = btrfs_item_ptr(leaf, path->slots[0],
822 struct btrfs_extent_item_v0);
823 num_refs = btrfs_extent_refs_v0(leaf, ei0);
824 /* FIXME: this isn't correct for data */
825 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
830 BUG_ON(num_refs == 0);
840 delayed_refs = &trans->transaction->delayed_refs;
841 spin_lock(&delayed_refs->lock);
842 head = btrfs_find_delayed_ref_head(trans, bytenr);
844 if (!mutex_trylock(&head->mutex)) {
845 atomic_inc(&head->node.refs);
846 spin_unlock(&delayed_refs->lock);
848 btrfs_release_path(path);
851 * Mutex was contended, block until it's released and try
854 mutex_lock(&head->mutex);
855 mutex_unlock(&head->mutex);
856 btrfs_put_delayed_ref(&head->node);
859 if (head->extent_op && head->extent_op->update_flags)
860 extent_flags |= head->extent_op->flags_to_set;
862 BUG_ON(num_refs == 0);
864 num_refs += head->node.ref_mod;
865 mutex_unlock(&head->mutex);
867 spin_unlock(&delayed_refs->lock);
869 WARN_ON(num_refs == 0);
873 *flags = extent_flags;
875 btrfs_free_path(path);
880 * Back reference rules. Back refs have three main goals:
882 * 1) differentiate between all holders of references to an extent so that
883 * when a reference is dropped we can make sure it was a valid reference
884 * before freeing the extent.
886 * 2) Provide enough information to quickly find the holders of an extent
887 * if we notice a given block is corrupted or bad.
889 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
890 * maintenance. This is actually the same as #2, but with a slightly
891 * different use case.
893 * There are two kinds of back refs. The implicit back refs is optimized
894 * for pointers in non-shared tree blocks. For a given pointer in a block,
895 * back refs of this kind provide information about the block's owner tree
896 * and the pointer's key. These information allow us to find the block by
897 * b-tree searching. The full back refs is for pointers in tree blocks not
898 * referenced by their owner trees. The location of tree block is recorded
899 * in the back refs. Actually the full back refs is generic, and can be
900 * used in all cases the implicit back refs is used. The major shortcoming
901 * of the full back refs is its overhead. Every time a tree block gets
902 * COWed, we have to update back refs entry for all pointers in it.
904 * For a newly allocated tree block, we use implicit back refs for
905 * pointers in it. This means most tree related operations only involve
906 * implicit back refs. For a tree block created in old transaction, the
907 * only way to drop a reference to it is COW it. So we can detect the
908 * event that tree block loses its owner tree's reference and do the
909 * back refs conversion.
911 * When a tree block is COW'd through a tree, there are four cases:
913 * The reference count of the block is one and the tree is the block's
914 * owner tree. Nothing to do in this case.
916 * The reference count of the block is one and the tree is not the
917 * block's owner tree. In this case, full back refs is used for pointers
918 * in the block. Remove these full back refs, add implicit back refs for
919 * every pointers in the new block.
921 * The reference count of the block is greater than one and the tree is
922 * the block's owner tree. In this case, implicit back refs is used for
923 * pointers in the block. Add full back refs for every pointers in the
924 * block, increase lower level extents' reference counts. The original
925 * implicit back refs are entailed to the new block.
927 * The reference count of the block is greater than one and the tree is
928 * not the block's owner tree. Add implicit back refs for every pointer in
929 * the new block, increase lower level extents' reference count.
931 * Back Reference Key composing:
933 * The key objectid corresponds to the first byte in the extent,
934 * The key type is used to differentiate between types of back refs.
935 * There are different meanings of the key offset for different types
938 * File extents can be referenced by:
940 * - multiple snapshots, subvolumes, or different generations in one subvol
941 * - different files inside a single subvolume
942 * - different offsets inside a file (bookend extents in file.c)
944 * The extent ref structure for the implicit back refs has fields for:
946 * - Objectid of the subvolume root
947 * - objectid of the file holding the reference
948 * - original offset in the file
949 * - how many bookend extents
951 * The key offset for the implicit back refs is hash of the first
954 * The extent ref structure for the full back refs has field for:
956 * - number of pointers in the tree leaf
958 * The key offset for the implicit back refs is the first byte of
961 * When a file extent is allocated, The implicit back refs is used.
962 * the fields are filled in:
964 * (root_key.objectid, inode objectid, offset in file, 1)
966 * When a file extent is removed file truncation, we find the
967 * corresponding implicit back refs and check the following fields:
969 * (btrfs_header_owner(leaf), inode objectid, offset in file)
971 * Btree extents can be referenced by:
973 * - Different subvolumes
975 * Both the implicit back refs and the full back refs for tree blocks
976 * only consist of key. The key offset for the implicit back refs is
977 * objectid of block's owner tree. The key offset for the full back refs
978 * is the first byte of parent block.
980 * When implicit back refs is used, information about the lowest key and
981 * level of the tree block are required. These information are stored in
982 * tree block info structure.
985 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
986 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root,
988 struct btrfs_path *path,
989 u64 owner, u32 extra_size)
991 struct btrfs_extent_item *item;
992 struct btrfs_extent_item_v0 *ei0;
993 struct btrfs_extent_ref_v0 *ref0;
994 struct btrfs_tree_block_info *bi;
995 struct extent_buffer *leaf;
996 struct btrfs_key key;
997 struct btrfs_key found_key;
998 u32 new_size = sizeof(*item);
1002 leaf = path->nodes[0];
1003 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1005 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1006 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1007 struct btrfs_extent_item_v0);
1008 refs = btrfs_extent_refs_v0(leaf, ei0);
1010 if (owner == (u64)-1) {
1012 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1013 ret = btrfs_next_leaf(root, path);
1016 BUG_ON(ret > 0); /* Corruption */
1017 leaf = path->nodes[0];
1019 btrfs_item_key_to_cpu(leaf, &found_key,
1021 BUG_ON(key.objectid != found_key.objectid);
1022 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1026 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_extent_ref_v0);
1028 owner = btrfs_ref_objectid_v0(leaf, ref0);
1032 btrfs_release_path(path);
1034 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1035 new_size += sizeof(*bi);
1037 new_size -= sizeof(*ei0);
1038 ret = btrfs_search_slot(trans, root, &key, path,
1039 new_size + extra_size, 1);
1042 BUG_ON(ret); /* Corruption */
1044 btrfs_extend_item(root, path, new_size);
1046 leaf = path->nodes[0];
1047 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1048 btrfs_set_extent_refs(leaf, item, refs);
1049 /* FIXME: get real generation */
1050 btrfs_set_extent_generation(leaf, item, 0);
1051 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1052 btrfs_set_extent_flags(leaf, item,
1053 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1054 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1055 bi = (struct btrfs_tree_block_info *)(item + 1);
1056 /* FIXME: get first key of the block */
1057 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1058 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1060 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1062 btrfs_mark_buffer_dirty(leaf);
1067 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1069 u32 high_crc = ~(u32)0;
1070 u32 low_crc = ~(u32)0;
1073 lenum = cpu_to_le64(root_objectid);
1074 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1075 lenum = cpu_to_le64(owner);
1076 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1077 lenum = cpu_to_le64(offset);
1078 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1080 return ((u64)high_crc << 31) ^ (u64)low_crc;
1083 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1084 struct btrfs_extent_data_ref *ref)
1086 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1087 btrfs_extent_data_ref_objectid(leaf, ref),
1088 btrfs_extent_data_ref_offset(leaf, ref));
1091 static int match_extent_data_ref(struct extent_buffer *leaf,
1092 struct btrfs_extent_data_ref *ref,
1093 u64 root_objectid, u64 owner, u64 offset)
1095 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1096 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1097 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1102 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1103 struct btrfs_root *root,
1104 struct btrfs_path *path,
1105 u64 bytenr, u64 parent,
1107 u64 owner, u64 offset)
1109 struct btrfs_key key;
1110 struct btrfs_extent_data_ref *ref;
1111 struct extent_buffer *leaf;
1117 key.objectid = bytenr;
1119 key.type = BTRFS_SHARED_DATA_REF_KEY;
1120 key.offset = parent;
1122 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1123 key.offset = hash_extent_data_ref(root_objectid,
1128 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1138 key.type = BTRFS_EXTENT_REF_V0_KEY;
1139 btrfs_release_path(path);
1140 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1151 leaf = path->nodes[0];
1152 nritems = btrfs_header_nritems(leaf);
1154 if (path->slots[0] >= nritems) {
1155 ret = btrfs_next_leaf(root, path);
1161 leaf = path->nodes[0];
1162 nritems = btrfs_header_nritems(leaf);
1166 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1167 if (key.objectid != bytenr ||
1168 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1171 ref = btrfs_item_ptr(leaf, path->slots[0],
1172 struct btrfs_extent_data_ref);
1174 if (match_extent_data_ref(leaf, ref, root_objectid,
1177 btrfs_release_path(path);
1189 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1190 struct btrfs_root *root,
1191 struct btrfs_path *path,
1192 u64 bytenr, u64 parent,
1193 u64 root_objectid, u64 owner,
1194 u64 offset, int refs_to_add)
1196 struct btrfs_key key;
1197 struct extent_buffer *leaf;
1202 key.objectid = bytenr;
1204 key.type = BTRFS_SHARED_DATA_REF_KEY;
1205 key.offset = parent;
1206 size = sizeof(struct btrfs_shared_data_ref);
1208 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1209 key.offset = hash_extent_data_ref(root_objectid,
1211 size = sizeof(struct btrfs_extent_data_ref);
1214 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1215 if (ret && ret != -EEXIST)
1218 leaf = path->nodes[0];
1220 struct btrfs_shared_data_ref *ref;
1221 ref = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_shared_data_ref);
1224 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1226 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1231 struct btrfs_extent_data_ref *ref;
1232 while (ret == -EEXIST) {
1233 ref = btrfs_item_ptr(leaf, path->slots[0],
1234 struct btrfs_extent_data_ref);
1235 if (match_extent_data_ref(leaf, ref, root_objectid,
1238 btrfs_release_path(path);
1240 ret = btrfs_insert_empty_item(trans, root, path, &key,
1242 if (ret && ret != -EEXIST)
1245 leaf = path->nodes[0];
1247 ref = btrfs_item_ptr(leaf, path->slots[0],
1248 struct btrfs_extent_data_ref);
1250 btrfs_set_extent_data_ref_root(leaf, ref,
1252 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1253 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1254 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1257 num_refs += refs_to_add;
1258 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1261 btrfs_mark_buffer_dirty(leaf);
1264 btrfs_release_path(path);
1268 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1269 struct btrfs_root *root,
1270 struct btrfs_path *path,
1273 struct btrfs_key key;
1274 struct btrfs_extent_data_ref *ref1 = NULL;
1275 struct btrfs_shared_data_ref *ref2 = NULL;
1276 struct extent_buffer *leaf;
1280 leaf = path->nodes[0];
1281 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1283 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1284 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_data_ref);
1286 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1287 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1288 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_shared_data_ref);
1290 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1291 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1292 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1293 struct btrfs_extent_ref_v0 *ref0;
1294 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1295 struct btrfs_extent_ref_v0);
1296 num_refs = btrfs_ref_count_v0(leaf, ref0);
1302 BUG_ON(num_refs < refs_to_drop);
1303 num_refs -= refs_to_drop;
1305 if (num_refs == 0) {
1306 ret = btrfs_del_item(trans, root, path);
1308 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1309 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1310 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1311 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1312 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1314 struct btrfs_extent_ref_v0 *ref0;
1315 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_ref_v0);
1317 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1320 btrfs_mark_buffer_dirty(leaf);
1325 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1326 struct btrfs_path *path,
1327 struct btrfs_extent_inline_ref *iref)
1329 struct btrfs_key key;
1330 struct extent_buffer *leaf;
1331 struct btrfs_extent_data_ref *ref1;
1332 struct btrfs_shared_data_ref *ref2;
1335 leaf = path->nodes[0];
1336 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1338 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1339 BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1341 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1344 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1346 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1347 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1348 struct btrfs_extent_data_ref);
1349 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1350 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1351 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_shared_data_ref);
1353 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1354 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1355 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1356 struct btrfs_extent_ref_v0 *ref0;
1357 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1358 struct btrfs_extent_ref_v0);
1359 num_refs = btrfs_ref_count_v0(leaf, ref0);
1367 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1368 struct btrfs_root *root,
1369 struct btrfs_path *path,
1370 u64 bytenr, u64 parent,
1373 struct btrfs_key key;
1376 key.objectid = bytenr;
1378 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1379 key.offset = parent;
1381 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1382 key.offset = root_objectid;
1385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1388 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1389 if (ret == -ENOENT && parent) {
1390 btrfs_release_path(path);
1391 key.type = BTRFS_EXTENT_REF_V0_KEY;
1392 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1400 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1401 struct btrfs_root *root,
1402 struct btrfs_path *path,
1403 u64 bytenr, u64 parent,
1406 struct btrfs_key key;
1409 key.objectid = bytenr;
1411 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1412 key.offset = parent;
1414 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1415 key.offset = root_objectid;
1418 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1419 btrfs_release_path(path);
1423 static inline int extent_ref_type(u64 parent, u64 owner)
1426 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1428 type = BTRFS_SHARED_BLOCK_REF_KEY;
1430 type = BTRFS_TREE_BLOCK_REF_KEY;
1433 type = BTRFS_SHARED_DATA_REF_KEY;
1435 type = BTRFS_EXTENT_DATA_REF_KEY;
1440 static int find_next_key(struct btrfs_path *path, int level,
1441 struct btrfs_key *key)
1444 for (; level < BTRFS_MAX_LEVEL; level++) {
1445 if (!path->nodes[level])
1447 if (path->slots[level] + 1 >=
1448 btrfs_header_nritems(path->nodes[level]))
1451 btrfs_item_key_to_cpu(path->nodes[level], key,
1452 path->slots[level] + 1);
1454 btrfs_node_key_to_cpu(path->nodes[level], key,
1455 path->slots[level] + 1);
1462 * look for inline back ref. if back ref is found, *ref_ret is set
1463 * to the address of inline back ref, and 0 is returned.
1465 * if back ref isn't found, *ref_ret is set to the address where it
1466 * should be inserted, and -ENOENT is returned.
1468 * if insert is true and there are too many inline back refs, the path
1469 * points to the extent item, and -EAGAIN is returned.
1471 * NOTE: inline back refs are ordered in the same way that back ref
1472 * items in the tree are ordered.
1474 static noinline_for_stack
1475 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1476 struct btrfs_root *root,
1477 struct btrfs_path *path,
1478 struct btrfs_extent_inline_ref **ref_ret,
1479 u64 bytenr, u64 num_bytes,
1480 u64 parent, u64 root_objectid,
1481 u64 owner, u64 offset, int insert)
1483 struct btrfs_key key;
1484 struct extent_buffer *leaf;
1485 struct btrfs_extent_item *ei;
1486 struct btrfs_extent_inline_ref *iref;
1496 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1499 key.objectid = bytenr;
1500 key.type = BTRFS_EXTENT_ITEM_KEY;
1501 key.offset = num_bytes;
1503 want = extent_ref_type(parent, owner);
1505 extra_size = btrfs_extent_inline_ref_size(want);
1506 path->keep_locks = 1;
1511 * Owner is our parent level, so we can just add one to get the level
1512 * for the block we are interested in.
1514 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1515 key.type = BTRFS_METADATA_ITEM_KEY;
1520 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1527 * We may be a newly converted file system which still has the old fat
1528 * extent entries for metadata, so try and see if we have one of those.
1530 if (ret > 0 && skinny_metadata) {
1531 skinny_metadata = false;
1532 if (path->slots[0]) {
1534 btrfs_item_key_to_cpu(path->nodes[0], &key,
1536 if (key.objectid == bytenr &&
1537 key.type == BTRFS_EXTENT_ITEM_KEY &&
1538 key.offset == num_bytes)
1542 key.type = BTRFS_EXTENT_ITEM_KEY;
1543 key.offset = num_bytes;
1544 btrfs_release_path(path);
1549 if (ret && !insert) {
1552 } else if (WARN_ON(ret)) {
1557 leaf = path->nodes[0];
1558 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1559 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1560 if (item_size < sizeof(*ei)) {
1565 ret = convert_extent_item_v0(trans, root, path, owner,
1571 leaf = path->nodes[0];
1572 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1575 BUG_ON(item_size < sizeof(*ei));
1577 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1578 flags = btrfs_extent_flags(leaf, ei);
1580 ptr = (unsigned long)(ei + 1);
1581 end = (unsigned long)ei + item_size;
1583 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1584 ptr += sizeof(struct btrfs_tree_block_info);
1594 iref = (struct btrfs_extent_inline_ref *)ptr;
1595 type = btrfs_extent_inline_ref_type(leaf, iref);
1599 ptr += btrfs_extent_inline_ref_size(type);
1603 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1604 struct btrfs_extent_data_ref *dref;
1605 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1606 if (match_extent_data_ref(leaf, dref, root_objectid,
1611 if (hash_extent_data_ref_item(leaf, dref) <
1612 hash_extent_data_ref(root_objectid, owner, offset))
1616 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1618 if (parent == ref_offset) {
1622 if (ref_offset < parent)
1625 if (root_objectid == ref_offset) {
1629 if (ref_offset < root_objectid)
1633 ptr += btrfs_extent_inline_ref_size(type);
1635 if (err == -ENOENT && insert) {
1636 if (item_size + extra_size >=
1637 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1642 * To add new inline back ref, we have to make sure
1643 * there is no corresponding back ref item.
1644 * For simplicity, we just do not add new inline back
1645 * ref if there is any kind of item for this block
1647 if (find_next_key(path, 0, &key) == 0 &&
1648 key.objectid == bytenr &&
1649 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1654 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1657 path->keep_locks = 0;
1658 btrfs_unlock_up_safe(path, 1);
1664 * helper to add new inline back ref
1666 static noinline_for_stack
1667 void setup_inline_extent_backref(struct btrfs_root *root,
1668 struct btrfs_path *path,
1669 struct btrfs_extent_inline_ref *iref,
1670 u64 parent, u64 root_objectid,
1671 u64 owner, u64 offset, int refs_to_add,
1672 struct btrfs_delayed_extent_op *extent_op)
1674 struct extent_buffer *leaf;
1675 struct btrfs_extent_item *ei;
1678 unsigned long item_offset;
1683 leaf = path->nodes[0];
1684 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1685 item_offset = (unsigned long)iref - (unsigned long)ei;
1687 type = extent_ref_type(parent, owner);
1688 size = btrfs_extent_inline_ref_size(type);
1690 btrfs_extend_item(root, path, size);
1692 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1693 refs = btrfs_extent_refs(leaf, ei);
1694 refs += refs_to_add;
1695 btrfs_set_extent_refs(leaf, ei, refs);
1697 __run_delayed_extent_op(extent_op, leaf, ei);
1699 ptr = (unsigned long)ei + item_offset;
1700 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1701 if (ptr < end - size)
1702 memmove_extent_buffer(leaf, ptr + size, ptr,
1705 iref = (struct btrfs_extent_inline_ref *)ptr;
1706 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1707 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1708 struct btrfs_extent_data_ref *dref;
1709 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1710 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1711 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1712 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1713 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1714 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1715 struct btrfs_shared_data_ref *sref;
1716 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1717 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1719 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1720 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1724 btrfs_mark_buffer_dirty(leaf);
1727 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1728 struct btrfs_root *root,
1729 struct btrfs_path *path,
1730 struct btrfs_extent_inline_ref **ref_ret,
1731 u64 bytenr, u64 num_bytes, u64 parent,
1732 u64 root_objectid, u64 owner, u64 offset)
1736 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1737 bytenr, num_bytes, parent,
1738 root_objectid, owner, offset, 0);
1742 btrfs_release_path(path);
1745 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1746 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1749 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1750 root_objectid, owner, offset);
1756 * helper to update/remove inline back ref
1758 static noinline_for_stack
1759 void update_inline_extent_backref(struct btrfs_root *root,
1760 struct btrfs_path *path,
1761 struct btrfs_extent_inline_ref *iref,
1763 struct btrfs_delayed_extent_op *extent_op)
1765 struct extent_buffer *leaf;
1766 struct btrfs_extent_item *ei;
1767 struct btrfs_extent_data_ref *dref = NULL;
1768 struct btrfs_shared_data_ref *sref = NULL;
1776 leaf = path->nodes[0];
1777 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1778 refs = btrfs_extent_refs(leaf, ei);
1779 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1780 refs += refs_to_mod;
1781 btrfs_set_extent_refs(leaf, ei, refs);
1783 __run_delayed_extent_op(extent_op, leaf, ei);
1785 type = btrfs_extent_inline_ref_type(leaf, iref);
1787 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1788 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1789 refs = btrfs_extent_data_ref_count(leaf, dref);
1790 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1791 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1792 refs = btrfs_shared_data_ref_count(leaf, sref);
1795 BUG_ON(refs_to_mod != -1);
1798 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1799 refs += refs_to_mod;
1802 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1803 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1805 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1807 size = btrfs_extent_inline_ref_size(type);
1808 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1809 ptr = (unsigned long)iref;
1810 end = (unsigned long)ei + item_size;
1811 if (ptr + size < end)
1812 memmove_extent_buffer(leaf, ptr, ptr + size,
1815 btrfs_truncate_item(root, path, item_size, 1);
1817 btrfs_mark_buffer_dirty(leaf);
1820 static noinline_for_stack
1821 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1822 struct btrfs_root *root,
1823 struct btrfs_path *path,
1824 u64 bytenr, u64 num_bytes, u64 parent,
1825 u64 root_objectid, u64 owner,
1826 u64 offset, int refs_to_add,
1827 struct btrfs_delayed_extent_op *extent_op)
1829 struct btrfs_extent_inline_ref *iref;
1832 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1833 bytenr, num_bytes, parent,
1834 root_objectid, owner, offset, 1);
1836 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1837 update_inline_extent_backref(root, path, iref,
1838 refs_to_add, extent_op);
1839 } else if (ret == -ENOENT) {
1840 setup_inline_extent_backref(root, path, iref, parent,
1841 root_objectid, owner, offset,
1842 refs_to_add, extent_op);
1848 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1849 struct btrfs_root *root,
1850 struct btrfs_path *path,
1851 u64 bytenr, u64 parent, u64 root_objectid,
1852 u64 owner, u64 offset, int refs_to_add)
1855 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1856 BUG_ON(refs_to_add != 1);
1857 ret = insert_tree_block_ref(trans, root, path, bytenr,
1858 parent, root_objectid);
1860 ret = insert_extent_data_ref(trans, root, path, bytenr,
1861 parent, root_objectid,
1862 owner, offset, refs_to_add);
1867 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1868 struct btrfs_root *root,
1869 struct btrfs_path *path,
1870 struct btrfs_extent_inline_ref *iref,
1871 int refs_to_drop, int is_data)
1875 BUG_ON(!is_data && refs_to_drop != 1);
1877 update_inline_extent_backref(root, path, iref,
1878 -refs_to_drop, NULL);
1879 } else if (is_data) {
1880 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1882 ret = btrfs_del_item(trans, root, path);
1887 static int btrfs_issue_discard(struct block_device *bdev,
1890 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1893 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894 u64 num_bytes, u64 *actual_bytes)
1897 u64 discarded_bytes = 0;
1898 struct btrfs_bio *bbio = NULL;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903 bytenr, &num_bytes, &bbio, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911 if (!stripe->dev->can_discard)
1914 ret = btrfs_issue_discard(stripe->dev->bdev,
1918 discarded_bytes += stripe->length;
1919 else if (ret != -EOPNOTSUPP)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1933 *actual_bytes = discarded_bytes;
1936 if (ret == -EOPNOTSUPP)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 u64 bytenr, u64 num_bytes, u64 parent,
1945 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1948 struct btrfs_fs_info *fs_info = root->fs_info;
1950 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956 parent, root_objectid, (int)owner,
1957 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1959 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961 parent, root_objectid, owner, offset,
1962 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root,
1969 u64 bytenr, u64 num_bytes,
1970 u64 parent, u64 root_objectid,
1971 u64 owner, u64 offset, int refs_to_add,
1972 struct btrfs_delayed_extent_op *extent_op)
1974 struct btrfs_path *path;
1975 struct extent_buffer *leaf;
1976 struct btrfs_extent_item *item;
1980 path = btrfs_alloc_path();
1985 path->leave_spinning = 1;
1986 /* this will setup the path even if it fails to insert the back ref */
1987 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1988 path, bytenr, num_bytes, parent,
1989 root_objectid, owner, offset,
1990 refs_to_add, extent_op);
1994 leaf = path->nodes[0];
1995 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1996 refs = btrfs_extent_refs(leaf, item);
1997 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1999 __run_delayed_extent_op(extent_op, leaf, item);
2001 btrfs_mark_buffer_dirty(leaf);
2002 btrfs_release_path(path);
2005 path->leave_spinning = 1;
2007 /* now insert the actual backref */
2008 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2009 path, bytenr, parent, root_objectid,
2010 owner, offset, refs_to_add);
2012 btrfs_abort_transaction(trans, root, ret);
2014 btrfs_free_path(path);
2018 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2019 struct btrfs_root *root,
2020 struct btrfs_delayed_ref_node *node,
2021 struct btrfs_delayed_extent_op *extent_op,
2022 int insert_reserved)
2025 struct btrfs_delayed_data_ref *ref;
2026 struct btrfs_key ins;
2031 ins.objectid = node->bytenr;
2032 ins.offset = node->num_bytes;
2033 ins.type = BTRFS_EXTENT_ITEM_KEY;
2035 ref = btrfs_delayed_node_to_data_ref(node);
2036 trace_run_delayed_data_ref(node, ref, node->action);
2038 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2039 parent = ref->parent;
2041 ref_root = ref->root;
2043 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2045 flags |= extent_op->flags_to_set;
2046 ret = alloc_reserved_file_extent(trans, root,
2047 parent, ref_root, flags,
2048 ref->objectid, ref->offset,
2049 &ins, node->ref_mod);
2050 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2051 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2052 node->num_bytes, parent,
2053 ref_root, ref->objectid,
2054 ref->offset, node->ref_mod,
2056 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2057 ret = __btrfs_free_extent(trans, root, node->bytenr,
2058 node->num_bytes, parent,
2059 ref_root, ref->objectid,
2060 ref->offset, node->ref_mod,
2068 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2069 struct extent_buffer *leaf,
2070 struct btrfs_extent_item *ei)
2072 u64 flags = btrfs_extent_flags(leaf, ei);
2073 if (extent_op->update_flags) {
2074 flags |= extent_op->flags_to_set;
2075 btrfs_set_extent_flags(leaf, ei, flags);
2078 if (extent_op->update_key) {
2079 struct btrfs_tree_block_info *bi;
2080 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2081 bi = (struct btrfs_tree_block_info *)(ei + 1);
2082 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2086 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_delayed_ref_node *node,
2089 struct btrfs_delayed_extent_op *extent_op)
2091 struct btrfs_key key;
2092 struct btrfs_path *path;
2093 struct btrfs_extent_item *ei;
2094 struct extent_buffer *leaf;
2098 int metadata = !extent_op->is_data;
2103 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2106 path = btrfs_alloc_path();
2110 key.objectid = node->bytenr;
2113 key.type = BTRFS_METADATA_ITEM_KEY;
2114 key.offset = extent_op->level;
2116 key.type = BTRFS_EXTENT_ITEM_KEY;
2117 key.offset = node->num_bytes;
2122 path->leave_spinning = 1;
2123 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2131 if (path->slots[0] > 0) {
2133 btrfs_item_key_to_cpu(path->nodes[0], &key,
2135 if (key.objectid == node->bytenr &&
2136 key.type == BTRFS_EXTENT_ITEM_KEY &&
2137 key.offset == node->num_bytes)
2141 btrfs_release_path(path);
2144 key.objectid = node->bytenr;
2145 key.offset = node->num_bytes;
2146 key.type = BTRFS_EXTENT_ITEM_KEY;
2155 leaf = path->nodes[0];
2156 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2157 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2158 if (item_size < sizeof(*ei)) {
2159 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2165 leaf = path->nodes[0];
2166 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2169 BUG_ON(item_size < sizeof(*ei));
2170 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2171 __run_delayed_extent_op(extent_op, leaf, ei);
2173 btrfs_mark_buffer_dirty(leaf);
2175 btrfs_free_path(path);
2179 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2180 struct btrfs_root *root,
2181 struct btrfs_delayed_ref_node *node,
2182 struct btrfs_delayed_extent_op *extent_op,
2183 int insert_reserved)
2186 struct btrfs_delayed_tree_ref *ref;
2187 struct btrfs_key ins;
2190 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2193 ref = btrfs_delayed_node_to_tree_ref(node);
2194 trace_run_delayed_tree_ref(node, ref, node->action);
2196 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2197 parent = ref->parent;
2199 ref_root = ref->root;
2201 ins.objectid = node->bytenr;
2202 if (skinny_metadata) {
2203 ins.offset = ref->level;
2204 ins.type = BTRFS_METADATA_ITEM_KEY;
2206 ins.offset = node->num_bytes;
2207 ins.type = BTRFS_EXTENT_ITEM_KEY;
2210 BUG_ON(node->ref_mod != 1);
2211 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2212 BUG_ON(!extent_op || !extent_op->update_flags);
2213 ret = alloc_reserved_tree_block(trans, root,
2215 extent_op->flags_to_set,
2218 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2219 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2220 node->num_bytes, parent, ref_root,
2221 ref->level, 0, 1, extent_op);
2222 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2223 ret = __btrfs_free_extent(trans, root, node->bytenr,
2224 node->num_bytes, parent, ref_root,
2225 ref->level, 0, 1, extent_op);
2232 /* helper function to actually process a single delayed ref entry */
2233 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2234 struct btrfs_root *root,
2235 struct btrfs_delayed_ref_node *node,
2236 struct btrfs_delayed_extent_op *extent_op,
2237 int insert_reserved)
2241 if (trans->aborted) {
2242 if (insert_reserved)
2243 btrfs_pin_extent(root, node->bytenr,
2244 node->num_bytes, 1);
2248 if (btrfs_delayed_ref_is_head(node)) {
2249 struct btrfs_delayed_ref_head *head;
2251 * we've hit the end of the chain and we were supposed
2252 * to insert this extent into the tree. But, it got
2253 * deleted before we ever needed to insert it, so all
2254 * we have to do is clean up the accounting
2257 head = btrfs_delayed_node_to_head(node);
2258 trace_run_delayed_ref_head(node, head, node->action);
2260 if (insert_reserved) {
2261 btrfs_pin_extent(root, node->bytenr,
2262 node->num_bytes, 1);
2263 if (head->is_data) {
2264 ret = btrfs_del_csums(trans, root,
2272 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2273 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2274 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2276 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2277 node->type == BTRFS_SHARED_DATA_REF_KEY)
2278 ret = run_delayed_data_ref(trans, root, node, extent_op,
2285 static noinline struct btrfs_delayed_ref_node *
2286 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2288 struct rb_node *node;
2289 struct btrfs_delayed_ref_node *ref;
2290 int action = BTRFS_ADD_DELAYED_REF;
2293 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2294 * this prevents ref count from going down to zero when
2295 * there still are pending delayed ref.
2297 node = rb_prev(&head->node.rb_node);
2301 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2303 if (ref->bytenr != head->node.bytenr)
2305 if (ref->action == action)
2307 node = rb_prev(node);
2309 if (action == BTRFS_ADD_DELAYED_REF) {
2310 action = BTRFS_DROP_DELAYED_REF;
2317 * Returns 0 on success or if called with an already aborted transaction.
2318 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2320 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2321 struct btrfs_root *root,
2322 struct list_head *cluster)
2324 struct btrfs_delayed_ref_root *delayed_refs;
2325 struct btrfs_delayed_ref_node *ref;
2326 struct btrfs_delayed_ref_head *locked_ref = NULL;
2327 struct btrfs_delayed_extent_op *extent_op;
2328 struct btrfs_fs_info *fs_info = root->fs_info;
2331 int must_insert_reserved = 0;
2333 delayed_refs = &trans->transaction->delayed_refs;
2336 /* pick a new head ref from the cluster list */
2337 if (list_empty(cluster))
2340 locked_ref = list_entry(cluster->next,
2341 struct btrfs_delayed_ref_head, cluster);
2343 /* grab the lock that says we are going to process
2344 * all the refs for this head */
2345 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2348 * we may have dropped the spin lock to get the head
2349 * mutex lock, and that might have given someone else
2350 * time to free the head. If that's true, it has been
2351 * removed from our list and we can move on.
2353 if (ret == -EAGAIN) {
2361 * We need to try and merge add/drops of the same ref since we
2362 * can run into issues with relocate dropping the implicit ref
2363 * and then it being added back again before the drop can
2364 * finish. If we merged anything we need to re-loop so we can
2367 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2371 * locked_ref is the head node, so we have to go one
2372 * node back for any delayed ref updates
2374 ref = select_delayed_ref(locked_ref);
2376 if (ref && ref->seq &&
2377 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2379 * there are still refs with lower seq numbers in the
2380 * process of being added. Don't run this ref yet.
2382 list_del_init(&locked_ref->cluster);
2383 btrfs_delayed_ref_unlock(locked_ref);
2385 delayed_refs->num_heads_ready++;
2386 spin_unlock(&delayed_refs->lock);
2388 spin_lock(&delayed_refs->lock);
2393 * record the must insert reserved flag before we
2394 * drop the spin lock.
2396 must_insert_reserved = locked_ref->must_insert_reserved;
2397 locked_ref->must_insert_reserved = 0;
2399 extent_op = locked_ref->extent_op;
2400 locked_ref->extent_op = NULL;
2403 /* All delayed refs have been processed, Go ahead
2404 * and send the head node to run_one_delayed_ref,
2405 * so that any accounting fixes can happen
2407 ref = &locked_ref->node;
2409 if (extent_op && must_insert_reserved) {
2410 btrfs_free_delayed_extent_op(extent_op);
2415 spin_unlock(&delayed_refs->lock);
2417 ret = run_delayed_extent_op(trans, root,
2419 btrfs_free_delayed_extent_op(extent_op);
2423 * Need to reset must_insert_reserved if
2424 * there was an error so the abort stuff
2425 * can cleanup the reserved space
2428 if (must_insert_reserved)
2429 locked_ref->must_insert_reserved = 1;
2430 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2431 spin_lock(&delayed_refs->lock);
2432 btrfs_delayed_ref_unlock(locked_ref);
2441 rb_erase(&ref->rb_node, &delayed_refs->root);
2442 if (btrfs_delayed_ref_is_head(ref)) {
2443 rb_erase(&locked_ref->href_node,
2444 &delayed_refs->href_root);
2446 delayed_refs->num_entries--;
2447 if (!btrfs_delayed_ref_is_head(ref)) {
2449 * when we play the delayed ref, also correct the
2452 switch (ref->action) {
2453 case BTRFS_ADD_DELAYED_REF:
2454 case BTRFS_ADD_DELAYED_EXTENT:
2455 locked_ref->node.ref_mod -= ref->ref_mod;
2457 case BTRFS_DROP_DELAYED_REF:
2458 locked_ref->node.ref_mod += ref->ref_mod;
2464 list_del_init(&locked_ref->cluster);
2466 spin_unlock(&delayed_refs->lock);
2468 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2469 must_insert_reserved);
2471 btrfs_free_delayed_extent_op(extent_op);
2473 btrfs_delayed_ref_unlock(locked_ref);
2474 btrfs_put_delayed_ref(ref);
2475 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2476 spin_lock(&delayed_refs->lock);
2481 * If this node is a head, that means all the refs in this head
2482 * have been dealt with, and we will pick the next head to deal
2483 * with, so we must unlock the head and drop it from the cluster
2484 * list before we release it.
2486 if (btrfs_delayed_ref_is_head(ref)) {
2487 btrfs_delayed_ref_unlock(locked_ref);
2490 btrfs_put_delayed_ref(ref);
2494 spin_lock(&delayed_refs->lock);
2499 #ifdef SCRAMBLE_DELAYED_REFS
2501 * Normally delayed refs get processed in ascending bytenr order. This
2502 * correlates in most cases to the order added. To expose dependencies on this
2503 * order, we start to process the tree in the middle instead of the beginning
2505 static u64 find_middle(struct rb_root *root)
2507 struct rb_node *n = root->rb_node;
2508 struct btrfs_delayed_ref_node *entry;
2511 u64 first = 0, last = 0;
2515 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2516 first = entry->bytenr;
2520 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2521 last = entry->bytenr;
2526 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2527 WARN_ON(!entry->in_tree);
2529 middle = entry->bytenr;
2542 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2543 struct btrfs_fs_info *fs_info)
2545 struct qgroup_update *qgroup_update;
2548 if (list_empty(&trans->qgroup_ref_list) !=
2549 !trans->delayed_ref_elem.seq) {
2550 /* list without seq or seq without list */
2552 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2553 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2554 (u32)(trans->delayed_ref_elem.seq >> 32),
2555 (u32)trans->delayed_ref_elem.seq);
2559 if (!trans->delayed_ref_elem.seq)
2562 while (!list_empty(&trans->qgroup_ref_list)) {
2563 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2564 struct qgroup_update, list);
2565 list_del(&qgroup_update->list);
2567 ret = btrfs_qgroup_account_ref(
2568 trans, fs_info, qgroup_update->node,
2569 qgroup_update->extent_op);
2570 kfree(qgroup_update);
2573 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2578 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2581 int val = atomic_read(&delayed_refs->ref_seq);
2583 if (val < seq || val >= seq + count)
2588 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2592 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2593 sizeof(struct btrfs_extent_inline_ref));
2594 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2595 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2598 * We don't ever fill up leaves all the way so multiply by 2 just to be
2599 * closer to what we're really going to want to ouse.
2601 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2604 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *root)
2607 struct btrfs_block_rsv *global_rsv;
2608 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2612 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2613 num_heads = heads_to_leaves(root, num_heads);
2615 num_bytes += (num_heads - 1) * root->leafsize;
2617 global_rsv = &root->fs_info->global_block_rsv;
2620 * If we can't allocate any more chunks lets make sure we have _lots_ of
2621 * wiggle room since running delayed refs can create more delayed refs.
2623 if (global_rsv->space_info->full)
2626 spin_lock(&global_rsv->lock);
2627 if (global_rsv->reserved <= num_bytes)
2629 spin_unlock(&global_rsv->lock);
2634 * this starts processing the delayed reference count updates and
2635 * extent insertions we have queued up so far. count can be
2636 * 0, which means to process everything in the tree at the start
2637 * of the run (but not newly added entries), or it can be some target
2638 * number you'd like to process.
2640 * Returns 0 on success or if called with an aborted transaction
2641 * Returns <0 on error and aborts the transaction
2643 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2644 struct btrfs_root *root, unsigned long count)
2646 struct rb_node *node;
2647 struct btrfs_delayed_ref_root *delayed_refs;
2648 struct btrfs_delayed_ref_head *head;
2649 struct list_head cluster;
2652 int run_all = count == (unsigned long)-1;
2656 /* We'll clean this up in btrfs_cleanup_transaction */
2660 if (root == root->fs_info->extent_root)
2661 root = root->fs_info->tree_root;
2663 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2665 delayed_refs = &trans->transaction->delayed_refs;
2666 INIT_LIST_HEAD(&cluster);
2668 count = delayed_refs->num_entries * 2;
2672 if (!run_all && !run_most) {
2674 int seq = atomic_read(&delayed_refs->ref_seq);
2677 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2679 DEFINE_WAIT(__wait);
2680 if (delayed_refs->flushing ||
2681 !btrfs_should_throttle_delayed_refs(trans, root))
2684 prepare_to_wait(&delayed_refs->wait, &__wait,
2685 TASK_UNINTERRUPTIBLE);
2687 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2690 finish_wait(&delayed_refs->wait, &__wait);
2692 if (!refs_newer(delayed_refs, seq, 256))
2697 finish_wait(&delayed_refs->wait, &__wait);
2703 atomic_inc(&delayed_refs->procs_running_refs);
2708 spin_lock(&delayed_refs->lock);
2710 #ifdef SCRAMBLE_DELAYED_REFS
2711 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2715 if (!(run_all || run_most) &&
2716 !btrfs_should_throttle_delayed_refs(trans, root))
2720 * go find something we can process in the rbtree. We start at
2721 * the beginning of the tree, and then build a cluster
2722 * of refs to process starting at the first one we are able to
2725 delayed_start = delayed_refs->run_delayed_start;
2726 ret = btrfs_find_ref_cluster(trans, &cluster,
2727 delayed_refs->run_delayed_start);
2731 ret = run_clustered_refs(trans, root, &cluster);
2733 btrfs_release_ref_cluster(&cluster);
2734 spin_unlock(&delayed_refs->lock);
2735 btrfs_abort_transaction(trans, root, ret);
2736 atomic_dec(&delayed_refs->procs_running_refs);
2737 wake_up(&delayed_refs->wait);
2741 atomic_add(ret, &delayed_refs->ref_seq);
2743 count -= min_t(unsigned long, ret, count);
2748 if (delayed_start >= delayed_refs->run_delayed_start) {
2751 * btrfs_find_ref_cluster looped. let's do one
2752 * more cycle. if we don't run any delayed ref
2753 * during that cycle (because we can't because
2754 * all of them are blocked), bail out.
2759 * no runnable refs left, stop trying
2766 /* refs were run, let's reset staleness detection */
2772 if (!list_empty(&trans->new_bgs)) {
2773 spin_unlock(&delayed_refs->lock);
2774 btrfs_create_pending_block_groups(trans, root);
2775 spin_lock(&delayed_refs->lock);
2778 node = rb_first(&delayed_refs->href_root);
2781 count = (unsigned long)-1;
2784 head = rb_entry(node, struct btrfs_delayed_ref_head,
2786 if (btrfs_delayed_ref_is_head(&head->node)) {
2787 struct btrfs_delayed_ref_node *ref;
2790 atomic_inc(&ref->refs);
2792 spin_unlock(&delayed_refs->lock);
2794 * Mutex was contended, block until it's
2795 * released and try again
2797 mutex_lock(&head->mutex);
2798 mutex_unlock(&head->mutex);
2800 btrfs_put_delayed_ref(ref);
2806 node = rb_next(node);
2808 spin_unlock(&delayed_refs->lock);
2809 schedule_timeout(1);
2813 atomic_dec(&delayed_refs->procs_running_refs);
2815 if (waitqueue_active(&delayed_refs->wait))
2816 wake_up(&delayed_refs->wait);
2818 spin_unlock(&delayed_refs->lock);
2819 assert_qgroups_uptodate(trans);
2823 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2824 struct btrfs_root *root,
2825 u64 bytenr, u64 num_bytes, u64 flags,
2826 int level, int is_data)
2828 struct btrfs_delayed_extent_op *extent_op;
2831 extent_op = btrfs_alloc_delayed_extent_op();
2835 extent_op->flags_to_set = flags;
2836 extent_op->update_flags = 1;
2837 extent_op->update_key = 0;
2838 extent_op->is_data = is_data ? 1 : 0;
2839 extent_op->level = level;
2841 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2842 num_bytes, extent_op);
2844 btrfs_free_delayed_extent_op(extent_op);
2848 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2849 struct btrfs_root *root,
2850 struct btrfs_path *path,
2851 u64 objectid, u64 offset, u64 bytenr)
2853 struct btrfs_delayed_ref_head *head;
2854 struct btrfs_delayed_ref_node *ref;
2855 struct btrfs_delayed_data_ref *data_ref;
2856 struct btrfs_delayed_ref_root *delayed_refs;
2857 struct rb_node *node;
2861 delayed_refs = &trans->transaction->delayed_refs;
2862 spin_lock(&delayed_refs->lock);
2863 head = btrfs_find_delayed_ref_head(trans, bytenr);
2867 if (!mutex_trylock(&head->mutex)) {
2868 atomic_inc(&head->node.refs);
2869 spin_unlock(&delayed_refs->lock);
2871 btrfs_release_path(path);
2874 * Mutex was contended, block until it's released and let
2877 mutex_lock(&head->mutex);
2878 mutex_unlock(&head->mutex);
2879 btrfs_put_delayed_ref(&head->node);
2883 node = rb_prev(&head->node.rb_node);
2887 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2889 if (ref->bytenr != bytenr)
2893 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2896 data_ref = btrfs_delayed_node_to_data_ref(ref);
2898 node = rb_prev(node);
2902 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2903 if (ref->bytenr == bytenr && ref->seq == seq)
2907 if (data_ref->root != root->root_key.objectid ||
2908 data_ref->objectid != objectid || data_ref->offset != offset)
2913 mutex_unlock(&head->mutex);
2915 spin_unlock(&delayed_refs->lock);
2919 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2920 struct btrfs_root *root,
2921 struct btrfs_path *path,
2922 u64 objectid, u64 offset, u64 bytenr)
2924 struct btrfs_root *extent_root = root->fs_info->extent_root;
2925 struct extent_buffer *leaf;
2926 struct btrfs_extent_data_ref *ref;
2927 struct btrfs_extent_inline_ref *iref;
2928 struct btrfs_extent_item *ei;
2929 struct btrfs_key key;
2933 key.objectid = bytenr;
2934 key.offset = (u64)-1;
2935 key.type = BTRFS_EXTENT_ITEM_KEY;
2937 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2940 BUG_ON(ret == 0); /* Corruption */
2943 if (path->slots[0] == 0)
2947 leaf = path->nodes[0];
2948 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2950 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2954 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2956 if (item_size < sizeof(*ei)) {
2957 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2961 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2963 if (item_size != sizeof(*ei) +
2964 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2967 if (btrfs_extent_generation(leaf, ei) <=
2968 btrfs_root_last_snapshot(&root->root_item))
2971 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2972 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2973 BTRFS_EXTENT_DATA_REF_KEY)
2976 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2977 if (btrfs_extent_refs(leaf, ei) !=
2978 btrfs_extent_data_ref_count(leaf, ref) ||
2979 btrfs_extent_data_ref_root(leaf, ref) !=
2980 root->root_key.objectid ||
2981 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2982 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2990 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2991 struct btrfs_root *root,
2992 u64 objectid, u64 offset, u64 bytenr)
2994 struct btrfs_path *path;
2998 path = btrfs_alloc_path();
3003 ret = check_committed_ref(trans, root, path, objectid,
3005 if (ret && ret != -ENOENT)
3008 ret2 = check_delayed_ref(trans, root, path, objectid,
3010 } while (ret2 == -EAGAIN);
3012 if (ret2 && ret2 != -ENOENT) {
3017 if (ret != -ENOENT || ret2 != -ENOENT)
3020 btrfs_free_path(path);
3021 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3026 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3027 struct btrfs_root *root,
3028 struct extent_buffer *buf,
3029 int full_backref, int inc, int for_cow)
3036 struct btrfs_key key;
3037 struct btrfs_file_extent_item *fi;
3041 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3042 u64, u64, u64, u64, u64, u64, int);
3044 ref_root = btrfs_header_owner(buf);
3045 nritems = btrfs_header_nritems(buf);
3046 level = btrfs_header_level(buf);
3048 if (!root->ref_cows && level == 0)
3052 process_func = btrfs_inc_extent_ref;
3054 process_func = btrfs_free_extent;
3057 parent = buf->start;
3061 for (i = 0; i < nritems; i++) {
3063 btrfs_item_key_to_cpu(buf, &key, i);
3064 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3066 fi = btrfs_item_ptr(buf, i,
3067 struct btrfs_file_extent_item);
3068 if (btrfs_file_extent_type(buf, fi) ==
3069 BTRFS_FILE_EXTENT_INLINE)
3071 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3075 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3076 key.offset -= btrfs_file_extent_offset(buf, fi);
3077 ret = process_func(trans, root, bytenr, num_bytes,
3078 parent, ref_root, key.objectid,
3079 key.offset, for_cow);
3083 bytenr = btrfs_node_blockptr(buf, i);
3084 num_bytes = btrfs_level_size(root, level - 1);
3085 ret = process_func(trans, root, bytenr, num_bytes,
3086 parent, ref_root, level - 1, 0,
3097 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3098 struct extent_buffer *buf, int full_backref, int for_cow)
3100 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3103 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3104 struct extent_buffer *buf, int full_backref, int for_cow)
3106 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3109 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3110 struct btrfs_root *root,
3111 struct btrfs_path *path,
3112 struct btrfs_block_group_cache *cache)
3115 struct btrfs_root *extent_root = root->fs_info->extent_root;
3117 struct extent_buffer *leaf;
3119 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3122 BUG_ON(ret); /* Corruption */
3124 leaf = path->nodes[0];
3125 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3126 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3127 btrfs_mark_buffer_dirty(leaf);
3128 btrfs_release_path(path);
3131 btrfs_abort_transaction(trans, root, ret);
3138 static struct btrfs_block_group_cache *
3139 next_block_group(struct btrfs_root *root,
3140 struct btrfs_block_group_cache *cache)
3142 struct rb_node *node;
3143 spin_lock(&root->fs_info->block_group_cache_lock);
3144 node = rb_next(&cache->cache_node);
3145 btrfs_put_block_group(cache);
3147 cache = rb_entry(node, struct btrfs_block_group_cache,
3149 btrfs_get_block_group(cache);
3152 spin_unlock(&root->fs_info->block_group_cache_lock);
3156 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3157 struct btrfs_trans_handle *trans,
3158 struct btrfs_path *path)
3160 struct btrfs_root *root = block_group->fs_info->tree_root;
3161 struct inode *inode = NULL;
3163 int dcs = BTRFS_DC_ERROR;
3169 * If this block group is smaller than 100 megs don't bother caching the
3172 if (block_group->key.offset < (100 * 1024 * 1024)) {
3173 spin_lock(&block_group->lock);
3174 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3175 spin_unlock(&block_group->lock);
3180 inode = lookup_free_space_inode(root, block_group, path);
3181 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3182 ret = PTR_ERR(inode);
3183 btrfs_release_path(path);
3187 if (IS_ERR(inode)) {
3191 if (block_group->ro)
3194 ret = create_free_space_inode(root, trans, block_group, path);
3200 /* We've already setup this transaction, go ahead and exit */
3201 if (block_group->cache_generation == trans->transid &&
3202 i_size_read(inode)) {
3203 dcs = BTRFS_DC_SETUP;
3208 * We want to set the generation to 0, that way if anything goes wrong
3209 * from here on out we know not to trust this cache when we load up next
3212 BTRFS_I(inode)->generation = 0;
3213 ret = btrfs_update_inode(trans, root, inode);
3216 if (i_size_read(inode) > 0) {
3217 ret = btrfs_check_trunc_cache_free_space(root,
3218 &root->fs_info->global_block_rsv);
3222 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3227 spin_lock(&block_group->lock);
3228 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3229 !btrfs_test_opt(root, SPACE_CACHE)) {
3231 * don't bother trying to write stuff out _if_
3232 * a) we're not cached,
3233 * b) we're with nospace_cache mount option.
3235 dcs = BTRFS_DC_WRITTEN;
3236 spin_unlock(&block_group->lock);
3239 spin_unlock(&block_group->lock);
3242 * Try to preallocate enough space based on how big the block group is.
3243 * Keep in mind this has to include any pinned space which could end up
3244 * taking up quite a bit since it's not folded into the other space
3247 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3252 num_pages *= PAGE_CACHE_SIZE;
3254 ret = btrfs_check_data_free_space(inode, num_pages);
3258 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3259 num_pages, num_pages,
3262 dcs = BTRFS_DC_SETUP;
3263 btrfs_free_reserved_data_space(inode, num_pages);
3268 btrfs_release_path(path);
3270 spin_lock(&block_group->lock);
3271 if (!ret && dcs == BTRFS_DC_SETUP)
3272 block_group->cache_generation = trans->transid;
3273 block_group->disk_cache_state = dcs;
3274 spin_unlock(&block_group->lock);
3279 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3280 struct btrfs_root *root)
3282 struct btrfs_block_group_cache *cache;
3284 struct btrfs_path *path;
3287 path = btrfs_alloc_path();
3293 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3295 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3297 cache = next_block_group(root, cache);
3305 err = cache_save_setup(cache, trans, path);
3306 last = cache->key.objectid + cache->key.offset;
3307 btrfs_put_block_group(cache);
3312 err = btrfs_run_delayed_refs(trans, root,
3314 if (err) /* File system offline */
3318 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3320 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3321 btrfs_put_block_group(cache);
3327 cache = next_block_group(root, cache);
3336 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3337 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3339 last = cache->key.objectid + cache->key.offset;
3341 err = write_one_cache_group(trans, root, path, cache);
3342 btrfs_put_block_group(cache);
3343 if (err) /* File system offline */
3349 * I don't think this is needed since we're just marking our
3350 * preallocated extent as written, but just in case it can't
3354 err = btrfs_run_delayed_refs(trans, root,
3356 if (err) /* File system offline */
3360 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3363 * Really this shouldn't happen, but it could if we
3364 * couldn't write the entire preallocated extent and
3365 * splitting the extent resulted in a new block.
3368 btrfs_put_block_group(cache);
3371 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3373 cache = next_block_group(root, cache);
3382 err = btrfs_write_out_cache(root, trans, cache, path);
3385 * If we didn't have an error then the cache state is still
3386 * NEED_WRITE, so we can set it to WRITTEN.
3388 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3389 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3390 last = cache->key.objectid + cache->key.offset;
3391 btrfs_put_block_group(cache);
3395 btrfs_free_path(path);
3399 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3401 struct btrfs_block_group_cache *block_group;
3404 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3405 if (!block_group || block_group->ro)
3408 btrfs_put_block_group(block_group);
3412 static const char *alloc_name(u64 flags)
3415 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3417 case BTRFS_BLOCK_GROUP_METADATA:
3419 case BTRFS_BLOCK_GROUP_DATA:
3421 case BTRFS_BLOCK_GROUP_SYSTEM:
3425 return "invalid-combination";
3429 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3430 u64 total_bytes, u64 bytes_used,
3431 struct btrfs_space_info **space_info)
3433 struct btrfs_space_info *found;
3438 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3439 BTRFS_BLOCK_GROUP_RAID10))
3444 found = __find_space_info(info, flags);
3446 spin_lock(&found->lock);
3447 found->total_bytes += total_bytes;
3448 found->disk_total += total_bytes * factor;
3449 found->bytes_used += bytes_used;
3450 found->disk_used += bytes_used * factor;
3452 spin_unlock(&found->lock);
3453 *space_info = found;
3456 found = kzalloc(sizeof(*found), GFP_NOFS);
3460 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3466 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3467 INIT_LIST_HEAD(&found->block_groups[i]);
3468 init_rwsem(&found->groups_sem);
3469 spin_lock_init(&found->lock);
3470 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3471 found->total_bytes = total_bytes;
3472 found->disk_total = total_bytes * factor;
3473 found->bytes_used = bytes_used;
3474 found->disk_used = bytes_used * factor;
3475 found->bytes_pinned = 0;
3476 found->bytes_reserved = 0;
3477 found->bytes_readonly = 0;
3478 found->bytes_may_use = 0;
3480 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3481 found->chunk_alloc = 0;
3483 init_waitqueue_head(&found->wait);
3485 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3486 info->space_info_kobj, "%s",
3487 alloc_name(found->flags));
3493 *space_info = found;
3494 list_add_rcu(&found->list, &info->space_info);
3495 if (flags & BTRFS_BLOCK_GROUP_DATA)
3496 info->data_sinfo = found;
3501 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3503 u64 extra_flags = chunk_to_extended(flags) &
3504 BTRFS_EXTENDED_PROFILE_MASK;
3506 write_seqlock(&fs_info->profiles_lock);
3507 if (flags & BTRFS_BLOCK_GROUP_DATA)
3508 fs_info->avail_data_alloc_bits |= extra_flags;
3509 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3510 fs_info->avail_metadata_alloc_bits |= extra_flags;
3511 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3512 fs_info->avail_system_alloc_bits |= extra_flags;
3513 write_sequnlock(&fs_info->profiles_lock);
3517 * returns target flags in extended format or 0 if restripe for this
3518 * chunk_type is not in progress
3520 * should be called with either volume_mutex or balance_lock held
3522 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3524 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3530 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3531 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3532 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3533 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3534 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3535 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3536 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3537 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3538 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3545 * @flags: available profiles in extended format (see ctree.h)
3547 * Returns reduced profile in chunk format. If profile changing is in
3548 * progress (either running or paused) picks the target profile (if it's
3549 * already available), otherwise falls back to plain reducing.
3551 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3554 * we add in the count of missing devices because we want
3555 * to make sure that any RAID levels on a degraded FS
3556 * continue to be honored.
3558 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3559 root->fs_info->fs_devices->missing_devices;
3564 * see if restripe for this chunk_type is in progress, if so
3565 * try to reduce to the target profile
3567 spin_lock(&root->fs_info->balance_lock);
3568 target = get_restripe_target(root->fs_info, flags);
3570 /* pick target profile only if it's already available */
3571 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3572 spin_unlock(&root->fs_info->balance_lock);
3573 return extended_to_chunk(target);
3576 spin_unlock(&root->fs_info->balance_lock);
3578 /* First, mask out the RAID levels which aren't possible */
3579 if (num_devices == 1)
3580 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3581 BTRFS_BLOCK_GROUP_RAID5);
3582 if (num_devices < 3)
3583 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3584 if (num_devices < 4)
3585 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3587 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3588 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3589 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3592 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3593 tmp = BTRFS_BLOCK_GROUP_RAID6;
3594 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3595 tmp = BTRFS_BLOCK_GROUP_RAID5;
3596 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3597 tmp = BTRFS_BLOCK_GROUP_RAID10;
3598 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3599 tmp = BTRFS_BLOCK_GROUP_RAID1;
3600 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3601 tmp = BTRFS_BLOCK_GROUP_RAID0;
3603 return extended_to_chunk(flags | tmp);
3606 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3611 seq = read_seqbegin(&root->fs_info->profiles_lock);
3613 if (flags & BTRFS_BLOCK_GROUP_DATA)
3614 flags |= root->fs_info->avail_data_alloc_bits;
3615 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3616 flags |= root->fs_info->avail_system_alloc_bits;
3617 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3618 flags |= root->fs_info->avail_metadata_alloc_bits;
3619 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3621 return btrfs_reduce_alloc_profile(root, flags);
3624 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3630 flags = BTRFS_BLOCK_GROUP_DATA;
3631 else if (root == root->fs_info->chunk_root)
3632 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3634 flags = BTRFS_BLOCK_GROUP_METADATA;
3636 ret = get_alloc_profile(root, flags);
3641 * This will check the space that the inode allocates from to make sure we have
3642 * enough space for bytes.
3644 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3646 struct btrfs_space_info *data_sinfo;
3647 struct btrfs_root *root = BTRFS_I(inode)->root;
3648 struct btrfs_fs_info *fs_info = root->fs_info;
3650 int ret = 0, committed = 0, alloc_chunk = 1;
3652 /* make sure bytes are sectorsize aligned */
3653 bytes = ALIGN(bytes, root->sectorsize);
3655 if (btrfs_is_free_space_inode(inode)) {
3657 ASSERT(current->journal_info);
3660 data_sinfo = fs_info->data_sinfo;
3665 /* make sure we have enough space to handle the data first */
3666 spin_lock(&data_sinfo->lock);
3667 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3668 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3669 data_sinfo->bytes_may_use;
3671 if (used + bytes > data_sinfo->total_bytes) {
3672 struct btrfs_trans_handle *trans;
3675 * if we don't have enough free bytes in this space then we need
3676 * to alloc a new chunk.
3678 if (!data_sinfo->full && alloc_chunk) {
3681 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3682 spin_unlock(&data_sinfo->lock);
3684 alloc_target = btrfs_get_alloc_profile(root, 1);
3686 * It is ugly that we don't call nolock join
3687 * transaction for the free space inode case here.
3688 * But it is safe because we only do the data space
3689 * reservation for the free space cache in the
3690 * transaction context, the common join transaction
3691 * just increase the counter of the current transaction
3692 * handler, doesn't try to acquire the trans_lock of
3695 trans = btrfs_join_transaction(root);
3697 return PTR_ERR(trans);
3699 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3701 CHUNK_ALLOC_NO_FORCE);
3702 btrfs_end_transaction(trans, root);
3711 data_sinfo = fs_info->data_sinfo;
3717 * If we don't have enough pinned space to deal with this
3718 * allocation don't bother committing the transaction.
3720 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3723 spin_unlock(&data_sinfo->lock);
3725 /* commit the current transaction and try again */
3728 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3731 trans = btrfs_join_transaction(root);
3733 return PTR_ERR(trans);
3734 ret = btrfs_commit_transaction(trans, root);
3740 trace_btrfs_space_reservation(root->fs_info,
3741 "space_info:enospc",
3742 data_sinfo->flags, bytes, 1);
3745 data_sinfo->bytes_may_use += bytes;
3746 trace_btrfs_space_reservation(root->fs_info, "space_info",
3747 data_sinfo->flags, bytes, 1);
3748 spin_unlock(&data_sinfo->lock);
3754 * Called if we need to clear a data reservation for this inode.
3756 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3758 struct btrfs_root *root = BTRFS_I(inode)->root;
3759 struct btrfs_space_info *data_sinfo;
3761 /* make sure bytes are sectorsize aligned */
3762 bytes = ALIGN(bytes, root->sectorsize);
3764 data_sinfo = root->fs_info->data_sinfo;
3765 spin_lock(&data_sinfo->lock);
3766 WARN_ON(data_sinfo->bytes_may_use < bytes);
3767 data_sinfo->bytes_may_use -= bytes;
3768 trace_btrfs_space_reservation(root->fs_info, "space_info",
3769 data_sinfo->flags, bytes, 0);
3770 spin_unlock(&data_sinfo->lock);
3773 static void force_metadata_allocation(struct btrfs_fs_info *info)
3775 struct list_head *head = &info->space_info;
3776 struct btrfs_space_info *found;
3779 list_for_each_entry_rcu(found, head, list) {
3780 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3781 found->force_alloc = CHUNK_ALLOC_FORCE;
3786 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3788 return (global->size << 1);
3791 static int should_alloc_chunk(struct btrfs_root *root,
3792 struct btrfs_space_info *sinfo, int force)
3794 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3795 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3796 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3799 if (force == CHUNK_ALLOC_FORCE)
3803 * We need to take into account the global rsv because for all intents
3804 * and purposes it's used space. Don't worry about locking the
3805 * global_rsv, it doesn't change except when the transaction commits.
3807 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3808 num_allocated += calc_global_rsv_need_space(global_rsv);
3811 * in limited mode, we want to have some free space up to
3812 * about 1% of the FS size.
3814 if (force == CHUNK_ALLOC_LIMITED) {
3815 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3816 thresh = max_t(u64, 64 * 1024 * 1024,
3817 div_factor_fine(thresh, 1));
3819 if (num_bytes - num_allocated < thresh)
3823 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3828 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3832 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3833 BTRFS_BLOCK_GROUP_RAID0 |
3834 BTRFS_BLOCK_GROUP_RAID5 |
3835 BTRFS_BLOCK_GROUP_RAID6))
3836 num_dev = root->fs_info->fs_devices->rw_devices;
3837 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3840 num_dev = 1; /* DUP or single */
3842 /* metadata for updaing devices and chunk tree */
3843 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3846 static void check_system_chunk(struct btrfs_trans_handle *trans,
3847 struct btrfs_root *root, u64 type)
3849 struct btrfs_space_info *info;
3853 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3854 spin_lock(&info->lock);
3855 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3856 info->bytes_reserved - info->bytes_readonly;
3857 spin_unlock(&info->lock);
3859 thresh = get_system_chunk_thresh(root, type);
3860 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3861 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3862 left, thresh, type);
3863 dump_space_info(info, 0, 0);
3866 if (left < thresh) {
3869 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3870 btrfs_alloc_chunk(trans, root, flags);
3874 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3875 struct btrfs_root *extent_root, u64 flags, int force)
3877 struct btrfs_space_info *space_info;
3878 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3879 int wait_for_alloc = 0;
3882 /* Don't re-enter if we're already allocating a chunk */
3883 if (trans->allocating_chunk)
3886 space_info = __find_space_info(extent_root->fs_info, flags);
3888 ret = update_space_info(extent_root->fs_info, flags,
3890 BUG_ON(ret); /* -ENOMEM */
3892 BUG_ON(!space_info); /* Logic error */
3895 spin_lock(&space_info->lock);
3896 if (force < space_info->force_alloc)
3897 force = space_info->force_alloc;
3898 if (space_info->full) {
3899 if (should_alloc_chunk(extent_root, space_info, force))
3903 spin_unlock(&space_info->lock);
3907 if (!should_alloc_chunk(extent_root, space_info, force)) {
3908 spin_unlock(&space_info->lock);
3910 } else if (space_info->chunk_alloc) {
3913 space_info->chunk_alloc = 1;
3916 spin_unlock(&space_info->lock);
3918 mutex_lock(&fs_info->chunk_mutex);
3921 * The chunk_mutex is held throughout the entirety of a chunk
3922 * allocation, so once we've acquired the chunk_mutex we know that the
3923 * other guy is done and we need to recheck and see if we should
3926 if (wait_for_alloc) {
3927 mutex_unlock(&fs_info->chunk_mutex);
3932 trans->allocating_chunk = true;
3935 * If we have mixed data/metadata chunks we want to make sure we keep
3936 * allocating mixed chunks instead of individual chunks.
3938 if (btrfs_mixed_space_info(space_info))
3939 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3942 * if we're doing a data chunk, go ahead and make sure that
3943 * we keep a reasonable number of metadata chunks allocated in the
3946 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3947 fs_info->data_chunk_allocations++;
3948 if (!(fs_info->data_chunk_allocations %
3949 fs_info->metadata_ratio))
3950 force_metadata_allocation(fs_info);
3954 * Check if we have enough space in SYSTEM chunk because we may need
3955 * to update devices.
3957 check_system_chunk(trans, extent_root, flags);
3959 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3960 trans->allocating_chunk = false;
3962 spin_lock(&space_info->lock);
3963 if (ret < 0 && ret != -ENOSPC)
3966 space_info->full = 1;
3970 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3972 space_info->chunk_alloc = 0;
3973 spin_unlock(&space_info->lock);
3974 mutex_unlock(&fs_info->chunk_mutex);
3978 static int can_overcommit(struct btrfs_root *root,
3979 struct btrfs_space_info *space_info, u64 bytes,
3980 enum btrfs_reserve_flush_enum flush)
3982 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3983 u64 profile = btrfs_get_alloc_profile(root, 0);
3988 used = space_info->bytes_used + space_info->bytes_reserved +
3989 space_info->bytes_pinned + space_info->bytes_readonly;
3992 * We only want to allow over committing if we have lots of actual space
3993 * free, but if we don't have enough space to handle the global reserve
3994 * space then we could end up having a real enospc problem when trying
3995 * to allocate a chunk or some other such important allocation.
3997 spin_lock(&global_rsv->lock);
3998 space_size = calc_global_rsv_need_space(global_rsv);
3999 spin_unlock(&global_rsv->lock);
4000 if (used + space_size >= space_info->total_bytes)
4003 used += space_info->bytes_may_use;
4005 spin_lock(&root->fs_info->free_chunk_lock);
4006 avail = root->fs_info->free_chunk_space;
4007 spin_unlock(&root->fs_info->free_chunk_lock);
4010 * If we have dup, raid1 or raid10 then only half of the free
4011 * space is actually useable. For raid56, the space info used
4012 * doesn't include the parity drive, so we don't have to
4015 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4016 BTRFS_BLOCK_GROUP_RAID1 |
4017 BTRFS_BLOCK_GROUP_RAID10))
4021 * If we aren't flushing all things, let us overcommit up to
4022 * 1/2th of the space. If we can flush, don't let us overcommit
4023 * too much, let it overcommit up to 1/8 of the space.
4025 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4030 if (used + bytes < space_info->total_bytes + avail)
4035 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4036 unsigned long nr_pages)
4038 struct super_block *sb = root->fs_info->sb;
4040 if (down_read_trylock(&sb->s_umount)) {
4041 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4042 up_read(&sb->s_umount);
4045 * We needn't worry the filesystem going from r/w to r/o though
4046 * we don't acquire ->s_umount mutex, because the filesystem
4047 * should guarantee the delalloc inodes list be empty after
4048 * the filesystem is readonly(all dirty pages are written to
4051 btrfs_start_delalloc_roots(root->fs_info, 0);
4052 if (!current->journal_info)
4053 btrfs_wait_ordered_roots(root->fs_info, -1);
4057 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4062 bytes = btrfs_calc_trans_metadata_size(root, 1);
4063 nr = (int)div64_u64(to_reclaim, bytes);
4069 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4072 * shrink metadata reservation for delalloc
4074 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4077 struct btrfs_block_rsv *block_rsv;
4078 struct btrfs_space_info *space_info;
4079 struct btrfs_trans_handle *trans;
4083 unsigned long nr_pages;
4086 enum btrfs_reserve_flush_enum flush;
4088 /* Calc the number of the pages we need flush for space reservation */
4089 items = calc_reclaim_items_nr(root, to_reclaim);
4090 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4092 trans = (struct btrfs_trans_handle *)current->journal_info;
4093 block_rsv = &root->fs_info->delalloc_block_rsv;
4094 space_info = block_rsv->space_info;
4096 delalloc_bytes = percpu_counter_sum_positive(
4097 &root->fs_info->delalloc_bytes);
4098 if (delalloc_bytes == 0) {
4102 btrfs_wait_ordered_roots(root->fs_info, items);
4107 while (delalloc_bytes && loops < 3) {
4108 max_reclaim = min(delalloc_bytes, to_reclaim);
4109 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4110 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4112 * We need to wait for the async pages to actually start before
4115 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4119 if (max_reclaim <= nr_pages)
4122 max_reclaim -= nr_pages;
4124 wait_event(root->fs_info->async_submit_wait,
4125 atomic_read(&root->fs_info->async_delalloc_pages) <=
4129 flush = BTRFS_RESERVE_FLUSH_ALL;
4131 flush = BTRFS_RESERVE_NO_FLUSH;
4132 spin_lock(&space_info->lock);
4133 if (can_overcommit(root, space_info, orig, flush)) {
4134 spin_unlock(&space_info->lock);
4137 spin_unlock(&space_info->lock);
4140 if (wait_ordered && !trans) {
4141 btrfs_wait_ordered_roots(root->fs_info, items);
4143 time_left = schedule_timeout_killable(1);
4147 delalloc_bytes = percpu_counter_sum_positive(
4148 &root->fs_info->delalloc_bytes);
4153 * maybe_commit_transaction - possibly commit the transaction if its ok to
4154 * @root - the root we're allocating for
4155 * @bytes - the number of bytes we want to reserve
4156 * @force - force the commit
4158 * This will check to make sure that committing the transaction will actually
4159 * get us somewhere and then commit the transaction if it does. Otherwise it
4160 * will return -ENOSPC.
4162 static int may_commit_transaction(struct btrfs_root *root,
4163 struct btrfs_space_info *space_info,
4164 u64 bytes, int force)
4166 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4167 struct btrfs_trans_handle *trans;
4169 trans = (struct btrfs_trans_handle *)current->journal_info;
4176 /* See if there is enough pinned space to make this reservation */
4177 spin_lock(&space_info->lock);
4178 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4180 spin_unlock(&space_info->lock);
4183 spin_unlock(&space_info->lock);
4186 * See if there is some space in the delayed insertion reservation for
4189 if (space_info != delayed_rsv->space_info)
4192 spin_lock(&space_info->lock);
4193 spin_lock(&delayed_rsv->lock);
4194 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4195 bytes - delayed_rsv->size) >= 0) {
4196 spin_unlock(&delayed_rsv->lock);
4197 spin_unlock(&space_info->lock);
4200 spin_unlock(&delayed_rsv->lock);
4201 spin_unlock(&space_info->lock);
4204 trans = btrfs_join_transaction(root);
4208 return btrfs_commit_transaction(trans, root);
4212 FLUSH_DELAYED_ITEMS_NR = 1,
4213 FLUSH_DELAYED_ITEMS = 2,
4215 FLUSH_DELALLOC_WAIT = 4,
4220 static int flush_space(struct btrfs_root *root,
4221 struct btrfs_space_info *space_info, u64 num_bytes,
4222 u64 orig_bytes, int state)
4224 struct btrfs_trans_handle *trans;
4229 case FLUSH_DELAYED_ITEMS_NR:
4230 case FLUSH_DELAYED_ITEMS:
4231 if (state == FLUSH_DELAYED_ITEMS_NR)
4232 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4236 trans = btrfs_join_transaction(root);
4237 if (IS_ERR(trans)) {
4238 ret = PTR_ERR(trans);
4241 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4242 btrfs_end_transaction(trans, root);
4244 case FLUSH_DELALLOC:
4245 case FLUSH_DELALLOC_WAIT:
4246 shrink_delalloc(root, num_bytes, orig_bytes,
4247 state == FLUSH_DELALLOC_WAIT);
4250 trans = btrfs_join_transaction(root);
4251 if (IS_ERR(trans)) {
4252 ret = PTR_ERR(trans);
4255 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4256 btrfs_get_alloc_profile(root, 0),
4257 CHUNK_ALLOC_NO_FORCE);
4258 btrfs_end_transaction(trans, root);
4263 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4273 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4274 * @root - the root we're allocating for
4275 * @block_rsv - the block_rsv we're allocating for
4276 * @orig_bytes - the number of bytes we want
4277 * @flush - whether or not we can flush to make our reservation
4279 * This will reserve orgi_bytes number of bytes from the space info associated
4280 * with the block_rsv. If there is not enough space it will make an attempt to
4281 * flush out space to make room. It will do this by flushing delalloc if
4282 * possible or committing the transaction. If flush is 0 then no attempts to
4283 * regain reservations will be made and this will fail if there is not enough
4286 static int reserve_metadata_bytes(struct btrfs_root *root,
4287 struct btrfs_block_rsv *block_rsv,
4289 enum btrfs_reserve_flush_enum flush)
4291 struct btrfs_space_info *space_info = block_rsv->space_info;
4293 u64 num_bytes = orig_bytes;
4294 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4296 bool flushing = false;
4300 spin_lock(&space_info->lock);
4302 * We only want to wait if somebody other than us is flushing and we
4303 * are actually allowed to flush all things.
4305 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4306 space_info->flush) {
4307 spin_unlock(&space_info->lock);
4309 * If we have a trans handle we can't wait because the flusher
4310 * may have to commit the transaction, which would mean we would
4311 * deadlock since we are waiting for the flusher to finish, but
4312 * hold the current transaction open.
4314 if (current->journal_info)
4316 ret = wait_event_killable(space_info->wait, !space_info->flush);
4317 /* Must have been killed, return */
4321 spin_lock(&space_info->lock);
4325 used = space_info->bytes_used + space_info->bytes_reserved +
4326 space_info->bytes_pinned + space_info->bytes_readonly +
4327 space_info->bytes_may_use;
4330 * The idea here is that we've not already over-reserved the block group
4331 * then we can go ahead and save our reservation first and then start
4332 * flushing if we need to. Otherwise if we've already overcommitted
4333 * lets start flushing stuff first and then come back and try to make
4336 if (used <= space_info->total_bytes) {
4337 if (used + orig_bytes <= space_info->total_bytes) {
4338 space_info->bytes_may_use += orig_bytes;
4339 trace_btrfs_space_reservation(root->fs_info,
4340 "space_info", space_info->flags, orig_bytes, 1);
4344 * Ok set num_bytes to orig_bytes since we aren't
4345 * overocmmitted, this way we only try and reclaim what
4348 num_bytes = orig_bytes;
4352 * Ok we're over committed, set num_bytes to the overcommitted
4353 * amount plus the amount of bytes that we need for this
4356 num_bytes = used - space_info->total_bytes +
4360 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4361 space_info->bytes_may_use += orig_bytes;
4362 trace_btrfs_space_reservation(root->fs_info, "space_info",
4363 space_info->flags, orig_bytes,
4369 * Couldn't make our reservation, save our place so while we're trying
4370 * to reclaim space we can actually use it instead of somebody else
4371 * stealing it from us.
4373 * We make the other tasks wait for the flush only when we can flush
4376 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4378 space_info->flush = 1;
4381 spin_unlock(&space_info->lock);
4383 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4386 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4391 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4392 * would happen. So skip delalloc flush.
4394 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4395 (flush_state == FLUSH_DELALLOC ||
4396 flush_state == FLUSH_DELALLOC_WAIT))
4397 flush_state = ALLOC_CHUNK;
4401 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4402 flush_state < COMMIT_TRANS)
4404 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4405 flush_state <= COMMIT_TRANS)
4409 if (ret == -ENOSPC &&
4410 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4411 struct btrfs_block_rsv *global_rsv =
4412 &root->fs_info->global_block_rsv;
4414 if (block_rsv != global_rsv &&
4415 !block_rsv_use_bytes(global_rsv, orig_bytes))
4419 trace_btrfs_space_reservation(root->fs_info,
4420 "space_info:enospc",
4421 space_info->flags, orig_bytes, 1);
4423 spin_lock(&space_info->lock);
4424 space_info->flush = 0;
4425 wake_up_all(&space_info->wait);
4426 spin_unlock(&space_info->lock);
4431 static struct btrfs_block_rsv *get_block_rsv(
4432 const struct btrfs_trans_handle *trans,
4433 const struct btrfs_root *root)
4435 struct btrfs_block_rsv *block_rsv = NULL;
4438 block_rsv = trans->block_rsv;
4440 if (root == root->fs_info->csum_root && trans->adding_csums)
4441 block_rsv = trans->block_rsv;
4443 if (root == root->fs_info->uuid_root)
4444 block_rsv = trans->block_rsv;
4447 block_rsv = root->block_rsv;
4450 block_rsv = &root->fs_info->empty_block_rsv;
4455 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4459 spin_lock(&block_rsv->lock);
4460 if (block_rsv->reserved >= num_bytes) {
4461 block_rsv->reserved -= num_bytes;
4462 if (block_rsv->reserved < block_rsv->size)
4463 block_rsv->full = 0;
4466 spin_unlock(&block_rsv->lock);
4470 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4471 u64 num_bytes, int update_size)
4473 spin_lock(&block_rsv->lock);
4474 block_rsv->reserved += num_bytes;
4476 block_rsv->size += num_bytes;
4477 else if (block_rsv->reserved >= block_rsv->size)
4478 block_rsv->full = 1;
4479 spin_unlock(&block_rsv->lock);
4482 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4483 struct btrfs_block_rsv *dest, u64 num_bytes,
4486 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4489 if (global_rsv->space_info != dest->space_info)
4492 spin_lock(&global_rsv->lock);
4493 min_bytes = div_factor(global_rsv->size, min_factor);
4494 if (global_rsv->reserved < min_bytes + num_bytes) {
4495 spin_unlock(&global_rsv->lock);
4498 global_rsv->reserved -= num_bytes;
4499 if (global_rsv->reserved < global_rsv->size)
4500 global_rsv->full = 0;
4501 spin_unlock(&global_rsv->lock);
4503 block_rsv_add_bytes(dest, num_bytes, 1);
4507 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4508 struct btrfs_block_rsv *block_rsv,
4509 struct btrfs_block_rsv *dest, u64 num_bytes)
4511 struct btrfs_space_info *space_info = block_rsv->space_info;
4513 spin_lock(&block_rsv->lock);
4514 if (num_bytes == (u64)-1)
4515 num_bytes = block_rsv->size;
4516 block_rsv->size -= num_bytes;
4517 if (block_rsv->reserved >= block_rsv->size) {
4518 num_bytes = block_rsv->reserved - block_rsv->size;
4519 block_rsv->reserved = block_rsv->size;
4520 block_rsv->full = 1;
4524 spin_unlock(&block_rsv->lock);
4526 if (num_bytes > 0) {
4528 spin_lock(&dest->lock);
4532 bytes_to_add = dest->size - dest->reserved;
4533 bytes_to_add = min(num_bytes, bytes_to_add);
4534 dest->reserved += bytes_to_add;
4535 if (dest->reserved >= dest->size)
4537 num_bytes -= bytes_to_add;
4539 spin_unlock(&dest->lock);
4542 spin_lock(&space_info->lock);
4543 space_info->bytes_may_use -= num_bytes;
4544 trace_btrfs_space_reservation(fs_info, "space_info",
4545 space_info->flags, num_bytes, 0);
4546 spin_unlock(&space_info->lock);
4551 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4552 struct btrfs_block_rsv *dst, u64 num_bytes)
4556 ret = block_rsv_use_bytes(src, num_bytes);
4560 block_rsv_add_bytes(dst, num_bytes, 1);
4564 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4566 memset(rsv, 0, sizeof(*rsv));
4567 spin_lock_init(&rsv->lock);
4571 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4572 unsigned short type)
4574 struct btrfs_block_rsv *block_rsv;
4575 struct btrfs_fs_info *fs_info = root->fs_info;
4577 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4581 btrfs_init_block_rsv(block_rsv, type);
4582 block_rsv->space_info = __find_space_info(fs_info,
4583 BTRFS_BLOCK_GROUP_METADATA);
4587 void btrfs_free_block_rsv(struct btrfs_root *root,
4588 struct btrfs_block_rsv *rsv)
4592 btrfs_block_rsv_release(root, rsv, (u64)-1);
4596 int btrfs_block_rsv_add(struct btrfs_root *root,
4597 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4598 enum btrfs_reserve_flush_enum flush)
4605 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4607 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4614 int btrfs_block_rsv_check(struct btrfs_root *root,
4615 struct btrfs_block_rsv *block_rsv, int min_factor)
4623 spin_lock(&block_rsv->lock);
4624 num_bytes = div_factor(block_rsv->size, min_factor);
4625 if (block_rsv->reserved >= num_bytes)
4627 spin_unlock(&block_rsv->lock);
4632 int btrfs_block_rsv_refill(struct btrfs_root *root,
4633 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4634 enum btrfs_reserve_flush_enum flush)
4642 spin_lock(&block_rsv->lock);
4643 num_bytes = min_reserved;
4644 if (block_rsv->reserved >= num_bytes)
4647 num_bytes -= block_rsv->reserved;
4648 spin_unlock(&block_rsv->lock);
4653 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4655 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4662 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4663 struct btrfs_block_rsv *dst_rsv,
4666 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4669 void btrfs_block_rsv_release(struct btrfs_root *root,
4670 struct btrfs_block_rsv *block_rsv,
4673 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4674 if (global_rsv->full || global_rsv == block_rsv ||
4675 block_rsv->space_info != global_rsv->space_info)
4677 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4682 * helper to calculate size of global block reservation.
4683 * the desired value is sum of space used by extent tree,
4684 * checksum tree and root tree
4686 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4688 struct btrfs_space_info *sinfo;
4692 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4694 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4695 spin_lock(&sinfo->lock);
4696 data_used = sinfo->bytes_used;
4697 spin_unlock(&sinfo->lock);
4699 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4700 spin_lock(&sinfo->lock);
4701 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4703 meta_used = sinfo->bytes_used;
4704 spin_unlock(&sinfo->lock);
4706 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4708 num_bytes += div64_u64(data_used + meta_used, 50);
4710 if (num_bytes * 3 > meta_used)
4711 num_bytes = div64_u64(meta_used, 3);
4713 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4716 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4718 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4719 struct btrfs_space_info *sinfo = block_rsv->space_info;
4722 num_bytes = calc_global_metadata_size(fs_info);
4724 spin_lock(&sinfo->lock);
4725 spin_lock(&block_rsv->lock);
4727 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4729 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4730 sinfo->bytes_reserved + sinfo->bytes_readonly +
4731 sinfo->bytes_may_use;
4733 if (sinfo->total_bytes > num_bytes) {
4734 num_bytes = sinfo->total_bytes - num_bytes;
4735 block_rsv->reserved += num_bytes;
4736 sinfo->bytes_may_use += num_bytes;
4737 trace_btrfs_space_reservation(fs_info, "space_info",
4738 sinfo->flags, num_bytes, 1);
4741 if (block_rsv->reserved >= block_rsv->size) {
4742 num_bytes = block_rsv->reserved - block_rsv->size;
4743 sinfo->bytes_may_use -= num_bytes;
4744 trace_btrfs_space_reservation(fs_info, "space_info",
4745 sinfo->flags, num_bytes, 0);
4746 block_rsv->reserved = block_rsv->size;
4747 block_rsv->full = 1;
4750 spin_unlock(&block_rsv->lock);
4751 spin_unlock(&sinfo->lock);
4754 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4756 struct btrfs_space_info *space_info;
4758 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4759 fs_info->chunk_block_rsv.space_info = space_info;
4761 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4762 fs_info->global_block_rsv.space_info = space_info;
4763 fs_info->delalloc_block_rsv.space_info = space_info;
4764 fs_info->trans_block_rsv.space_info = space_info;
4765 fs_info->empty_block_rsv.space_info = space_info;
4766 fs_info->delayed_block_rsv.space_info = space_info;
4768 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4769 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4770 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4771 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4772 if (fs_info->quota_root)
4773 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4774 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4776 update_global_block_rsv(fs_info);
4779 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4781 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4783 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4784 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4785 WARN_ON(fs_info->trans_block_rsv.size > 0);
4786 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4787 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4788 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4789 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4790 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4793 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4794 struct btrfs_root *root)
4796 if (!trans->block_rsv)
4799 if (!trans->bytes_reserved)
4802 trace_btrfs_space_reservation(root->fs_info, "transaction",
4803 trans->transid, trans->bytes_reserved, 0);
4804 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4805 trans->bytes_reserved = 0;
4808 /* Can only return 0 or -ENOSPC */
4809 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4810 struct inode *inode)
4812 struct btrfs_root *root = BTRFS_I(inode)->root;
4813 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4814 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4817 * We need to hold space in order to delete our orphan item once we've
4818 * added it, so this takes the reservation so we can release it later
4819 * when we are truly done with the orphan item.
4821 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4822 trace_btrfs_space_reservation(root->fs_info, "orphan",
4823 btrfs_ino(inode), num_bytes, 1);
4824 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4827 void btrfs_orphan_release_metadata(struct inode *inode)
4829 struct btrfs_root *root = BTRFS_I(inode)->root;
4830 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4831 trace_btrfs_space_reservation(root->fs_info, "orphan",
4832 btrfs_ino(inode), num_bytes, 0);
4833 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4837 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4838 * root: the root of the parent directory
4839 * rsv: block reservation
4840 * items: the number of items that we need do reservation
4841 * qgroup_reserved: used to return the reserved size in qgroup
4843 * This function is used to reserve the space for snapshot/subvolume
4844 * creation and deletion. Those operations are different with the
4845 * common file/directory operations, they change two fs/file trees
4846 * and root tree, the number of items that the qgroup reserves is
4847 * different with the free space reservation. So we can not use
4848 * the space reseravtion mechanism in start_transaction().
4850 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4851 struct btrfs_block_rsv *rsv,
4853 u64 *qgroup_reserved,
4854 bool use_global_rsv)
4858 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4860 if (root->fs_info->quota_enabled) {
4861 /* One for parent inode, two for dir entries */
4862 num_bytes = 3 * root->leafsize;
4863 ret = btrfs_qgroup_reserve(root, num_bytes);
4870 *qgroup_reserved = num_bytes;
4872 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4873 rsv->space_info = __find_space_info(root->fs_info,
4874 BTRFS_BLOCK_GROUP_METADATA);
4875 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4876 BTRFS_RESERVE_FLUSH_ALL);
4878 if (ret == -ENOSPC && use_global_rsv)
4879 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4882 if (*qgroup_reserved)
4883 btrfs_qgroup_free(root, *qgroup_reserved);
4889 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4890 struct btrfs_block_rsv *rsv,
4891 u64 qgroup_reserved)
4893 btrfs_block_rsv_release(root, rsv, (u64)-1);
4894 if (qgroup_reserved)
4895 btrfs_qgroup_free(root, qgroup_reserved);
4899 * drop_outstanding_extent - drop an outstanding extent
4900 * @inode: the inode we're dropping the extent for
4902 * This is called when we are freeing up an outstanding extent, either called
4903 * after an error or after an extent is written. This will return the number of
4904 * reserved extents that need to be freed. This must be called with
4905 * BTRFS_I(inode)->lock held.
4907 static unsigned drop_outstanding_extent(struct inode *inode)
4909 unsigned drop_inode_space = 0;
4910 unsigned dropped_extents = 0;
4912 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4913 BTRFS_I(inode)->outstanding_extents--;
4915 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4916 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4917 &BTRFS_I(inode)->runtime_flags))
4918 drop_inode_space = 1;
4921 * If we have more or the same amount of outsanding extents than we have
4922 * reserved then we need to leave the reserved extents count alone.
4924 if (BTRFS_I(inode)->outstanding_extents >=
4925 BTRFS_I(inode)->reserved_extents)
4926 return drop_inode_space;
4928 dropped_extents = BTRFS_I(inode)->reserved_extents -
4929 BTRFS_I(inode)->outstanding_extents;
4930 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4931 return dropped_extents + drop_inode_space;
4935 * calc_csum_metadata_size - return the amount of metada space that must be
4936 * reserved/free'd for the given bytes.
4937 * @inode: the inode we're manipulating
4938 * @num_bytes: the number of bytes in question
4939 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4941 * This adjusts the number of csum_bytes in the inode and then returns the
4942 * correct amount of metadata that must either be reserved or freed. We
4943 * calculate how many checksums we can fit into one leaf and then divide the
4944 * number of bytes that will need to be checksumed by this value to figure out
4945 * how many checksums will be required. If we are adding bytes then the number
4946 * may go up and we will return the number of additional bytes that must be
4947 * reserved. If it is going down we will return the number of bytes that must
4950 * This must be called with BTRFS_I(inode)->lock held.
4952 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4955 struct btrfs_root *root = BTRFS_I(inode)->root;
4957 int num_csums_per_leaf;
4961 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4962 BTRFS_I(inode)->csum_bytes == 0)
4965 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4967 BTRFS_I(inode)->csum_bytes += num_bytes;
4969 BTRFS_I(inode)->csum_bytes -= num_bytes;
4970 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4971 num_csums_per_leaf = (int)div64_u64(csum_size,
4972 sizeof(struct btrfs_csum_item) +
4973 sizeof(struct btrfs_disk_key));
4974 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4975 num_csums = num_csums + num_csums_per_leaf - 1;
4976 num_csums = num_csums / num_csums_per_leaf;
4978 old_csums = old_csums + num_csums_per_leaf - 1;
4979 old_csums = old_csums / num_csums_per_leaf;
4981 /* No change, no need to reserve more */
4982 if (old_csums == num_csums)
4986 return btrfs_calc_trans_metadata_size(root,
4987 num_csums - old_csums);
4989 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4992 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4994 struct btrfs_root *root = BTRFS_I(inode)->root;
4995 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4998 unsigned nr_extents = 0;
4999 int extra_reserve = 0;
5000 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5002 bool delalloc_lock = true;
5006 /* If we are a free space inode we need to not flush since we will be in
5007 * the middle of a transaction commit. We also don't need the delalloc
5008 * mutex since we won't race with anybody. We need this mostly to make
5009 * lockdep shut its filthy mouth.
5011 if (btrfs_is_free_space_inode(inode)) {
5012 flush = BTRFS_RESERVE_NO_FLUSH;
5013 delalloc_lock = false;
5016 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5017 btrfs_transaction_in_commit(root->fs_info))
5018 schedule_timeout(1);
5021 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5023 num_bytes = ALIGN(num_bytes, root->sectorsize);
5025 spin_lock(&BTRFS_I(inode)->lock);
5026 BTRFS_I(inode)->outstanding_extents++;
5028 if (BTRFS_I(inode)->outstanding_extents >
5029 BTRFS_I(inode)->reserved_extents)
5030 nr_extents = BTRFS_I(inode)->outstanding_extents -
5031 BTRFS_I(inode)->reserved_extents;
5034 * Add an item to reserve for updating the inode when we complete the
5037 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5038 &BTRFS_I(inode)->runtime_flags)) {
5043 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5044 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5045 csum_bytes = BTRFS_I(inode)->csum_bytes;
5046 spin_unlock(&BTRFS_I(inode)->lock);
5048 if (root->fs_info->quota_enabled) {
5049 ret = btrfs_qgroup_reserve(root, num_bytes +
5050 nr_extents * root->leafsize);
5055 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5056 if (unlikely(ret)) {
5057 if (root->fs_info->quota_enabled)
5058 btrfs_qgroup_free(root, num_bytes +
5059 nr_extents * root->leafsize);
5063 spin_lock(&BTRFS_I(inode)->lock);
5064 if (extra_reserve) {
5065 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5066 &BTRFS_I(inode)->runtime_flags);
5069 BTRFS_I(inode)->reserved_extents += nr_extents;
5070 spin_unlock(&BTRFS_I(inode)->lock);
5073 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5076 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5077 btrfs_ino(inode), to_reserve, 1);
5078 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5083 spin_lock(&BTRFS_I(inode)->lock);
5084 dropped = drop_outstanding_extent(inode);
5086 * If the inodes csum_bytes is the same as the original
5087 * csum_bytes then we know we haven't raced with any free()ers
5088 * so we can just reduce our inodes csum bytes and carry on.
5090 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5091 calc_csum_metadata_size(inode, num_bytes, 0);
5093 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5097 * This is tricky, but first we need to figure out how much we
5098 * free'd from any free-ers that occured during this
5099 * reservation, so we reset ->csum_bytes to the csum_bytes
5100 * before we dropped our lock, and then call the free for the
5101 * number of bytes that were freed while we were trying our
5104 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5105 BTRFS_I(inode)->csum_bytes = csum_bytes;
5106 to_free = calc_csum_metadata_size(inode, bytes, 0);
5110 * Now we need to see how much we would have freed had we not
5111 * been making this reservation and our ->csum_bytes were not
5112 * artificially inflated.
5114 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5115 bytes = csum_bytes - orig_csum_bytes;
5116 bytes = calc_csum_metadata_size(inode, bytes, 0);
5119 * Now reset ->csum_bytes to what it should be. If bytes is
5120 * more than to_free then we would have free'd more space had we
5121 * not had an artificially high ->csum_bytes, so we need to free
5122 * the remainder. If bytes is the same or less then we don't
5123 * need to do anything, the other free-ers did the correct
5126 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5127 if (bytes > to_free)
5128 to_free = bytes - to_free;
5132 spin_unlock(&BTRFS_I(inode)->lock);
5134 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5137 btrfs_block_rsv_release(root, block_rsv, to_free);
5138 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5139 btrfs_ino(inode), to_free, 0);
5142 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5147 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5148 * @inode: the inode to release the reservation for
5149 * @num_bytes: the number of bytes we're releasing
5151 * This will release the metadata reservation for an inode. This can be called
5152 * once we complete IO for a given set of bytes to release their metadata
5155 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5157 struct btrfs_root *root = BTRFS_I(inode)->root;
5161 num_bytes = ALIGN(num_bytes, root->sectorsize);
5162 spin_lock(&BTRFS_I(inode)->lock);
5163 dropped = drop_outstanding_extent(inode);
5166 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5167 spin_unlock(&BTRFS_I(inode)->lock);
5169 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5171 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5172 btrfs_ino(inode), to_free, 0);
5173 if (root->fs_info->quota_enabled) {
5174 btrfs_qgroup_free(root, num_bytes +
5175 dropped * root->leafsize);
5178 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5183 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5184 * @inode: inode we're writing to
5185 * @num_bytes: the number of bytes we want to allocate
5187 * This will do the following things
5189 * o reserve space in the data space info for num_bytes
5190 * o reserve space in the metadata space info based on number of outstanding
5191 * extents and how much csums will be needed
5192 * o add to the inodes ->delalloc_bytes
5193 * o add it to the fs_info's delalloc inodes list.
5195 * This will return 0 for success and -ENOSPC if there is no space left.
5197 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5201 ret = btrfs_check_data_free_space(inode, num_bytes);
5205 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5207 btrfs_free_reserved_data_space(inode, num_bytes);
5215 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5216 * @inode: inode we're releasing space for
5217 * @num_bytes: the number of bytes we want to free up
5219 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5220 * called in the case that we don't need the metadata AND data reservations
5221 * anymore. So if there is an error or we insert an inline extent.
5223 * This function will release the metadata space that was not used and will
5224 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5225 * list if there are no delalloc bytes left.
5227 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5229 btrfs_delalloc_release_metadata(inode, num_bytes);
5230 btrfs_free_reserved_data_space(inode, num_bytes);
5233 static int update_block_group(struct btrfs_root *root,
5234 u64 bytenr, u64 num_bytes, int alloc)
5236 struct btrfs_block_group_cache *cache = NULL;
5237 struct btrfs_fs_info *info = root->fs_info;
5238 u64 total = num_bytes;
5243 /* block accounting for super block */
5244 spin_lock(&info->delalloc_root_lock);
5245 old_val = btrfs_super_bytes_used(info->super_copy);
5247 old_val += num_bytes;
5249 old_val -= num_bytes;
5250 btrfs_set_super_bytes_used(info->super_copy, old_val);
5251 spin_unlock(&info->delalloc_root_lock);
5254 cache = btrfs_lookup_block_group(info, bytenr);
5257 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5258 BTRFS_BLOCK_GROUP_RAID1 |
5259 BTRFS_BLOCK_GROUP_RAID10))
5264 * If this block group has free space cache written out, we
5265 * need to make sure to load it if we are removing space. This
5266 * is because we need the unpinning stage to actually add the
5267 * space back to the block group, otherwise we will leak space.
5269 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5270 cache_block_group(cache, 1);
5272 byte_in_group = bytenr - cache->key.objectid;
5273 WARN_ON(byte_in_group > cache->key.offset);
5275 spin_lock(&cache->space_info->lock);
5276 spin_lock(&cache->lock);
5278 if (btrfs_test_opt(root, SPACE_CACHE) &&
5279 cache->disk_cache_state < BTRFS_DC_CLEAR)
5280 cache->disk_cache_state = BTRFS_DC_CLEAR;
5283 old_val = btrfs_block_group_used(&cache->item);
5284 num_bytes = min(total, cache->key.offset - byte_in_group);
5286 old_val += num_bytes;
5287 btrfs_set_block_group_used(&cache->item, old_val);
5288 cache->reserved -= num_bytes;
5289 cache->space_info->bytes_reserved -= num_bytes;
5290 cache->space_info->bytes_used += num_bytes;
5291 cache->space_info->disk_used += num_bytes * factor;
5292 spin_unlock(&cache->lock);
5293 spin_unlock(&cache->space_info->lock);
5295 old_val -= num_bytes;
5296 btrfs_set_block_group_used(&cache->item, old_val);
5297 cache->pinned += num_bytes;
5298 cache->space_info->bytes_pinned += num_bytes;
5299 cache->space_info->bytes_used -= num_bytes;
5300 cache->space_info->disk_used -= num_bytes * factor;
5301 spin_unlock(&cache->lock);
5302 spin_unlock(&cache->space_info->lock);
5304 set_extent_dirty(info->pinned_extents,
5305 bytenr, bytenr + num_bytes - 1,
5306 GFP_NOFS | __GFP_NOFAIL);
5308 btrfs_put_block_group(cache);
5310 bytenr += num_bytes;
5315 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5317 struct btrfs_block_group_cache *cache;
5320 spin_lock(&root->fs_info->block_group_cache_lock);
5321 bytenr = root->fs_info->first_logical_byte;
5322 spin_unlock(&root->fs_info->block_group_cache_lock);
5324 if (bytenr < (u64)-1)
5327 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5331 bytenr = cache->key.objectid;
5332 btrfs_put_block_group(cache);
5337 static int pin_down_extent(struct btrfs_root *root,
5338 struct btrfs_block_group_cache *cache,
5339 u64 bytenr, u64 num_bytes, int reserved)
5341 spin_lock(&cache->space_info->lock);
5342 spin_lock(&cache->lock);
5343 cache->pinned += num_bytes;
5344 cache->space_info->bytes_pinned += num_bytes;
5346 cache->reserved -= num_bytes;
5347 cache->space_info->bytes_reserved -= num_bytes;
5349 spin_unlock(&cache->lock);
5350 spin_unlock(&cache->space_info->lock);
5352 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5353 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5355 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5360 * this function must be called within transaction
5362 int btrfs_pin_extent(struct btrfs_root *root,
5363 u64 bytenr, u64 num_bytes, int reserved)
5365 struct btrfs_block_group_cache *cache;
5367 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5368 BUG_ON(!cache); /* Logic error */
5370 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5372 btrfs_put_block_group(cache);
5377 * this function must be called within transaction
5379 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5380 u64 bytenr, u64 num_bytes)
5382 struct btrfs_block_group_cache *cache;
5385 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5390 * pull in the free space cache (if any) so that our pin
5391 * removes the free space from the cache. We have load_only set
5392 * to one because the slow code to read in the free extents does check
5393 * the pinned extents.
5395 cache_block_group(cache, 1);
5397 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5399 /* remove us from the free space cache (if we're there at all) */
5400 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5401 btrfs_put_block_group(cache);
5405 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5408 struct btrfs_block_group_cache *block_group;
5409 struct btrfs_caching_control *caching_ctl;
5411 block_group = btrfs_lookup_block_group(root->fs_info, start);
5415 cache_block_group(block_group, 0);
5416 caching_ctl = get_caching_control(block_group);
5420 BUG_ON(!block_group_cache_done(block_group));
5421 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5423 mutex_lock(&caching_ctl->mutex);
5425 if (start >= caching_ctl->progress) {
5426 ret = add_excluded_extent(root, start, num_bytes);
5427 } else if (start + num_bytes <= caching_ctl->progress) {
5428 ret = btrfs_remove_free_space(block_group,
5431 num_bytes = caching_ctl->progress - start;
5432 ret = btrfs_remove_free_space(block_group,
5437 num_bytes = (start + num_bytes) -
5438 caching_ctl->progress;
5439 start = caching_ctl->progress;
5440 ret = add_excluded_extent(root, start, num_bytes);
5443 mutex_unlock(&caching_ctl->mutex);
5444 put_caching_control(caching_ctl);
5446 btrfs_put_block_group(block_group);
5450 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5451 struct extent_buffer *eb)
5453 struct btrfs_file_extent_item *item;
5454 struct btrfs_key key;
5458 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5461 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5462 btrfs_item_key_to_cpu(eb, &key, i);
5463 if (key.type != BTRFS_EXTENT_DATA_KEY)
5465 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5466 found_type = btrfs_file_extent_type(eb, item);
5467 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5469 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5471 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5472 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5473 __exclude_logged_extent(log, key.objectid, key.offset);
5480 * btrfs_update_reserved_bytes - update the block_group and space info counters
5481 * @cache: The cache we are manipulating
5482 * @num_bytes: The number of bytes in question
5483 * @reserve: One of the reservation enums
5485 * This is called by the allocator when it reserves space, or by somebody who is
5486 * freeing space that was never actually used on disk. For example if you
5487 * reserve some space for a new leaf in transaction A and before transaction A
5488 * commits you free that leaf, you call this with reserve set to 0 in order to
5489 * clear the reservation.
5491 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5492 * ENOSPC accounting. For data we handle the reservation through clearing the
5493 * delalloc bits in the io_tree. We have to do this since we could end up
5494 * allocating less disk space for the amount of data we have reserved in the
5495 * case of compression.
5497 * If this is a reservation and the block group has become read only we cannot
5498 * make the reservation and return -EAGAIN, otherwise this function always
5501 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5502 u64 num_bytes, int reserve)
5504 struct btrfs_space_info *space_info = cache->space_info;
5507 spin_lock(&space_info->lock);
5508 spin_lock(&cache->lock);
5509 if (reserve != RESERVE_FREE) {
5513 cache->reserved += num_bytes;
5514 space_info->bytes_reserved += num_bytes;
5515 if (reserve == RESERVE_ALLOC) {
5516 trace_btrfs_space_reservation(cache->fs_info,
5517 "space_info", space_info->flags,
5519 space_info->bytes_may_use -= num_bytes;
5524 space_info->bytes_readonly += num_bytes;
5525 cache->reserved -= num_bytes;
5526 space_info->bytes_reserved -= num_bytes;
5528 spin_unlock(&cache->lock);
5529 spin_unlock(&space_info->lock);
5533 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5534 struct btrfs_root *root)
5536 struct btrfs_fs_info *fs_info = root->fs_info;
5537 struct btrfs_caching_control *next;
5538 struct btrfs_caching_control *caching_ctl;
5539 struct btrfs_block_group_cache *cache;
5540 struct btrfs_space_info *space_info;
5542 down_write(&fs_info->extent_commit_sem);
5544 list_for_each_entry_safe(caching_ctl, next,
5545 &fs_info->caching_block_groups, list) {
5546 cache = caching_ctl->block_group;
5547 if (block_group_cache_done(cache)) {
5548 cache->last_byte_to_unpin = (u64)-1;
5549 list_del_init(&caching_ctl->list);
5550 put_caching_control(caching_ctl);
5552 cache->last_byte_to_unpin = caching_ctl->progress;
5556 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5557 fs_info->pinned_extents = &fs_info->freed_extents[1];
5559 fs_info->pinned_extents = &fs_info->freed_extents[0];
5561 up_write(&fs_info->extent_commit_sem);
5563 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5564 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5566 update_global_block_rsv(fs_info);
5569 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5571 struct btrfs_fs_info *fs_info = root->fs_info;
5572 struct btrfs_block_group_cache *cache = NULL;
5573 struct btrfs_space_info *space_info;
5574 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5578 while (start <= end) {
5581 start >= cache->key.objectid + cache->key.offset) {
5583 btrfs_put_block_group(cache);
5584 cache = btrfs_lookup_block_group(fs_info, start);
5585 BUG_ON(!cache); /* Logic error */
5588 len = cache->key.objectid + cache->key.offset - start;
5589 len = min(len, end + 1 - start);
5591 if (start < cache->last_byte_to_unpin) {
5592 len = min(len, cache->last_byte_to_unpin - start);
5593 btrfs_add_free_space(cache, start, len);
5597 space_info = cache->space_info;
5599 spin_lock(&space_info->lock);
5600 spin_lock(&cache->lock);
5601 cache->pinned -= len;
5602 space_info->bytes_pinned -= len;
5604 space_info->bytes_readonly += len;
5607 spin_unlock(&cache->lock);
5608 if (!readonly && global_rsv->space_info == space_info) {
5609 spin_lock(&global_rsv->lock);
5610 if (!global_rsv->full) {
5611 len = min(len, global_rsv->size -
5612 global_rsv->reserved);
5613 global_rsv->reserved += len;
5614 space_info->bytes_may_use += len;
5615 if (global_rsv->reserved >= global_rsv->size)
5616 global_rsv->full = 1;
5618 spin_unlock(&global_rsv->lock);
5620 spin_unlock(&space_info->lock);
5624 btrfs_put_block_group(cache);
5628 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5629 struct btrfs_root *root)
5631 struct btrfs_fs_info *fs_info = root->fs_info;
5632 struct extent_io_tree *unpin;
5640 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5641 unpin = &fs_info->freed_extents[1];
5643 unpin = &fs_info->freed_extents[0];
5646 ret = find_first_extent_bit(unpin, 0, &start, &end,
5647 EXTENT_DIRTY, NULL);
5651 if (btrfs_test_opt(root, DISCARD))
5652 ret = btrfs_discard_extent(root, start,
5653 end + 1 - start, NULL);
5655 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5656 unpin_extent_range(root, start, end);
5663 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5664 u64 owner, u64 root_objectid)
5666 struct btrfs_space_info *space_info;
5669 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5670 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5671 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5673 flags = BTRFS_BLOCK_GROUP_METADATA;
5675 flags = BTRFS_BLOCK_GROUP_DATA;
5678 space_info = __find_space_info(fs_info, flags);
5679 BUG_ON(!space_info); /* Logic bug */
5680 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5684 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5685 struct btrfs_root *root,
5686 u64 bytenr, u64 num_bytes, u64 parent,
5687 u64 root_objectid, u64 owner_objectid,
5688 u64 owner_offset, int refs_to_drop,
5689 struct btrfs_delayed_extent_op *extent_op)
5691 struct btrfs_key key;
5692 struct btrfs_path *path;
5693 struct btrfs_fs_info *info = root->fs_info;
5694 struct btrfs_root *extent_root = info->extent_root;
5695 struct extent_buffer *leaf;
5696 struct btrfs_extent_item *ei;
5697 struct btrfs_extent_inline_ref *iref;
5700 int extent_slot = 0;
5701 int found_extent = 0;
5705 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5708 path = btrfs_alloc_path();
5713 path->leave_spinning = 1;
5715 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5716 BUG_ON(!is_data && refs_to_drop != 1);
5719 skinny_metadata = 0;
5721 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5722 bytenr, num_bytes, parent,
5723 root_objectid, owner_objectid,
5726 extent_slot = path->slots[0];
5727 while (extent_slot >= 0) {
5728 btrfs_item_key_to_cpu(path->nodes[0], &key,
5730 if (key.objectid != bytenr)
5732 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5733 key.offset == num_bytes) {
5737 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5738 key.offset == owner_objectid) {
5742 if (path->slots[0] - extent_slot > 5)
5746 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5747 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5748 if (found_extent && item_size < sizeof(*ei))
5751 if (!found_extent) {
5753 ret = remove_extent_backref(trans, extent_root, path,
5757 btrfs_abort_transaction(trans, extent_root, ret);
5760 btrfs_release_path(path);
5761 path->leave_spinning = 1;
5763 key.objectid = bytenr;
5764 key.type = BTRFS_EXTENT_ITEM_KEY;
5765 key.offset = num_bytes;
5767 if (!is_data && skinny_metadata) {
5768 key.type = BTRFS_METADATA_ITEM_KEY;
5769 key.offset = owner_objectid;
5772 ret = btrfs_search_slot(trans, extent_root,
5774 if (ret > 0 && skinny_metadata && path->slots[0]) {
5776 * Couldn't find our skinny metadata item,
5777 * see if we have ye olde extent item.
5780 btrfs_item_key_to_cpu(path->nodes[0], &key,
5782 if (key.objectid == bytenr &&
5783 key.type == BTRFS_EXTENT_ITEM_KEY &&
5784 key.offset == num_bytes)
5788 if (ret > 0 && skinny_metadata) {
5789 skinny_metadata = false;
5790 key.type = BTRFS_EXTENT_ITEM_KEY;
5791 key.offset = num_bytes;
5792 btrfs_release_path(path);
5793 ret = btrfs_search_slot(trans, extent_root,
5798 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5801 btrfs_print_leaf(extent_root,
5805 btrfs_abort_transaction(trans, extent_root, ret);
5808 extent_slot = path->slots[0];
5810 } else if (WARN_ON(ret == -ENOENT)) {
5811 btrfs_print_leaf(extent_root, path->nodes[0]);
5813 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5814 bytenr, parent, root_objectid, owner_objectid,
5817 btrfs_abort_transaction(trans, extent_root, ret);
5821 leaf = path->nodes[0];
5822 item_size = btrfs_item_size_nr(leaf, extent_slot);
5823 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5824 if (item_size < sizeof(*ei)) {
5825 BUG_ON(found_extent || extent_slot != path->slots[0]);
5826 ret = convert_extent_item_v0(trans, extent_root, path,
5829 btrfs_abort_transaction(trans, extent_root, ret);
5833 btrfs_release_path(path);
5834 path->leave_spinning = 1;
5836 key.objectid = bytenr;
5837 key.type = BTRFS_EXTENT_ITEM_KEY;
5838 key.offset = num_bytes;
5840 ret = btrfs_search_slot(trans, extent_root, &key, path,
5843 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5845 btrfs_print_leaf(extent_root, path->nodes[0]);
5848 btrfs_abort_transaction(trans, extent_root, ret);
5852 extent_slot = path->slots[0];
5853 leaf = path->nodes[0];
5854 item_size = btrfs_item_size_nr(leaf, extent_slot);
5857 BUG_ON(item_size < sizeof(*ei));
5858 ei = btrfs_item_ptr(leaf, extent_slot,
5859 struct btrfs_extent_item);
5860 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5861 key.type == BTRFS_EXTENT_ITEM_KEY) {
5862 struct btrfs_tree_block_info *bi;
5863 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5864 bi = (struct btrfs_tree_block_info *)(ei + 1);
5865 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5868 refs = btrfs_extent_refs(leaf, ei);
5869 if (refs < refs_to_drop) {
5870 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5871 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5873 btrfs_abort_transaction(trans, extent_root, ret);
5876 refs -= refs_to_drop;
5880 __run_delayed_extent_op(extent_op, leaf, ei);
5882 * In the case of inline back ref, reference count will
5883 * be updated by remove_extent_backref
5886 BUG_ON(!found_extent);
5888 btrfs_set_extent_refs(leaf, ei, refs);
5889 btrfs_mark_buffer_dirty(leaf);
5892 ret = remove_extent_backref(trans, extent_root, path,
5896 btrfs_abort_transaction(trans, extent_root, ret);
5900 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5904 BUG_ON(is_data && refs_to_drop !=
5905 extent_data_ref_count(root, path, iref));
5907 BUG_ON(path->slots[0] != extent_slot);
5909 BUG_ON(path->slots[0] != extent_slot + 1);
5910 path->slots[0] = extent_slot;
5915 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5918 btrfs_abort_transaction(trans, extent_root, ret);
5921 btrfs_release_path(path);
5924 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5926 btrfs_abort_transaction(trans, extent_root, ret);
5931 ret = update_block_group(root, bytenr, num_bytes, 0);
5933 btrfs_abort_transaction(trans, extent_root, ret);
5938 btrfs_free_path(path);
5943 * when we free an block, it is possible (and likely) that we free the last
5944 * delayed ref for that extent as well. This searches the delayed ref tree for
5945 * a given extent, and if there are no other delayed refs to be processed, it
5946 * removes it from the tree.
5948 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5949 struct btrfs_root *root, u64 bytenr)
5951 struct btrfs_delayed_ref_head *head;
5952 struct btrfs_delayed_ref_root *delayed_refs;
5953 struct btrfs_delayed_ref_node *ref;
5954 struct rb_node *node;
5957 delayed_refs = &trans->transaction->delayed_refs;
5958 spin_lock(&delayed_refs->lock);
5959 head = btrfs_find_delayed_ref_head(trans, bytenr);
5963 node = rb_prev(&head->node.rb_node);
5967 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5969 /* there are still entries for this ref, we can't drop it */
5970 if (ref->bytenr == bytenr)
5973 if (head->extent_op) {
5974 if (!head->must_insert_reserved)
5976 btrfs_free_delayed_extent_op(head->extent_op);
5977 head->extent_op = NULL;
5981 * waiting for the lock here would deadlock. If someone else has it
5982 * locked they are already in the process of dropping it anyway
5984 if (!mutex_trylock(&head->mutex))
5988 * at this point we have a head with no other entries. Go
5989 * ahead and process it.
5991 head->node.in_tree = 0;
5992 rb_erase(&head->node.rb_node, &delayed_refs->root);
5993 rb_erase(&head->href_node, &delayed_refs->href_root);
5995 delayed_refs->num_entries--;
5998 * we don't take a ref on the node because we're removing it from the
5999 * tree, so we just steal the ref the tree was holding.
6001 delayed_refs->num_heads--;
6002 if (list_empty(&head->cluster))
6003 delayed_refs->num_heads_ready--;
6005 list_del_init(&head->cluster);
6006 spin_unlock(&delayed_refs->lock);
6008 BUG_ON(head->extent_op);
6009 if (head->must_insert_reserved)
6012 mutex_unlock(&head->mutex);
6013 btrfs_put_delayed_ref(&head->node);
6016 spin_unlock(&delayed_refs->lock);
6020 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6021 struct btrfs_root *root,
6022 struct extent_buffer *buf,
6023 u64 parent, int last_ref)
6025 struct btrfs_block_group_cache *cache = NULL;
6029 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6030 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6031 buf->start, buf->len,
6032 parent, root->root_key.objectid,
6033 btrfs_header_level(buf),
6034 BTRFS_DROP_DELAYED_REF, NULL, 0);
6035 BUG_ON(ret); /* -ENOMEM */
6041 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6043 if (btrfs_header_generation(buf) == trans->transid) {
6044 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6045 ret = check_ref_cleanup(trans, root, buf->start);
6050 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6051 pin_down_extent(root, cache, buf->start, buf->len, 1);
6055 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6057 btrfs_add_free_space(cache, buf->start, buf->len);
6058 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6059 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6064 add_pinned_bytes(root->fs_info, buf->len,
6065 btrfs_header_level(buf),
6066 root->root_key.objectid);
6069 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6072 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6073 btrfs_put_block_group(cache);
6076 /* Can return -ENOMEM */
6077 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6078 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6079 u64 owner, u64 offset, int for_cow)
6082 struct btrfs_fs_info *fs_info = root->fs_info;
6084 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6087 * tree log blocks never actually go into the extent allocation
6088 * tree, just update pinning info and exit early.
6090 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6091 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6092 /* unlocks the pinned mutex */
6093 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6095 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6096 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6098 parent, root_objectid, (int)owner,
6099 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6101 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6103 parent, root_objectid, owner,
6104 offset, BTRFS_DROP_DELAYED_REF,
6110 static u64 stripe_align(struct btrfs_root *root,
6111 struct btrfs_block_group_cache *cache,
6112 u64 val, u64 num_bytes)
6114 u64 ret = ALIGN(val, root->stripesize);
6119 * when we wait for progress in the block group caching, its because
6120 * our allocation attempt failed at least once. So, we must sleep
6121 * and let some progress happen before we try again.
6123 * This function will sleep at least once waiting for new free space to
6124 * show up, and then it will check the block group free space numbers
6125 * for our min num_bytes. Another option is to have it go ahead
6126 * and look in the rbtree for a free extent of a given size, but this
6129 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6130 * any of the information in this block group.
6132 static noinline void
6133 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6136 struct btrfs_caching_control *caching_ctl;
6138 caching_ctl = get_caching_control(cache);
6142 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6143 (cache->free_space_ctl->free_space >= num_bytes));
6145 put_caching_control(caching_ctl);
6149 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6151 struct btrfs_caching_control *caching_ctl;
6154 caching_ctl = get_caching_control(cache);
6156 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6158 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6159 if (cache->cached == BTRFS_CACHE_ERROR)
6161 put_caching_control(caching_ctl);
6165 int __get_raid_index(u64 flags)
6167 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6168 return BTRFS_RAID_RAID10;
6169 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6170 return BTRFS_RAID_RAID1;
6171 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6172 return BTRFS_RAID_DUP;
6173 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6174 return BTRFS_RAID_RAID0;
6175 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6176 return BTRFS_RAID_RAID5;
6177 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6178 return BTRFS_RAID_RAID6;
6180 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6183 int get_block_group_index(struct btrfs_block_group_cache *cache)
6185 return __get_raid_index(cache->flags);
6188 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6189 [BTRFS_RAID_RAID10] = "raid10",
6190 [BTRFS_RAID_RAID1] = "raid1",
6191 [BTRFS_RAID_DUP] = "dup",
6192 [BTRFS_RAID_RAID0] = "raid0",
6193 [BTRFS_RAID_SINGLE] = "single",
6194 [BTRFS_RAID_RAID5] = "raid5",
6195 [BTRFS_RAID_RAID6] = "raid6",
6198 const char *get_raid_name(enum btrfs_raid_types type)
6200 if (type >= BTRFS_NR_RAID_TYPES)
6203 return btrfs_raid_type_names[type];
6206 enum btrfs_loop_type {
6207 LOOP_CACHING_NOWAIT = 0,
6208 LOOP_CACHING_WAIT = 1,
6209 LOOP_ALLOC_CHUNK = 2,
6210 LOOP_NO_EMPTY_SIZE = 3,
6214 * walks the btree of allocated extents and find a hole of a given size.
6215 * The key ins is changed to record the hole:
6216 * ins->objectid == start position
6217 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6218 * ins->offset == the size of the hole.
6219 * Any available blocks before search_start are skipped.
6221 * If there is no suitable free space, we will record the max size of
6222 * the free space extent currently.
6224 static noinline int find_free_extent(struct btrfs_root *orig_root,
6225 u64 num_bytes, u64 empty_size,
6226 u64 hint_byte, struct btrfs_key *ins,
6230 struct btrfs_root *root = orig_root->fs_info->extent_root;
6231 struct btrfs_free_cluster *last_ptr = NULL;
6232 struct btrfs_block_group_cache *block_group = NULL;
6233 struct btrfs_block_group_cache *used_block_group;
6234 u64 search_start = 0;
6235 u64 max_extent_size = 0;
6236 int empty_cluster = 2 * 1024 * 1024;
6237 struct btrfs_space_info *space_info;
6239 int index = __get_raid_index(flags);
6240 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6241 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6242 bool failed_cluster_refill = false;
6243 bool failed_alloc = false;
6244 bool use_cluster = true;
6245 bool have_caching_bg = false;
6247 WARN_ON(num_bytes < root->sectorsize);
6248 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6252 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6254 space_info = __find_space_info(root->fs_info, flags);
6256 btrfs_err(root->fs_info, "No space info for %llu", flags);
6261 * If the space info is for both data and metadata it means we have a
6262 * small filesystem and we can't use the clustering stuff.
6264 if (btrfs_mixed_space_info(space_info))
6265 use_cluster = false;
6267 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6268 last_ptr = &root->fs_info->meta_alloc_cluster;
6269 if (!btrfs_test_opt(root, SSD))
6270 empty_cluster = 64 * 1024;
6273 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6274 btrfs_test_opt(root, SSD)) {
6275 last_ptr = &root->fs_info->data_alloc_cluster;
6279 spin_lock(&last_ptr->lock);
6280 if (last_ptr->block_group)
6281 hint_byte = last_ptr->window_start;
6282 spin_unlock(&last_ptr->lock);
6285 search_start = max(search_start, first_logical_byte(root, 0));
6286 search_start = max(search_start, hint_byte);
6291 if (search_start == hint_byte) {
6292 block_group = btrfs_lookup_block_group(root->fs_info,
6294 used_block_group = block_group;
6296 * we don't want to use the block group if it doesn't match our
6297 * allocation bits, or if its not cached.
6299 * However if we are re-searching with an ideal block group
6300 * picked out then we don't care that the block group is cached.
6302 if (block_group && block_group_bits(block_group, flags) &&
6303 block_group->cached != BTRFS_CACHE_NO) {
6304 down_read(&space_info->groups_sem);
6305 if (list_empty(&block_group->list) ||
6308 * someone is removing this block group,
6309 * we can't jump into the have_block_group
6310 * target because our list pointers are not
6313 btrfs_put_block_group(block_group);
6314 up_read(&space_info->groups_sem);
6316 index = get_block_group_index(block_group);
6317 goto have_block_group;
6319 } else if (block_group) {
6320 btrfs_put_block_group(block_group);
6324 have_caching_bg = false;
6325 down_read(&space_info->groups_sem);
6326 list_for_each_entry(block_group, &space_info->block_groups[index],
6331 used_block_group = block_group;
6332 btrfs_get_block_group(block_group);
6333 search_start = block_group->key.objectid;
6336 * this can happen if we end up cycling through all the
6337 * raid types, but we want to make sure we only allocate
6338 * for the proper type.
6340 if (!block_group_bits(block_group, flags)) {
6341 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6342 BTRFS_BLOCK_GROUP_RAID1 |
6343 BTRFS_BLOCK_GROUP_RAID5 |
6344 BTRFS_BLOCK_GROUP_RAID6 |
6345 BTRFS_BLOCK_GROUP_RAID10;
6348 * if they asked for extra copies and this block group
6349 * doesn't provide them, bail. This does allow us to
6350 * fill raid0 from raid1.
6352 if ((flags & extra) && !(block_group->flags & extra))
6357 cached = block_group_cache_done(block_group);
6358 if (unlikely(!cached)) {
6359 ret = cache_block_group(block_group, 0);
6364 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6366 if (unlikely(block_group->ro))
6370 * Ok we want to try and use the cluster allocator, so
6374 unsigned long aligned_cluster;
6376 * the refill lock keeps out other
6377 * people trying to start a new cluster
6379 spin_lock(&last_ptr->refill_lock);
6380 used_block_group = last_ptr->block_group;
6381 if (used_block_group != block_group &&
6382 (!used_block_group ||
6383 used_block_group->ro ||
6384 !block_group_bits(used_block_group, flags))) {
6385 used_block_group = block_group;
6386 goto refill_cluster;
6389 if (used_block_group != block_group)
6390 btrfs_get_block_group(used_block_group);
6392 offset = btrfs_alloc_from_cluster(used_block_group,
6395 used_block_group->key.objectid,
6398 /* we have a block, we're done */
6399 spin_unlock(&last_ptr->refill_lock);
6400 trace_btrfs_reserve_extent_cluster(root,
6401 block_group, search_start, num_bytes);
6405 WARN_ON(last_ptr->block_group != used_block_group);
6406 if (used_block_group != block_group) {
6407 btrfs_put_block_group(used_block_group);
6408 used_block_group = block_group;
6411 BUG_ON(used_block_group != block_group);
6412 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6413 * set up a new clusters, so lets just skip it
6414 * and let the allocator find whatever block
6415 * it can find. If we reach this point, we
6416 * will have tried the cluster allocator
6417 * plenty of times and not have found
6418 * anything, so we are likely way too
6419 * fragmented for the clustering stuff to find
6422 * However, if the cluster is taken from the
6423 * current block group, release the cluster
6424 * first, so that we stand a better chance of
6425 * succeeding in the unclustered
6427 if (loop >= LOOP_NO_EMPTY_SIZE &&
6428 last_ptr->block_group != block_group) {
6429 spin_unlock(&last_ptr->refill_lock);
6430 goto unclustered_alloc;
6434 * this cluster didn't work out, free it and
6437 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6439 if (loop >= LOOP_NO_EMPTY_SIZE) {
6440 spin_unlock(&last_ptr->refill_lock);
6441 goto unclustered_alloc;
6444 aligned_cluster = max_t(unsigned long,
6445 empty_cluster + empty_size,
6446 block_group->full_stripe_len);
6448 /* allocate a cluster in this block group */
6449 ret = btrfs_find_space_cluster(root, block_group,
6450 last_ptr, search_start,
6455 * now pull our allocation out of this
6458 offset = btrfs_alloc_from_cluster(block_group,
6464 /* we found one, proceed */
6465 spin_unlock(&last_ptr->refill_lock);
6466 trace_btrfs_reserve_extent_cluster(root,
6467 block_group, search_start,
6471 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6472 && !failed_cluster_refill) {
6473 spin_unlock(&last_ptr->refill_lock);
6475 failed_cluster_refill = true;
6476 wait_block_group_cache_progress(block_group,
6477 num_bytes + empty_cluster + empty_size);
6478 goto have_block_group;
6482 * at this point we either didn't find a cluster
6483 * or we weren't able to allocate a block from our
6484 * cluster. Free the cluster we've been trying
6485 * to use, and go to the next block group
6487 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6488 spin_unlock(&last_ptr->refill_lock);
6493 spin_lock(&block_group->free_space_ctl->tree_lock);
6495 block_group->free_space_ctl->free_space <
6496 num_bytes + empty_cluster + empty_size) {
6497 if (block_group->free_space_ctl->free_space >
6500 block_group->free_space_ctl->free_space;
6501 spin_unlock(&block_group->free_space_ctl->tree_lock);
6504 spin_unlock(&block_group->free_space_ctl->tree_lock);
6506 offset = btrfs_find_space_for_alloc(block_group, search_start,
6507 num_bytes, empty_size,
6510 * If we didn't find a chunk, and we haven't failed on this
6511 * block group before, and this block group is in the middle of
6512 * caching and we are ok with waiting, then go ahead and wait
6513 * for progress to be made, and set failed_alloc to true.
6515 * If failed_alloc is true then we've already waited on this
6516 * block group once and should move on to the next block group.
6518 if (!offset && !failed_alloc && !cached &&
6519 loop > LOOP_CACHING_NOWAIT) {
6520 wait_block_group_cache_progress(block_group,
6521 num_bytes + empty_size);
6522 failed_alloc = true;
6523 goto have_block_group;
6524 } else if (!offset) {
6526 have_caching_bg = true;
6530 search_start = stripe_align(root, used_block_group,
6533 /* move on to the next group */
6534 if (search_start + num_bytes >
6535 used_block_group->key.objectid + used_block_group->key.offset) {
6536 btrfs_add_free_space(used_block_group, offset, num_bytes);
6540 if (offset < search_start)
6541 btrfs_add_free_space(used_block_group, offset,
6542 search_start - offset);
6543 BUG_ON(offset > search_start);
6545 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6547 if (ret == -EAGAIN) {
6548 btrfs_add_free_space(used_block_group, offset, num_bytes);
6552 /* we are all good, lets return */
6553 ins->objectid = search_start;
6554 ins->offset = num_bytes;
6556 trace_btrfs_reserve_extent(orig_root, block_group,
6557 search_start, num_bytes);
6558 if (used_block_group != block_group)
6559 btrfs_put_block_group(used_block_group);
6560 btrfs_put_block_group(block_group);
6563 failed_cluster_refill = false;
6564 failed_alloc = false;
6565 BUG_ON(index != get_block_group_index(block_group));
6566 if (used_block_group != block_group)
6567 btrfs_put_block_group(used_block_group);
6568 btrfs_put_block_group(block_group);
6570 up_read(&space_info->groups_sem);
6572 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6575 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6579 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6580 * caching kthreads as we move along
6581 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6582 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6583 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6586 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6589 if (loop == LOOP_ALLOC_CHUNK) {
6590 struct btrfs_trans_handle *trans;
6592 trans = btrfs_join_transaction(root);
6593 if (IS_ERR(trans)) {
6594 ret = PTR_ERR(trans);
6598 ret = do_chunk_alloc(trans, root, flags,
6601 * Do not bail out on ENOSPC since we
6602 * can do more things.
6604 if (ret < 0 && ret != -ENOSPC)
6605 btrfs_abort_transaction(trans,
6609 btrfs_end_transaction(trans, root);
6614 if (loop == LOOP_NO_EMPTY_SIZE) {
6620 } else if (!ins->objectid) {
6622 } else if (ins->objectid) {
6627 ins->offset = max_extent_size;
6631 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6632 int dump_block_groups)
6634 struct btrfs_block_group_cache *cache;
6637 spin_lock(&info->lock);
6638 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6640 info->total_bytes - info->bytes_used - info->bytes_pinned -
6641 info->bytes_reserved - info->bytes_readonly,
6642 (info->full) ? "" : "not ");
6643 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6644 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6645 info->total_bytes, info->bytes_used, info->bytes_pinned,
6646 info->bytes_reserved, info->bytes_may_use,
6647 info->bytes_readonly);
6648 spin_unlock(&info->lock);
6650 if (!dump_block_groups)
6653 down_read(&info->groups_sem);
6655 list_for_each_entry(cache, &info->block_groups[index], list) {
6656 spin_lock(&cache->lock);
6657 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6658 cache->key.objectid, cache->key.offset,
6659 btrfs_block_group_used(&cache->item), cache->pinned,
6660 cache->reserved, cache->ro ? "[readonly]" : "");
6661 btrfs_dump_free_space(cache, bytes);
6662 spin_unlock(&cache->lock);
6664 if (++index < BTRFS_NR_RAID_TYPES)
6666 up_read(&info->groups_sem);
6669 int btrfs_reserve_extent(struct btrfs_root *root,
6670 u64 num_bytes, u64 min_alloc_size,
6671 u64 empty_size, u64 hint_byte,
6672 struct btrfs_key *ins, int is_data)
6674 bool final_tried = false;
6678 flags = btrfs_get_alloc_profile(root, is_data);
6680 WARN_ON(num_bytes < root->sectorsize);
6681 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6684 if (ret == -ENOSPC) {
6685 if (!final_tried && ins->offset) {
6686 num_bytes = min(num_bytes >> 1, ins->offset);
6687 num_bytes = round_down(num_bytes, root->sectorsize);
6688 num_bytes = max(num_bytes, min_alloc_size);
6689 if (num_bytes == min_alloc_size)
6692 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6693 struct btrfs_space_info *sinfo;
6695 sinfo = __find_space_info(root->fs_info, flags);
6696 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6699 dump_space_info(sinfo, num_bytes, 1);
6706 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6707 u64 start, u64 len, int pin)
6709 struct btrfs_block_group_cache *cache;
6712 cache = btrfs_lookup_block_group(root->fs_info, start);
6714 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6719 if (btrfs_test_opt(root, DISCARD))
6720 ret = btrfs_discard_extent(root, start, len, NULL);
6723 pin_down_extent(root, cache, start, len, 1);
6725 btrfs_add_free_space(cache, start, len);
6726 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6728 btrfs_put_block_group(cache);
6730 trace_btrfs_reserved_extent_free(root, start, len);
6735 int btrfs_free_reserved_extent(struct btrfs_root *root,
6738 return __btrfs_free_reserved_extent(root, start, len, 0);
6741 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6744 return __btrfs_free_reserved_extent(root, start, len, 1);
6747 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6748 struct btrfs_root *root,
6749 u64 parent, u64 root_objectid,
6750 u64 flags, u64 owner, u64 offset,
6751 struct btrfs_key *ins, int ref_mod)
6754 struct btrfs_fs_info *fs_info = root->fs_info;
6755 struct btrfs_extent_item *extent_item;
6756 struct btrfs_extent_inline_ref *iref;
6757 struct btrfs_path *path;
6758 struct extent_buffer *leaf;
6763 type = BTRFS_SHARED_DATA_REF_KEY;
6765 type = BTRFS_EXTENT_DATA_REF_KEY;
6767 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6769 path = btrfs_alloc_path();
6773 path->leave_spinning = 1;
6774 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6777 btrfs_free_path(path);
6781 leaf = path->nodes[0];
6782 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6783 struct btrfs_extent_item);
6784 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6785 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6786 btrfs_set_extent_flags(leaf, extent_item,
6787 flags | BTRFS_EXTENT_FLAG_DATA);
6789 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6790 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6792 struct btrfs_shared_data_ref *ref;
6793 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6794 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6795 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6797 struct btrfs_extent_data_ref *ref;
6798 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6799 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6800 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6801 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6802 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6805 btrfs_mark_buffer_dirty(path->nodes[0]);
6806 btrfs_free_path(path);
6808 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6809 if (ret) { /* -ENOENT, logic error */
6810 btrfs_err(fs_info, "update block group failed for %llu %llu",
6811 ins->objectid, ins->offset);
6814 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6818 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6819 struct btrfs_root *root,
6820 u64 parent, u64 root_objectid,
6821 u64 flags, struct btrfs_disk_key *key,
6822 int level, struct btrfs_key *ins)
6825 struct btrfs_fs_info *fs_info = root->fs_info;
6826 struct btrfs_extent_item *extent_item;
6827 struct btrfs_tree_block_info *block_info;
6828 struct btrfs_extent_inline_ref *iref;
6829 struct btrfs_path *path;
6830 struct extent_buffer *leaf;
6831 u32 size = sizeof(*extent_item) + sizeof(*iref);
6832 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6835 if (!skinny_metadata)
6836 size += sizeof(*block_info);
6838 path = btrfs_alloc_path();
6840 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6845 path->leave_spinning = 1;
6846 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6849 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6851 btrfs_free_path(path);
6855 leaf = path->nodes[0];
6856 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6857 struct btrfs_extent_item);
6858 btrfs_set_extent_refs(leaf, extent_item, 1);
6859 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6860 btrfs_set_extent_flags(leaf, extent_item,
6861 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6863 if (skinny_metadata) {
6864 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6866 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6867 btrfs_set_tree_block_key(leaf, block_info, key);
6868 btrfs_set_tree_block_level(leaf, block_info, level);
6869 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6873 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6874 btrfs_set_extent_inline_ref_type(leaf, iref,
6875 BTRFS_SHARED_BLOCK_REF_KEY);
6876 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6878 btrfs_set_extent_inline_ref_type(leaf, iref,
6879 BTRFS_TREE_BLOCK_REF_KEY);
6880 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6883 btrfs_mark_buffer_dirty(leaf);
6884 btrfs_free_path(path);
6886 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6887 if (ret) { /* -ENOENT, logic error */
6888 btrfs_err(fs_info, "update block group failed for %llu %llu",
6889 ins->objectid, ins->offset);
6893 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6897 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6898 struct btrfs_root *root,
6899 u64 root_objectid, u64 owner,
6900 u64 offset, struct btrfs_key *ins)
6904 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6906 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6908 root_objectid, owner, offset,
6909 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6914 * this is used by the tree logging recovery code. It records that
6915 * an extent has been allocated and makes sure to clear the free
6916 * space cache bits as well
6918 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6919 struct btrfs_root *root,
6920 u64 root_objectid, u64 owner, u64 offset,
6921 struct btrfs_key *ins)
6924 struct btrfs_block_group_cache *block_group;
6927 * Mixed block groups will exclude before processing the log so we only
6928 * need to do the exlude dance if this fs isn't mixed.
6930 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6931 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6936 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6940 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6941 RESERVE_ALLOC_NO_ACCOUNT);
6942 BUG_ON(ret); /* logic error */
6943 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6944 0, owner, offset, ins, 1);
6945 btrfs_put_block_group(block_group);
6949 static struct extent_buffer *
6950 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6951 u64 bytenr, u32 blocksize, int level)
6953 struct extent_buffer *buf;
6955 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6957 return ERR_PTR(-ENOMEM);
6958 btrfs_set_header_generation(buf, trans->transid);
6959 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6960 btrfs_tree_lock(buf);
6961 clean_tree_block(trans, root, buf);
6962 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6964 btrfs_set_lock_blocking(buf);
6965 btrfs_set_buffer_uptodate(buf);
6967 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6969 * we allow two log transactions at a time, use different
6970 * EXENT bit to differentiate dirty pages.
6972 if (root->log_transid % 2 == 0)
6973 set_extent_dirty(&root->dirty_log_pages, buf->start,
6974 buf->start + buf->len - 1, GFP_NOFS);
6976 set_extent_new(&root->dirty_log_pages, buf->start,
6977 buf->start + buf->len - 1, GFP_NOFS);
6979 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6980 buf->start + buf->len - 1, GFP_NOFS);
6982 trans->blocks_used++;
6983 /* this returns a buffer locked for blocking */
6987 static struct btrfs_block_rsv *
6988 use_block_rsv(struct btrfs_trans_handle *trans,
6989 struct btrfs_root *root, u32 blocksize)
6991 struct btrfs_block_rsv *block_rsv;
6992 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6994 bool global_updated = false;
6996 block_rsv = get_block_rsv(trans, root);
6998 if (unlikely(block_rsv->size == 0))
7001 ret = block_rsv_use_bytes(block_rsv, blocksize);
7005 if (block_rsv->failfast)
7006 return ERR_PTR(ret);
7008 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7009 global_updated = true;
7010 update_global_block_rsv(root->fs_info);
7014 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7015 static DEFINE_RATELIMIT_STATE(_rs,
7016 DEFAULT_RATELIMIT_INTERVAL * 10,
7017 /*DEFAULT_RATELIMIT_BURST*/ 1);
7018 if (__ratelimit(&_rs))
7020 "btrfs: block rsv returned %d\n", ret);
7023 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7024 BTRFS_RESERVE_NO_FLUSH);
7028 * If we couldn't reserve metadata bytes try and use some from
7029 * the global reserve if its space type is the same as the global
7032 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7033 block_rsv->space_info == global_rsv->space_info) {
7034 ret = block_rsv_use_bytes(global_rsv, blocksize);
7038 return ERR_PTR(ret);
7041 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7042 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7044 block_rsv_add_bytes(block_rsv, blocksize, 0);
7045 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7049 * finds a free extent and does all the dirty work required for allocation
7050 * returns the key for the extent through ins, and a tree buffer for
7051 * the first block of the extent through buf.
7053 * returns the tree buffer or NULL.
7055 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7056 struct btrfs_root *root, u32 blocksize,
7057 u64 parent, u64 root_objectid,
7058 struct btrfs_disk_key *key, int level,
7059 u64 hint, u64 empty_size)
7061 struct btrfs_key ins;
7062 struct btrfs_block_rsv *block_rsv;
7063 struct extent_buffer *buf;
7066 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7069 block_rsv = use_block_rsv(trans, root, blocksize);
7070 if (IS_ERR(block_rsv))
7071 return ERR_CAST(block_rsv);
7073 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7074 empty_size, hint, &ins, 0);
7076 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7077 return ERR_PTR(ret);
7080 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7082 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7084 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7086 parent = ins.objectid;
7087 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7091 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7092 struct btrfs_delayed_extent_op *extent_op;
7093 extent_op = btrfs_alloc_delayed_extent_op();
7094 BUG_ON(!extent_op); /* -ENOMEM */
7096 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7098 memset(&extent_op->key, 0, sizeof(extent_op->key));
7099 extent_op->flags_to_set = flags;
7100 if (skinny_metadata)
7101 extent_op->update_key = 0;
7103 extent_op->update_key = 1;
7104 extent_op->update_flags = 1;
7105 extent_op->is_data = 0;
7106 extent_op->level = level;
7108 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7110 ins.offset, parent, root_objectid,
7111 level, BTRFS_ADD_DELAYED_EXTENT,
7113 BUG_ON(ret); /* -ENOMEM */
7118 struct walk_control {
7119 u64 refs[BTRFS_MAX_LEVEL];
7120 u64 flags[BTRFS_MAX_LEVEL];
7121 struct btrfs_key update_progress;
7132 #define DROP_REFERENCE 1
7133 #define UPDATE_BACKREF 2
7135 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7136 struct btrfs_root *root,
7137 struct walk_control *wc,
7138 struct btrfs_path *path)
7146 struct btrfs_key key;
7147 struct extent_buffer *eb;
7152 if (path->slots[wc->level] < wc->reada_slot) {
7153 wc->reada_count = wc->reada_count * 2 / 3;
7154 wc->reada_count = max(wc->reada_count, 2);
7156 wc->reada_count = wc->reada_count * 3 / 2;
7157 wc->reada_count = min_t(int, wc->reada_count,
7158 BTRFS_NODEPTRS_PER_BLOCK(root));
7161 eb = path->nodes[wc->level];
7162 nritems = btrfs_header_nritems(eb);
7163 blocksize = btrfs_level_size(root, wc->level - 1);
7165 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7166 if (nread >= wc->reada_count)
7170 bytenr = btrfs_node_blockptr(eb, slot);
7171 generation = btrfs_node_ptr_generation(eb, slot);
7173 if (slot == path->slots[wc->level])
7176 if (wc->stage == UPDATE_BACKREF &&
7177 generation <= root->root_key.offset)
7180 /* We don't lock the tree block, it's OK to be racy here */
7181 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7182 wc->level - 1, 1, &refs,
7184 /* We don't care about errors in readahead. */
7189 if (wc->stage == DROP_REFERENCE) {
7193 if (wc->level == 1 &&
7194 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7196 if (!wc->update_ref ||
7197 generation <= root->root_key.offset)
7199 btrfs_node_key_to_cpu(eb, &key, slot);
7200 ret = btrfs_comp_cpu_keys(&key,
7201 &wc->update_progress);
7205 if (wc->level == 1 &&
7206 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7210 ret = readahead_tree_block(root, bytenr, blocksize,
7216 wc->reada_slot = slot;
7220 * helper to process tree block while walking down the tree.
7222 * when wc->stage == UPDATE_BACKREF, this function updates
7223 * back refs for pointers in the block.
7225 * NOTE: return value 1 means we should stop walking down.
7227 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7228 struct btrfs_root *root,
7229 struct btrfs_path *path,
7230 struct walk_control *wc, int lookup_info)
7232 int level = wc->level;
7233 struct extent_buffer *eb = path->nodes[level];
7234 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7237 if (wc->stage == UPDATE_BACKREF &&
7238 btrfs_header_owner(eb) != root->root_key.objectid)
7242 * when reference count of tree block is 1, it won't increase
7243 * again. once full backref flag is set, we never clear it.
7246 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7247 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7248 BUG_ON(!path->locks[level]);
7249 ret = btrfs_lookup_extent_info(trans, root,
7250 eb->start, level, 1,
7253 BUG_ON(ret == -ENOMEM);
7256 BUG_ON(wc->refs[level] == 0);
7259 if (wc->stage == DROP_REFERENCE) {
7260 if (wc->refs[level] > 1)
7263 if (path->locks[level] && !wc->keep_locks) {
7264 btrfs_tree_unlock_rw(eb, path->locks[level]);
7265 path->locks[level] = 0;
7270 /* wc->stage == UPDATE_BACKREF */
7271 if (!(wc->flags[level] & flag)) {
7272 BUG_ON(!path->locks[level]);
7273 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7274 BUG_ON(ret); /* -ENOMEM */
7275 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7276 BUG_ON(ret); /* -ENOMEM */
7277 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7279 btrfs_header_level(eb), 0);
7280 BUG_ON(ret); /* -ENOMEM */
7281 wc->flags[level] |= flag;
7285 * the block is shared by multiple trees, so it's not good to
7286 * keep the tree lock
7288 if (path->locks[level] && level > 0) {
7289 btrfs_tree_unlock_rw(eb, path->locks[level]);
7290 path->locks[level] = 0;
7296 * helper to process tree block pointer.
7298 * when wc->stage == DROP_REFERENCE, this function checks
7299 * reference count of the block pointed to. if the block
7300 * is shared and we need update back refs for the subtree
7301 * rooted at the block, this function changes wc->stage to
7302 * UPDATE_BACKREF. if the block is shared and there is no
7303 * need to update back, this function drops the reference
7306 * NOTE: return value 1 means we should stop walking down.
7308 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7309 struct btrfs_root *root,
7310 struct btrfs_path *path,
7311 struct walk_control *wc, int *lookup_info)
7317 struct btrfs_key key;
7318 struct extent_buffer *next;
7319 int level = wc->level;
7323 generation = btrfs_node_ptr_generation(path->nodes[level],
7324 path->slots[level]);
7326 * if the lower level block was created before the snapshot
7327 * was created, we know there is no need to update back refs
7330 if (wc->stage == UPDATE_BACKREF &&
7331 generation <= root->root_key.offset) {
7336 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7337 blocksize = btrfs_level_size(root, level - 1);
7339 next = btrfs_find_tree_block(root, bytenr, blocksize);
7341 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7344 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7348 btrfs_tree_lock(next);
7349 btrfs_set_lock_blocking(next);
7351 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7352 &wc->refs[level - 1],
7353 &wc->flags[level - 1]);
7355 btrfs_tree_unlock(next);
7359 if (unlikely(wc->refs[level - 1] == 0)) {
7360 btrfs_err(root->fs_info, "Missing references.");
7365 if (wc->stage == DROP_REFERENCE) {
7366 if (wc->refs[level - 1] > 1) {
7368 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7371 if (!wc->update_ref ||
7372 generation <= root->root_key.offset)
7375 btrfs_node_key_to_cpu(path->nodes[level], &key,
7376 path->slots[level]);
7377 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7381 wc->stage = UPDATE_BACKREF;
7382 wc->shared_level = level - 1;
7386 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7390 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7391 btrfs_tree_unlock(next);
7392 free_extent_buffer(next);
7398 if (reada && level == 1)
7399 reada_walk_down(trans, root, wc, path);
7400 next = read_tree_block(root, bytenr, blocksize, generation);
7401 if (!next || !extent_buffer_uptodate(next)) {
7402 free_extent_buffer(next);
7405 btrfs_tree_lock(next);
7406 btrfs_set_lock_blocking(next);
7410 BUG_ON(level != btrfs_header_level(next));
7411 path->nodes[level] = next;
7412 path->slots[level] = 0;
7413 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7419 wc->refs[level - 1] = 0;
7420 wc->flags[level - 1] = 0;
7421 if (wc->stage == DROP_REFERENCE) {
7422 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7423 parent = path->nodes[level]->start;
7425 BUG_ON(root->root_key.objectid !=
7426 btrfs_header_owner(path->nodes[level]));
7430 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7431 root->root_key.objectid, level - 1, 0, 0);
7432 BUG_ON(ret); /* -ENOMEM */
7434 btrfs_tree_unlock(next);
7435 free_extent_buffer(next);
7441 * helper to process tree block while walking up the tree.
7443 * when wc->stage == DROP_REFERENCE, this function drops
7444 * reference count on the block.
7446 * when wc->stage == UPDATE_BACKREF, this function changes
7447 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7448 * to UPDATE_BACKREF previously while processing the block.
7450 * NOTE: return value 1 means we should stop walking up.
7452 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7453 struct btrfs_root *root,
7454 struct btrfs_path *path,
7455 struct walk_control *wc)
7458 int level = wc->level;
7459 struct extent_buffer *eb = path->nodes[level];
7462 if (wc->stage == UPDATE_BACKREF) {
7463 BUG_ON(wc->shared_level < level);
7464 if (level < wc->shared_level)
7467 ret = find_next_key(path, level + 1, &wc->update_progress);
7471 wc->stage = DROP_REFERENCE;
7472 wc->shared_level = -1;
7473 path->slots[level] = 0;
7476 * check reference count again if the block isn't locked.
7477 * we should start walking down the tree again if reference
7480 if (!path->locks[level]) {
7482 btrfs_tree_lock(eb);
7483 btrfs_set_lock_blocking(eb);
7484 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7486 ret = btrfs_lookup_extent_info(trans, root,
7487 eb->start, level, 1,
7491 btrfs_tree_unlock_rw(eb, path->locks[level]);
7492 path->locks[level] = 0;
7495 BUG_ON(wc->refs[level] == 0);
7496 if (wc->refs[level] == 1) {
7497 btrfs_tree_unlock_rw(eb, path->locks[level]);
7498 path->locks[level] = 0;
7504 /* wc->stage == DROP_REFERENCE */
7505 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7507 if (wc->refs[level] == 1) {
7509 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7510 ret = btrfs_dec_ref(trans, root, eb, 1,
7513 ret = btrfs_dec_ref(trans, root, eb, 0,
7515 BUG_ON(ret); /* -ENOMEM */
7517 /* make block locked assertion in clean_tree_block happy */
7518 if (!path->locks[level] &&
7519 btrfs_header_generation(eb) == trans->transid) {
7520 btrfs_tree_lock(eb);
7521 btrfs_set_lock_blocking(eb);
7522 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7524 clean_tree_block(trans, root, eb);
7527 if (eb == root->node) {
7528 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7531 BUG_ON(root->root_key.objectid !=
7532 btrfs_header_owner(eb));
7534 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7535 parent = path->nodes[level + 1]->start;
7537 BUG_ON(root->root_key.objectid !=
7538 btrfs_header_owner(path->nodes[level + 1]));
7541 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7543 wc->refs[level] = 0;
7544 wc->flags[level] = 0;
7548 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7549 struct btrfs_root *root,
7550 struct btrfs_path *path,
7551 struct walk_control *wc)
7553 int level = wc->level;
7554 int lookup_info = 1;
7557 while (level >= 0) {
7558 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7565 if (path->slots[level] >=
7566 btrfs_header_nritems(path->nodes[level]))
7569 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7571 path->slots[level]++;
7580 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7581 struct btrfs_root *root,
7582 struct btrfs_path *path,
7583 struct walk_control *wc, int max_level)
7585 int level = wc->level;
7588 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7589 while (level < max_level && path->nodes[level]) {
7591 if (path->slots[level] + 1 <
7592 btrfs_header_nritems(path->nodes[level])) {
7593 path->slots[level]++;
7596 ret = walk_up_proc(trans, root, path, wc);
7600 if (path->locks[level]) {
7601 btrfs_tree_unlock_rw(path->nodes[level],
7602 path->locks[level]);
7603 path->locks[level] = 0;
7605 free_extent_buffer(path->nodes[level]);
7606 path->nodes[level] = NULL;
7614 * drop a subvolume tree.
7616 * this function traverses the tree freeing any blocks that only
7617 * referenced by the tree.
7619 * when a shared tree block is found. this function decreases its
7620 * reference count by one. if update_ref is true, this function
7621 * also make sure backrefs for the shared block and all lower level
7622 * blocks are properly updated.
7624 * If called with for_reloc == 0, may exit early with -EAGAIN
7626 int btrfs_drop_snapshot(struct btrfs_root *root,
7627 struct btrfs_block_rsv *block_rsv, int update_ref,
7630 struct btrfs_path *path;
7631 struct btrfs_trans_handle *trans;
7632 struct btrfs_root *tree_root = root->fs_info->tree_root;
7633 struct btrfs_root_item *root_item = &root->root_item;
7634 struct walk_control *wc;
7635 struct btrfs_key key;
7639 bool root_dropped = false;
7641 path = btrfs_alloc_path();
7647 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7649 btrfs_free_path(path);
7654 trans = btrfs_start_transaction(tree_root, 0);
7655 if (IS_ERR(trans)) {
7656 err = PTR_ERR(trans);
7661 trans->block_rsv = block_rsv;
7663 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7664 level = btrfs_header_level(root->node);
7665 path->nodes[level] = btrfs_lock_root_node(root);
7666 btrfs_set_lock_blocking(path->nodes[level]);
7667 path->slots[level] = 0;
7668 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7669 memset(&wc->update_progress, 0,
7670 sizeof(wc->update_progress));
7672 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7673 memcpy(&wc->update_progress, &key,
7674 sizeof(wc->update_progress));
7676 level = root_item->drop_level;
7678 path->lowest_level = level;
7679 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7680 path->lowest_level = 0;
7688 * unlock our path, this is safe because only this
7689 * function is allowed to delete this snapshot
7691 btrfs_unlock_up_safe(path, 0);
7693 level = btrfs_header_level(root->node);
7695 btrfs_tree_lock(path->nodes[level]);
7696 btrfs_set_lock_blocking(path->nodes[level]);
7697 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7699 ret = btrfs_lookup_extent_info(trans, root,
7700 path->nodes[level]->start,
7701 level, 1, &wc->refs[level],
7707 BUG_ON(wc->refs[level] == 0);
7709 if (level == root_item->drop_level)
7712 btrfs_tree_unlock(path->nodes[level]);
7713 path->locks[level] = 0;
7714 WARN_ON(wc->refs[level] != 1);
7720 wc->shared_level = -1;
7721 wc->stage = DROP_REFERENCE;
7722 wc->update_ref = update_ref;
7724 wc->for_reloc = for_reloc;
7725 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7729 ret = walk_down_tree(trans, root, path, wc);
7735 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7742 BUG_ON(wc->stage != DROP_REFERENCE);
7746 if (wc->stage == DROP_REFERENCE) {
7748 btrfs_node_key(path->nodes[level],
7749 &root_item->drop_progress,
7750 path->slots[level]);
7751 root_item->drop_level = level;
7754 BUG_ON(wc->level == 0);
7755 if (btrfs_should_end_transaction(trans, tree_root) ||
7756 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7757 ret = btrfs_update_root(trans, tree_root,
7761 btrfs_abort_transaction(trans, tree_root, ret);
7766 btrfs_end_transaction_throttle(trans, tree_root);
7767 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7768 pr_debug("btrfs: drop snapshot early exit\n");
7773 trans = btrfs_start_transaction(tree_root, 0);
7774 if (IS_ERR(trans)) {
7775 err = PTR_ERR(trans);
7779 trans->block_rsv = block_rsv;
7782 btrfs_release_path(path);
7786 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7788 btrfs_abort_transaction(trans, tree_root, ret);
7792 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7793 ret = btrfs_find_root(tree_root, &root->root_key, path,
7796 btrfs_abort_transaction(trans, tree_root, ret);
7799 } else if (ret > 0) {
7800 /* if we fail to delete the orphan item this time
7801 * around, it'll get picked up the next time.
7803 * The most common failure here is just -ENOENT.
7805 btrfs_del_orphan_item(trans, tree_root,
7806 root->root_key.objectid);
7810 if (root->in_radix) {
7811 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7813 free_extent_buffer(root->node);
7814 free_extent_buffer(root->commit_root);
7815 btrfs_put_fs_root(root);
7817 root_dropped = true;
7819 btrfs_end_transaction_throttle(trans, tree_root);
7822 btrfs_free_path(path);
7825 * So if we need to stop dropping the snapshot for whatever reason we
7826 * need to make sure to add it back to the dead root list so that we
7827 * keep trying to do the work later. This also cleans up roots if we
7828 * don't have it in the radix (like when we recover after a power fail
7829 * or unmount) so we don't leak memory.
7831 if (!for_reloc && root_dropped == false)
7832 btrfs_add_dead_root(root);
7834 btrfs_std_error(root->fs_info, err);
7839 * drop subtree rooted at tree block 'node'.
7841 * NOTE: this function will unlock and release tree block 'node'
7842 * only used by relocation code
7844 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7845 struct btrfs_root *root,
7846 struct extent_buffer *node,
7847 struct extent_buffer *parent)
7849 struct btrfs_path *path;
7850 struct walk_control *wc;
7856 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7858 path = btrfs_alloc_path();
7862 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7864 btrfs_free_path(path);
7868 btrfs_assert_tree_locked(parent);
7869 parent_level = btrfs_header_level(parent);
7870 extent_buffer_get(parent);
7871 path->nodes[parent_level] = parent;
7872 path->slots[parent_level] = btrfs_header_nritems(parent);
7874 btrfs_assert_tree_locked(node);
7875 level = btrfs_header_level(node);
7876 path->nodes[level] = node;
7877 path->slots[level] = 0;
7878 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7880 wc->refs[parent_level] = 1;
7881 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7883 wc->shared_level = -1;
7884 wc->stage = DROP_REFERENCE;
7888 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7891 wret = walk_down_tree(trans, root, path, wc);
7897 wret = walk_up_tree(trans, root, path, wc, parent_level);
7905 btrfs_free_path(path);
7909 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7915 * if restripe for this chunk_type is on pick target profile and
7916 * return, otherwise do the usual balance
7918 stripped = get_restripe_target(root->fs_info, flags);
7920 return extended_to_chunk(stripped);
7923 * we add in the count of missing devices because we want
7924 * to make sure that any RAID levels on a degraded FS
7925 * continue to be honored.
7927 num_devices = root->fs_info->fs_devices->rw_devices +
7928 root->fs_info->fs_devices->missing_devices;
7930 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7931 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7932 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7934 if (num_devices == 1) {
7935 stripped |= BTRFS_BLOCK_GROUP_DUP;
7936 stripped = flags & ~stripped;
7938 /* turn raid0 into single device chunks */
7939 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7942 /* turn mirroring into duplication */
7943 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7944 BTRFS_BLOCK_GROUP_RAID10))
7945 return stripped | BTRFS_BLOCK_GROUP_DUP;
7947 /* they already had raid on here, just return */
7948 if (flags & stripped)
7951 stripped |= BTRFS_BLOCK_GROUP_DUP;
7952 stripped = flags & ~stripped;
7954 /* switch duplicated blocks with raid1 */
7955 if (flags & BTRFS_BLOCK_GROUP_DUP)
7956 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7958 /* this is drive concat, leave it alone */
7964 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7966 struct btrfs_space_info *sinfo = cache->space_info;
7968 u64 min_allocable_bytes;
7973 * We need some metadata space and system metadata space for
7974 * allocating chunks in some corner cases until we force to set
7975 * it to be readonly.
7978 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7980 min_allocable_bytes = 1 * 1024 * 1024;
7982 min_allocable_bytes = 0;
7984 spin_lock(&sinfo->lock);
7985 spin_lock(&cache->lock);
7992 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7993 cache->bytes_super - btrfs_block_group_used(&cache->item);
7995 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7996 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7997 min_allocable_bytes <= sinfo->total_bytes) {
7998 sinfo->bytes_readonly += num_bytes;
8003 spin_unlock(&cache->lock);
8004 spin_unlock(&sinfo->lock);
8008 int btrfs_set_block_group_ro(struct btrfs_root *root,
8009 struct btrfs_block_group_cache *cache)
8012 struct btrfs_trans_handle *trans;
8018 trans = btrfs_join_transaction(root);
8020 return PTR_ERR(trans);
8022 alloc_flags = update_block_group_flags(root, cache->flags);
8023 if (alloc_flags != cache->flags) {
8024 ret = do_chunk_alloc(trans, root, alloc_flags,
8030 ret = set_block_group_ro(cache, 0);
8033 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8034 ret = do_chunk_alloc(trans, root, alloc_flags,
8038 ret = set_block_group_ro(cache, 0);
8040 btrfs_end_transaction(trans, root);
8044 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8045 struct btrfs_root *root, u64 type)
8047 u64 alloc_flags = get_alloc_profile(root, type);
8048 return do_chunk_alloc(trans, root, alloc_flags,
8053 * helper to account the unused space of all the readonly block group in the
8054 * list. takes mirrors into account.
8056 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8058 struct btrfs_block_group_cache *block_group;
8062 list_for_each_entry(block_group, groups_list, list) {
8063 spin_lock(&block_group->lock);
8065 if (!block_group->ro) {
8066 spin_unlock(&block_group->lock);
8070 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8071 BTRFS_BLOCK_GROUP_RAID10 |
8072 BTRFS_BLOCK_GROUP_DUP))
8077 free_bytes += (block_group->key.offset -
8078 btrfs_block_group_used(&block_group->item)) *
8081 spin_unlock(&block_group->lock);
8088 * helper to account the unused space of all the readonly block group in the
8089 * space_info. takes mirrors into account.
8091 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8096 spin_lock(&sinfo->lock);
8098 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8099 if (!list_empty(&sinfo->block_groups[i]))
8100 free_bytes += __btrfs_get_ro_block_group_free_space(
8101 &sinfo->block_groups[i]);
8103 spin_unlock(&sinfo->lock);
8108 void btrfs_set_block_group_rw(struct btrfs_root *root,
8109 struct btrfs_block_group_cache *cache)
8111 struct btrfs_space_info *sinfo = cache->space_info;
8116 spin_lock(&sinfo->lock);
8117 spin_lock(&cache->lock);
8118 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8119 cache->bytes_super - btrfs_block_group_used(&cache->item);
8120 sinfo->bytes_readonly -= num_bytes;
8122 spin_unlock(&cache->lock);
8123 spin_unlock(&sinfo->lock);
8127 * checks to see if its even possible to relocate this block group.
8129 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8130 * ok to go ahead and try.
8132 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8134 struct btrfs_block_group_cache *block_group;
8135 struct btrfs_space_info *space_info;
8136 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8137 struct btrfs_device *device;
8138 struct btrfs_trans_handle *trans;
8147 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8149 /* odd, couldn't find the block group, leave it alone */
8153 min_free = btrfs_block_group_used(&block_group->item);
8155 /* no bytes used, we're good */
8159 space_info = block_group->space_info;
8160 spin_lock(&space_info->lock);
8162 full = space_info->full;
8165 * if this is the last block group we have in this space, we can't
8166 * relocate it unless we're able to allocate a new chunk below.
8168 * Otherwise, we need to make sure we have room in the space to handle
8169 * all of the extents from this block group. If we can, we're good
8171 if ((space_info->total_bytes != block_group->key.offset) &&
8172 (space_info->bytes_used + space_info->bytes_reserved +
8173 space_info->bytes_pinned + space_info->bytes_readonly +
8174 min_free < space_info->total_bytes)) {
8175 spin_unlock(&space_info->lock);
8178 spin_unlock(&space_info->lock);
8181 * ok we don't have enough space, but maybe we have free space on our
8182 * devices to allocate new chunks for relocation, so loop through our
8183 * alloc devices and guess if we have enough space. if this block
8184 * group is going to be restriped, run checks against the target
8185 * profile instead of the current one.
8197 target = get_restripe_target(root->fs_info, block_group->flags);
8199 index = __get_raid_index(extended_to_chunk(target));
8202 * this is just a balance, so if we were marked as full
8203 * we know there is no space for a new chunk
8208 index = get_block_group_index(block_group);
8211 if (index == BTRFS_RAID_RAID10) {
8215 } else if (index == BTRFS_RAID_RAID1) {
8217 } else if (index == BTRFS_RAID_DUP) {
8220 } else if (index == BTRFS_RAID_RAID0) {
8221 dev_min = fs_devices->rw_devices;
8222 do_div(min_free, dev_min);
8225 /* We need to do this so that we can look at pending chunks */
8226 trans = btrfs_join_transaction(root);
8227 if (IS_ERR(trans)) {
8228 ret = PTR_ERR(trans);
8232 mutex_lock(&root->fs_info->chunk_mutex);
8233 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8237 * check to make sure we can actually find a chunk with enough
8238 * space to fit our block group in.
8240 if (device->total_bytes > device->bytes_used + min_free &&
8241 !device->is_tgtdev_for_dev_replace) {
8242 ret = find_free_dev_extent(trans, device, min_free,
8247 if (dev_nr >= dev_min)
8253 mutex_unlock(&root->fs_info->chunk_mutex);
8254 btrfs_end_transaction(trans, root);
8256 btrfs_put_block_group(block_group);
8260 static int find_first_block_group(struct btrfs_root *root,
8261 struct btrfs_path *path, struct btrfs_key *key)
8264 struct btrfs_key found_key;
8265 struct extent_buffer *leaf;
8268 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8273 slot = path->slots[0];
8274 leaf = path->nodes[0];
8275 if (slot >= btrfs_header_nritems(leaf)) {
8276 ret = btrfs_next_leaf(root, path);
8283 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8285 if (found_key.objectid >= key->objectid &&
8286 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8296 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8298 struct btrfs_block_group_cache *block_group;
8302 struct inode *inode;
8304 block_group = btrfs_lookup_first_block_group(info, last);
8305 while (block_group) {
8306 spin_lock(&block_group->lock);
8307 if (block_group->iref)
8309 spin_unlock(&block_group->lock);
8310 block_group = next_block_group(info->tree_root,
8320 inode = block_group->inode;
8321 block_group->iref = 0;
8322 block_group->inode = NULL;
8323 spin_unlock(&block_group->lock);
8325 last = block_group->key.objectid + block_group->key.offset;
8326 btrfs_put_block_group(block_group);
8330 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8332 struct btrfs_block_group_cache *block_group;
8333 struct btrfs_space_info *space_info;
8334 struct btrfs_caching_control *caching_ctl;
8337 down_write(&info->extent_commit_sem);
8338 while (!list_empty(&info->caching_block_groups)) {
8339 caching_ctl = list_entry(info->caching_block_groups.next,
8340 struct btrfs_caching_control, list);
8341 list_del(&caching_ctl->list);
8342 put_caching_control(caching_ctl);
8344 up_write(&info->extent_commit_sem);
8346 spin_lock(&info->block_group_cache_lock);
8347 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8348 block_group = rb_entry(n, struct btrfs_block_group_cache,
8350 rb_erase(&block_group->cache_node,
8351 &info->block_group_cache_tree);
8352 spin_unlock(&info->block_group_cache_lock);
8354 down_write(&block_group->space_info->groups_sem);
8355 list_del(&block_group->list);
8356 up_write(&block_group->space_info->groups_sem);
8358 if (block_group->cached == BTRFS_CACHE_STARTED)
8359 wait_block_group_cache_done(block_group);
8362 * We haven't cached this block group, which means we could
8363 * possibly have excluded extents on this block group.
8365 if (block_group->cached == BTRFS_CACHE_NO ||
8366 block_group->cached == BTRFS_CACHE_ERROR)
8367 free_excluded_extents(info->extent_root, block_group);
8369 btrfs_remove_free_space_cache(block_group);
8370 btrfs_put_block_group(block_group);
8372 spin_lock(&info->block_group_cache_lock);
8374 spin_unlock(&info->block_group_cache_lock);
8376 /* now that all the block groups are freed, go through and
8377 * free all the space_info structs. This is only called during
8378 * the final stages of unmount, and so we know nobody is
8379 * using them. We call synchronize_rcu() once before we start,
8380 * just to be on the safe side.
8384 release_global_block_rsv(info);
8386 while (!list_empty(&info->space_info)) {
8389 space_info = list_entry(info->space_info.next,
8390 struct btrfs_space_info,
8392 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8393 if (WARN_ON(space_info->bytes_pinned > 0 ||
8394 space_info->bytes_reserved > 0 ||
8395 space_info->bytes_may_use > 0)) {
8396 dump_space_info(space_info, 0, 0);
8399 list_del(&space_info->list);
8400 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8401 struct kobject *kobj;
8402 kobj = &space_info->block_group_kobjs[i];
8408 kobject_del(&space_info->kobj);
8409 kobject_put(&space_info->kobj);
8414 static void __link_block_group(struct btrfs_space_info *space_info,
8415 struct btrfs_block_group_cache *cache)
8417 int index = get_block_group_index(cache);
8419 down_write(&space_info->groups_sem);
8420 if (list_empty(&space_info->block_groups[index])) {
8421 struct kobject *kobj = &space_info->block_group_kobjs[index];
8424 kobject_get(&space_info->kobj); /* put in release */
8425 ret = kobject_init_and_add(kobj, &btrfs_raid_ktype,
8427 get_raid_name(index));
8429 pr_warn("btrfs: failed to add kobject for block cache. ignoring.\n");
8430 kobject_put(&space_info->kobj);
8433 list_add_tail(&cache->list, &space_info->block_groups[index]);
8434 up_write(&space_info->groups_sem);
8437 int btrfs_read_block_groups(struct btrfs_root *root)
8439 struct btrfs_path *path;
8441 struct btrfs_block_group_cache *cache;
8442 struct btrfs_fs_info *info = root->fs_info;
8443 struct btrfs_space_info *space_info;
8444 struct btrfs_key key;
8445 struct btrfs_key found_key;
8446 struct extent_buffer *leaf;
8450 root = info->extent_root;
8453 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8454 path = btrfs_alloc_path();
8459 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8460 if (btrfs_test_opt(root, SPACE_CACHE) &&
8461 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8463 if (btrfs_test_opt(root, CLEAR_CACHE))
8467 ret = find_first_block_group(root, path, &key);
8472 leaf = path->nodes[0];
8473 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8474 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8479 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8481 if (!cache->free_space_ctl) {
8487 atomic_set(&cache->count, 1);
8488 spin_lock_init(&cache->lock);
8489 cache->fs_info = info;
8490 INIT_LIST_HEAD(&cache->list);
8491 INIT_LIST_HEAD(&cache->cluster_list);
8495 * When we mount with old space cache, we need to
8496 * set BTRFS_DC_CLEAR and set dirty flag.
8498 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8499 * truncate the old free space cache inode and
8501 * b) Setting 'dirty flag' makes sure that we flush
8502 * the new space cache info onto disk.
8504 cache->disk_cache_state = BTRFS_DC_CLEAR;
8505 if (btrfs_test_opt(root, SPACE_CACHE))
8509 read_extent_buffer(leaf, &cache->item,
8510 btrfs_item_ptr_offset(leaf, path->slots[0]),
8511 sizeof(cache->item));
8512 memcpy(&cache->key, &found_key, sizeof(found_key));
8514 key.objectid = found_key.objectid + found_key.offset;
8515 btrfs_release_path(path);
8516 cache->flags = btrfs_block_group_flags(&cache->item);
8517 cache->sectorsize = root->sectorsize;
8518 cache->full_stripe_len = btrfs_full_stripe_len(root,
8519 &root->fs_info->mapping_tree,
8520 found_key.objectid);
8521 btrfs_init_free_space_ctl(cache);
8524 * We need to exclude the super stripes now so that the space
8525 * info has super bytes accounted for, otherwise we'll think
8526 * we have more space than we actually do.
8528 ret = exclude_super_stripes(root, cache);
8531 * We may have excluded something, so call this just in
8534 free_excluded_extents(root, cache);
8535 kfree(cache->free_space_ctl);
8541 * check for two cases, either we are full, and therefore
8542 * don't need to bother with the caching work since we won't
8543 * find any space, or we are empty, and we can just add all
8544 * the space in and be done with it. This saves us _alot_ of
8545 * time, particularly in the full case.
8547 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8548 cache->last_byte_to_unpin = (u64)-1;
8549 cache->cached = BTRFS_CACHE_FINISHED;
8550 free_excluded_extents(root, cache);
8551 } else if (btrfs_block_group_used(&cache->item) == 0) {
8552 cache->last_byte_to_unpin = (u64)-1;
8553 cache->cached = BTRFS_CACHE_FINISHED;
8554 add_new_free_space(cache, root->fs_info,
8556 found_key.objectid +
8558 free_excluded_extents(root, cache);
8561 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8563 btrfs_remove_free_space_cache(cache);
8564 btrfs_put_block_group(cache);
8568 ret = update_space_info(info, cache->flags, found_key.offset,
8569 btrfs_block_group_used(&cache->item),
8572 btrfs_remove_free_space_cache(cache);
8573 spin_lock(&info->block_group_cache_lock);
8574 rb_erase(&cache->cache_node,
8575 &info->block_group_cache_tree);
8576 spin_unlock(&info->block_group_cache_lock);
8577 btrfs_put_block_group(cache);
8581 cache->space_info = space_info;
8582 spin_lock(&cache->space_info->lock);
8583 cache->space_info->bytes_readonly += cache->bytes_super;
8584 spin_unlock(&cache->space_info->lock);
8586 __link_block_group(space_info, cache);
8588 set_avail_alloc_bits(root->fs_info, cache->flags);
8589 if (btrfs_chunk_readonly(root, cache->key.objectid))
8590 set_block_group_ro(cache, 1);
8593 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8594 if (!(get_alloc_profile(root, space_info->flags) &
8595 (BTRFS_BLOCK_GROUP_RAID10 |
8596 BTRFS_BLOCK_GROUP_RAID1 |
8597 BTRFS_BLOCK_GROUP_RAID5 |
8598 BTRFS_BLOCK_GROUP_RAID6 |
8599 BTRFS_BLOCK_GROUP_DUP)))
8602 * avoid allocating from un-mirrored block group if there are
8603 * mirrored block groups.
8605 list_for_each_entry(cache,
8606 &space_info->block_groups[BTRFS_RAID_RAID0],
8608 set_block_group_ro(cache, 1);
8609 list_for_each_entry(cache,
8610 &space_info->block_groups[BTRFS_RAID_SINGLE],
8612 set_block_group_ro(cache, 1);
8615 init_global_block_rsv(info);
8618 btrfs_free_path(path);
8622 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8623 struct btrfs_root *root)
8625 struct btrfs_block_group_cache *block_group, *tmp;
8626 struct btrfs_root *extent_root = root->fs_info->extent_root;
8627 struct btrfs_block_group_item item;
8628 struct btrfs_key key;
8631 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8633 list_del_init(&block_group->new_bg_list);
8638 spin_lock(&block_group->lock);
8639 memcpy(&item, &block_group->item, sizeof(item));
8640 memcpy(&key, &block_group->key, sizeof(key));
8641 spin_unlock(&block_group->lock);
8643 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8646 btrfs_abort_transaction(trans, extent_root, ret);
8647 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8648 key.objectid, key.offset);
8650 btrfs_abort_transaction(trans, extent_root, ret);
8654 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8655 struct btrfs_root *root, u64 bytes_used,
8656 u64 type, u64 chunk_objectid, u64 chunk_offset,
8660 struct btrfs_root *extent_root;
8661 struct btrfs_block_group_cache *cache;
8663 extent_root = root->fs_info->extent_root;
8665 root->fs_info->last_trans_log_full_commit = trans->transid;
8667 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8670 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8672 if (!cache->free_space_ctl) {
8677 cache->key.objectid = chunk_offset;
8678 cache->key.offset = size;
8679 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8680 cache->sectorsize = root->sectorsize;
8681 cache->fs_info = root->fs_info;
8682 cache->full_stripe_len = btrfs_full_stripe_len(root,
8683 &root->fs_info->mapping_tree,
8686 atomic_set(&cache->count, 1);
8687 spin_lock_init(&cache->lock);
8688 INIT_LIST_HEAD(&cache->list);
8689 INIT_LIST_HEAD(&cache->cluster_list);
8690 INIT_LIST_HEAD(&cache->new_bg_list);
8692 btrfs_init_free_space_ctl(cache);
8694 btrfs_set_block_group_used(&cache->item, bytes_used);
8695 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8696 cache->flags = type;
8697 btrfs_set_block_group_flags(&cache->item, type);
8699 cache->last_byte_to_unpin = (u64)-1;
8700 cache->cached = BTRFS_CACHE_FINISHED;
8701 ret = exclude_super_stripes(root, cache);
8704 * We may have excluded something, so call this just in
8707 free_excluded_extents(root, cache);
8708 kfree(cache->free_space_ctl);
8713 add_new_free_space(cache, root->fs_info, chunk_offset,
8714 chunk_offset + size);
8716 free_excluded_extents(root, cache);
8718 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8720 btrfs_remove_free_space_cache(cache);
8721 btrfs_put_block_group(cache);
8725 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8726 &cache->space_info);
8728 btrfs_remove_free_space_cache(cache);
8729 spin_lock(&root->fs_info->block_group_cache_lock);
8730 rb_erase(&cache->cache_node,
8731 &root->fs_info->block_group_cache_tree);
8732 spin_unlock(&root->fs_info->block_group_cache_lock);
8733 btrfs_put_block_group(cache);
8736 update_global_block_rsv(root->fs_info);
8738 spin_lock(&cache->space_info->lock);
8739 cache->space_info->bytes_readonly += cache->bytes_super;
8740 spin_unlock(&cache->space_info->lock);
8742 __link_block_group(cache->space_info, cache);
8744 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8746 set_avail_alloc_bits(extent_root->fs_info, type);
8751 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8753 u64 extra_flags = chunk_to_extended(flags) &
8754 BTRFS_EXTENDED_PROFILE_MASK;
8756 write_seqlock(&fs_info->profiles_lock);
8757 if (flags & BTRFS_BLOCK_GROUP_DATA)
8758 fs_info->avail_data_alloc_bits &= ~extra_flags;
8759 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8760 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8761 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8762 fs_info->avail_system_alloc_bits &= ~extra_flags;
8763 write_sequnlock(&fs_info->profiles_lock);
8766 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8767 struct btrfs_root *root, u64 group_start)
8769 struct btrfs_path *path;
8770 struct btrfs_block_group_cache *block_group;
8771 struct btrfs_free_cluster *cluster;
8772 struct btrfs_root *tree_root = root->fs_info->tree_root;
8773 struct btrfs_key key;
8774 struct inode *inode;
8779 root = root->fs_info->extent_root;
8781 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8782 BUG_ON(!block_group);
8783 BUG_ON(!block_group->ro);
8786 * Free the reserved super bytes from this block group before
8789 free_excluded_extents(root, block_group);
8791 memcpy(&key, &block_group->key, sizeof(key));
8792 index = get_block_group_index(block_group);
8793 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8794 BTRFS_BLOCK_GROUP_RAID1 |
8795 BTRFS_BLOCK_GROUP_RAID10))
8800 /* make sure this block group isn't part of an allocation cluster */
8801 cluster = &root->fs_info->data_alloc_cluster;
8802 spin_lock(&cluster->refill_lock);
8803 btrfs_return_cluster_to_free_space(block_group, cluster);
8804 spin_unlock(&cluster->refill_lock);
8807 * make sure this block group isn't part of a metadata
8808 * allocation cluster
8810 cluster = &root->fs_info->meta_alloc_cluster;
8811 spin_lock(&cluster->refill_lock);
8812 btrfs_return_cluster_to_free_space(block_group, cluster);
8813 spin_unlock(&cluster->refill_lock);
8815 path = btrfs_alloc_path();
8821 inode = lookup_free_space_inode(tree_root, block_group, path);
8822 if (!IS_ERR(inode)) {
8823 ret = btrfs_orphan_add(trans, inode);
8825 btrfs_add_delayed_iput(inode);
8829 /* One for the block groups ref */
8830 spin_lock(&block_group->lock);
8831 if (block_group->iref) {
8832 block_group->iref = 0;
8833 block_group->inode = NULL;
8834 spin_unlock(&block_group->lock);
8837 spin_unlock(&block_group->lock);
8839 /* One for our lookup ref */
8840 btrfs_add_delayed_iput(inode);
8843 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8844 key.offset = block_group->key.objectid;
8847 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8851 btrfs_release_path(path);
8853 ret = btrfs_del_item(trans, tree_root, path);
8856 btrfs_release_path(path);
8859 spin_lock(&root->fs_info->block_group_cache_lock);
8860 rb_erase(&block_group->cache_node,
8861 &root->fs_info->block_group_cache_tree);
8863 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8864 root->fs_info->first_logical_byte = (u64)-1;
8865 spin_unlock(&root->fs_info->block_group_cache_lock);
8867 down_write(&block_group->space_info->groups_sem);
8869 * we must use list_del_init so people can check to see if they
8870 * are still on the list after taking the semaphore
8872 list_del_init(&block_group->list);
8873 if (list_empty(&block_group->space_info->block_groups[index])) {
8874 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8875 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8876 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8878 up_write(&block_group->space_info->groups_sem);
8880 if (block_group->cached == BTRFS_CACHE_STARTED)
8881 wait_block_group_cache_done(block_group);
8883 btrfs_remove_free_space_cache(block_group);
8885 spin_lock(&block_group->space_info->lock);
8886 block_group->space_info->total_bytes -= block_group->key.offset;
8887 block_group->space_info->bytes_readonly -= block_group->key.offset;
8888 block_group->space_info->disk_total -= block_group->key.offset * factor;
8889 spin_unlock(&block_group->space_info->lock);
8891 memcpy(&key, &block_group->key, sizeof(key));
8893 btrfs_clear_space_info_full(root->fs_info);
8895 btrfs_put_block_group(block_group);
8896 btrfs_put_block_group(block_group);
8898 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8904 ret = btrfs_del_item(trans, root, path);
8906 btrfs_free_path(path);
8910 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8912 struct btrfs_space_info *space_info;
8913 struct btrfs_super_block *disk_super;
8919 disk_super = fs_info->super_copy;
8920 if (!btrfs_super_root(disk_super))
8923 features = btrfs_super_incompat_flags(disk_super);
8924 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8927 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8928 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8933 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8934 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8936 flags = BTRFS_BLOCK_GROUP_METADATA;
8937 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8941 flags = BTRFS_BLOCK_GROUP_DATA;
8942 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8948 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8950 return unpin_extent_range(root, start, end);
8953 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8954 u64 num_bytes, u64 *actual_bytes)
8956 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8959 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8961 struct btrfs_fs_info *fs_info = root->fs_info;
8962 struct btrfs_block_group_cache *cache = NULL;
8967 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8971 * try to trim all FS space, our block group may start from non-zero.
8973 if (range->len == total_bytes)
8974 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8976 cache = btrfs_lookup_block_group(fs_info, range->start);
8979 if (cache->key.objectid >= (range->start + range->len)) {
8980 btrfs_put_block_group(cache);
8984 start = max(range->start, cache->key.objectid);
8985 end = min(range->start + range->len,
8986 cache->key.objectid + cache->key.offset);
8988 if (end - start >= range->minlen) {
8989 if (!block_group_cache_done(cache)) {
8990 ret = cache_block_group(cache, 0);
8992 btrfs_put_block_group(cache);
8995 ret = wait_block_group_cache_done(cache);
8997 btrfs_put_block_group(cache);
9001 ret = btrfs_trim_block_group(cache,
9007 trimmed += group_trimmed;
9009 btrfs_put_block_group(cache);
9014 cache = next_block_group(fs_info->tree_root, cache);
9017 range->len = trimmed;