2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE = 0,
57 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 static int update_block_group(struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
108 int btrfs_pin_extent(struct btrfs_root *root,
109 u64 bytenr, u64 num_bytes, int reserved);
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 if (block_group->key.objectid < cache->key.objectid) {
159 } else if (block_group->key.objectid > cache->key.objectid) {
162 spin_unlock(&info->block_group_cache_lock);
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
174 spin_unlock(&info->block_group_cache_lock);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
187 struct btrfs_block_group_cache *cache, *ret = NULL;
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
195 cache = rb_entry(n, struct btrfs_block_group_cache,
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
220 spin_unlock(&info->block_group_cache_lock);
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (cache->cached != BTRFS_CACHE_STARTED) {
315 spin_unlock(&cache->lock);
319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache->caching_ctl) {
321 spin_unlock(&cache->lock);
325 ctl = cache->caching_ctl;
326 atomic_inc(&ctl->count);
327 spin_unlock(&cache->lock);
331 static void put_caching_control(struct btrfs_caching_control *ctl)
333 if (atomic_dec_and_test(&ctl->count))
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
343 struct btrfs_fs_info *info, u64 start, u64 end)
345 u64 extent_start, extent_end, size, total_added = 0;
348 while (start < end) {
349 ret = find_first_extent_bit(info->pinned_extents, start,
350 &extent_start, &extent_end,
351 EXTENT_DIRTY | EXTENT_UPTODATE,
356 if (extent_start <= start) {
357 start = extent_end + 1;
358 } else if (extent_start > start && extent_start < end) {
359 size = extent_start - start;
361 ret = btrfs_add_free_space(block_group, start,
363 BUG_ON(ret); /* -ENOMEM or logic error */
364 start = extent_end + 1;
373 ret = btrfs_add_free_space(block_group, start, size);
374 BUG_ON(ret); /* -ENOMEM or logic error */
380 static noinline void caching_thread(struct btrfs_work *work)
382 struct btrfs_block_group_cache *block_group;
383 struct btrfs_fs_info *fs_info;
384 struct btrfs_caching_control *caching_ctl;
385 struct btrfs_root *extent_root;
386 struct btrfs_path *path;
387 struct extent_buffer *leaf;
388 struct btrfs_key key;
394 caching_ctl = container_of(work, struct btrfs_caching_control, work);
395 block_group = caching_ctl->block_group;
396 fs_info = block_group->fs_info;
397 extent_root = fs_info->extent_root;
399 path = btrfs_alloc_path();
403 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
411 path->skip_locking = 1;
412 path->search_commit_root = 1;
417 key.type = BTRFS_EXTENT_ITEM_KEY;
419 mutex_lock(&caching_ctl->mutex);
420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info->extent_commit_sem);
424 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
428 leaf = path->nodes[0];
429 nritems = btrfs_header_nritems(leaf);
432 if (btrfs_fs_closing(fs_info) > 1) {
437 if (path->slots[0] < nritems) {
438 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
440 ret = find_next_key(path, 0, &key);
444 if (need_resched()) {
445 caching_ctl->progress = last;
446 btrfs_release_path(path);
447 up_read(&fs_info->extent_commit_sem);
448 mutex_unlock(&caching_ctl->mutex);
453 ret = btrfs_next_leaf(extent_root, path);
458 leaf = path->nodes[0];
459 nritems = btrfs_header_nritems(leaf);
463 if (key.objectid < last) {
466 key.type = BTRFS_EXTENT_ITEM_KEY;
468 caching_ctl->progress = last;
469 btrfs_release_path(path);
473 if (key.objectid < block_group->key.objectid) {
478 if (key.objectid >= block_group->key.objectid +
479 block_group->key.offset)
482 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 key.type == BTRFS_METADATA_ITEM_KEY) {
484 total_found += add_new_free_space(block_group,
487 if (key.type == BTRFS_METADATA_ITEM_KEY)
488 last = key.objectid +
489 fs_info->tree_root->leafsize;
491 last = key.objectid + key.offset;
493 if (total_found > (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl->wait);
502 total_found += add_new_free_space(block_group, fs_info, last,
503 block_group->key.objectid +
504 block_group->key.offset);
505 caching_ctl->progress = (u64)-1;
507 spin_lock(&block_group->lock);
508 block_group->caching_ctl = NULL;
509 block_group->cached = BTRFS_CACHE_FINISHED;
510 spin_unlock(&block_group->lock);
513 btrfs_free_path(path);
514 up_read(&fs_info->extent_commit_sem);
516 free_excluded_extents(extent_root, block_group);
518 mutex_unlock(&caching_ctl->mutex);
521 spin_lock(&block_group->lock);
522 block_group->caching_ctl = NULL;
523 block_group->cached = BTRFS_CACHE_ERROR;
524 spin_unlock(&block_group->lock);
526 wake_up(&caching_ctl->wait);
528 put_caching_control(caching_ctl);
529 btrfs_put_block_group(block_group);
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
536 struct btrfs_fs_info *fs_info = cache->fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
544 INIT_LIST_HEAD(&caching_ctl->list);
545 mutex_init(&caching_ctl->mutex);
546 init_waitqueue_head(&caching_ctl->wait);
547 caching_ctl->block_group = cache;
548 caching_ctl->progress = cache->key.objectid;
549 atomic_set(&caching_ctl->count, 1);
550 caching_ctl->work.func = caching_thread;
552 spin_lock(&cache->lock);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 ret = load_free_space_cache(fs_info, cache);
593 spin_lock(&cache->lock);
595 cache->caching_ctl = NULL;
596 cache->cached = BTRFS_CACHE_FINISHED;
597 cache->last_byte_to_unpin = (u64)-1;
599 if (load_cache_only) {
600 cache->caching_ctl = NULL;
601 cache->cached = BTRFS_CACHE_NO;
603 cache->cached = BTRFS_CACHE_STARTED;
606 spin_unlock(&cache->lock);
607 wake_up(&caching_ctl->wait);
609 put_caching_control(caching_ctl);
610 free_excluded_extents(fs_info->extent_root, cache);
615 * We are not going to do the fast caching, set cached to the
616 * appropriate value and wakeup any waiters.
618 spin_lock(&cache->lock);
619 if (load_cache_only) {
620 cache->caching_ctl = NULL;
621 cache->cached = BTRFS_CACHE_NO;
623 cache->cached = BTRFS_CACHE_STARTED;
625 spin_unlock(&cache->lock);
626 wake_up(&caching_ctl->wait);
629 if (load_cache_only) {
630 put_caching_control(caching_ctl);
634 down_write(&fs_info->extent_commit_sem);
635 atomic_inc(&caching_ctl->count);
636 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
637 up_write(&fs_info->extent_commit_sem);
639 btrfs_get_block_group(cache);
641 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
647 * return the block group that starts at or after bytenr
649 static struct btrfs_block_group_cache *
650 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
652 struct btrfs_block_group_cache *cache;
654 cache = block_group_cache_tree_search(info, bytenr, 0);
660 * return the block group that contains the given bytenr
662 struct btrfs_block_group_cache *btrfs_lookup_block_group(
663 struct btrfs_fs_info *info,
666 struct btrfs_block_group_cache *cache;
668 cache = block_group_cache_tree_search(info, bytenr, 1);
673 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
676 struct list_head *head = &info->space_info;
677 struct btrfs_space_info *found;
679 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
682 list_for_each_entry_rcu(found, head, list) {
683 if (found->flags & flags) {
693 * after adding space to the filesystem, we need to clear the full flags
694 * on all the space infos.
696 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
698 struct list_head *head = &info->space_info;
699 struct btrfs_space_info *found;
702 list_for_each_entry_rcu(found, head, list)
707 /* simple helper to search for an existing extent at a given offset */
708 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
711 struct btrfs_key key;
712 struct btrfs_path *path;
714 path = btrfs_alloc_path();
718 key.objectid = start;
720 key.type = BTRFS_EXTENT_ITEM_KEY;
721 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
724 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
725 if (key.objectid == start &&
726 key.type == BTRFS_METADATA_ITEM_KEY)
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->leafsize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
784 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
789 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
790 if (path->slots[0]) {
792 btrfs_item_key_to_cpu(path->nodes[0], &key,
794 if (key.objectid == bytenr &&
795 key.type == BTRFS_EXTENT_ITEM_KEY &&
796 key.offset == root->leafsize)
800 key.objectid = bytenr;
801 key.type = BTRFS_EXTENT_ITEM_KEY;
802 key.offset = root->leafsize;
803 btrfs_release_path(path);
809 leaf = path->nodes[0];
810 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
811 if (item_size >= sizeof(*ei)) {
812 ei = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item);
814 num_refs = btrfs_extent_refs(leaf, ei);
815 extent_flags = btrfs_extent_flags(leaf, ei);
817 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
818 struct btrfs_extent_item_v0 *ei0;
819 BUG_ON(item_size != sizeof(*ei0));
820 ei0 = btrfs_item_ptr(leaf, path->slots[0],
821 struct btrfs_extent_item_v0);
822 num_refs = btrfs_extent_refs_v0(leaf, ei0);
823 /* FIXME: this isn't correct for data */
824 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
829 BUG_ON(num_refs == 0);
839 delayed_refs = &trans->transaction->delayed_refs;
840 spin_lock(&delayed_refs->lock);
841 head = btrfs_find_delayed_ref_head(trans, bytenr);
843 if (!mutex_trylock(&head->mutex)) {
844 atomic_inc(&head->node.refs);
845 spin_unlock(&delayed_refs->lock);
847 btrfs_release_path(path);
850 * Mutex was contended, block until it's released and try
853 mutex_lock(&head->mutex);
854 mutex_unlock(&head->mutex);
855 btrfs_put_delayed_ref(&head->node);
858 if (head->extent_op && head->extent_op->update_flags)
859 extent_flags |= head->extent_op->flags_to_set;
861 BUG_ON(num_refs == 0);
863 num_refs += head->node.ref_mod;
864 mutex_unlock(&head->mutex);
866 spin_unlock(&delayed_refs->lock);
868 WARN_ON(num_refs == 0);
872 *flags = extent_flags;
874 btrfs_free_path(path);
879 * Back reference rules. Back refs have three main goals:
881 * 1) differentiate between all holders of references to an extent so that
882 * when a reference is dropped we can make sure it was a valid reference
883 * before freeing the extent.
885 * 2) Provide enough information to quickly find the holders of an extent
886 * if we notice a given block is corrupted or bad.
888 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
889 * maintenance. This is actually the same as #2, but with a slightly
890 * different use case.
892 * There are two kinds of back refs. The implicit back refs is optimized
893 * for pointers in non-shared tree blocks. For a given pointer in a block,
894 * back refs of this kind provide information about the block's owner tree
895 * and the pointer's key. These information allow us to find the block by
896 * b-tree searching. The full back refs is for pointers in tree blocks not
897 * referenced by their owner trees. The location of tree block is recorded
898 * in the back refs. Actually the full back refs is generic, and can be
899 * used in all cases the implicit back refs is used. The major shortcoming
900 * of the full back refs is its overhead. Every time a tree block gets
901 * COWed, we have to update back refs entry for all pointers in it.
903 * For a newly allocated tree block, we use implicit back refs for
904 * pointers in it. This means most tree related operations only involve
905 * implicit back refs. For a tree block created in old transaction, the
906 * only way to drop a reference to it is COW it. So we can detect the
907 * event that tree block loses its owner tree's reference and do the
908 * back refs conversion.
910 * When a tree block is COW'd through a tree, there are four cases:
912 * The reference count of the block is one and the tree is the block's
913 * owner tree. Nothing to do in this case.
915 * The reference count of the block is one and the tree is not the
916 * block's owner tree. In this case, full back refs is used for pointers
917 * in the block. Remove these full back refs, add implicit back refs for
918 * every pointers in the new block.
920 * The reference count of the block is greater than one and the tree is
921 * the block's owner tree. In this case, implicit back refs is used for
922 * pointers in the block. Add full back refs for every pointers in the
923 * block, increase lower level extents' reference counts. The original
924 * implicit back refs are entailed to the new block.
926 * The reference count of the block is greater than one and the tree is
927 * not the block's owner tree. Add implicit back refs for every pointer in
928 * the new block, increase lower level extents' reference count.
930 * Back Reference Key composing:
932 * The key objectid corresponds to the first byte in the extent,
933 * The key type is used to differentiate between types of back refs.
934 * There are different meanings of the key offset for different types
937 * File extents can be referenced by:
939 * - multiple snapshots, subvolumes, or different generations in one subvol
940 * - different files inside a single subvolume
941 * - different offsets inside a file (bookend extents in file.c)
943 * The extent ref structure for the implicit back refs has fields for:
945 * - Objectid of the subvolume root
946 * - objectid of the file holding the reference
947 * - original offset in the file
948 * - how many bookend extents
950 * The key offset for the implicit back refs is hash of the first
953 * The extent ref structure for the full back refs has field for:
955 * - number of pointers in the tree leaf
957 * The key offset for the implicit back refs is the first byte of
960 * When a file extent is allocated, The implicit back refs is used.
961 * the fields are filled in:
963 * (root_key.objectid, inode objectid, offset in file, 1)
965 * When a file extent is removed file truncation, we find the
966 * corresponding implicit back refs and check the following fields:
968 * (btrfs_header_owner(leaf), inode objectid, offset in file)
970 * Btree extents can be referenced by:
972 * - Different subvolumes
974 * Both the implicit back refs and the full back refs for tree blocks
975 * only consist of key. The key offset for the implicit back refs is
976 * objectid of block's owner tree. The key offset for the full back refs
977 * is the first byte of parent block.
979 * When implicit back refs is used, information about the lowest key and
980 * level of the tree block are required. These information are stored in
981 * tree block info structure.
984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
985 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root,
987 struct btrfs_path *path,
988 u64 owner, u32 extra_size)
990 struct btrfs_extent_item *item;
991 struct btrfs_extent_item_v0 *ei0;
992 struct btrfs_extent_ref_v0 *ref0;
993 struct btrfs_tree_block_info *bi;
994 struct extent_buffer *leaf;
995 struct btrfs_key key;
996 struct btrfs_key found_key;
997 u32 new_size = sizeof(*item);
1001 leaf = path->nodes[0];
1002 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1004 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1005 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_extent_item_v0);
1007 refs = btrfs_extent_refs_v0(leaf, ei0);
1009 if (owner == (u64)-1) {
1011 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1012 ret = btrfs_next_leaf(root, path);
1015 BUG_ON(ret > 0); /* Corruption */
1016 leaf = path->nodes[0];
1018 btrfs_item_key_to_cpu(leaf, &found_key,
1020 BUG_ON(key.objectid != found_key.objectid);
1021 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1025 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1026 struct btrfs_extent_ref_v0);
1027 owner = btrfs_ref_objectid_v0(leaf, ref0);
1031 btrfs_release_path(path);
1033 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1034 new_size += sizeof(*bi);
1036 new_size -= sizeof(*ei0);
1037 ret = btrfs_search_slot(trans, root, &key, path,
1038 new_size + extra_size, 1);
1041 BUG_ON(ret); /* Corruption */
1043 btrfs_extend_item(root, path, new_size);
1045 leaf = path->nodes[0];
1046 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1047 btrfs_set_extent_refs(leaf, item, refs);
1048 /* FIXME: get real generation */
1049 btrfs_set_extent_generation(leaf, item, 0);
1050 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1051 btrfs_set_extent_flags(leaf, item,
1052 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1053 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1054 bi = (struct btrfs_tree_block_info *)(item + 1);
1055 /* FIXME: get first key of the block */
1056 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1057 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1059 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1061 btrfs_mark_buffer_dirty(leaf);
1066 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1068 u32 high_crc = ~(u32)0;
1069 u32 low_crc = ~(u32)0;
1072 lenum = cpu_to_le64(root_objectid);
1073 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1074 lenum = cpu_to_le64(owner);
1075 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1076 lenum = cpu_to_le64(offset);
1077 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1079 return ((u64)high_crc << 31) ^ (u64)low_crc;
1082 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1083 struct btrfs_extent_data_ref *ref)
1085 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1086 btrfs_extent_data_ref_objectid(leaf, ref),
1087 btrfs_extent_data_ref_offset(leaf, ref));
1090 static int match_extent_data_ref(struct extent_buffer *leaf,
1091 struct btrfs_extent_data_ref *ref,
1092 u64 root_objectid, u64 owner, u64 offset)
1094 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1095 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1096 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1101 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1102 struct btrfs_root *root,
1103 struct btrfs_path *path,
1104 u64 bytenr, u64 parent,
1106 u64 owner, u64 offset)
1108 struct btrfs_key key;
1109 struct btrfs_extent_data_ref *ref;
1110 struct extent_buffer *leaf;
1116 key.objectid = bytenr;
1118 key.type = BTRFS_SHARED_DATA_REF_KEY;
1119 key.offset = parent;
1121 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1122 key.offset = hash_extent_data_ref(root_objectid,
1127 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1136 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1137 key.type = BTRFS_EXTENT_REF_V0_KEY;
1138 btrfs_release_path(path);
1139 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1150 leaf = path->nodes[0];
1151 nritems = btrfs_header_nritems(leaf);
1153 if (path->slots[0] >= nritems) {
1154 ret = btrfs_next_leaf(root, path);
1160 leaf = path->nodes[0];
1161 nritems = btrfs_header_nritems(leaf);
1165 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1166 if (key.objectid != bytenr ||
1167 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1170 ref = btrfs_item_ptr(leaf, path->slots[0],
1171 struct btrfs_extent_data_ref);
1173 if (match_extent_data_ref(leaf, ref, root_objectid,
1176 btrfs_release_path(path);
1188 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1189 struct btrfs_root *root,
1190 struct btrfs_path *path,
1191 u64 bytenr, u64 parent,
1192 u64 root_objectid, u64 owner,
1193 u64 offset, int refs_to_add)
1195 struct btrfs_key key;
1196 struct extent_buffer *leaf;
1201 key.objectid = bytenr;
1203 key.type = BTRFS_SHARED_DATA_REF_KEY;
1204 key.offset = parent;
1205 size = sizeof(struct btrfs_shared_data_ref);
1207 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1208 key.offset = hash_extent_data_ref(root_objectid,
1210 size = sizeof(struct btrfs_extent_data_ref);
1213 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1214 if (ret && ret != -EEXIST)
1217 leaf = path->nodes[0];
1219 struct btrfs_shared_data_ref *ref;
1220 ref = btrfs_item_ptr(leaf, path->slots[0],
1221 struct btrfs_shared_data_ref);
1223 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1225 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1226 num_refs += refs_to_add;
1227 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1230 struct btrfs_extent_data_ref *ref;
1231 while (ret == -EEXIST) {
1232 ref = btrfs_item_ptr(leaf, path->slots[0],
1233 struct btrfs_extent_data_ref);
1234 if (match_extent_data_ref(leaf, ref, root_objectid,
1237 btrfs_release_path(path);
1239 ret = btrfs_insert_empty_item(trans, root, path, &key,
1241 if (ret && ret != -EEXIST)
1244 leaf = path->nodes[0];
1246 ref = btrfs_item_ptr(leaf, path->slots[0],
1247 struct btrfs_extent_data_ref);
1249 btrfs_set_extent_data_ref_root(leaf, ref,
1251 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1252 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1253 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1255 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1256 num_refs += refs_to_add;
1257 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1260 btrfs_mark_buffer_dirty(leaf);
1263 btrfs_release_path(path);
1267 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1268 struct btrfs_root *root,
1269 struct btrfs_path *path,
1272 struct btrfs_key key;
1273 struct btrfs_extent_data_ref *ref1 = NULL;
1274 struct btrfs_shared_data_ref *ref2 = NULL;
1275 struct extent_buffer *leaf;
1279 leaf = path->nodes[0];
1280 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1282 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1283 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_data_ref);
1285 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1286 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1287 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_shared_data_ref);
1289 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1290 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1291 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1292 struct btrfs_extent_ref_v0 *ref0;
1293 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1294 struct btrfs_extent_ref_v0);
1295 num_refs = btrfs_ref_count_v0(leaf, ref0);
1301 BUG_ON(num_refs < refs_to_drop);
1302 num_refs -= refs_to_drop;
1304 if (num_refs == 0) {
1305 ret = btrfs_del_item(trans, root, path);
1307 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1308 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1309 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1310 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1311 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1313 struct btrfs_extent_ref_v0 *ref0;
1314 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1315 struct btrfs_extent_ref_v0);
1316 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1319 btrfs_mark_buffer_dirty(leaf);
1324 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1325 struct btrfs_path *path,
1326 struct btrfs_extent_inline_ref *iref)
1328 struct btrfs_key key;
1329 struct extent_buffer *leaf;
1330 struct btrfs_extent_data_ref *ref1;
1331 struct btrfs_shared_data_ref *ref2;
1334 leaf = path->nodes[0];
1335 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1337 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1338 BTRFS_EXTENT_DATA_REF_KEY) {
1339 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1340 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1342 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1343 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1345 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1346 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1347 struct btrfs_extent_data_ref);
1348 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1349 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1350 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_shared_data_ref);
1352 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1353 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1354 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1355 struct btrfs_extent_ref_v0 *ref0;
1356 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1357 struct btrfs_extent_ref_v0);
1358 num_refs = btrfs_ref_count_v0(leaf, ref0);
1366 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1367 struct btrfs_root *root,
1368 struct btrfs_path *path,
1369 u64 bytenr, u64 parent,
1372 struct btrfs_key key;
1375 key.objectid = bytenr;
1377 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1378 key.offset = parent;
1380 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1381 key.offset = root_objectid;
1384 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1387 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1388 if (ret == -ENOENT && parent) {
1389 btrfs_release_path(path);
1390 key.type = BTRFS_EXTENT_REF_V0_KEY;
1391 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1399 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1400 struct btrfs_root *root,
1401 struct btrfs_path *path,
1402 u64 bytenr, u64 parent,
1405 struct btrfs_key key;
1408 key.objectid = bytenr;
1410 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1411 key.offset = parent;
1413 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1414 key.offset = root_objectid;
1417 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1418 btrfs_release_path(path);
1422 static inline int extent_ref_type(u64 parent, u64 owner)
1425 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1427 type = BTRFS_SHARED_BLOCK_REF_KEY;
1429 type = BTRFS_TREE_BLOCK_REF_KEY;
1432 type = BTRFS_SHARED_DATA_REF_KEY;
1434 type = BTRFS_EXTENT_DATA_REF_KEY;
1439 static int find_next_key(struct btrfs_path *path, int level,
1440 struct btrfs_key *key)
1443 for (; level < BTRFS_MAX_LEVEL; level++) {
1444 if (!path->nodes[level])
1446 if (path->slots[level] + 1 >=
1447 btrfs_header_nritems(path->nodes[level]))
1450 btrfs_item_key_to_cpu(path->nodes[level], key,
1451 path->slots[level] + 1);
1453 btrfs_node_key_to_cpu(path->nodes[level], key,
1454 path->slots[level] + 1);
1461 * look for inline back ref. if back ref is found, *ref_ret is set
1462 * to the address of inline back ref, and 0 is returned.
1464 * if back ref isn't found, *ref_ret is set to the address where it
1465 * should be inserted, and -ENOENT is returned.
1467 * if insert is true and there are too many inline back refs, the path
1468 * points to the extent item, and -EAGAIN is returned.
1470 * NOTE: inline back refs are ordered in the same way that back ref
1471 * items in the tree are ordered.
1473 static noinline_for_stack
1474 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1475 struct btrfs_root *root,
1476 struct btrfs_path *path,
1477 struct btrfs_extent_inline_ref **ref_ret,
1478 u64 bytenr, u64 num_bytes,
1479 u64 parent, u64 root_objectid,
1480 u64 owner, u64 offset, int insert)
1482 struct btrfs_key key;
1483 struct extent_buffer *leaf;
1484 struct btrfs_extent_item *ei;
1485 struct btrfs_extent_inline_ref *iref;
1495 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1498 key.objectid = bytenr;
1499 key.type = BTRFS_EXTENT_ITEM_KEY;
1500 key.offset = num_bytes;
1502 want = extent_ref_type(parent, owner);
1504 extra_size = btrfs_extent_inline_ref_size(want);
1505 path->keep_locks = 1;
1510 * Owner is our parent level, so we can just add one to get the level
1511 * for the block we are interested in.
1513 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1514 key.type = BTRFS_METADATA_ITEM_KEY;
1519 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1526 * We may be a newly converted file system which still has the old fat
1527 * extent entries for metadata, so try and see if we have one of those.
1529 if (ret > 0 && skinny_metadata) {
1530 skinny_metadata = false;
1531 if (path->slots[0]) {
1533 btrfs_item_key_to_cpu(path->nodes[0], &key,
1535 if (key.objectid == bytenr &&
1536 key.type == BTRFS_EXTENT_ITEM_KEY &&
1537 key.offset == num_bytes)
1541 key.type = BTRFS_EXTENT_ITEM_KEY;
1542 key.offset = num_bytes;
1543 btrfs_release_path(path);
1548 if (ret && !insert) {
1551 } else if (WARN_ON(ret)) {
1556 leaf = path->nodes[0];
1557 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1558 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1559 if (item_size < sizeof(*ei)) {
1564 ret = convert_extent_item_v0(trans, root, path, owner,
1570 leaf = path->nodes[0];
1571 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1574 BUG_ON(item_size < sizeof(*ei));
1576 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1577 flags = btrfs_extent_flags(leaf, ei);
1579 ptr = (unsigned long)(ei + 1);
1580 end = (unsigned long)ei + item_size;
1582 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1583 ptr += sizeof(struct btrfs_tree_block_info);
1593 iref = (struct btrfs_extent_inline_ref *)ptr;
1594 type = btrfs_extent_inline_ref_type(leaf, iref);
1598 ptr += btrfs_extent_inline_ref_size(type);
1602 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1603 struct btrfs_extent_data_ref *dref;
1604 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1605 if (match_extent_data_ref(leaf, dref, root_objectid,
1610 if (hash_extent_data_ref_item(leaf, dref) <
1611 hash_extent_data_ref(root_objectid, owner, offset))
1615 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1617 if (parent == ref_offset) {
1621 if (ref_offset < parent)
1624 if (root_objectid == ref_offset) {
1628 if (ref_offset < root_objectid)
1632 ptr += btrfs_extent_inline_ref_size(type);
1634 if (err == -ENOENT && insert) {
1635 if (item_size + extra_size >=
1636 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1641 * To add new inline back ref, we have to make sure
1642 * there is no corresponding back ref item.
1643 * For simplicity, we just do not add new inline back
1644 * ref if there is any kind of item for this block
1646 if (find_next_key(path, 0, &key) == 0 &&
1647 key.objectid == bytenr &&
1648 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1653 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1656 path->keep_locks = 0;
1657 btrfs_unlock_up_safe(path, 1);
1663 * helper to add new inline back ref
1665 static noinline_for_stack
1666 void setup_inline_extent_backref(struct btrfs_root *root,
1667 struct btrfs_path *path,
1668 struct btrfs_extent_inline_ref *iref,
1669 u64 parent, u64 root_objectid,
1670 u64 owner, u64 offset, int refs_to_add,
1671 struct btrfs_delayed_extent_op *extent_op)
1673 struct extent_buffer *leaf;
1674 struct btrfs_extent_item *ei;
1677 unsigned long item_offset;
1682 leaf = path->nodes[0];
1683 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1684 item_offset = (unsigned long)iref - (unsigned long)ei;
1686 type = extent_ref_type(parent, owner);
1687 size = btrfs_extent_inline_ref_size(type);
1689 btrfs_extend_item(root, path, size);
1691 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1692 refs = btrfs_extent_refs(leaf, ei);
1693 refs += refs_to_add;
1694 btrfs_set_extent_refs(leaf, ei, refs);
1696 __run_delayed_extent_op(extent_op, leaf, ei);
1698 ptr = (unsigned long)ei + item_offset;
1699 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1700 if (ptr < end - size)
1701 memmove_extent_buffer(leaf, ptr + size, ptr,
1704 iref = (struct btrfs_extent_inline_ref *)ptr;
1705 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1706 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1707 struct btrfs_extent_data_ref *dref;
1708 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1709 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1710 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1711 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1712 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1713 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1714 struct btrfs_shared_data_ref *sref;
1715 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1716 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1719 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1723 btrfs_mark_buffer_dirty(leaf);
1726 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1727 struct btrfs_root *root,
1728 struct btrfs_path *path,
1729 struct btrfs_extent_inline_ref **ref_ret,
1730 u64 bytenr, u64 num_bytes, u64 parent,
1731 u64 root_objectid, u64 owner, u64 offset)
1735 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1736 bytenr, num_bytes, parent,
1737 root_objectid, owner, offset, 0);
1741 btrfs_release_path(path);
1744 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1745 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1748 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1749 root_objectid, owner, offset);
1755 * helper to update/remove inline back ref
1757 static noinline_for_stack
1758 void update_inline_extent_backref(struct btrfs_root *root,
1759 struct btrfs_path *path,
1760 struct btrfs_extent_inline_ref *iref,
1762 struct btrfs_delayed_extent_op *extent_op)
1764 struct extent_buffer *leaf;
1765 struct btrfs_extent_item *ei;
1766 struct btrfs_extent_data_ref *dref = NULL;
1767 struct btrfs_shared_data_ref *sref = NULL;
1775 leaf = path->nodes[0];
1776 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1777 refs = btrfs_extent_refs(leaf, ei);
1778 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1779 refs += refs_to_mod;
1780 btrfs_set_extent_refs(leaf, ei, refs);
1782 __run_delayed_extent_op(extent_op, leaf, ei);
1784 type = btrfs_extent_inline_ref_type(leaf, iref);
1786 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1787 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1788 refs = btrfs_extent_data_ref_count(leaf, dref);
1789 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1790 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1791 refs = btrfs_shared_data_ref_count(leaf, sref);
1794 BUG_ON(refs_to_mod != -1);
1797 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1798 refs += refs_to_mod;
1801 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1802 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1804 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1806 size = btrfs_extent_inline_ref_size(type);
1807 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1808 ptr = (unsigned long)iref;
1809 end = (unsigned long)ei + item_size;
1810 if (ptr + size < end)
1811 memmove_extent_buffer(leaf, ptr, ptr + size,
1814 btrfs_truncate_item(root, path, item_size, 1);
1816 btrfs_mark_buffer_dirty(leaf);
1819 static noinline_for_stack
1820 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1821 struct btrfs_root *root,
1822 struct btrfs_path *path,
1823 u64 bytenr, u64 num_bytes, u64 parent,
1824 u64 root_objectid, u64 owner,
1825 u64 offset, int refs_to_add,
1826 struct btrfs_delayed_extent_op *extent_op)
1828 struct btrfs_extent_inline_ref *iref;
1831 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1832 bytenr, num_bytes, parent,
1833 root_objectid, owner, offset, 1);
1835 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1836 update_inline_extent_backref(root, path, iref,
1837 refs_to_add, extent_op);
1838 } else if (ret == -ENOENT) {
1839 setup_inline_extent_backref(root, path, iref, parent,
1840 root_objectid, owner, offset,
1841 refs_to_add, extent_op);
1847 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1848 struct btrfs_root *root,
1849 struct btrfs_path *path,
1850 u64 bytenr, u64 parent, u64 root_objectid,
1851 u64 owner, u64 offset, int refs_to_add)
1854 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1855 BUG_ON(refs_to_add != 1);
1856 ret = insert_tree_block_ref(trans, root, path, bytenr,
1857 parent, root_objectid);
1859 ret = insert_extent_data_ref(trans, root, path, bytenr,
1860 parent, root_objectid,
1861 owner, offset, refs_to_add);
1866 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1867 struct btrfs_root *root,
1868 struct btrfs_path *path,
1869 struct btrfs_extent_inline_ref *iref,
1870 int refs_to_drop, int is_data)
1874 BUG_ON(!is_data && refs_to_drop != 1);
1876 update_inline_extent_backref(root, path, iref,
1877 -refs_to_drop, NULL);
1878 } else if (is_data) {
1879 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1881 ret = btrfs_del_item(trans, root, path);
1886 static int btrfs_issue_discard(struct block_device *bdev,
1889 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1892 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1893 u64 num_bytes, u64 *actual_bytes)
1896 u64 discarded_bytes = 0;
1897 struct btrfs_bio *bbio = NULL;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1902 bytenr, &num_bytes, &bbio, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe *stripe = bbio->stripes;
1909 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1910 if (!stripe->dev->can_discard)
1913 ret = btrfs_issue_discard(stripe->dev->bdev,
1917 discarded_bytes += stripe->length;
1918 else if (ret != -EOPNOTSUPP)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1932 *actual_bytes = discarded_bytes;
1935 if (ret == -EOPNOTSUPP)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 u64 bytenr, u64 num_bytes, u64 parent,
1944 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1947 struct btrfs_fs_info *fs_info = root->fs_info;
1949 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1950 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1952 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1953 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1955 parent, root_objectid, (int)owner,
1956 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1958 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1960 parent, root_objectid, owner, offset,
1961 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1966 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1967 struct btrfs_root *root,
1968 u64 bytenr, u64 num_bytes,
1969 u64 parent, u64 root_objectid,
1970 u64 owner, u64 offset, int refs_to_add,
1971 struct btrfs_delayed_extent_op *extent_op)
1973 struct btrfs_path *path;
1974 struct extent_buffer *leaf;
1975 struct btrfs_extent_item *item;
1979 path = btrfs_alloc_path();
1984 path->leave_spinning = 1;
1985 /* this will setup the path even if it fails to insert the back ref */
1986 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1987 path, bytenr, num_bytes, parent,
1988 root_objectid, owner, offset,
1989 refs_to_add, extent_op);
1993 leaf = path->nodes[0];
1994 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1995 refs = btrfs_extent_refs(leaf, item);
1996 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1998 __run_delayed_extent_op(extent_op, leaf, item);
2000 btrfs_mark_buffer_dirty(leaf);
2001 btrfs_release_path(path);
2004 path->leave_spinning = 1;
2006 /* now insert the actual backref */
2007 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2008 path, bytenr, parent, root_objectid,
2009 owner, offset, refs_to_add);
2011 btrfs_abort_transaction(trans, root, ret);
2013 btrfs_free_path(path);
2017 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2018 struct btrfs_root *root,
2019 struct btrfs_delayed_ref_node *node,
2020 struct btrfs_delayed_extent_op *extent_op,
2021 int insert_reserved)
2024 struct btrfs_delayed_data_ref *ref;
2025 struct btrfs_key ins;
2030 ins.objectid = node->bytenr;
2031 ins.offset = node->num_bytes;
2032 ins.type = BTRFS_EXTENT_ITEM_KEY;
2034 ref = btrfs_delayed_node_to_data_ref(node);
2035 trace_run_delayed_data_ref(node, ref, node->action);
2037 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2038 parent = ref->parent;
2040 ref_root = ref->root;
2042 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2044 flags |= extent_op->flags_to_set;
2045 ret = alloc_reserved_file_extent(trans, root,
2046 parent, ref_root, flags,
2047 ref->objectid, ref->offset,
2048 &ins, node->ref_mod);
2049 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2050 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2051 node->num_bytes, parent,
2052 ref_root, ref->objectid,
2053 ref->offset, node->ref_mod,
2055 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2056 ret = __btrfs_free_extent(trans, root, node->bytenr,
2057 node->num_bytes, parent,
2058 ref_root, ref->objectid,
2059 ref->offset, node->ref_mod,
2067 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2068 struct extent_buffer *leaf,
2069 struct btrfs_extent_item *ei)
2071 u64 flags = btrfs_extent_flags(leaf, ei);
2072 if (extent_op->update_flags) {
2073 flags |= extent_op->flags_to_set;
2074 btrfs_set_extent_flags(leaf, ei, flags);
2077 if (extent_op->update_key) {
2078 struct btrfs_tree_block_info *bi;
2079 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2080 bi = (struct btrfs_tree_block_info *)(ei + 1);
2081 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2085 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2086 struct btrfs_root *root,
2087 struct btrfs_delayed_ref_node *node,
2088 struct btrfs_delayed_extent_op *extent_op)
2090 struct btrfs_key key;
2091 struct btrfs_path *path;
2092 struct btrfs_extent_item *ei;
2093 struct extent_buffer *leaf;
2097 int metadata = !extent_op->is_data;
2102 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2105 path = btrfs_alloc_path();
2109 key.objectid = node->bytenr;
2112 key.type = BTRFS_METADATA_ITEM_KEY;
2113 key.offset = extent_op->level;
2115 key.type = BTRFS_EXTENT_ITEM_KEY;
2116 key.offset = node->num_bytes;
2121 path->leave_spinning = 1;
2122 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2130 if (path->slots[0] > 0) {
2132 btrfs_item_key_to_cpu(path->nodes[0], &key,
2134 if (key.objectid == node->bytenr &&
2135 key.type == BTRFS_EXTENT_ITEM_KEY &&
2136 key.offset == node->num_bytes)
2140 btrfs_release_path(path);
2143 key.objectid = node->bytenr;
2144 key.offset = node->num_bytes;
2145 key.type = BTRFS_EXTENT_ITEM_KEY;
2154 leaf = path->nodes[0];
2155 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2156 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2157 if (item_size < sizeof(*ei)) {
2158 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2164 leaf = path->nodes[0];
2165 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2168 BUG_ON(item_size < sizeof(*ei));
2169 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2170 __run_delayed_extent_op(extent_op, leaf, ei);
2172 btrfs_mark_buffer_dirty(leaf);
2174 btrfs_free_path(path);
2178 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2179 struct btrfs_root *root,
2180 struct btrfs_delayed_ref_node *node,
2181 struct btrfs_delayed_extent_op *extent_op,
2182 int insert_reserved)
2185 struct btrfs_delayed_tree_ref *ref;
2186 struct btrfs_key ins;
2189 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2192 ref = btrfs_delayed_node_to_tree_ref(node);
2193 trace_run_delayed_tree_ref(node, ref, node->action);
2195 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2196 parent = ref->parent;
2198 ref_root = ref->root;
2200 ins.objectid = node->bytenr;
2201 if (skinny_metadata) {
2202 ins.offset = ref->level;
2203 ins.type = BTRFS_METADATA_ITEM_KEY;
2205 ins.offset = node->num_bytes;
2206 ins.type = BTRFS_EXTENT_ITEM_KEY;
2209 BUG_ON(node->ref_mod != 1);
2210 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2211 BUG_ON(!extent_op || !extent_op->update_flags);
2212 ret = alloc_reserved_tree_block(trans, root,
2214 extent_op->flags_to_set,
2217 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2218 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2219 node->num_bytes, parent, ref_root,
2220 ref->level, 0, 1, extent_op);
2221 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2222 ret = __btrfs_free_extent(trans, root, node->bytenr,
2223 node->num_bytes, parent, ref_root,
2224 ref->level, 0, 1, extent_op);
2231 /* helper function to actually process a single delayed ref entry */
2232 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2233 struct btrfs_root *root,
2234 struct btrfs_delayed_ref_node *node,
2235 struct btrfs_delayed_extent_op *extent_op,
2236 int insert_reserved)
2240 if (trans->aborted) {
2241 if (insert_reserved)
2242 btrfs_pin_extent(root, node->bytenr,
2243 node->num_bytes, 1);
2247 if (btrfs_delayed_ref_is_head(node)) {
2248 struct btrfs_delayed_ref_head *head;
2250 * we've hit the end of the chain and we were supposed
2251 * to insert this extent into the tree. But, it got
2252 * deleted before we ever needed to insert it, so all
2253 * we have to do is clean up the accounting
2256 head = btrfs_delayed_node_to_head(node);
2257 trace_run_delayed_ref_head(node, head, node->action);
2259 if (insert_reserved) {
2260 btrfs_pin_extent(root, node->bytenr,
2261 node->num_bytes, 1);
2262 if (head->is_data) {
2263 ret = btrfs_del_csums(trans, root,
2271 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2272 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2273 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2275 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2276 node->type == BTRFS_SHARED_DATA_REF_KEY)
2277 ret = run_delayed_data_ref(trans, root, node, extent_op,
2284 static noinline struct btrfs_delayed_ref_node *
2285 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2287 struct rb_node *node;
2288 struct btrfs_delayed_ref_node *ref;
2289 int action = BTRFS_ADD_DELAYED_REF;
2292 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2293 * this prevents ref count from going down to zero when
2294 * there still are pending delayed ref.
2296 node = rb_prev(&head->node.rb_node);
2300 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2302 if (ref->bytenr != head->node.bytenr)
2304 if (ref->action == action)
2306 node = rb_prev(node);
2308 if (action == BTRFS_ADD_DELAYED_REF) {
2309 action = BTRFS_DROP_DELAYED_REF;
2316 * Returns 0 on success or if called with an already aborted transaction.
2317 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2319 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2320 struct btrfs_root *root,
2321 struct list_head *cluster)
2323 struct btrfs_delayed_ref_root *delayed_refs;
2324 struct btrfs_delayed_ref_node *ref;
2325 struct btrfs_delayed_ref_head *locked_ref = NULL;
2326 struct btrfs_delayed_extent_op *extent_op;
2327 struct btrfs_fs_info *fs_info = root->fs_info;
2330 int must_insert_reserved = 0;
2332 delayed_refs = &trans->transaction->delayed_refs;
2335 /* pick a new head ref from the cluster list */
2336 if (list_empty(cluster))
2339 locked_ref = list_entry(cluster->next,
2340 struct btrfs_delayed_ref_head, cluster);
2342 /* grab the lock that says we are going to process
2343 * all the refs for this head */
2344 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2347 * we may have dropped the spin lock to get the head
2348 * mutex lock, and that might have given someone else
2349 * time to free the head. If that's true, it has been
2350 * removed from our list and we can move on.
2352 if (ret == -EAGAIN) {
2360 * We need to try and merge add/drops of the same ref since we
2361 * can run into issues with relocate dropping the implicit ref
2362 * and then it being added back again before the drop can
2363 * finish. If we merged anything we need to re-loop so we can
2366 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2370 * locked_ref is the head node, so we have to go one
2371 * node back for any delayed ref updates
2373 ref = select_delayed_ref(locked_ref);
2375 if (ref && ref->seq &&
2376 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2378 * there are still refs with lower seq numbers in the
2379 * process of being added. Don't run this ref yet.
2381 list_del_init(&locked_ref->cluster);
2382 btrfs_delayed_ref_unlock(locked_ref);
2384 delayed_refs->num_heads_ready++;
2385 spin_unlock(&delayed_refs->lock);
2387 spin_lock(&delayed_refs->lock);
2392 * record the must insert reserved flag before we
2393 * drop the spin lock.
2395 must_insert_reserved = locked_ref->must_insert_reserved;
2396 locked_ref->must_insert_reserved = 0;
2398 extent_op = locked_ref->extent_op;
2399 locked_ref->extent_op = NULL;
2402 /* All delayed refs have been processed, Go ahead
2403 * and send the head node to run_one_delayed_ref,
2404 * so that any accounting fixes can happen
2406 ref = &locked_ref->node;
2408 if (extent_op && must_insert_reserved) {
2409 btrfs_free_delayed_extent_op(extent_op);
2414 spin_unlock(&delayed_refs->lock);
2416 ret = run_delayed_extent_op(trans, root,
2418 btrfs_free_delayed_extent_op(extent_op);
2422 * Need to reset must_insert_reserved if
2423 * there was an error so the abort stuff
2424 * can cleanup the reserved space
2427 if (must_insert_reserved)
2428 locked_ref->must_insert_reserved = 1;
2429 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2430 spin_lock(&delayed_refs->lock);
2431 btrfs_delayed_ref_unlock(locked_ref);
2440 rb_erase(&ref->rb_node, &delayed_refs->root);
2441 delayed_refs->num_entries--;
2442 if (!btrfs_delayed_ref_is_head(ref)) {
2444 * when we play the delayed ref, also correct the
2447 switch (ref->action) {
2448 case BTRFS_ADD_DELAYED_REF:
2449 case BTRFS_ADD_DELAYED_EXTENT:
2450 locked_ref->node.ref_mod -= ref->ref_mod;
2452 case BTRFS_DROP_DELAYED_REF:
2453 locked_ref->node.ref_mod += ref->ref_mod;
2459 list_del_init(&locked_ref->cluster);
2461 spin_unlock(&delayed_refs->lock);
2463 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2464 must_insert_reserved);
2466 btrfs_free_delayed_extent_op(extent_op);
2468 btrfs_delayed_ref_unlock(locked_ref);
2469 btrfs_put_delayed_ref(ref);
2470 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2471 spin_lock(&delayed_refs->lock);
2476 * If this node is a head, that means all the refs in this head
2477 * have been dealt with, and we will pick the next head to deal
2478 * with, so we must unlock the head and drop it from the cluster
2479 * list before we release it.
2481 if (btrfs_delayed_ref_is_head(ref)) {
2482 btrfs_delayed_ref_unlock(locked_ref);
2485 btrfs_put_delayed_ref(ref);
2489 spin_lock(&delayed_refs->lock);
2494 #ifdef SCRAMBLE_DELAYED_REFS
2496 * Normally delayed refs get processed in ascending bytenr order. This
2497 * correlates in most cases to the order added. To expose dependencies on this
2498 * order, we start to process the tree in the middle instead of the beginning
2500 static u64 find_middle(struct rb_root *root)
2502 struct rb_node *n = root->rb_node;
2503 struct btrfs_delayed_ref_node *entry;
2506 u64 first = 0, last = 0;
2510 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2511 first = entry->bytenr;
2515 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2516 last = entry->bytenr;
2521 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2522 WARN_ON(!entry->in_tree);
2524 middle = entry->bytenr;
2537 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2538 struct btrfs_fs_info *fs_info)
2540 struct qgroup_update *qgroup_update;
2543 if (list_empty(&trans->qgroup_ref_list) !=
2544 !trans->delayed_ref_elem.seq) {
2545 /* list without seq or seq without list */
2547 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2548 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2549 (u32)(trans->delayed_ref_elem.seq >> 32),
2550 (u32)trans->delayed_ref_elem.seq);
2554 if (!trans->delayed_ref_elem.seq)
2557 while (!list_empty(&trans->qgroup_ref_list)) {
2558 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2559 struct qgroup_update, list);
2560 list_del(&qgroup_update->list);
2562 ret = btrfs_qgroup_account_ref(
2563 trans, fs_info, qgroup_update->node,
2564 qgroup_update->extent_op);
2565 kfree(qgroup_update);
2568 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2573 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2576 int val = atomic_read(&delayed_refs->ref_seq);
2578 if (val < seq || val >= seq + count)
2583 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2587 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2588 sizeof(struct btrfs_extent_inline_ref));
2589 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2590 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2593 * We don't ever fill up leaves all the way so multiply by 2 just to be
2594 * closer to what we're really going to want to ouse.
2596 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2599 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2600 struct btrfs_root *root)
2602 struct btrfs_block_rsv *global_rsv;
2603 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2607 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2608 num_heads = heads_to_leaves(root, num_heads);
2610 num_bytes += (num_heads - 1) * root->leafsize;
2612 global_rsv = &root->fs_info->global_block_rsv;
2615 * If we can't allocate any more chunks lets make sure we have _lots_ of
2616 * wiggle room since running delayed refs can create more delayed refs.
2618 if (global_rsv->space_info->full)
2621 spin_lock(&global_rsv->lock);
2622 if (global_rsv->reserved <= num_bytes)
2624 spin_unlock(&global_rsv->lock);
2629 * this starts processing the delayed reference count updates and
2630 * extent insertions we have queued up so far. count can be
2631 * 0, which means to process everything in the tree at the start
2632 * of the run (but not newly added entries), or it can be some target
2633 * number you'd like to process.
2635 * Returns 0 on success or if called with an aborted transaction
2636 * Returns <0 on error and aborts the transaction
2638 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2639 struct btrfs_root *root, unsigned long count)
2641 struct rb_node *node;
2642 struct btrfs_delayed_ref_root *delayed_refs;
2643 struct btrfs_delayed_ref_node *ref;
2644 struct list_head cluster;
2647 int run_all = count == (unsigned long)-1;
2651 /* We'll clean this up in btrfs_cleanup_transaction */
2655 if (root == root->fs_info->extent_root)
2656 root = root->fs_info->tree_root;
2658 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2660 delayed_refs = &trans->transaction->delayed_refs;
2661 INIT_LIST_HEAD(&cluster);
2663 count = delayed_refs->num_entries * 2;
2667 if (!run_all && !run_most) {
2669 int seq = atomic_read(&delayed_refs->ref_seq);
2672 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2674 DEFINE_WAIT(__wait);
2675 if (delayed_refs->flushing ||
2676 !btrfs_should_throttle_delayed_refs(trans, root))
2679 prepare_to_wait(&delayed_refs->wait, &__wait,
2680 TASK_UNINTERRUPTIBLE);
2682 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2685 finish_wait(&delayed_refs->wait, &__wait);
2687 if (!refs_newer(delayed_refs, seq, 256))
2692 finish_wait(&delayed_refs->wait, &__wait);
2698 atomic_inc(&delayed_refs->procs_running_refs);
2703 spin_lock(&delayed_refs->lock);
2705 #ifdef SCRAMBLE_DELAYED_REFS
2706 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2710 if (!(run_all || run_most) &&
2711 !btrfs_should_throttle_delayed_refs(trans, root))
2715 * go find something we can process in the rbtree. We start at
2716 * the beginning of the tree, and then build a cluster
2717 * of refs to process starting at the first one we are able to
2720 delayed_start = delayed_refs->run_delayed_start;
2721 ret = btrfs_find_ref_cluster(trans, &cluster,
2722 delayed_refs->run_delayed_start);
2726 ret = run_clustered_refs(trans, root, &cluster);
2728 btrfs_release_ref_cluster(&cluster);
2729 spin_unlock(&delayed_refs->lock);
2730 btrfs_abort_transaction(trans, root, ret);
2731 atomic_dec(&delayed_refs->procs_running_refs);
2732 wake_up(&delayed_refs->wait);
2736 atomic_add(ret, &delayed_refs->ref_seq);
2738 count -= min_t(unsigned long, ret, count);
2743 if (delayed_start >= delayed_refs->run_delayed_start) {
2746 * btrfs_find_ref_cluster looped. let's do one
2747 * more cycle. if we don't run any delayed ref
2748 * during that cycle (because we can't because
2749 * all of them are blocked), bail out.
2754 * no runnable refs left, stop trying
2761 /* refs were run, let's reset staleness detection */
2767 if (!list_empty(&trans->new_bgs)) {
2768 spin_unlock(&delayed_refs->lock);
2769 btrfs_create_pending_block_groups(trans, root);
2770 spin_lock(&delayed_refs->lock);
2773 node = rb_first(&delayed_refs->root);
2776 count = (unsigned long)-1;
2779 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2781 if (btrfs_delayed_ref_is_head(ref)) {
2782 struct btrfs_delayed_ref_head *head;
2784 head = btrfs_delayed_node_to_head(ref);
2785 atomic_inc(&ref->refs);
2787 spin_unlock(&delayed_refs->lock);
2789 * Mutex was contended, block until it's
2790 * released and try again
2792 mutex_lock(&head->mutex);
2793 mutex_unlock(&head->mutex);
2795 btrfs_put_delayed_ref(ref);
2799 node = rb_next(node);
2801 spin_unlock(&delayed_refs->lock);
2802 schedule_timeout(1);
2806 atomic_dec(&delayed_refs->procs_running_refs);
2808 if (waitqueue_active(&delayed_refs->wait))
2809 wake_up(&delayed_refs->wait);
2811 spin_unlock(&delayed_refs->lock);
2812 assert_qgroups_uptodate(trans);
2816 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2817 struct btrfs_root *root,
2818 u64 bytenr, u64 num_bytes, u64 flags,
2819 int level, int is_data)
2821 struct btrfs_delayed_extent_op *extent_op;
2824 extent_op = btrfs_alloc_delayed_extent_op();
2828 extent_op->flags_to_set = flags;
2829 extent_op->update_flags = 1;
2830 extent_op->update_key = 0;
2831 extent_op->is_data = is_data ? 1 : 0;
2832 extent_op->level = level;
2834 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2835 num_bytes, extent_op);
2837 btrfs_free_delayed_extent_op(extent_op);
2841 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2842 struct btrfs_root *root,
2843 struct btrfs_path *path,
2844 u64 objectid, u64 offset, u64 bytenr)
2846 struct btrfs_delayed_ref_head *head;
2847 struct btrfs_delayed_ref_node *ref;
2848 struct btrfs_delayed_data_ref *data_ref;
2849 struct btrfs_delayed_ref_root *delayed_refs;
2850 struct rb_node *node;
2854 delayed_refs = &trans->transaction->delayed_refs;
2855 spin_lock(&delayed_refs->lock);
2856 head = btrfs_find_delayed_ref_head(trans, bytenr);
2860 if (!mutex_trylock(&head->mutex)) {
2861 atomic_inc(&head->node.refs);
2862 spin_unlock(&delayed_refs->lock);
2864 btrfs_release_path(path);
2867 * Mutex was contended, block until it's released and let
2870 mutex_lock(&head->mutex);
2871 mutex_unlock(&head->mutex);
2872 btrfs_put_delayed_ref(&head->node);
2876 node = rb_prev(&head->node.rb_node);
2880 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2882 if (ref->bytenr != bytenr)
2886 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2889 data_ref = btrfs_delayed_node_to_data_ref(ref);
2891 node = rb_prev(node);
2895 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2896 if (ref->bytenr == bytenr && ref->seq == seq)
2900 if (data_ref->root != root->root_key.objectid ||
2901 data_ref->objectid != objectid || data_ref->offset != offset)
2906 mutex_unlock(&head->mutex);
2908 spin_unlock(&delayed_refs->lock);
2912 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2913 struct btrfs_root *root,
2914 struct btrfs_path *path,
2915 u64 objectid, u64 offset, u64 bytenr)
2917 struct btrfs_root *extent_root = root->fs_info->extent_root;
2918 struct extent_buffer *leaf;
2919 struct btrfs_extent_data_ref *ref;
2920 struct btrfs_extent_inline_ref *iref;
2921 struct btrfs_extent_item *ei;
2922 struct btrfs_key key;
2926 key.objectid = bytenr;
2927 key.offset = (u64)-1;
2928 key.type = BTRFS_EXTENT_ITEM_KEY;
2930 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2933 BUG_ON(ret == 0); /* Corruption */
2936 if (path->slots[0] == 0)
2940 leaf = path->nodes[0];
2941 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2943 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2947 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2948 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2949 if (item_size < sizeof(*ei)) {
2950 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2954 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2956 if (item_size != sizeof(*ei) +
2957 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2960 if (btrfs_extent_generation(leaf, ei) <=
2961 btrfs_root_last_snapshot(&root->root_item))
2964 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2965 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2966 BTRFS_EXTENT_DATA_REF_KEY)
2969 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2970 if (btrfs_extent_refs(leaf, ei) !=
2971 btrfs_extent_data_ref_count(leaf, ref) ||
2972 btrfs_extent_data_ref_root(leaf, ref) !=
2973 root->root_key.objectid ||
2974 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2975 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2983 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2984 struct btrfs_root *root,
2985 u64 objectid, u64 offset, u64 bytenr)
2987 struct btrfs_path *path;
2991 path = btrfs_alloc_path();
2996 ret = check_committed_ref(trans, root, path, objectid,
2998 if (ret && ret != -ENOENT)
3001 ret2 = check_delayed_ref(trans, root, path, objectid,
3003 } while (ret2 == -EAGAIN);
3005 if (ret2 && ret2 != -ENOENT) {
3010 if (ret != -ENOENT || ret2 != -ENOENT)
3013 btrfs_free_path(path);
3014 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3019 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3020 struct btrfs_root *root,
3021 struct extent_buffer *buf,
3022 int full_backref, int inc, int for_cow)
3029 struct btrfs_key key;
3030 struct btrfs_file_extent_item *fi;
3034 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3035 u64, u64, u64, u64, u64, u64, int);
3037 ref_root = btrfs_header_owner(buf);
3038 nritems = btrfs_header_nritems(buf);
3039 level = btrfs_header_level(buf);
3041 if (!root->ref_cows && level == 0)
3045 process_func = btrfs_inc_extent_ref;
3047 process_func = btrfs_free_extent;
3050 parent = buf->start;
3054 for (i = 0; i < nritems; i++) {
3056 btrfs_item_key_to_cpu(buf, &key, i);
3057 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3059 fi = btrfs_item_ptr(buf, i,
3060 struct btrfs_file_extent_item);
3061 if (btrfs_file_extent_type(buf, fi) ==
3062 BTRFS_FILE_EXTENT_INLINE)
3064 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3068 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3069 key.offset -= btrfs_file_extent_offset(buf, fi);
3070 ret = process_func(trans, root, bytenr, num_bytes,
3071 parent, ref_root, key.objectid,
3072 key.offset, for_cow);
3076 bytenr = btrfs_node_blockptr(buf, i);
3077 num_bytes = btrfs_level_size(root, level - 1);
3078 ret = process_func(trans, root, bytenr, num_bytes,
3079 parent, ref_root, level - 1, 0,
3090 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3091 struct extent_buffer *buf, int full_backref, int for_cow)
3093 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3096 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3097 struct extent_buffer *buf, int full_backref, int for_cow)
3099 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3102 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3103 struct btrfs_root *root,
3104 struct btrfs_path *path,
3105 struct btrfs_block_group_cache *cache)
3108 struct btrfs_root *extent_root = root->fs_info->extent_root;
3110 struct extent_buffer *leaf;
3112 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3115 BUG_ON(ret); /* Corruption */
3117 leaf = path->nodes[0];
3118 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3119 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3120 btrfs_mark_buffer_dirty(leaf);
3121 btrfs_release_path(path);
3124 btrfs_abort_transaction(trans, root, ret);
3131 static struct btrfs_block_group_cache *
3132 next_block_group(struct btrfs_root *root,
3133 struct btrfs_block_group_cache *cache)
3135 struct rb_node *node;
3136 spin_lock(&root->fs_info->block_group_cache_lock);
3137 node = rb_next(&cache->cache_node);
3138 btrfs_put_block_group(cache);
3140 cache = rb_entry(node, struct btrfs_block_group_cache,
3142 btrfs_get_block_group(cache);
3145 spin_unlock(&root->fs_info->block_group_cache_lock);
3149 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3150 struct btrfs_trans_handle *trans,
3151 struct btrfs_path *path)
3153 struct btrfs_root *root = block_group->fs_info->tree_root;
3154 struct inode *inode = NULL;
3156 int dcs = BTRFS_DC_ERROR;
3162 * If this block group is smaller than 100 megs don't bother caching the
3165 if (block_group->key.offset < (100 * 1024 * 1024)) {
3166 spin_lock(&block_group->lock);
3167 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3168 spin_unlock(&block_group->lock);
3173 inode = lookup_free_space_inode(root, block_group, path);
3174 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3175 ret = PTR_ERR(inode);
3176 btrfs_release_path(path);
3180 if (IS_ERR(inode)) {
3184 if (block_group->ro)
3187 ret = create_free_space_inode(root, trans, block_group, path);
3193 /* We've already setup this transaction, go ahead and exit */
3194 if (block_group->cache_generation == trans->transid &&
3195 i_size_read(inode)) {
3196 dcs = BTRFS_DC_SETUP;
3201 * We want to set the generation to 0, that way if anything goes wrong
3202 * from here on out we know not to trust this cache when we load up next
3205 BTRFS_I(inode)->generation = 0;
3206 ret = btrfs_update_inode(trans, root, inode);
3209 if (i_size_read(inode) > 0) {
3210 ret = btrfs_check_trunc_cache_free_space(root,
3211 &root->fs_info->global_block_rsv);
3215 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3220 spin_lock(&block_group->lock);
3221 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3222 !btrfs_test_opt(root, SPACE_CACHE)) {
3224 * don't bother trying to write stuff out _if_
3225 * a) we're not cached,
3226 * b) we're with nospace_cache mount option.
3228 dcs = BTRFS_DC_WRITTEN;
3229 spin_unlock(&block_group->lock);
3232 spin_unlock(&block_group->lock);
3235 * Try to preallocate enough space based on how big the block group is.
3236 * Keep in mind this has to include any pinned space which could end up
3237 * taking up quite a bit since it's not folded into the other space
3240 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3245 num_pages *= PAGE_CACHE_SIZE;
3247 ret = btrfs_check_data_free_space(inode, num_pages);
3251 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3252 num_pages, num_pages,
3255 dcs = BTRFS_DC_SETUP;
3256 btrfs_free_reserved_data_space(inode, num_pages);
3261 btrfs_release_path(path);
3263 spin_lock(&block_group->lock);
3264 if (!ret && dcs == BTRFS_DC_SETUP)
3265 block_group->cache_generation = trans->transid;
3266 block_group->disk_cache_state = dcs;
3267 spin_unlock(&block_group->lock);
3272 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3273 struct btrfs_root *root)
3275 struct btrfs_block_group_cache *cache;
3277 struct btrfs_path *path;
3280 path = btrfs_alloc_path();
3286 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3288 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3290 cache = next_block_group(root, cache);
3298 err = cache_save_setup(cache, trans, path);
3299 last = cache->key.objectid + cache->key.offset;
3300 btrfs_put_block_group(cache);
3305 err = btrfs_run_delayed_refs(trans, root,
3307 if (err) /* File system offline */
3311 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3313 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3314 btrfs_put_block_group(cache);
3320 cache = next_block_group(root, cache);
3329 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3330 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3332 last = cache->key.objectid + cache->key.offset;
3334 err = write_one_cache_group(trans, root, path, cache);
3335 btrfs_put_block_group(cache);
3336 if (err) /* File system offline */
3342 * I don't think this is needed since we're just marking our
3343 * preallocated extent as written, but just in case it can't
3347 err = btrfs_run_delayed_refs(trans, root,
3349 if (err) /* File system offline */
3353 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3356 * Really this shouldn't happen, but it could if we
3357 * couldn't write the entire preallocated extent and
3358 * splitting the extent resulted in a new block.
3361 btrfs_put_block_group(cache);
3364 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3366 cache = next_block_group(root, cache);
3375 err = btrfs_write_out_cache(root, trans, cache, path);
3378 * If we didn't have an error then the cache state is still
3379 * NEED_WRITE, so we can set it to WRITTEN.
3381 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3382 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3383 last = cache->key.objectid + cache->key.offset;
3384 btrfs_put_block_group(cache);
3388 btrfs_free_path(path);
3392 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3394 struct btrfs_block_group_cache *block_group;
3397 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3398 if (!block_group || block_group->ro)
3401 btrfs_put_block_group(block_group);
3405 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3406 u64 total_bytes, u64 bytes_used,
3407 struct btrfs_space_info **space_info)
3409 struct btrfs_space_info *found;
3414 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3415 BTRFS_BLOCK_GROUP_RAID10))
3420 found = __find_space_info(info, flags);
3422 spin_lock(&found->lock);
3423 found->total_bytes += total_bytes;
3424 found->disk_total += total_bytes * factor;
3425 found->bytes_used += bytes_used;
3426 found->disk_used += bytes_used * factor;
3428 spin_unlock(&found->lock);
3429 *space_info = found;
3432 found = kzalloc(sizeof(*found), GFP_NOFS);
3436 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3442 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3443 INIT_LIST_HEAD(&found->block_groups[i]);
3444 init_rwsem(&found->groups_sem);
3445 spin_lock_init(&found->lock);
3446 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
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;
3451 found->bytes_pinned = 0;
3452 found->bytes_reserved = 0;
3453 found->bytes_readonly = 0;
3454 found->bytes_may_use = 0;
3456 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3457 found->chunk_alloc = 0;
3459 init_waitqueue_head(&found->wait);
3460 *space_info = found;
3461 list_add_rcu(&found->list, &info->space_info);
3462 if (flags & BTRFS_BLOCK_GROUP_DATA)
3463 info->data_sinfo = found;
3467 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3469 u64 extra_flags = chunk_to_extended(flags) &
3470 BTRFS_EXTENDED_PROFILE_MASK;
3472 write_seqlock(&fs_info->profiles_lock);
3473 if (flags & BTRFS_BLOCK_GROUP_DATA)
3474 fs_info->avail_data_alloc_bits |= extra_flags;
3475 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3476 fs_info->avail_metadata_alloc_bits |= extra_flags;
3477 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3478 fs_info->avail_system_alloc_bits |= extra_flags;
3479 write_sequnlock(&fs_info->profiles_lock);
3483 * returns target flags in extended format or 0 if restripe for this
3484 * chunk_type is not in progress
3486 * should be called with either volume_mutex or balance_lock held
3488 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3490 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3496 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3497 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3498 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3499 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3500 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3501 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3502 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3503 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3504 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3511 * @flags: available profiles in extended format (see ctree.h)
3513 * Returns reduced profile in chunk format. If profile changing is in
3514 * progress (either running or paused) picks the target profile (if it's
3515 * already available), otherwise falls back to plain reducing.
3517 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3520 * we add in the count of missing devices because we want
3521 * to make sure that any RAID levels on a degraded FS
3522 * continue to be honored.
3524 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3525 root->fs_info->fs_devices->missing_devices;
3530 * see if restripe for this chunk_type is in progress, if so
3531 * try to reduce to the target profile
3533 spin_lock(&root->fs_info->balance_lock);
3534 target = get_restripe_target(root->fs_info, flags);
3536 /* pick target profile only if it's already available */
3537 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3538 spin_unlock(&root->fs_info->balance_lock);
3539 return extended_to_chunk(target);
3542 spin_unlock(&root->fs_info->balance_lock);
3544 /* First, mask out the RAID levels which aren't possible */
3545 if (num_devices == 1)
3546 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3547 BTRFS_BLOCK_GROUP_RAID5);
3548 if (num_devices < 3)
3549 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3550 if (num_devices < 4)
3551 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3553 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3554 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3555 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3558 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3559 tmp = BTRFS_BLOCK_GROUP_RAID6;
3560 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3561 tmp = BTRFS_BLOCK_GROUP_RAID5;
3562 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3563 tmp = BTRFS_BLOCK_GROUP_RAID10;
3564 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3565 tmp = BTRFS_BLOCK_GROUP_RAID1;
3566 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3567 tmp = BTRFS_BLOCK_GROUP_RAID0;
3569 return extended_to_chunk(flags | tmp);
3572 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3577 seq = read_seqbegin(&root->fs_info->profiles_lock);
3579 if (flags & BTRFS_BLOCK_GROUP_DATA)
3580 flags |= root->fs_info->avail_data_alloc_bits;
3581 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3582 flags |= root->fs_info->avail_system_alloc_bits;
3583 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3584 flags |= root->fs_info->avail_metadata_alloc_bits;
3585 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3587 return btrfs_reduce_alloc_profile(root, flags);
3590 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3596 flags = BTRFS_BLOCK_GROUP_DATA;
3597 else if (root == root->fs_info->chunk_root)
3598 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3600 flags = BTRFS_BLOCK_GROUP_METADATA;
3602 ret = get_alloc_profile(root, flags);
3607 * This will check the space that the inode allocates from to make sure we have
3608 * enough space for bytes.
3610 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3612 struct btrfs_space_info *data_sinfo;
3613 struct btrfs_root *root = BTRFS_I(inode)->root;
3614 struct btrfs_fs_info *fs_info = root->fs_info;
3616 int ret = 0, committed = 0, alloc_chunk = 1;
3618 /* make sure bytes are sectorsize aligned */
3619 bytes = ALIGN(bytes, root->sectorsize);
3621 if (btrfs_is_free_space_inode(inode)) {
3623 ASSERT(current->journal_info);
3626 data_sinfo = fs_info->data_sinfo;
3631 /* make sure we have enough space to handle the data first */
3632 spin_lock(&data_sinfo->lock);
3633 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3634 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3635 data_sinfo->bytes_may_use;
3637 if (used + bytes > data_sinfo->total_bytes) {
3638 struct btrfs_trans_handle *trans;
3641 * if we don't have enough free bytes in this space then we need
3642 * to alloc a new chunk.
3644 if (!data_sinfo->full && alloc_chunk) {
3647 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3648 spin_unlock(&data_sinfo->lock);
3650 alloc_target = btrfs_get_alloc_profile(root, 1);
3652 * It is ugly that we don't call nolock join
3653 * transaction for the free space inode case here.
3654 * But it is safe because we only do the data space
3655 * reservation for the free space cache in the
3656 * transaction context, the common join transaction
3657 * just increase the counter of the current transaction
3658 * handler, doesn't try to acquire the trans_lock of
3661 trans = btrfs_join_transaction(root);
3663 return PTR_ERR(trans);
3665 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3667 CHUNK_ALLOC_NO_FORCE);
3668 btrfs_end_transaction(trans, root);
3677 data_sinfo = fs_info->data_sinfo;
3683 * If we don't have enough pinned space to deal with this
3684 * allocation don't bother committing the transaction.
3686 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3689 spin_unlock(&data_sinfo->lock);
3691 /* commit the current transaction and try again */
3694 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3697 trans = btrfs_join_transaction(root);
3699 return PTR_ERR(trans);
3700 ret = btrfs_commit_transaction(trans, root);
3706 trace_btrfs_space_reservation(root->fs_info,
3707 "space_info:enospc",
3708 data_sinfo->flags, bytes, 1);
3711 data_sinfo->bytes_may_use += bytes;
3712 trace_btrfs_space_reservation(root->fs_info, "space_info",
3713 data_sinfo->flags, bytes, 1);
3714 spin_unlock(&data_sinfo->lock);
3720 * Called if we need to clear a data reservation for this inode.
3722 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3724 struct btrfs_root *root = BTRFS_I(inode)->root;
3725 struct btrfs_space_info *data_sinfo;
3727 /* make sure bytes are sectorsize aligned */
3728 bytes = ALIGN(bytes, root->sectorsize);
3730 data_sinfo = root->fs_info->data_sinfo;
3731 spin_lock(&data_sinfo->lock);
3732 WARN_ON(data_sinfo->bytes_may_use < bytes);
3733 data_sinfo->bytes_may_use -= bytes;
3734 trace_btrfs_space_reservation(root->fs_info, "space_info",
3735 data_sinfo->flags, bytes, 0);
3736 spin_unlock(&data_sinfo->lock);
3739 static void force_metadata_allocation(struct btrfs_fs_info *info)
3741 struct list_head *head = &info->space_info;
3742 struct btrfs_space_info *found;
3745 list_for_each_entry_rcu(found, head, list) {
3746 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3747 found->force_alloc = CHUNK_ALLOC_FORCE;
3752 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3754 return (global->size << 1);
3757 static int should_alloc_chunk(struct btrfs_root *root,
3758 struct btrfs_space_info *sinfo, int force)
3760 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3761 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3762 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3765 if (force == CHUNK_ALLOC_FORCE)
3769 * We need to take into account the global rsv because for all intents
3770 * and purposes it's used space. Don't worry about locking the
3771 * global_rsv, it doesn't change except when the transaction commits.
3773 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3774 num_allocated += calc_global_rsv_need_space(global_rsv);
3777 * in limited mode, we want to have some free space up to
3778 * about 1% of the FS size.
3780 if (force == CHUNK_ALLOC_LIMITED) {
3781 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3782 thresh = max_t(u64, 64 * 1024 * 1024,
3783 div_factor_fine(thresh, 1));
3785 if (num_bytes - num_allocated < thresh)
3789 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3794 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3798 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3799 BTRFS_BLOCK_GROUP_RAID0 |
3800 BTRFS_BLOCK_GROUP_RAID5 |
3801 BTRFS_BLOCK_GROUP_RAID6))
3802 num_dev = root->fs_info->fs_devices->rw_devices;
3803 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3806 num_dev = 1; /* DUP or single */
3808 /* metadata for updaing devices and chunk tree */
3809 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3812 static void check_system_chunk(struct btrfs_trans_handle *trans,
3813 struct btrfs_root *root, u64 type)
3815 struct btrfs_space_info *info;
3819 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3820 spin_lock(&info->lock);
3821 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3822 info->bytes_reserved - info->bytes_readonly;
3823 spin_unlock(&info->lock);
3825 thresh = get_system_chunk_thresh(root, type);
3826 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3827 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3828 left, thresh, type);
3829 dump_space_info(info, 0, 0);
3832 if (left < thresh) {
3835 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3836 btrfs_alloc_chunk(trans, root, flags);
3840 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3841 struct btrfs_root *extent_root, u64 flags, int force)
3843 struct btrfs_space_info *space_info;
3844 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3845 int wait_for_alloc = 0;
3848 /* Don't re-enter if we're already allocating a chunk */
3849 if (trans->allocating_chunk)
3852 space_info = __find_space_info(extent_root->fs_info, flags);
3854 ret = update_space_info(extent_root->fs_info, flags,
3856 BUG_ON(ret); /* -ENOMEM */
3858 BUG_ON(!space_info); /* Logic error */
3861 spin_lock(&space_info->lock);
3862 if (force < space_info->force_alloc)
3863 force = space_info->force_alloc;
3864 if (space_info->full) {
3865 if (should_alloc_chunk(extent_root, space_info, force))
3869 spin_unlock(&space_info->lock);
3873 if (!should_alloc_chunk(extent_root, space_info, force)) {
3874 spin_unlock(&space_info->lock);
3876 } else if (space_info->chunk_alloc) {
3879 space_info->chunk_alloc = 1;
3882 spin_unlock(&space_info->lock);
3884 mutex_lock(&fs_info->chunk_mutex);
3887 * The chunk_mutex is held throughout the entirety of a chunk
3888 * allocation, so once we've acquired the chunk_mutex we know that the
3889 * other guy is done and we need to recheck and see if we should
3892 if (wait_for_alloc) {
3893 mutex_unlock(&fs_info->chunk_mutex);
3898 trans->allocating_chunk = true;
3901 * If we have mixed data/metadata chunks we want to make sure we keep
3902 * allocating mixed chunks instead of individual chunks.
3904 if (btrfs_mixed_space_info(space_info))
3905 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3908 * if we're doing a data chunk, go ahead and make sure that
3909 * we keep a reasonable number of metadata chunks allocated in the
3912 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3913 fs_info->data_chunk_allocations++;
3914 if (!(fs_info->data_chunk_allocations %
3915 fs_info->metadata_ratio))
3916 force_metadata_allocation(fs_info);
3920 * Check if we have enough space in SYSTEM chunk because we may need
3921 * to update devices.
3923 check_system_chunk(trans, extent_root, flags);
3925 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3926 trans->allocating_chunk = false;
3928 spin_lock(&space_info->lock);
3929 if (ret < 0 && ret != -ENOSPC)
3932 space_info->full = 1;
3936 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3938 space_info->chunk_alloc = 0;
3939 spin_unlock(&space_info->lock);
3940 mutex_unlock(&fs_info->chunk_mutex);
3944 static int can_overcommit(struct btrfs_root *root,
3945 struct btrfs_space_info *space_info, u64 bytes,
3946 enum btrfs_reserve_flush_enum flush)
3948 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3949 u64 profile = btrfs_get_alloc_profile(root, 0);
3954 used = space_info->bytes_used + space_info->bytes_reserved +
3955 space_info->bytes_pinned + space_info->bytes_readonly;
3958 * We only want to allow over committing if we have lots of actual space
3959 * free, but if we don't have enough space to handle the global reserve
3960 * space then we could end up having a real enospc problem when trying
3961 * to allocate a chunk or some other such important allocation.
3963 spin_lock(&global_rsv->lock);
3964 space_size = calc_global_rsv_need_space(global_rsv);
3965 spin_unlock(&global_rsv->lock);
3966 if (used + space_size >= space_info->total_bytes)
3969 used += space_info->bytes_may_use;
3971 spin_lock(&root->fs_info->free_chunk_lock);
3972 avail = root->fs_info->free_chunk_space;
3973 spin_unlock(&root->fs_info->free_chunk_lock);
3976 * If we have dup, raid1 or raid10 then only half of the free
3977 * space is actually useable. For raid56, the space info used
3978 * doesn't include the parity drive, so we don't have to
3981 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3982 BTRFS_BLOCK_GROUP_RAID1 |
3983 BTRFS_BLOCK_GROUP_RAID10))
3987 * If we aren't flushing all things, let us overcommit up to
3988 * 1/2th of the space. If we can flush, don't let us overcommit
3989 * too much, let it overcommit up to 1/8 of the space.
3991 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3996 if (used + bytes < space_info->total_bytes + avail)
4001 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4002 unsigned long nr_pages)
4004 struct super_block *sb = root->fs_info->sb;
4006 if (down_read_trylock(&sb->s_umount)) {
4007 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4008 up_read(&sb->s_umount);
4011 * We needn't worry the filesystem going from r/w to r/o though
4012 * we don't acquire ->s_umount mutex, because the filesystem
4013 * should guarantee the delalloc inodes list be empty after
4014 * the filesystem is readonly(all dirty pages are written to
4017 btrfs_start_delalloc_roots(root->fs_info, 0);
4018 if (!current->journal_info)
4019 btrfs_wait_ordered_roots(root->fs_info, -1);
4023 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4028 bytes = btrfs_calc_trans_metadata_size(root, 1);
4029 nr = (int)div64_u64(to_reclaim, bytes);
4035 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4038 * shrink metadata reservation for delalloc
4040 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4043 struct btrfs_block_rsv *block_rsv;
4044 struct btrfs_space_info *space_info;
4045 struct btrfs_trans_handle *trans;
4049 unsigned long nr_pages;
4052 enum btrfs_reserve_flush_enum flush;
4054 /* Calc the number of the pages we need flush for space reservation */
4055 items = calc_reclaim_items_nr(root, to_reclaim);
4056 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4058 trans = (struct btrfs_trans_handle *)current->journal_info;
4059 block_rsv = &root->fs_info->delalloc_block_rsv;
4060 space_info = block_rsv->space_info;
4062 delalloc_bytes = percpu_counter_sum_positive(
4063 &root->fs_info->delalloc_bytes);
4064 if (delalloc_bytes == 0) {
4068 btrfs_wait_ordered_roots(root->fs_info, items);
4073 while (delalloc_bytes && loops < 3) {
4074 max_reclaim = min(delalloc_bytes, to_reclaim);
4075 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4076 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4078 * We need to wait for the async pages to actually start before
4081 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4085 if (max_reclaim <= nr_pages)
4088 max_reclaim -= nr_pages;
4090 wait_event(root->fs_info->async_submit_wait,
4091 atomic_read(&root->fs_info->async_delalloc_pages) <=
4095 flush = BTRFS_RESERVE_FLUSH_ALL;
4097 flush = BTRFS_RESERVE_NO_FLUSH;
4098 spin_lock(&space_info->lock);
4099 if (can_overcommit(root, space_info, orig, flush)) {
4100 spin_unlock(&space_info->lock);
4103 spin_unlock(&space_info->lock);
4106 if (wait_ordered && !trans) {
4107 btrfs_wait_ordered_roots(root->fs_info, items);
4109 time_left = schedule_timeout_killable(1);
4113 delalloc_bytes = percpu_counter_sum_positive(
4114 &root->fs_info->delalloc_bytes);
4119 * maybe_commit_transaction - possibly commit the transaction if its ok to
4120 * @root - the root we're allocating for
4121 * @bytes - the number of bytes we want to reserve
4122 * @force - force the commit
4124 * This will check to make sure that committing the transaction will actually
4125 * get us somewhere and then commit the transaction if it does. Otherwise it
4126 * will return -ENOSPC.
4128 static int may_commit_transaction(struct btrfs_root *root,
4129 struct btrfs_space_info *space_info,
4130 u64 bytes, int force)
4132 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4133 struct btrfs_trans_handle *trans;
4135 trans = (struct btrfs_trans_handle *)current->journal_info;
4142 /* See if there is enough pinned space to make this reservation */
4143 spin_lock(&space_info->lock);
4144 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4146 spin_unlock(&space_info->lock);
4149 spin_unlock(&space_info->lock);
4152 * See if there is some space in the delayed insertion reservation for
4155 if (space_info != delayed_rsv->space_info)
4158 spin_lock(&space_info->lock);
4159 spin_lock(&delayed_rsv->lock);
4160 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4161 bytes - delayed_rsv->size) >= 0) {
4162 spin_unlock(&delayed_rsv->lock);
4163 spin_unlock(&space_info->lock);
4166 spin_unlock(&delayed_rsv->lock);
4167 spin_unlock(&space_info->lock);
4170 trans = btrfs_join_transaction(root);
4174 return btrfs_commit_transaction(trans, root);
4178 FLUSH_DELAYED_ITEMS_NR = 1,
4179 FLUSH_DELAYED_ITEMS = 2,
4181 FLUSH_DELALLOC_WAIT = 4,
4186 static int flush_space(struct btrfs_root *root,
4187 struct btrfs_space_info *space_info, u64 num_bytes,
4188 u64 orig_bytes, int state)
4190 struct btrfs_trans_handle *trans;
4195 case FLUSH_DELAYED_ITEMS_NR:
4196 case FLUSH_DELAYED_ITEMS:
4197 if (state == FLUSH_DELAYED_ITEMS_NR)
4198 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4202 trans = btrfs_join_transaction(root);
4203 if (IS_ERR(trans)) {
4204 ret = PTR_ERR(trans);
4207 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4208 btrfs_end_transaction(trans, root);
4210 case FLUSH_DELALLOC:
4211 case FLUSH_DELALLOC_WAIT:
4212 shrink_delalloc(root, num_bytes, orig_bytes,
4213 state == FLUSH_DELALLOC_WAIT);
4216 trans = btrfs_join_transaction(root);
4217 if (IS_ERR(trans)) {
4218 ret = PTR_ERR(trans);
4221 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4222 btrfs_get_alloc_profile(root, 0),
4223 CHUNK_ALLOC_NO_FORCE);
4224 btrfs_end_transaction(trans, root);
4229 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4239 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4240 * @root - the root we're allocating for
4241 * @block_rsv - the block_rsv we're allocating for
4242 * @orig_bytes - the number of bytes we want
4243 * @flush - whether or not we can flush to make our reservation
4245 * This will reserve orgi_bytes number of bytes from the space info associated
4246 * with the block_rsv. If there is not enough space it will make an attempt to
4247 * flush out space to make room. It will do this by flushing delalloc if
4248 * possible or committing the transaction. If flush is 0 then no attempts to
4249 * regain reservations will be made and this will fail if there is not enough
4252 static int reserve_metadata_bytes(struct btrfs_root *root,
4253 struct btrfs_block_rsv *block_rsv,
4255 enum btrfs_reserve_flush_enum flush)
4257 struct btrfs_space_info *space_info = block_rsv->space_info;
4259 u64 num_bytes = orig_bytes;
4260 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4262 bool flushing = false;
4266 spin_lock(&space_info->lock);
4268 * We only want to wait if somebody other than us is flushing and we
4269 * are actually allowed to flush all things.
4271 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4272 space_info->flush) {
4273 spin_unlock(&space_info->lock);
4275 * If we have a trans handle we can't wait because the flusher
4276 * may have to commit the transaction, which would mean we would
4277 * deadlock since we are waiting for the flusher to finish, but
4278 * hold the current transaction open.
4280 if (current->journal_info)
4282 ret = wait_event_killable(space_info->wait, !space_info->flush);
4283 /* Must have been killed, return */
4287 spin_lock(&space_info->lock);
4291 used = space_info->bytes_used + space_info->bytes_reserved +
4292 space_info->bytes_pinned + space_info->bytes_readonly +
4293 space_info->bytes_may_use;
4296 * The idea here is that we've not already over-reserved the block group
4297 * then we can go ahead and save our reservation first and then start
4298 * flushing if we need to. Otherwise if we've already overcommitted
4299 * lets start flushing stuff first and then come back and try to make
4302 if (used <= space_info->total_bytes) {
4303 if (used + orig_bytes <= space_info->total_bytes) {
4304 space_info->bytes_may_use += orig_bytes;
4305 trace_btrfs_space_reservation(root->fs_info,
4306 "space_info", space_info->flags, orig_bytes, 1);
4310 * Ok set num_bytes to orig_bytes since we aren't
4311 * overocmmitted, this way we only try and reclaim what
4314 num_bytes = orig_bytes;
4318 * Ok we're over committed, set num_bytes to the overcommitted
4319 * amount plus the amount of bytes that we need for this
4322 num_bytes = used - space_info->total_bytes +
4326 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4327 space_info->bytes_may_use += orig_bytes;
4328 trace_btrfs_space_reservation(root->fs_info, "space_info",
4329 space_info->flags, orig_bytes,
4335 * Couldn't make our reservation, save our place so while we're trying
4336 * to reclaim space we can actually use it instead of somebody else
4337 * stealing it from us.
4339 * We make the other tasks wait for the flush only when we can flush
4342 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4344 space_info->flush = 1;
4347 spin_unlock(&space_info->lock);
4349 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4352 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4357 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4358 * would happen. So skip delalloc flush.
4360 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4361 (flush_state == FLUSH_DELALLOC ||
4362 flush_state == FLUSH_DELALLOC_WAIT))
4363 flush_state = ALLOC_CHUNK;
4367 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4368 flush_state < COMMIT_TRANS)
4370 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4371 flush_state <= COMMIT_TRANS)
4375 if (ret == -ENOSPC &&
4376 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4377 struct btrfs_block_rsv *global_rsv =
4378 &root->fs_info->global_block_rsv;
4380 if (block_rsv != global_rsv &&
4381 !block_rsv_use_bytes(global_rsv, orig_bytes))
4385 trace_btrfs_space_reservation(root->fs_info,
4386 "space_info:enospc",
4387 space_info->flags, orig_bytes, 1);
4389 spin_lock(&space_info->lock);
4390 space_info->flush = 0;
4391 wake_up_all(&space_info->wait);
4392 spin_unlock(&space_info->lock);
4397 static struct btrfs_block_rsv *get_block_rsv(
4398 const struct btrfs_trans_handle *trans,
4399 const struct btrfs_root *root)
4401 struct btrfs_block_rsv *block_rsv = NULL;
4404 block_rsv = trans->block_rsv;
4406 if (root == root->fs_info->csum_root && trans->adding_csums)
4407 block_rsv = trans->block_rsv;
4409 if (root == root->fs_info->uuid_root)
4410 block_rsv = trans->block_rsv;
4413 block_rsv = root->block_rsv;
4416 block_rsv = &root->fs_info->empty_block_rsv;
4421 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4425 spin_lock(&block_rsv->lock);
4426 if (block_rsv->reserved >= num_bytes) {
4427 block_rsv->reserved -= num_bytes;
4428 if (block_rsv->reserved < block_rsv->size)
4429 block_rsv->full = 0;
4432 spin_unlock(&block_rsv->lock);
4436 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4437 u64 num_bytes, int update_size)
4439 spin_lock(&block_rsv->lock);
4440 block_rsv->reserved += num_bytes;
4442 block_rsv->size += num_bytes;
4443 else if (block_rsv->reserved >= block_rsv->size)
4444 block_rsv->full = 1;
4445 spin_unlock(&block_rsv->lock);
4448 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4449 struct btrfs_block_rsv *dest, u64 num_bytes,
4452 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4455 if (global_rsv->space_info != dest->space_info)
4458 spin_lock(&global_rsv->lock);
4459 min_bytes = div_factor(global_rsv->size, min_factor);
4460 if (global_rsv->reserved < min_bytes + num_bytes) {
4461 spin_unlock(&global_rsv->lock);
4464 global_rsv->reserved -= num_bytes;
4465 if (global_rsv->reserved < global_rsv->size)
4466 global_rsv->full = 0;
4467 spin_unlock(&global_rsv->lock);
4469 block_rsv_add_bytes(dest, num_bytes, 1);
4473 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4474 struct btrfs_block_rsv *block_rsv,
4475 struct btrfs_block_rsv *dest, u64 num_bytes)
4477 struct btrfs_space_info *space_info = block_rsv->space_info;
4479 spin_lock(&block_rsv->lock);
4480 if (num_bytes == (u64)-1)
4481 num_bytes = block_rsv->size;
4482 block_rsv->size -= num_bytes;
4483 if (block_rsv->reserved >= block_rsv->size) {
4484 num_bytes = block_rsv->reserved - block_rsv->size;
4485 block_rsv->reserved = block_rsv->size;
4486 block_rsv->full = 1;
4490 spin_unlock(&block_rsv->lock);
4492 if (num_bytes > 0) {
4494 spin_lock(&dest->lock);
4498 bytes_to_add = dest->size - dest->reserved;
4499 bytes_to_add = min(num_bytes, bytes_to_add);
4500 dest->reserved += bytes_to_add;
4501 if (dest->reserved >= dest->size)
4503 num_bytes -= bytes_to_add;
4505 spin_unlock(&dest->lock);
4508 spin_lock(&space_info->lock);
4509 space_info->bytes_may_use -= num_bytes;
4510 trace_btrfs_space_reservation(fs_info, "space_info",
4511 space_info->flags, num_bytes, 0);
4512 spin_unlock(&space_info->lock);
4517 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4518 struct btrfs_block_rsv *dst, u64 num_bytes)
4522 ret = block_rsv_use_bytes(src, num_bytes);
4526 block_rsv_add_bytes(dst, num_bytes, 1);
4530 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4532 memset(rsv, 0, sizeof(*rsv));
4533 spin_lock_init(&rsv->lock);
4537 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4538 unsigned short type)
4540 struct btrfs_block_rsv *block_rsv;
4541 struct btrfs_fs_info *fs_info = root->fs_info;
4543 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4547 btrfs_init_block_rsv(block_rsv, type);
4548 block_rsv->space_info = __find_space_info(fs_info,
4549 BTRFS_BLOCK_GROUP_METADATA);
4553 void btrfs_free_block_rsv(struct btrfs_root *root,
4554 struct btrfs_block_rsv *rsv)
4558 btrfs_block_rsv_release(root, rsv, (u64)-1);
4562 int btrfs_block_rsv_add(struct btrfs_root *root,
4563 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4564 enum btrfs_reserve_flush_enum flush)
4571 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4573 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4580 int btrfs_block_rsv_check(struct btrfs_root *root,
4581 struct btrfs_block_rsv *block_rsv, int min_factor)
4589 spin_lock(&block_rsv->lock);
4590 num_bytes = div_factor(block_rsv->size, min_factor);
4591 if (block_rsv->reserved >= num_bytes)
4593 spin_unlock(&block_rsv->lock);
4598 int btrfs_block_rsv_refill(struct btrfs_root *root,
4599 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4600 enum btrfs_reserve_flush_enum flush)
4608 spin_lock(&block_rsv->lock);
4609 num_bytes = min_reserved;
4610 if (block_rsv->reserved >= num_bytes)
4613 num_bytes -= block_rsv->reserved;
4614 spin_unlock(&block_rsv->lock);
4619 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4621 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4628 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4629 struct btrfs_block_rsv *dst_rsv,
4632 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4635 void btrfs_block_rsv_release(struct btrfs_root *root,
4636 struct btrfs_block_rsv *block_rsv,
4639 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4640 if (global_rsv->full || global_rsv == block_rsv ||
4641 block_rsv->space_info != global_rsv->space_info)
4643 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4648 * helper to calculate size of global block reservation.
4649 * the desired value is sum of space used by extent tree,
4650 * checksum tree and root tree
4652 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4654 struct btrfs_space_info *sinfo;
4658 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4660 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4661 spin_lock(&sinfo->lock);
4662 data_used = sinfo->bytes_used;
4663 spin_unlock(&sinfo->lock);
4665 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4666 spin_lock(&sinfo->lock);
4667 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4669 meta_used = sinfo->bytes_used;
4670 spin_unlock(&sinfo->lock);
4672 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4674 num_bytes += div64_u64(data_used + meta_used, 50);
4676 if (num_bytes * 3 > meta_used)
4677 num_bytes = div64_u64(meta_used, 3);
4679 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4682 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4684 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4685 struct btrfs_space_info *sinfo = block_rsv->space_info;
4688 num_bytes = calc_global_metadata_size(fs_info);
4690 spin_lock(&sinfo->lock);
4691 spin_lock(&block_rsv->lock);
4693 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4695 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4696 sinfo->bytes_reserved + sinfo->bytes_readonly +
4697 sinfo->bytes_may_use;
4699 if (sinfo->total_bytes > num_bytes) {
4700 num_bytes = sinfo->total_bytes - num_bytes;
4701 block_rsv->reserved += num_bytes;
4702 sinfo->bytes_may_use += num_bytes;
4703 trace_btrfs_space_reservation(fs_info, "space_info",
4704 sinfo->flags, num_bytes, 1);
4707 if (block_rsv->reserved >= block_rsv->size) {
4708 num_bytes = block_rsv->reserved - block_rsv->size;
4709 sinfo->bytes_may_use -= num_bytes;
4710 trace_btrfs_space_reservation(fs_info, "space_info",
4711 sinfo->flags, num_bytes, 0);
4712 block_rsv->reserved = block_rsv->size;
4713 block_rsv->full = 1;
4716 spin_unlock(&block_rsv->lock);
4717 spin_unlock(&sinfo->lock);
4720 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4722 struct btrfs_space_info *space_info;
4724 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4725 fs_info->chunk_block_rsv.space_info = space_info;
4727 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4728 fs_info->global_block_rsv.space_info = space_info;
4729 fs_info->delalloc_block_rsv.space_info = space_info;
4730 fs_info->trans_block_rsv.space_info = space_info;
4731 fs_info->empty_block_rsv.space_info = space_info;
4732 fs_info->delayed_block_rsv.space_info = space_info;
4734 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4735 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4736 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4737 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4738 if (fs_info->quota_root)
4739 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4740 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4742 update_global_block_rsv(fs_info);
4745 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4747 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4749 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4750 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4751 WARN_ON(fs_info->trans_block_rsv.size > 0);
4752 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4753 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4754 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4755 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4756 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4759 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4760 struct btrfs_root *root)
4762 if (!trans->block_rsv)
4765 if (!trans->bytes_reserved)
4768 trace_btrfs_space_reservation(root->fs_info, "transaction",
4769 trans->transid, trans->bytes_reserved, 0);
4770 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4771 trans->bytes_reserved = 0;
4774 /* Can only return 0 or -ENOSPC */
4775 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4776 struct inode *inode)
4778 struct btrfs_root *root = BTRFS_I(inode)->root;
4779 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4780 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4783 * We need to hold space in order to delete our orphan item once we've
4784 * added it, so this takes the reservation so we can release it later
4785 * when we are truly done with the orphan item.
4787 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4788 trace_btrfs_space_reservation(root->fs_info, "orphan",
4789 btrfs_ino(inode), num_bytes, 1);
4790 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4793 void btrfs_orphan_release_metadata(struct inode *inode)
4795 struct btrfs_root *root = BTRFS_I(inode)->root;
4796 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4797 trace_btrfs_space_reservation(root->fs_info, "orphan",
4798 btrfs_ino(inode), num_bytes, 0);
4799 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4803 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4804 * root: the root of the parent directory
4805 * rsv: block reservation
4806 * items: the number of items that we need do reservation
4807 * qgroup_reserved: used to return the reserved size in qgroup
4809 * This function is used to reserve the space for snapshot/subvolume
4810 * creation and deletion. Those operations are different with the
4811 * common file/directory operations, they change two fs/file trees
4812 * and root tree, the number of items that the qgroup reserves is
4813 * different with the free space reservation. So we can not use
4814 * the space reseravtion mechanism in start_transaction().
4816 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4817 struct btrfs_block_rsv *rsv,
4819 u64 *qgroup_reserved,
4820 bool use_global_rsv)
4824 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4826 if (root->fs_info->quota_enabled) {
4827 /* One for parent inode, two for dir entries */
4828 num_bytes = 3 * root->leafsize;
4829 ret = btrfs_qgroup_reserve(root, num_bytes);
4836 *qgroup_reserved = num_bytes;
4838 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4839 rsv->space_info = __find_space_info(root->fs_info,
4840 BTRFS_BLOCK_GROUP_METADATA);
4841 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4842 BTRFS_RESERVE_FLUSH_ALL);
4844 if (ret == -ENOSPC && use_global_rsv)
4845 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4848 if (*qgroup_reserved)
4849 btrfs_qgroup_free(root, *qgroup_reserved);
4855 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4856 struct btrfs_block_rsv *rsv,
4857 u64 qgroup_reserved)
4859 btrfs_block_rsv_release(root, rsv, (u64)-1);
4860 if (qgroup_reserved)
4861 btrfs_qgroup_free(root, qgroup_reserved);
4865 * drop_outstanding_extent - drop an outstanding extent
4866 * @inode: the inode we're dropping the extent for
4868 * This is called when we are freeing up an outstanding extent, either called
4869 * after an error or after an extent is written. This will return the number of
4870 * reserved extents that need to be freed. This must be called with
4871 * BTRFS_I(inode)->lock held.
4873 static unsigned drop_outstanding_extent(struct inode *inode)
4875 unsigned drop_inode_space = 0;
4876 unsigned dropped_extents = 0;
4878 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4879 BTRFS_I(inode)->outstanding_extents--;
4881 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4882 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4883 &BTRFS_I(inode)->runtime_flags))
4884 drop_inode_space = 1;
4887 * If we have more or the same amount of outsanding extents than we have
4888 * reserved then we need to leave the reserved extents count alone.
4890 if (BTRFS_I(inode)->outstanding_extents >=
4891 BTRFS_I(inode)->reserved_extents)
4892 return drop_inode_space;
4894 dropped_extents = BTRFS_I(inode)->reserved_extents -
4895 BTRFS_I(inode)->outstanding_extents;
4896 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4897 return dropped_extents + drop_inode_space;
4901 * calc_csum_metadata_size - return the amount of metada space that must be
4902 * reserved/free'd for the given bytes.
4903 * @inode: the inode we're manipulating
4904 * @num_bytes: the number of bytes in question
4905 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4907 * This adjusts the number of csum_bytes in the inode and then returns the
4908 * correct amount of metadata that must either be reserved or freed. We
4909 * calculate how many checksums we can fit into one leaf and then divide the
4910 * number of bytes that will need to be checksumed by this value to figure out
4911 * how many checksums will be required. If we are adding bytes then the number
4912 * may go up and we will return the number of additional bytes that must be
4913 * reserved. If it is going down we will return the number of bytes that must
4916 * This must be called with BTRFS_I(inode)->lock held.
4918 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4921 struct btrfs_root *root = BTRFS_I(inode)->root;
4923 int num_csums_per_leaf;
4927 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4928 BTRFS_I(inode)->csum_bytes == 0)
4931 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4933 BTRFS_I(inode)->csum_bytes += num_bytes;
4935 BTRFS_I(inode)->csum_bytes -= num_bytes;
4936 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4937 num_csums_per_leaf = (int)div64_u64(csum_size,
4938 sizeof(struct btrfs_csum_item) +
4939 sizeof(struct btrfs_disk_key));
4940 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4941 num_csums = num_csums + num_csums_per_leaf - 1;
4942 num_csums = num_csums / num_csums_per_leaf;
4944 old_csums = old_csums + num_csums_per_leaf - 1;
4945 old_csums = old_csums / num_csums_per_leaf;
4947 /* No change, no need to reserve more */
4948 if (old_csums == num_csums)
4952 return btrfs_calc_trans_metadata_size(root,
4953 num_csums - old_csums);
4955 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4958 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4960 struct btrfs_root *root = BTRFS_I(inode)->root;
4961 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4964 unsigned nr_extents = 0;
4965 int extra_reserve = 0;
4966 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4968 bool delalloc_lock = true;
4972 /* If we are a free space inode we need to not flush since we will be in
4973 * the middle of a transaction commit. We also don't need the delalloc
4974 * mutex since we won't race with anybody. We need this mostly to make
4975 * lockdep shut its filthy mouth.
4977 if (btrfs_is_free_space_inode(inode)) {
4978 flush = BTRFS_RESERVE_NO_FLUSH;
4979 delalloc_lock = false;
4982 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4983 btrfs_transaction_in_commit(root->fs_info))
4984 schedule_timeout(1);
4987 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4989 num_bytes = ALIGN(num_bytes, root->sectorsize);
4991 spin_lock(&BTRFS_I(inode)->lock);
4992 BTRFS_I(inode)->outstanding_extents++;
4994 if (BTRFS_I(inode)->outstanding_extents >
4995 BTRFS_I(inode)->reserved_extents)
4996 nr_extents = BTRFS_I(inode)->outstanding_extents -
4997 BTRFS_I(inode)->reserved_extents;
5000 * Add an item to reserve for updating the inode when we complete the
5003 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5004 &BTRFS_I(inode)->runtime_flags)) {
5009 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5010 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5011 csum_bytes = BTRFS_I(inode)->csum_bytes;
5012 spin_unlock(&BTRFS_I(inode)->lock);
5014 if (root->fs_info->quota_enabled) {
5015 ret = btrfs_qgroup_reserve(root, num_bytes +
5016 nr_extents * root->leafsize);
5021 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5022 if (unlikely(ret)) {
5023 if (root->fs_info->quota_enabled)
5024 btrfs_qgroup_free(root, num_bytes +
5025 nr_extents * root->leafsize);
5029 spin_lock(&BTRFS_I(inode)->lock);
5030 if (extra_reserve) {
5031 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5032 &BTRFS_I(inode)->runtime_flags);
5035 BTRFS_I(inode)->reserved_extents += nr_extents;
5036 spin_unlock(&BTRFS_I(inode)->lock);
5039 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5042 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5043 btrfs_ino(inode), to_reserve, 1);
5044 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5049 spin_lock(&BTRFS_I(inode)->lock);
5050 dropped = drop_outstanding_extent(inode);
5052 * If the inodes csum_bytes is the same as the original
5053 * csum_bytes then we know we haven't raced with any free()ers
5054 * so we can just reduce our inodes csum bytes and carry on.
5056 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5057 calc_csum_metadata_size(inode, num_bytes, 0);
5059 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5063 * This is tricky, but first we need to figure out how much we
5064 * free'd from any free-ers that occured during this
5065 * reservation, so we reset ->csum_bytes to the csum_bytes
5066 * before we dropped our lock, and then call the free for the
5067 * number of bytes that were freed while we were trying our
5070 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5071 BTRFS_I(inode)->csum_bytes = csum_bytes;
5072 to_free = calc_csum_metadata_size(inode, bytes, 0);
5076 * Now we need to see how much we would have freed had we not
5077 * been making this reservation and our ->csum_bytes were not
5078 * artificially inflated.
5080 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5081 bytes = csum_bytes - orig_csum_bytes;
5082 bytes = calc_csum_metadata_size(inode, bytes, 0);
5085 * Now reset ->csum_bytes to what it should be. If bytes is
5086 * more than to_free then we would have free'd more space had we
5087 * not had an artificially high ->csum_bytes, so we need to free
5088 * the remainder. If bytes is the same or less then we don't
5089 * need to do anything, the other free-ers did the correct
5092 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5093 if (bytes > to_free)
5094 to_free = bytes - to_free;
5098 spin_unlock(&BTRFS_I(inode)->lock);
5100 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5103 btrfs_block_rsv_release(root, block_rsv, to_free);
5104 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5105 btrfs_ino(inode), to_free, 0);
5108 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5113 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5114 * @inode: the inode to release the reservation for
5115 * @num_bytes: the number of bytes we're releasing
5117 * This will release the metadata reservation for an inode. This can be called
5118 * once we complete IO for a given set of bytes to release their metadata
5121 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5123 struct btrfs_root *root = BTRFS_I(inode)->root;
5127 num_bytes = ALIGN(num_bytes, root->sectorsize);
5128 spin_lock(&BTRFS_I(inode)->lock);
5129 dropped = drop_outstanding_extent(inode);
5132 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5133 spin_unlock(&BTRFS_I(inode)->lock);
5135 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5137 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5138 btrfs_ino(inode), to_free, 0);
5139 if (root->fs_info->quota_enabled) {
5140 btrfs_qgroup_free(root, num_bytes +
5141 dropped * root->leafsize);
5144 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5149 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5150 * @inode: inode we're writing to
5151 * @num_bytes: the number of bytes we want to allocate
5153 * This will do the following things
5155 * o reserve space in the data space info for num_bytes
5156 * o reserve space in the metadata space info based on number of outstanding
5157 * extents and how much csums will be needed
5158 * o add to the inodes ->delalloc_bytes
5159 * o add it to the fs_info's delalloc inodes list.
5161 * This will return 0 for success and -ENOSPC if there is no space left.
5163 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5167 ret = btrfs_check_data_free_space(inode, num_bytes);
5171 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5173 btrfs_free_reserved_data_space(inode, num_bytes);
5181 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5182 * @inode: inode we're releasing space for
5183 * @num_bytes: the number of bytes we want to free up
5185 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5186 * called in the case that we don't need the metadata AND data reservations
5187 * anymore. So if there is an error or we insert an inline extent.
5189 * This function will release the metadata space that was not used and will
5190 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5191 * list if there are no delalloc bytes left.
5193 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5195 btrfs_delalloc_release_metadata(inode, num_bytes);
5196 btrfs_free_reserved_data_space(inode, num_bytes);
5199 static int update_block_group(struct btrfs_root *root,
5200 u64 bytenr, u64 num_bytes, int alloc)
5202 struct btrfs_block_group_cache *cache = NULL;
5203 struct btrfs_fs_info *info = root->fs_info;
5204 u64 total = num_bytes;
5209 /* block accounting for super block */
5210 spin_lock(&info->delalloc_root_lock);
5211 old_val = btrfs_super_bytes_used(info->super_copy);
5213 old_val += num_bytes;
5215 old_val -= num_bytes;
5216 btrfs_set_super_bytes_used(info->super_copy, old_val);
5217 spin_unlock(&info->delalloc_root_lock);
5220 cache = btrfs_lookup_block_group(info, bytenr);
5223 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5224 BTRFS_BLOCK_GROUP_RAID1 |
5225 BTRFS_BLOCK_GROUP_RAID10))
5230 * If this block group has free space cache written out, we
5231 * need to make sure to load it if we are removing space. This
5232 * is because we need the unpinning stage to actually add the
5233 * space back to the block group, otherwise we will leak space.
5235 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5236 cache_block_group(cache, 1);
5238 byte_in_group = bytenr - cache->key.objectid;
5239 WARN_ON(byte_in_group > cache->key.offset);
5241 spin_lock(&cache->space_info->lock);
5242 spin_lock(&cache->lock);
5244 if (btrfs_test_opt(root, SPACE_CACHE) &&
5245 cache->disk_cache_state < BTRFS_DC_CLEAR)
5246 cache->disk_cache_state = BTRFS_DC_CLEAR;
5249 old_val = btrfs_block_group_used(&cache->item);
5250 num_bytes = min(total, cache->key.offset - byte_in_group);
5252 old_val += num_bytes;
5253 btrfs_set_block_group_used(&cache->item, old_val);
5254 cache->reserved -= num_bytes;
5255 cache->space_info->bytes_reserved -= num_bytes;
5256 cache->space_info->bytes_used += num_bytes;
5257 cache->space_info->disk_used += num_bytes * factor;
5258 spin_unlock(&cache->lock);
5259 spin_unlock(&cache->space_info->lock);
5261 old_val -= num_bytes;
5262 btrfs_set_block_group_used(&cache->item, old_val);
5263 cache->pinned += num_bytes;
5264 cache->space_info->bytes_pinned += num_bytes;
5265 cache->space_info->bytes_used -= num_bytes;
5266 cache->space_info->disk_used -= num_bytes * factor;
5267 spin_unlock(&cache->lock);
5268 spin_unlock(&cache->space_info->lock);
5270 set_extent_dirty(info->pinned_extents,
5271 bytenr, bytenr + num_bytes - 1,
5272 GFP_NOFS | __GFP_NOFAIL);
5274 btrfs_put_block_group(cache);
5276 bytenr += num_bytes;
5281 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5283 struct btrfs_block_group_cache *cache;
5286 spin_lock(&root->fs_info->block_group_cache_lock);
5287 bytenr = root->fs_info->first_logical_byte;
5288 spin_unlock(&root->fs_info->block_group_cache_lock);
5290 if (bytenr < (u64)-1)
5293 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5297 bytenr = cache->key.objectid;
5298 btrfs_put_block_group(cache);
5303 static int pin_down_extent(struct btrfs_root *root,
5304 struct btrfs_block_group_cache *cache,
5305 u64 bytenr, u64 num_bytes, int reserved)
5307 spin_lock(&cache->space_info->lock);
5308 spin_lock(&cache->lock);
5309 cache->pinned += num_bytes;
5310 cache->space_info->bytes_pinned += num_bytes;
5312 cache->reserved -= num_bytes;
5313 cache->space_info->bytes_reserved -= num_bytes;
5315 spin_unlock(&cache->lock);
5316 spin_unlock(&cache->space_info->lock);
5318 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5319 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5321 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5326 * this function must be called within transaction
5328 int btrfs_pin_extent(struct btrfs_root *root,
5329 u64 bytenr, u64 num_bytes, int reserved)
5331 struct btrfs_block_group_cache *cache;
5333 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5334 BUG_ON(!cache); /* Logic error */
5336 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5338 btrfs_put_block_group(cache);
5343 * this function must be called within transaction
5345 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5346 u64 bytenr, u64 num_bytes)
5348 struct btrfs_block_group_cache *cache;
5351 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5356 * pull in the free space cache (if any) so that our pin
5357 * removes the free space from the cache. We have load_only set
5358 * to one because the slow code to read in the free extents does check
5359 * the pinned extents.
5361 cache_block_group(cache, 1);
5363 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5365 /* remove us from the free space cache (if we're there at all) */
5366 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5367 btrfs_put_block_group(cache);
5371 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5374 struct btrfs_block_group_cache *block_group;
5375 struct btrfs_caching_control *caching_ctl;
5377 block_group = btrfs_lookup_block_group(root->fs_info, start);
5381 cache_block_group(block_group, 0);
5382 caching_ctl = get_caching_control(block_group);
5386 BUG_ON(!block_group_cache_done(block_group));
5387 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5389 mutex_lock(&caching_ctl->mutex);
5391 if (start >= caching_ctl->progress) {
5392 ret = add_excluded_extent(root, start, num_bytes);
5393 } else if (start + num_bytes <= caching_ctl->progress) {
5394 ret = btrfs_remove_free_space(block_group,
5397 num_bytes = caching_ctl->progress - start;
5398 ret = btrfs_remove_free_space(block_group,
5403 num_bytes = (start + num_bytes) -
5404 caching_ctl->progress;
5405 start = caching_ctl->progress;
5406 ret = add_excluded_extent(root, start, num_bytes);
5409 mutex_unlock(&caching_ctl->mutex);
5410 put_caching_control(caching_ctl);
5412 btrfs_put_block_group(block_group);
5416 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5417 struct extent_buffer *eb)
5419 struct btrfs_file_extent_item *item;
5420 struct btrfs_key key;
5424 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5427 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5428 btrfs_item_key_to_cpu(eb, &key, i);
5429 if (key.type != BTRFS_EXTENT_DATA_KEY)
5431 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5432 found_type = btrfs_file_extent_type(eb, item);
5433 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5435 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5437 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5438 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5439 __exclude_logged_extent(log, key.objectid, key.offset);
5446 * btrfs_update_reserved_bytes - update the block_group and space info counters
5447 * @cache: The cache we are manipulating
5448 * @num_bytes: The number of bytes in question
5449 * @reserve: One of the reservation enums
5451 * This is called by the allocator when it reserves space, or by somebody who is
5452 * freeing space that was never actually used on disk. For example if you
5453 * reserve some space for a new leaf in transaction A and before transaction A
5454 * commits you free that leaf, you call this with reserve set to 0 in order to
5455 * clear the reservation.
5457 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5458 * ENOSPC accounting. For data we handle the reservation through clearing the
5459 * delalloc bits in the io_tree. We have to do this since we could end up
5460 * allocating less disk space for the amount of data we have reserved in the
5461 * case of compression.
5463 * If this is a reservation and the block group has become read only we cannot
5464 * make the reservation and return -EAGAIN, otherwise this function always
5467 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5468 u64 num_bytes, int reserve)
5470 struct btrfs_space_info *space_info = cache->space_info;
5473 spin_lock(&space_info->lock);
5474 spin_lock(&cache->lock);
5475 if (reserve != RESERVE_FREE) {
5479 cache->reserved += num_bytes;
5480 space_info->bytes_reserved += num_bytes;
5481 if (reserve == RESERVE_ALLOC) {
5482 trace_btrfs_space_reservation(cache->fs_info,
5483 "space_info", space_info->flags,
5485 space_info->bytes_may_use -= num_bytes;
5490 space_info->bytes_readonly += num_bytes;
5491 cache->reserved -= num_bytes;
5492 space_info->bytes_reserved -= num_bytes;
5494 spin_unlock(&cache->lock);
5495 spin_unlock(&space_info->lock);
5499 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5500 struct btrfs_root *root)
5502 struct btrfs_fs_info *fs_info = root->fs_info;
5503 struct btrfs_caching_control *next;
5504 struct btrfs_caching_control *caching_ctl;
5505 struct btrfs_block_group_cache *cache;
5506 struct btrfs_space_info *space_info;
5508 down_write(&fs_info->extent_commit_sem);
5510 list_for_each_entry_safe(caching_ctl, next,
5511 &fs_info->caching_block_groups, list) {
5512 cache = caching_ctl->block_group;
5513 if (block_group_cache_done(cache)) {
5514 cache->last_byte_to_unpin = (u64)-1;
5515 list_del_init(&caching_ctl->list);
5516 put_caching_control(caching_ctl);
5518 cache->last_byte_to_unpin = caching_ctl->progress;
5522 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5523 fs_info->pinned_extents = &fs_info->freed_extents[1];
5525 fs_info->pinned_extents = &fs_info->freed_extents[0];
5527 up_write(&fs_info->extent_commit_sem);
5529 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5530 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5532 update_global_block_rsv(fs_info);
5535 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5537 struct btrfs_fs_info *fs_info = root->fs_info;
5538 struct btrfs_block_group_cache *cache = NULL;
5539 struct btrfs_space_info *space_info;
5540 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5544 while (start <= end) {
5547 start >= cache->key.objectid + cache->key.offset) {
5549 btrfs_put_block_group(cache);
5550 cache = btrfs_lookup_block_group(fs_info, start);
5551 BUG_ON(!cache); /* Logic error */
5554 len = cache->key.objectid + cache->key.offset - start;
5555 len = min(len, end + 1 - start);
5557 if (start < cache->last_byte_to_unpin) {
5558 len = min(len, cache->last_byte_to_unpin - start);
5559 btrfs_add_free_space(cache, start, len);
5563 space_info = cache->space_info;
5565 spin_lock(&space_info->lock);
5566 spin_lock(&cache->lock);
5567 cache->pinned -= len;
5568 space_info->bytes_pinned -= len;
5570 space_info->bytes_readonly += len;
5573 spin_unlock(&cache->lock);
5574 if (!readonly && global_rsv->space_info == space_info) {
5575 spin_lock(&global_rsv->lock);
5576 if (!global_rsv->full) {
5577 len = min(len, global_rsv->size -
5578 global_rsv->reserved);
5579 global_rsv->reserved += len;
5580 space_info->bytes_may_use += len;
5581 if (global_rsv->reserved >= global_rsv->size)
5582 global_rsv->full = 1;
5584 spin_unlock(&global_rsv->lock);
5586 spin_unlock(&space_info->lock);
5590 btrfs_put_block_group(cache);
5594 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5595 struct btrfs_root *root)
5597 struct btrfs_fs_info *fs_info = root->fs_info;
5598 struct extent_io_tree *unpin;
5606 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5607 unpin = &fs_info->freed_extents[1];
5609 unpin = &fs_info->freed_extents[0];
5612 ret = find_first_extent_bit(unpin, 0, &start, &end,
5613 EXTENT_DIRTY, NULL);
5617 if (btrfs_test_opt(root, DISCARD))
5618 ret = btrfs_discard_extent(root, start,
5619 end + 1 - start, NULL);
5621 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5622 unpin_extent_range(root, start, end);
5629 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5630 u64 owner, u64 root_objectid)
5632 struct btrfs_space_info *space_info;
5635 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5636 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5637 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5639 flags = BTRFS_BLOCK_GROUP_METADATA;
5641 flags = BTRFS_BLOCK_GROUP_DATA;
5644 space_info = __find_space_info(fs_info, flags);
5645 BUG_ON(!space_info); /* Logic bug */
5646 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5650 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5651 struct btrfs_root *root,
5652 u64 bytenr, u64 num_bytes, u64 parent,
5653 u64 root_objectid, u64 owner_objectid,
5654 u64 owner_offset, int refs_to_drop,
5655 struct btrfs_delayed_extent_op *extent_op)
5657 struct btrfs_key key;
5658 struct btrfs_path *path;
5659 struct btrfs_fs_info *info = root->fs_info;
5660 struct btrfs_root *extent_root = info->extent_root;
5661 struct extent_buffer *leaf;
5662 struct btrfs_extent_item *ei;
5663 struct btrfs_extent_inline_ref *iref;
5666 int extent_slot = 0;
5667 int found_extent = 0;
5671 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5674 path = btrfs_alloc_path();
5679 path->leave_spinning = 1;
5681 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5682 BUG_ON(!is_data && refs_to_drop != 1);
5685 skinny_metadata = 0;
5687 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5688 bytenr, num_bytes, parent,
5689 root_objectid, owner_objectid,
5692 extent_slot = path->slots[0];
5693 while (extent_slot >= 0) {
5694 btrfs_item_key_to_cpu(path->nodes[0], &key,
5696 if (key.objectid != bytenr)
5698 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5699 key.offset == num_bytes) {
5703 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5704 key.offset == owner_objectid) {
5708 if (path->slots[0] - extent_slot > 5)
5712 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5713 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5714 if (found_extent && item_size < sizeof(*ei))
5717 if (!found_extent) {
5719 ret = remove_extent_backref(trans, extent_root, path,
5723 btrfs_abort_transaction(trans, extent_root, ret);
5726 btrfs_release_path(path);
5727 path->leave_spinning = 1;
5729 key.objectid = bytenr;
5730 key.type = BTRFS_EXTENT_ITEM_KEY;
5731 key.offset = num_bytes;
5733 if (!is_data && skinny_metadata) {
5734 key.type = BTRFS_METADATA_ITEM_KEY;
5735 key.offset = owner_objectid;
5738 ret = btrfs_search_slot(trans, extent_root,
5740 if (ret > 0 && skinny_metadata && path->slots[0]) {
5742 * Couldn't find our skinny metadata item,
5743 * see if we have ye olde extent item.
5746 btrfs_item_key_to_cpu(path->nodes[0], &key,
5748 if (key.objectid == bytenr &&
5749 key.type == BTRFS_EXTENT_ITEM_KEY &&
5750 key.offset == num_bytes)
5754 if (ret > 0 && skinny_metadata) {
5755 skinny_metadata = false;
5756 key.type = BTRFS_EXTENT_ITEM_KEY;
5757 key.offset = num_bytes;
5758 btrfs_release_path(path);
5759 ret = btrfs_search_slot(trans, extent_root,
5764 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5767 btrfs_print_leaf(extent_root,
5771 btrfs_abort_transaction(trans, extent_root, ret);
5774 extent_slot = path->slots[0];
5776 } else if (WARN_ON(ret == -ENOENT)) {
5777 btrfs_print_leaf(extent_root, path->nodes[0]);
5779 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5780 bytenr, parent, root_objectid, owner_objectid,
5783 btrfs_abort_transaction(trans, extent_root, ret);
5787 leaf = path->nodes[0];
5788 item_size = btrfs_item_size_nr(leaf, extent_slot);
5789 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5790 if (item_size < sizeof(*ei)) {
5791 BUG_ON(found_extent || extent_slot != path->slots[0]);
5792 ret = convert_extent_item_v0(trans, extent_root, path,
5795 btrfs_abort_transaction(trans, extent_root, ret);
5799 btrfs_release_path(path);
5800 path->leave_spinning = 1;
5802 key.objectid = bytenr;
5803 key.type = BTRFS_EXTENT_ITEM_KEY;
5804 key.offset = num_bytes;
5806 ret = btrfs_search_slot(trans, extent_root, &key, path,
5809 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5811 btrfs_print_leaf(extent_root, path->nodes[0]);
5814 btrfs_abort_transaction(trans, extent_root, ret);
5818 extent_slot = path->slots[0];
5819 leaf = path->nodes[0];
5820 item_size = btrfs_item_size_nr(leaf, extent_slot);
5823 BUG_ON(item_size < sizeof(*ei));
5824 ei = btrfs_item_ptr(leaf, extent_slot,
5825 struct btrfs_extent_item);
5826 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5827 key.type == BTRFS_EXTENT_ITEM_KEY) {
5828 struct btrfs_tree_block_info *bi;
5829 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5830 bi = (struct btrfs_tree_block_info *)(ei + 1);
5831 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5834 refs = btrfs_extent_refs(leaf, ei);
5835 if (refs < refs_to_drop) {
5836 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5837 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5839 btrfs_abort_transaction(trans, extent_root, ret);
5842 refs -= refs_to_drop;
5846 __run_delayed_extent_op(extent_op, leaf, ei);
5848 * In the case of inline back ref, reference count will
5849 * be updated by remove_extent_backref
5852 BUG_ON(!found_extent);
5854 btrfs_set_extent_refs(leaf, ei, refs);
5855 btrfs_mark_buffer_dirty(leaf);
5858 ret = remove_extent_backref(trans, extent_root, path,
5862 btrfs_abort_transaction(trans, extent_root, ret);
5866 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5870 BUG_ON(is_data && refs_to_drop !=
5871 extent_data_ref_count(root, path, iref));
5873 BUG_ON(path->slots[0] != extent_slot);
5875 BUG_ON(path->slots[0] != extent_slot + 1);
5876 path->slots[0] = extent_slot;
5881 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5884 btrfs_abort_transaction(trans, extent_root, ret);
5887 btrfs_release_path(path);
5890 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5892 btrfs_abort_transaction(trans, extent_root, ret);
5897 ret = update_block_group(root, bytenr, num_bytes, 0);
5899 btrfs_abort_transaction(trans, extent_root, ret);
5904 btrfs_free_path(path);
5909 * when we free an block, it is possible (and likely) that we free the last
5910 * delayed ref for that extent as well. This searches the delayed ref tree for
5911 * a given extent, and if there are no other delayed refs to be processed, it
5912 * removes it from the tree.
5914 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5915 struct btrfs_root *root, u64 bytenr)
5917 struct btrfs_delayed_ref_head *head;
5918 struct btrfs_delayed_ref_root *delayed_refs;
5919 struct btrfs_delayed_ref_node *ref;
5920 struct rb_node *node;
5923 delayed_refs = &trans->transaction->delayed_refs;
5924 spin_lock(&delayed_refs->lock);
5925 head = btrfs_find_delayed_ref_head(trans, bytenr);
5929 node = rb_prev(&head->node.rb_node);
5933 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5935 /* there are still entries for this ref, we can't drop it */
5936 if (ref->bytenr == bytenr)
5939 if (head->extent_op) {
5940 if (!head->must_insert_reserved)
5942 btrfs_free_delayed_extent_op(head->extent_op);
5943 head->extent_op = NULL;
5947 * waiting for the lock here would deadlock. If someone else has it
5948 * locked they are already in the process of dropping it anyway
5950 if (!mutex_trylock(&head->mutex))
5954 * at this point we have a head with no other entries. Go
5955 * ahead and process it.
5957 head->node.in_tree = 0;
5958 rb_erase(&head->node.rb_node, &delayed_refs->root);
5960 delayed_refs->num_entries--;
5963 * we don't take a ref on the node because we're removing it from the
5964 * tree, so we just steal the ref the tree was holding.
5966 delayed_refs->num_heads--;
5967 if (list_empty(&head->cluster))
5968 delayed_refs->num_heads_ready--;
5970 list_del_init(&head->cluster);
5971 spin_unlock(&delayed_refs->lock);
5973 BUG_ON(head->extent_op);
5974 if (head->must_insert_reserved)
5977 mutex_unlock(&head->mutex);
5978 btrfs_put_delayed_ref(&head->node);
5981 spin_unlock(&delayed_refs->lock);
5985 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5986 struct btrfs_root *root,
5987 struct extent_buffer *buf,
5988 u64 parent, int last_ref)
5990 struct btrfs_block_group_cache *cache = NULL;
5994 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5995 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5996 buf->start, buf->len,
5997 parent, root->root_key.objectid,
5998 btrfs_header_level(buf),
5999 BTRFS_DROP_DELAYED_REF, NULL, 0);
6000 BUG_ON(ret); /* -ENOMEM */
6006 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6008 if (btrfs_header_generation(buf) == trans->transid) {
6009 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6010 ret = check_ref_cleanup(trans, root, buf->start);
6015 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6016 pin_down_extent(root, cache, buf->start, buf->len, 1);
6020 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6022 btrfs_add_free_space(cache, buf->start, buf->len);
6023 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6024 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6029 add_pinned_bytes(root->fs_info, buf->len,
6030 btrfs_header_level(buf),
6031 root->root_key.objectid);
6034 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6037 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6038 btrfs_put_block_group(cache);
6041 /* Can return -ENOMEM */
6042 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6043 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6044 u64 owner, u64 offset, int for_cow)
6047 struct btrfs_fs_info *fs_info = root->fs_info;
6049 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6052 * tree log blocks never actually go into the extent allocation
6053 * tree, just update pinning info and exit early.
6055 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6056 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6057 /* unlocks the pinned mutex */
6058 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6060 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6061 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6063 parent, root_objectid, (int)owner,
6064 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6066 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6068 parent, root_objectid, owner,
6069 offset, BTRFS_DROP_DELAYED_REF,
6075 static u64 stripe_align(struct btrfs_root *root,
6076 struct btrfs_block_group_cache *cache,
6077 u64 val, u64 num_bytes)
6079 u64 ret = ALIGN(val, root->stripesize);
6084 * when we wait for progress in the block group caching, its because
6085 * our allocation attempt failed at least once. So, we must sleep
6086 * and let some progress happen before we try again.
6088 * This function will sleep at least once waiting for new free space to
6089 * show up, and then it will check the block group free space numbers
6090 * for our min num_bytes. Another option is to have it go ahead
6091 * and look in the rbtree for a free extent of a given size, but this
6094 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6095 * any of the information in this block group.
6097 static noinline void
6098 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6101 struct btrfs_caching_control *caching_ctl;
6103 caching_ctl = get_caching_control(cache);
6107 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6108 (cache->free_space_ctl->free_space >= num_bytes));
6110 put_caching_control(caching_ctl);
6114 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6116 struct btrfs_caching_control *caching_ctl;
6119 caching_ctl = get_caching_control(cache);
6121 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6123 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6124 if (cache->cached == BTRFS_CACHE_ERROR)
6126 put_caching_control(caching_ctl);
6130 int __get_raid_index(u64 flags)
6132 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6133 return BTRFS_RAID_RAID10;
6134 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6135 return BTRFS_RAID_RAID1;
6136 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6137 return BTRFS_RAID_DUP;
6138 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6139 return BTRFS_RAID_RAID0;
6140 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6141 return BTRFS_RAID_RAID5;
6142 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6143 return BTRFS_RAID_RAID6;
6145 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6148 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6150 return __get_raid_index(cache->flags);
6153 enum btrfs_loop_type {
6154 LOOP_CACHING_NOWAIT = 0,
6155 LOOP_CACHING_WAIT = 1,
6156 LOOP_ALLOC_CHUNK = 2,
6157 LOOP_NO_EMPTY_SIZE = 3,
6161 * walks the btree of allocated extents and find a hole of a given size.
6162 * The key ins is changed to record the hole:
6163 * ins->objectid == start position
6164 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6165 * ins->offset == the size of the hole.
6166 * Any available blocks before search_start are skipped.
6168 * If there is no suitable free space, we will record the max size of
6169 * the free space extent currently.
6171 static noinline int find_free_extent(struct btrfs_root *orig_root,
6172 u64 num_bytes, u64 empty_size,
6173 u64 hint_byte, struct btrfs_key *ins,
6177 struct btrfs_root *root = orig_root->fs_info->extent_root;
6178 struct btrfs_free_cluster *last_ptr = NULL;
6179 struct btrfs_block_group_cache *block_group = NULL;
6180 struct btrfs_block_group_cache *used_block_group;
6181 u64 search_start = 0;
6182 u64 max_extent_size = 0;
6183 int empty_cluster = 2 * 1024 * 1024;
6184 struct btrfs_space_info *space_info;
6186 int index = __get_raid_index(flags);
6187 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6188 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6189 bool found_uncached_bg = false;
6190 bool failed_cluster_refill = false;
6191 bool failed_alloc = false;
6192 bool use_cluster = true;
6193 bool have_caching_bg = false;
6195 WARN_ON(num_bytes < root->sectorsize);
6196 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6200 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6202 space_info = __find_space_info(root->fs_info, flags);
6204 btrfs_err(root->fs_info, "No space info for %llu", flags);
6209 * If the space info is for both data and metadata it means we have a
6210 * small filesystem and we can't use the clustering stuff.
6212 if (btrfs_mixed_space_info(space_info))
6213 use_cluster = false;
6215 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6216 last_ptr = &root->fs_info->meta_alloc_cluster;
6217 if (!btrfs_test_opt(root, SSD))
6218 empty_cluster = 64 * 1024;
6221 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6222 btrfs_test_opt(root, SSD)) {
6223 last_ptr = &root->fs_info->data_alloc_cluster;
6227 spin_lock(&last_ptr->lock);
6228 if (last_ptr->block_group)
6229 hint_byte = last_ptr->window_start;
6230 spin_unlock(&last_ptr->lock);
6233 search_start = max(search_start, first_logical_byte(root, 0));
6234 search_start = max(search_start, hint_byte);
6239 if (search_start == hint_byte) {
6240 block_group = btrfs_lookup_block_group(root->fs_info,
6242 used_block_group = block_group;
6244 * we don't want to use the block group if it doesn't match our
6245 * allocation bits, or if its not cached.
6247 * However if we are re-searching with an ideal block group
6248 * picked out then we don't care that the block group is cached.
6250 if (block_group && block_group_bits(block_group, flags) &&
6251 block_group->cached != BTRFS_CACHE_NO) {
6252 down_read(&space_info->groups_sem);
6253 if (list_empty(&block_group->list) ||
6256 * someone is removing this block group,
6257 * we can't jump into the have_block_group
6258 * target because our list pointers are not
6261 btrfs_put_block_group(block_group);
6262 up_read(&space_info->groups_sem);
6264 index = get_block_group_index(block_group);
6265 goto have_block_group;
6267 } else if (block_group) {
6268 btrfs_put_block_group(block_group);
6272 have_caching_bg = false;
6273 down_read(&space_info->groups_sem);
6274 list_for_each_entry(block_group, &space_info->block_groups[index],
6279 used_block_group = block_group;
6280 btrfs_get_block_group(block_group);
6281 search_start = block_group->key.objectid;
6284 * this can happen if we end up cycling through all the
6285 * raid types, but we want to make sure we only allocate
6286 * for the proper type.
6288 if (!block_group_bits(block_group, flags)) {
6289 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6290 BTRFS_BLOCK_GROUP_RAID1 |
6291 BTRFS_BLOCK_GROUP_RAID5 |
6292 BTRFS_BLOCK_GROUP_RAID6 |
6293 BTRFS_BLOCK_GROUP_RAID10;
6296 * if they asked for extra copies and this block group
6297 * doesn't provide them, bail. This does allow us to
6298 * fill raid0 from raid1.
6300 if ((flags & extra) && !(block_group->flags & extra))
6305 cached = block_group_cache_done(block_group);
6306 if (unlikely(!cached)) {
6307 found_uncached_bg = true;
6308 ret = cache_block_group(block_group, 0);
6313 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6315 if (unlikely(block_group->ro))
6319 * Ok we want to try and use the cluster allocator, so
6323 unsigned long aligned_cluster;
6325 * the refill lock keeps out other
6326 * people trying to start a new cluster
6328 spin_lock(&last_ptr->refill_lock);
6329 used_block_group = last_ptr->block_group;
6330 if (used_block_group != block_group &&
6331 (!used_block_group ||
6332 used_block_group->ro ||
6333 !block_group_bits(used_block_group, flags))) {
6334 used_block_group = block_group;
6335 goto refill_cluster;
6338 if (used_block_group != block_group)
6339 btrfs_get_block_group(used_block_group);
6341 offset = btrfs_alloc_from_cluster(used_block_group,
6344 used_block_group->key.objectid,
6347 /* we have a block, we're done */
6348 spin_unlock(&last_ptr->refill_lock);
6349 trace_btrfs_reserve_extent_cluster(root,
6350 block_group, search_start, num_bytes);
6354 WARN_ON(last_ptr->block_group != used_block_group);
6355 if (used_block_group != block_group) {
6356 btrfs_put_block_group(used_block_group);
6357 used_block_group = block_group;
6360 BUG_ON(used_block_group != block_group);
6361 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6362 * set up a new clusters, so lets just skip it
6363 * and let the allocator find whatever block
6364 * it can find. If we reach this point, we
6365 * will have tried the cluster allocator
6366 * plenty of times and not have found
6367 * anything, so we are likely way too
6368 * fragmented for the clustering stuff to find
6371 * However, if the cluster is taken from the
6372 * current block group, release the cluster
6373 * first, so that we stand a better chance of
6374 * succeeding in the unclustered
6376 if (loop >= LOOP_NO_EMPTY_SIZE &&
6377 last_ptr->block_group != block_group) {
6378 spin_unlock(&last_ptr->refill_lock);
6379 goto unclustered_alloc;
6383 * this cluster didn't work out, free it and
6386 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6388 if (loop >= LOOP_NO_EMPTY_SIZE) {
6389 spin_unlock(&last_ptr->refill_lock);
6390 goto unclustered_alloc;
6393 aligned_cluster = max_t(unsigned long,
6394 empty_cluster + empty_size,
6395 block_group->full_stripe_len);
6397 /* allocate a cluster in this block group */
6398 ret = btrfs_find_space_cluster(root, block_group,
6399 last_ptr, search_start,
6404 * now pull our allocation out of this
6407 offset = btrfs_alloc_from_cluster(block_group,
6413 /* we found one, proceed */
6414 spin_unlock(&last_ptr->refill_lock);
6415 trace_btrfs_reserve_extent_cluster(root,
6416 block_group, search_start,
6420 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6421 && !failed_cluster_refill) {
6422 spin_unlock(&last_ptr->refill_lock);
6424 failed_cluster_refill = true;
6425 wait_block_group_cache_progress(block_group,
6426 num_bytes + empty_cluster + empty_size);
6427 goto have_block_group;
6431 * at this point we either didn't find a cluster
6432 * or we weren't able to allocate a block from our
6433 * cluster. Free the cluster we've been trying
6434 * to use, and go to the next block group
6436 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6437 spin_unlock(&last_ptr->refill_lock);
6442 spin_lock(&block_group->free_space_ctl->tree_lock);
6444 block_group->free_space_ctl->free_space <
6445 num_bytes + empty_cluster + empty_size) {
6446 if (block_group->free_space_ctl->free_space >
6449 block_group->free_space_ctl->free_space;
6450 spin_unlock(&block_group->free_space_ctl->tree_lock);
6453 spin_unlock(&block_group->free_space_ctl->tree_lock);
6455 offset = btrfs_find_space_for_alloc(block_group, search_start,
6456 num_bytes, empty_size,
6459 * If we didn't find a chunk, and we haven't failed on this
6460 * block group before, and this block group is in the middle of
6461 * caching and we are ok with waiting, then go ahead and wait
6462 * for progress to be made, and set failed_alloc to true.
6464 * If failed_alloc is true then we've already waited on this
6465 * block group once and should move on to the next block group.
6467 if (!offset && !failed_alloc && !cached &&
6468 loop > LOOP_CACHING_NOWAIT) {
6469 wait_block_group_cache_progress(block_group,
6470 num_bytes + empty_size);
6471 failed_alloc = true;
6472 goto have_block_group;
6473 } else if (!offset) {
6475 have_caching_bg = true;
6479 search_start = stripe_align(root, used_block_group,
6482 /* move on to the next group */
6483 if (search_start + num_bytes >
6484 used_block_group->key.objectid + used_block_group->key.offset) {
6485 btrfs_add_free_space(used_block_group, offset, num_bytes);
6489 if (offset < search_start)
6490 btrfs_add_free_space(used_block_group, offset,
6491 search_start - offset);
6492 BUG_ON(offset > search_start);
6494 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6496 if (ret == -EAGAIN) {
6497 btrfs_add_free_space(used_block_group, offset, num_bytes);
6501 /* we are all good, lets return */
6502 ins->objectid = search_start;
6503 ins->offset = num_bytes;
6505 trace_btrfs_reserve_extent(orig_root, block_group,
6506 search_start, num_bytes);
6507 if (used_block_group != block_group)
6508 btrfs_put_block_group(used_block_group);
6509 btrfs_put_block_group(block_group);
6512 failed_cluster_refill = false;
6513 failed_alloc = false;
6514 BUG_ON(index != get_block_group_index(block_group));
6515 if (used_block_group != block_group)
6516 btrfs_put_block_group(used_block_group);
6517 btrfs_put_block_group(block_group);
6519 up_read(&space_info->groups_sem);
6521 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6524 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6528 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6529 * caching kthreads as we move along
6530 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6531 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6532 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6535 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6538 if (loop == LOOP_ALLOC_CHUNK) {
6539 struct btrfs_trans_handle *trans;
6541 trans = btrfs_join_transaction(root);
6542 if (IS_ERR(trans)) {
6543 ret = PTR_ERR(trans);
6547 ret = do_chunk_alloc(trans, root, flags,
6550 * Do not bail out on ENOSPC since we
6551 * can do more things.
6553 if (ret < 0 && ret != -ENOSPC)
6554 btrfs_abort_transaction(trans,
6558 btrfs_end_transaction(trans, root);
6563 if (loop == LOOP_NO_EMPTY_SIZE) {
6569 } else if (!ins->objectid) {
6571 } else if (ins->objectid) {
6576 ins->offset = max_extent_size;
6580 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6581 int dump_block_groups)
6583 struct btrfs_block_group_cache *cache;
6586 spin_lock(&info->lock);
6587 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6589 info->total_bytes - info->bytes_used - info->bytes_pinned -
6590 info->bytes_reserved - info->bytes_readonly,
6591 (info->full) ? "" : "not ");
6592 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6593 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6594 info->total_bytes, info->bytes_used, info->bytes_pinned,
6595 info->bytes_reserved, info->bytes_may_use,
6596 info->bytes_readonly);
6597 spin_unlock(&info->lock);
6599 if (!dump_block_groups)
6602 down_read(&info->groups_sem);
6604 list_for_each_entry(cache, &info->block_groups[index], list) {
6605 spin_lock(&cache->lock);
6606 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6607 cache->key.objectid, cache->key.offset,
6608 btrfs_block_group_used(&cache->item), cache->pinned,
6609 cache->reserved, cache->ro ? "[readonly]" : "");
6610 btrfs_dump_free_space(cache, bytes);
6611 spin_unlock(&cache->lock);
6613 if (++index < BTRFS_NR_RAID_TYPES)
6615 up_read(&info->groups_sem);
6618 int btrfs_reserve_extent(struct btrfs_root *root,
6619 u64 num_bytes, u64 min_alloc_size,
6620 u64 empty_size, u64 hint_byte,
6621 struct btrfs_key *ins, int is_data)
6623 bool final_tried = false;
6627 flags = btrfs_get_alloc_profile(root, is_data);
6629 WARN_ON(num_bytes < root->sectorsize);
6630 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6633 if (ret == -ENOSPC) {
6634 if (!final_tried && ins->offset) {
6635 num_bytes = min(num_bytes >> 1, ins->offset);
6636 num_bytes = round_down(num_bytes, root->sectorsize);
6637 num_bytes = max(num_bytes, min_alloc_size);
6638 if (num_bytes == min_alloc_size)
6641 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6642 struct btrfs_space_info *sinfo;
6644 sinfo = __find_space_info(root->fs_info, flags);
6645 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6648 dump_space_info(sinfo, num_bytes, 1);
6655 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6656 u64 start, u64 len, int pin)
6658 struct btrfs_block_group_cache *cache;
6661 cache = btrfs_lookup_block_group(root->fs_info, start);
6663 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6668 if (btrfs_test_opt(root, DISCARD))
6669 ret = btrfs_discard_extent(root, start, len, NULL);
6672 pin_down_extent(root, cache, start, len, 1);
6674 btrfs_add_free_space(cache, start, len);
6675 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6677 btrfs_put_block_group(cache);
6679 trace_btrfs_reserved_extent_free(root, start, len);
6684 int btrfs_free_reserved_extent(struct btrfs_root *root,
6687 return __btrfs_free_reserved_extent(root, start, len, 0);
6690 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6693 return __btrfs_free_reserved_extent(root, start, len, 1);
6696 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6697 struct btrfs_root *root,
6698 u64 parent, u64 root_objectid,
6699 u64 flags, u64 owner, u64 offset,
6700 struct btrfs_key *ins, int ref_mod)
6703 struct btrfs_fs_info *fs_info = root->fs_info;
6704 struct btrfs_extent_item *extent_item;
6705 struct btrfs_extent_inline_ref *iref;
6706 struct btrfs_path *path;
6707 struct extent_buffer *leaf;
6712 type = BTRFS_SHARED_DATA_REF_KEY;
6714 type = BTRFS_EXTENT_DATA_REF_KEY;
6716 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6718 path = btrfs_alloc_path();
6722 path->leave_spinning = 1;
6723 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6726 btrfs_free_path(path);
6730 leaf = path->nodes[0];
6731 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6732 struct btrfs_extent_item);
6733 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6734 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6735 btrfs_set_extent_flags(leaf, extent_item,
6736 flags | BTRFS_EXTENT_FLAG_DATA);
6738 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6739 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6741 struct btrfs_shared_data_ref *ref;
6742 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6743 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6744 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6746 struct btrfs_extent_data_ref *ref;
6747 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6748 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6749 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6750 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6751 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6754 btrfs_mark_buffer_dirty(path->nodes[0]);
6755 btrfs_free_path(path);
6757 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6758 if (ret) { /* -ENOENT, logic error */
6759 btrfs_err(fs_info, "update block group failed for %llu %llu",
6760 ins->objectid, ins->offset);
6763 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6767 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6768 struct btrfs_root *root,
6769 u64 parent, u64 root_objectid,
6770 u64 flags, struct btrfs_disk_key *key,
6771 int level, struct btrfs_key *ins)
6774 struct btrfs_fs_info *fs_info = root->fs_info;
6775 struct btrfs_extent_item *extent_item;
6776 struct btrfs_tree_block_info *block_info;
6777 struct btrfs_extent_inline_ref *iref;
6778 struct btrfs_path *path;
6779 struct extent_buffer *leaf;
6780 u32 size = sizeof(*extent_item) + sizeof(*iref);
6781 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6784 if (!skinny_metadata)
6785 size += sizeof(*block_info);
6787 path = btrfs_alloc_path();
6789 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6794 path->leave_spinning = 1;
6795 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6798 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6800 btrfs_free_path(path);
6804 leaf = path->nodes[0];
6805 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6806 struct btrfs_extent_item);
6807 btrfs_set_extent_refs(leaf, extent_item, 1);
6808 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6809 btrfs_set_extent_flags(leaf, extent_item,
6810 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6812 if (skinny_metadata) {
6813 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6815 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6816 btrfs_set_tree_block_key(leaf, block_info, key);
6817 btrfs_set_tree_block_level(leaf, block_info, level);
6818 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6822 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6823 btrfs_set_extent_inline_ref_type(leaf, iref,
6824 BTRFS_SHARED_BLOCK_REF_KEY);
6825 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6827 btrfs_set_extent_inline_ref_type(leaf, iref,
6828 BTRFS_TREE_BLOCK_REF_KEY);
6829 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6832 btrfs_mark_buffer_dirty(leaf);
6833 btrfs_free_path(path);
6835 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6836 if (ret) { /* -ENOENT, logic error */
6837 btrfs_err(fs_info, "update block group failed for %llu %llu",
6838 ins->objectid, ins->offset);
6842 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6846 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6847 struct btrfs_root *root,
6848 u64 root_objectid, u64 owner,
6849 u64 offset, struct btrfs_key *ins)
6853 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6855 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6857 root_objectid, owner, offset,
6858 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6863 * this is used by the tree logging recovery code. It records that
6864 * an extent has been allocated and makes sure to clear the free
6865 * space cache bits as well
6867 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6868 struct btrfs_root *root,
6869 u64 root_objectid, u64 owner, u64 offset,
6870 struct btrfs_key *ins)
6873 struct btrfs_block_group_cache *block_group;
6876 * Mixed block groups will exclude before processing the log so we only
6877 * need to do the exlude dance if this fs isn't mixed.
6879 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6880 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6885 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6889 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6890 RESERVE_ALLOC_NO_ACCOUNT);
6891 BUG_ON(ret); /* logic error */
6892 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6893 0, owner, offset, ins, 1);
6894 btrfs_put_block_group(block_group);
6898 static struct extent_buffer *
6899 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6900 u64 bytenr, u32 blocksize, int level)
6902 struct extent_buffer *buf;
6904 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6906 return ERR_PTR(-ENOMEM);
6907 btrfs_set_header_generation(buf, trans->transid);
6908 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6909 btrfs_tree_lock(buf);
6910 clean_tree_block(trans, root, buf);
6911 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6913 btrfs_set_lock_blocking(buf);
6914 btrfs_set_buffer_uptodate(buf);
6916 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6918 * we allow two log transactions at a time, use different
6919 * EXENT bit to differentiate dirty pages.
6921 if (root->log_transid % 2 == 0)
6922 set_extent_dirty(&root->dirty_log_pages, buf->start,
6923 buf->start + buf->len - 1, GFP_NOFS);
6925 set_extent_new(&root->dirty_log_pages, buf->start,
6926 buf->start + buf->len - 1, GFP_NOFS);
6928 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6929 buf->start + buf->len - 1, GFP_NOFS);
6931 trans->blocks_used++;
6932 /* this returns a buffer locked for blocking */
6936 static struct btrfs_block_rsv *
6937 use_block_rsv(struct btrfs_trans_handle *trans,
6938 struct btrfs_root *root, u32 blocksize)
6940 struct btrfs_block_rsv *block_rsv;
6941 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6943 bool global_updated = false;
6945 block_rsv = get_block_rsv(trans, root);
6947 if (unlikely(block_rsv->size == 0))
6950 ret = block_rsv_use_bytes(block_rsv, blocksize);
6954 if (block_rsv->failfast)
6955 return ERR_PTR(ret);
6957 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6958 global_updated = true;
6959 update_global_block_rsv(root->fs_info);
6963 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6964 static DEFINE_RATELIMIT_STATE(_rs,
6965 DEFAULT_RATELIMIT_INTERVAL * 10,
6966 /*DEFAULT_RATELIMIT_BURST*/ 1);
6967 if (__ratelimit(&_rs))
6969 "btrfs: block rsv returned %d\n", ret);
6972 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6973 BTRFS_RESERVE_NO_FLUSH);
6977 * If we couldn't reserve metadata bytes try and use some from
6978 * the global reserve if its space type is the same as the global
6981 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6982 block_rsv->space_info == global_rsv->space_info) {
6983 ret = block_rsv_use_bytes(global_rsv, blocksize);
6987 return ERR_PTR(ret);
6990 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6991 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6993 block_rsv_add_bytes(block_rsv, blocksize, 0);
6994 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6998 * finds a free extent and does all the dirty work required for allocation
6999 * returns the key for the extent through ins, and a tree buffer for
7000 * the first block of the extent through buf.
7002 * returns the tree buffer or NULL.
7004 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7005 struct btrfs_root *root, u32 blocksize,
7006 u64 parent, u64 root_objectid,
7007 struct btrfs_disk_key *key, int level,
7008 u64 hint, u64 empty_size)
7010 struct btrfs_key ins;
7011 struct btrfs_block_rsv *block_rsv;
7012 struct extent_buffer *buf;
7015 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7018 block_rsv = use_block_rsv(trans, root, blocksize);
7019 if (IS_ERR(block_rsv))
7020 return ERR_CAST(block_rsv);
7022 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7023 empty_size, hint, &ins, 0);
7025 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7026 return ERR_PTR(ret);
7029 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7031 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7033 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7035 parent = ins.objectid;
7036 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7040 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7041 struct btrfs_delayed_extent_op *extent_op;
7042 extent_op = btrfs_alloc_delayed_extent_op();
7043 BUG_ON(!extent_op); /* -ENOMEM */
7045 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7047 memset(&extent_op->key, 0, sizeof(extent_op->key));
7048 extent_op->flags_to_set = flags;
7049 if (skinny_metadata)
7050 extent_op->update_key = 0;
7052 extent_op->update_key = 1;
7053 extent_op->update_flags = 1;
7054 extent_op->is_data = 0;
7055 extent_op->level = level;
7057 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7059 ins.offset, parent, root_objectid,
7060 level, BTRFS_ADD_DELAYED_EXTENT,
7062 BUG_ON(ret); /* -ENOMEM */
7067 struct walk_control {
7068 u64 refs[BTRFS_MAX_LEVEL];
7069 u64 flags[BTRFS_MAX_LEVEL];
7070 struct btrfs_key update_progress;
7081 #define DROP_REFERENCE 1
7082 #define UPDATE_BACKREF 2
7084 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7085 struct btrfs_root *root,
7086 struct walk_control *wc,
7087 struct btrfs_path *path)
7095 struct btrfs_key key;
7096 struct extent_buffer *eb;
7101 if (path->slots[wc->level] < wc->reada_slot) {
7102 wc->reada_count = wc->reada_count * 2 / 3;
7103 wc->reada_count = max(wc->reada_count, 2);
7105 wc->reada_count = wc->reada_count * 3 / 2;
7106 wc->reada_count = min_t(int, wc->reada_count,
7107 BTRFS_NODEPTRS_PER_BLOCK(root));
7110 eb = path->nodes[wc->level];
7111 nritems = btrfs_header_nritems(eb);
7112 blocksize = btrfs_level_size(root, wc->level - 1);
7114 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7115 if (nread >= wc->reada_count)
7119 bytenr = btrfs_node_blockptr(eb, slot);
7120 generation = btrfs_node_ptr_generation(eb, slot);
7122 if (slot == path->slots[wc->level])
7125 if (wc->stage == UPDATE_BACKREF &&
7126 generation <= root->root_key.offset)
7129 /* We don't lock the tree block, it's OK to be racy here */
7130 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7131 wc->level - 1, 1, &refs,
7133 /* We don't care about errors in readahead. */
7138 if (wc->stage == DROP_REFERENCE) {
7142 if (wc->level == 1 &&
7143 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7145 if (!wc->update_ref ||
7146 generation <= root->root_key.offset)
7148 btrfs_node_key_to_cpu(eb, &key, slot);
7149 ret = btrfs_comp_cpu_keys(&key,
7150 &wc->update_progress);
7154 if (wc->level == 1 &&
7155 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7159 ret = readahead_tree_block(root, bytenr, blocksize,
7165 wc->reada_slot = slot;
7169 * helper to process tree block while walking down the tree.
7171 * when wc->stage == UPDATE_BACKREF, this function updates
7172 * back refs for pointers in the block.
7174 * NOTE: return value 1 means we should stop walking down.
7176 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7177 struct btrfs_root *root,
7178 struct btrfs_path *path,
7179 struct walk_control *wc, int lookup_info)
7181 int level = wc->level;
7182 struct extent_buffer *eb = path->nodes[level];
7183 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7186 if (wc->stage == UPDATE_BACKREF &&
7187 btrfs_header_owner(eb) != root->root_key.objectid)
7191 * when reference count of tree block is 1, it won't increase
7192 * again. once full backref flag is set, we never clear it.
7195 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7196 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7197 BUG_ON(!path->locks[level]);
7198 ret = btrfs_lookup_extent_info(trans, root,
7199 eb->start, level, 1,
7202 BUG_ON(ret == -ENOMEM);
7205 BUG_ON(wc->refs[level] == 0);
7208 if (wc->stage == DROP_REFERENCE) {
7209 if (wc->refs[level] > 1)
7212 if (path->locks[level] && !wc->keep_locks) {
7213 btrfs_tree_unlock_rw(eb, path->locks[level]);
7214 path->locks[level] = 0;
7219 /* wc->stage == UPDATE_BACKREF */
7220 if (!(wc->flags[level] & flag)) {
7221 BUG_ON(!path->locks[level]);
7222 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7223 BUG_ON(ret); /* -ENOMEM */
7224 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7225 BUG_ON(ret); /* -ENOMEM */
7226 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7228 btrfs_header_level(eb), 0);
7229 BUG_ON(ret); /* -ENOMEM */
7230 wc->flags[level] |= flag;
7234 * the block is shared by multiple trees, so it's not good to
7235 * keep the tree lock
7237 if (path->locks[level] && level > 0) {
7238 btrfs_tree_unlock_rw(eb, path->locks[level]);
7239 path->locks[level] = 0;
7245 * helper to process tree block pointer.
7247 * when wc->stage == DROP_REFERENCE, this function checks
7248 * reference count of the block pointed to. if the block
7249 * is shared and we need update back refs for the subtree
7250 * rooted at the block, this function changes wc->stage to
7251 * UPDATE_BACKREF. if the block is shared and there is no
7252 * need to update back, this function drops the reference
7255 * NOTE: return value 1 means we should stop walking down.
7257 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7258 struct btrfs_root *root,
7259 struct btrfs_path *path,
7260 struct walk_control *wc, int *lookup_info)
7266 struct btrfs_key key;
7267 struct extent_buffer *next;
7268 int level = wc->level;
7272 generation = btrfs_node_ptr_generation(path->nodes[level],
7273 path->slots[level]);
7275 * if the lower level block was created before the snapshot
7276 * was created, we know there is no need to update back refs
7279 if (wc->stage == UPDATE_BACKREF &&
7280 generation <= root->root_key.offset) {
7285 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7286 blocksize = btrfs_level_size(root, level - 1);
7288 next = btrfs_find_tree_block(root, bytenr, blocksize);
7290 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7293 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7297 btrfs_tree_lock(next);
7298 btrfs_set_lock_blocking(next);
7300 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7301 &wc->refs[level - 1],
7302 &wc->flags[level - 1]);
7304 btrfs_tree_unlock(next);
7308 if (unlikely(wc->refs[level - 1] == 0)) {
7309 btrfs_err(root->fs_info, "Missing references.");
7314 if (wc->stage == DROP_REFERENCE) {
7315 if (wc->refs[level - 1] > 1) {
7317 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7320 if (!wc->update_ref ||
7321 generation <= root->root_key.offset)
7324 btrfs_node_key_to_cpu(path->nodes[level], &key,
7325 path->slots[level]);
7326 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7330 wc->stage = UPDATE_BACKREF;
7331 wc->shared_level = level - 1;
7335 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7339 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7340 btrfs_tree_unlock(next);
7341 free_extent_buffer(next);
7347 if (reada && level == 1)
7348 reada_walk_down(trans, root, wc, path);
7349 next = read_tree_block(root, bytenr, blocksize, generation);
7350 if (!next || !extent_buffer_uptodate(next)) {
7351 free_extent_buffer(next);
7354 btrfs_tree_lock(next);
7355 btrfs_set_lock_blocking(next);
7359 BUG_ON(level != btrfs_header_level(next));
7360 path->nodes[level] = next;
7361 path->slots[level] = 0;
7362 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7368 wc->refs[level - 1] = 0;
7369 wc->flags[level - 1] = 0;
7370 if (wc->stage == DROP_REFERENCE) {
7371 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7372 parent = path->nodes[level]->start;
7374 BUG_ON(root->root_key.objectid !=
7375 btrfs_header_owner(path->nodes[level]));
7379 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7380 root->root_key.objectid, level - 1, 0, 0);
7381 BUG_ON(ret); /* -ENOMEM */
7383 btrfs_tree_unlock(next);
7384 free_extent_buffer(next);
7390 * helper to process tree block while walking up the tree.
7392 * when wc->stage == DROP_REFERENCE, this function drops
7393 * reference count on the block.
7395 * when wc->stage == UPDATE_BACKREF, this function changes
7396 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7397 * to UPDATE_BACKREF previously while processing the block.
7399 * NOTE: return value 1 means we should stop walking up.
7401 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7402 struct btrfs_root *root,
7403 struct btrfs_path *path,
7404 struct walk_control *wc)
7407 int level = wc->level;
7408 struct extent_buffer *eb = path->nodes[level];
7411 if (wc->stage == UPDATE_BACKREF) {
7412 BUG_ON(wc->shared_level < level);
7413 if (level < wc->shared_level)
7416 ret = find_next_key(path, level + 1, &wc->update_progress);
7420 wc->stage = DROP_REFERENCE;
7421 wc->shared_level = -1;
7422 path->slots[level] = 0;
7425 * check reference count again if the block isn't locked.
7426 * we should start walking down the tree again if reference
7429 if (!path->locks[level]) {
7431 btrfs_tree_lock(eb);
7432 btrfs_set_lock_blocking(eb);
7433 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7435 ret = btrfs_lookup_extent_info(trans, root,
7436 eb->start, level, 1,
7440 btrfs_tree_unlock_rw(eb, path->locks[level]);
7441 path->locks[level] = 0;
7444 BUG_ON(wc->refs[level] == 0);
7445 if (wc->refs[level] == 1) {
7446 btrfs_tree_unlock_rw(eb, path->locks[level]);
7447 path->locks[level] = 0;
7453 /* wc->stage == DROP_REFERENCE */
7454 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7456 if (wc->refs[level] == 1) {
7458 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7459 ret = btrfs_dec_ref(trans, root, eb, 1,
7462 ret = btrfs_dec_ref(trans, root, eb, 0,
7464 BUG_ON(ret); /* -ENOMEM */
7466 /* make block locked assertion in clean_tree_block happy */
7467 if (!path->locks[level] &&
7468 btrfs_header_generation(eb) == trans->transid) {
7469 btrfs_tree_lock(eb);
7470 btrfs_set_lock_blocking(eb);
7471 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7473 clean_tree_block(trans, root, eb);
7476 if (eb == root->node) {
7477 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7480 BUG_ON(root->root_key.objectid !=
7481 btrfs_header_owner(eb));
7483 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7484 parent = path->nodes[level + 1]->start;
7486 BUG_ON(root->root_key.objectid !=
7487 btrfs_header_owner(path->nodes[level + 1]));
7490 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7492 wc->refs[level] = 0;
7493 wc->flags[level] = 0;
7497 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7498 struct btrfs_root *root,
7499 struct btrfs_path *path,
7500 struct walk_control *wc)
7502 int level = wc->level;
7503 int lookup_info = 1;
7506 while (level >= 0) {
7507 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7514 if (path->slots[level] >=
7515 btrfs_header_nritems(path->nodes[level]))
7518 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7520 path->slots[level]++;
7529 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7530 struct btrfs_root *root,
7531 struct btrfs_path *path,
7532 struct walk_control *wc, int max_level)
7534 int level = wc->level;
7537 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7538 while (level < max_level && path->nodes[level]) {
7540 if (path->slots[level] + 1 <
7541 btrfs_header_nritems(path->nodes[level])) {
7542 path->slots[level]++;
7545 ret = walk_up_proc(trans, root, path, wc);
7549 if (path->locks[level]) {
7550 btrfs_tree_unlock_rw(path->nodes[level],
7551 path->locks[level]);
7552 path->locks[level] = 0;
7554 free_extent_buffer(path->nodes[level]);
7555 path->nodes[level] = NULL;
7563 * drop a subvolume tree.
7565 * this function traverses the tree freeing any blocks that only
7566 * referenced by the tree.
7568 * when a shared tree block is found. this function decreases its
7569 * reference count by one. if update_ref is true, this function
7570 * also make sure backrefs for the shared block and all lower level
7571 * blocks are properly updated.
7573 * If called with for_reloc == 0, may exit early with -EAGAIN
7575 int btrfs_drop_snapshot(struct btrfs_root *root,
7576 struct btrfs_block_rsv *block_rsv, int update_ref,
7579 struct btrfs_path *path;
7580 struct btrfs_trans_handle *trans;
7581 struct btrfs_root *tree_root = root->fs_info->tree_root;
7582 struct btrfs_root_item *root_item = &root->root_item;
7583 struct walk_control *wc;
7584 struct btrfs_key key;
7588 bool root_dropped = false;
7590 path = btrfs_alloc_path();
7596 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7598 btrfs_free_path(path);
7603 trans = btrfs_start_transaction(tree_root, 0);
7604 if (IS_ERR(trans)) {
7605 err = PTR_ERR(trans);
7610 trans->block_rsv = block_rsv;
7612 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7613 level = btrfs_header_level(root->node);
7614 path->nodes[level] = btrfs_lock_root_node(root);
7615 btrfs_set_lock_blocking(path->nodes[level]);
7616 path->slots[level] = 0;
7617 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7618 memset(&wc->update_progress, 0,
7619 sizeof(wc->update_progress));
7621 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7622 memcpy(&wc->update_progress, &key,
7623 sizeof(wc->update_progress));
7625 level = root_item->drop_level;
7627 path->lowest_level = level;
7628 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7629 path->lowest_level = 0;
7637 * unlock our path, this is safe because only this
7638 * function is allowed to delete this snapshot
7640 btrfs_unlock_up_safe(path, 0);
7642 level = btrfs_header_level(root->node);
7644 btrfs_tree_lock(path->nodes[level]);
7645 btrfs_set_lock_blocking(path->nodes[level]);
7646 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7648 ret = btrfs_lookup_extent_info(trans, root,
7649 path->nodes[level]->start,
7650 level, 1, &wc->refs[level],
7656 BUG_ON(wc->refs[level] == 0);
7658 if (level == root_item->drop_level)
7661 btrfs_tree_unlock(path->nodes[level]);
7662 path->locks[level] = 0;
7663 WARN_ON(wc->refs[level] != 1);
7669 wc->shared_level = -1;
7670 wc->stage = DROP_REFERENCE;
7671 wc->update_ref = update_ref;
7673 wc->for_reloc = for_reloc;
7674 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7678 ret = walk_down_tree(trans, root, path, wc);
7684 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7691 BUG_ON(wc->stage != DROP_REFERENCE);
7695 if (wc->stage == DROP_REFERENCE) {
7697 btrfs_node_key(path->nodes[level],
7698 &root_item->drop_progress,
7699 path->slots[level]);
7700 root_item->drop_level = level;
7703 BUG_ON(wc->level == 0);
7704 if (btrfs_should_end_transaction(trans, tree_root) ||
7705 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7706 ret = btrfs_update_root(trans, tree_root,
7710 btrfs_abort_transaction(trans, tree_root, ret);
7715 btrfs_end_transaction_throttle(trans, tree_root);
7716 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7717 pr_debug("btrfs: drop snapshot early exit\n");
7722 trans = btrfs_start_transaction(tree_root, 0);
7723 if (IS_ERR(trans)) {
7724 err = PTR_ERR(trans);
7728 trans->block_rsv = block_rsv;
7731 btrfs_release_path(path);
7735 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7737 btrfs_abort_transaction(trans, tree_root, ret);
7741 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7742 ret = btrfs_find_root(tree_root, &root->root_key, path,
7745 btrfs_abort_transaction(trans, tree_root, ret);
7748 } else if (ret > 0) {
7749 /* if we fail to delete the orphan item this time
7750 * around, it'll get picked up the next time.
7752 * The most common failure here is just -ENOENT.
7754 btrfs_del_orphan_item(trans, tree_root,
7755 root->root_key.objectid);
7759 if (root->in_radix) {
7760 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7762 free_extent_buffer(root->node);
7763 free_extent_buffer(root->commit_root);
7764 btrfs_put_fs_root(root);
7766 root_dropped = true;
7768 btrfs_end_transaction_throttle(trans, tree_root);
7771 btrfs_free_path(path);
7774 * So if we need to stop dropping the snapshot for whatever reason we
7775 * need to make sure to add it back to the dead root list so that we
7776 * keep trying to do the work later. This also cleans up roots if we
7777 * don't have it in the radix (like when we recover after a power fail
7778 * or unmount) so we don't leak memory.
7780 if (!for_reloc && root_dropped == false)
7781 btrfs_add_dead_root(root);
7783 btrfs_std_error(root->fs_info, err);
7788 * drop subtree rooted at tree block 'node'.
7790 * NOTE: this function will unlock and release tree block 'node'
7791 * only used by relocation code
7793 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7794 struct btrfs_root *root,
7795 struct extent_buffer *node,
7796 struct extent_buffer *parent)
7798 struct btrfs_path *path;
7799 struct walk_control *wc;
7805 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7807 path = btrfs_alloc_path();
7811 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7813 btrfs_free_path(path);
7817 btrfs_assert_tree_locked(parent);
7818 parent_level = btrfs_header_level(parent);
7819 extent_buffer_get(parent);
7820 path->nodes[parent_level] = parent;
7821 path->slots[parent_level] = btrfs_header_nritems(parent);
7823 btrfs_assert_tree_locked(node);
7824 level = btrfs_header_level(node);
7825 path->nodes[level] = node;
7826 path->slots[level] = 0;
7827 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7829 wc->refs[parent_level] = 1;
7830 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7832 wc->shared_level = -1;
7833 wc->stage = DROP_REFERENCE;
7837 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7840 wret = walk_down_tree(trans, root, path, wc);
7846 wret = walk_up_tree(trans, root, path, wc, parent_level);
7854 btrfs_free_path(path);
7858 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7864 * if restripe for this chunk_type is on pick target profile and
7865 * return, otherwise do the usual balance
7867 stripped = get_restripe_target(root->fs_info, flags);
7869 return extended_to_chunk(stripped);
7872 * we add in the count of missing devices because we want
7873 * to make sure that any RAID levels on a degraded FS
7874 * continue to be honored.
7876 num_devices = root->fs_info->fs_devices->rw_devices +
7877 root->fs_info->fs_devices->missing_devices;
7879 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7880 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7881 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7883 if (num_devices == 1) {
7884 stripped |= BTRFS_BLOCK_GROUP_DUP;
7885 stripped = flags & ~stripped;
7887 /* turn raid0 into single device chunks */
7888 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7891 /* turn mirroring into duplication */
7892 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7893 BTRFS_BLOCK_GROUP_RAID10))
7894 return stripped | BTRFS_BLOCK_GROUP_DUP;
7896 /* they already had raid on here, just return */
7897 if (flags & stripped)
7900 stripped |= BTRFS_BLOCK_GROUP_DUP;
7901 stripped = flags & ~stripped;
7903 /* switch duplicated blocks with raid1 */
7904 if (flags & BTRFS_BLOCK_GROUP_DUP)
7905 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7907 /* this is drive concat, leave it alone */
7913 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7915 struct btrfs_space_info *sinfo = cache->space_info;
7917 u64 min_allocable_bytes;
7922 * We need some metadata space and system metadata space for
7923 * allocating chunks in some corner cases until we force to set
7924 * it to be readonly.
7927 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7929 min_allocable_bytes = 1 * 1024 * 1024;
7931 min_allocable_bytes = 0;
7933 spin_lock(&sinfo->lock);
7934 spin_lock(&cache->lock);
7941 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7942 cache->bytes_super - btrfs_block_group_used(&cache->item);
7944 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7945 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7946 min_allocable_bytes <= sinfo->total_bytes) {
7947 sinfo->bytes_readonly += num_bytes;
7952 spin_unlock(&cache->lock);
7953 spin_unlock(&sinfo->lock);
7957 int btrfs_set_block_group_ro(struct btrfs_root *root,
7958 struct btrfs_block_group_cache *cache)
7961 struct btrfs_trans_handle *trans;
7967 trans = btrfs_join_transaction(root);
7969 return PTR_ERR(trans);
7971 alloc_flags = update_block_group_flags(root, cache->flags);
7972 if (alloc_flags != cache->flags) {
7973 ret = do_chunk_alloc(trans, root, alloc_flags,
7979 ret = set_block_group_ro(cache, 0);
7982 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7983 ret = do_chunk_alloc(trans, root, alloc_flags,
7987 ret = set_block_group_ro(cache, 0);
7989 btrfs_end_transaction(trans, root);
7993 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7994 struct btrfs_root *root, u64 type)
7996 u64 alloc_flags = get_alloc_profile(root, type);
7997 return do_chunk_alloc(trans, root, alloc_flags,
8002 * helper to account the unused space of all the readonly block group in the
8003 * list. takes mirrors into account.
8005 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8007 struct btrfs_block_group_cache *block_group;
8011 list_for_each_entry(block_group, groups_list, list) {
8012 spin_lock(&block_group->lock);
8014 if (!block_group->ro) {
8015 spin_unlock(&block_group->lock);
8019 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8020 BTRFS_BLOCK_GROUP_RAID10 |
8021 BTRFS_BLOCK_GROUP_DUP))
8026 free_bytes += (block_group->key.offset -
8027 btrfs_block_group_used(&block_group->item)) *
8030 spin_unlock(&block_group->lock);
8037 * helper to account the unused space of all the readonly block group in the
8038 * space_info. takes mirrors into account.
8040 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8045 spin_lock(&sinfo->lock);
8047 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8048 if (!list_empty(&sinfo->block_groups[i]))
8049 free_bytes += __btrfs_get_ro_block_group_free_space(
8050 &sinfo->block_groups[i]);
8052 spin_unlock(&sinfo->lock);
8057 void btrfs_set_block_group_rw(struct btrfs_root *root,
8058 struct btrfs_block_group_cache *cache)
8060 struct btrfs_space_info *sinfo = cache->space_info;
8065 spin_lock(&sinfo->lock);
8066 spin_lock(&cache->lock);
8067 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8068 cache->bytes_super - btrfs_block_group_used(&cache->item);
8069 sinfo->bytes_readonly -= num_bytes;
8071 spin_unlock(&cache->lock);
8072 spin_unlock(&sinfo->lock);
8076 * checks to see if its even possible to relocate this block group.
8078 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8079 * ok to go ahead and try.
8081 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8083 struct btrfs_block_group_cache *block_group;
8084 struct btrfs_space_info *space_info;
8085 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8086 struct btrfs_device *device;
8087 struct btrfs_trans_handle *trans;
8096 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8098 /* odd, couldn't find the block group, leave it alone */
8102 min_free = btrfs_block_group_used(&block_group->item);
8104 /* no bytes used, we're good */
8108 space_info = block_group->space_info;
8109 spin_lock(&space_info->lock);
8111 full = space_info->full;
8114 * if this is the last block group we have in this space, we can't
8115 * relocate it unless we're able to allocate a new chunk below.
8117 * Otherwise, we need to make sure we have room in the space to handle
8118 * all of the extents from this block group. If we can, we're good
8120 if ((space_info->total_bytes != block_group->key.offset) &&
8121 (space_info->bytes_used + space_info->bytes_reserved +
8122 space_info->bytes_pinned + space_info->bytes_readonly +
8123 min_free < space_info->total_bytes)) {
8124 spin_unlock(&space_info->lock);
8127 spin_unlock(&space_info->lock);
8130 * ok we don't have enough space, but maybe we have free space on our
8131 * devices to allocate new chunks for relocation, so loop through our
8132 * alloc devices and guess if we have enough space. if this block
8133 * group is going to be restriped, run checks against the target
8134 * profile instead of the current one.
8146 target = get_restripe_target(root->fs_info, block_group->flags);
8148 index = __get_raid_index(extended_to_chunk(target));
8151 * this is just a balance, so if we were marked as full
8152 * we know there is no space for a new chunk
8157 index = get_block_group_index(block_group);
8160 if (index == BTRFS_RAID_RAID10) {
8164 } else if (index == BTRFS_RAID_RAID1) {
8166 } else if (index == BTRFS_RAID_DUP) {
8169 } else if (index == BTRFS_RAID_RAID0) {
8170 dev_min = fs_devices->rw_devices;
8171 do_div(min_free, dev_min);
8174 /* We need to do this so that we can look at pending chunks */
8175 trans = btrfs_join_transaction(root);
8176 if (IS_ERR(trans)) {
8177 ret = PTR_ERR(trans);
8181 mutex_lock(&root->fs_info->chunk_mutex);
8182 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8186 * check to make sure we can actually find a chunk with enough
8187 * space to fit our block group in.
8189 if (device->total_bytes > device->bytes_used + min_free &&
8190 !device->is_tgtdev_for_dev_replace) {
8191 ret = find_free_dev_extent(trans, device, min_free,
8196 if (dev_nr >= dev_min)
8202 mutex_unlock(&root->fs_info->chunk_mutex);
8203 btrfs_end_transaction(trans, root);
8205 btrfs_put_block_group(block_group);
8209 static int find_first_block_group(struct btrfs_root *root,
8210 struct btrfs_path *path, struct btrfs_key *key)
8213 struct btrfs_key found_key;
8214 struct extent_buffer *leaf;
8217 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8222 slot = path->slots[0];
8223 leaf = path->nodes[0];
8224 if (slot >= btrfs_header_nritems(leaf)) {
8225 ret = btrfs_next_leaf(root, path);
8232 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8234 if (found_key.objectid >= key->objectid &&
8235 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8245 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8247 struct btrfs_block_group_cache *block_group;
8251 struct inode *inode;
8253 block_group = btrfs_lookup_first_block_group(info, last);
8254 while (block_group) {
8255 spin_lock(&block_group->lock);
8256 if (block_group->iref)
8258 spin_unlock(&block_group->lock);
8259 block_group = next_block_group(info->tree_root,
8269 inode = block_group->inode;
8270 block_group->iref = 0;
8271 block_group->inode = NULL;
8272 spin_unlock(&block_group->lock);
8274 last = block_group->key.objectid + block_group->key.offset;
8275 btrfs_put_block_group(block_group);
8279 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8281 struct btrfs_block_group_cache *block_group;
8282 struct btrfs_space_info *space_info;
8283 struct btrfs_caching_control *caching_ctl;
8286 down_write(&info->extent_commit_sem);
8287 while (!list_empty(&info->caching_block_groups)) {
8288 caching_ctl = list_entry(info->caching_block_groups.next,
8289 struct btrfs_caching_control, list);
8290 list_del(&caching_ctl->list);
8291 put_caching_control(caching_ctl);
8293 up_write(&info->extent_commit_sem);
8295 spin_lock(&info->block_group_cache_lock);
8296 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8297 block_group = rb_entry(n, struct btrfs_block_group_cache,
8299 rb_erase(&block_group->cache_node,
8300 &info->block_group_cache_tree);
8301 spin_unlock(&info->block_group_cache_lock);
8303 down_write(&block_group->space_info->groups_sem);
8304 list_del(&block_group->list);
8305 up_write(&block_group->space_info->groups_sem);
8307 if (block_group->cached == BTRFS_CACHE_STARTED)
8308 wait_block_group_cache_done(block_group);
8311 * We haven't cached this block group, which means we could
8312 * possibly have excluded extents on this block group.
8314 if (block_group->cached == BTRFS_CACHE_NO ||
8315 block_group->cached == BTRFS_CACHE_ERROR)
8316 free_excluded_extents(info->extent_root, block_group);
8318 btrfs_remove_free_space_cache(block_group);
8319 btrfs_put_block_group(block_group);
8321 spin_lock(&info->block_group_cache_lock);
8323 spin_unlock(&info->block_group_cache_lock);
8325 /* now that all the block groups are freed, go through and
8326 * free all the space_info structs. This is only called during
8327 * the final stages of unmount, and so we know nobody is
8328 * using them. We call synchronize_rcu() once before we start,
8329 * just to be on the safe side.
8333 release_global_block_rsv(info);
8335 while (!list_empty(&info->space_info)) {
8336 space_info = list_entry(info->space_info.next,
8337 struct btrfs_space_info,
8339 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8340 if (WARN_ON(space_info->bytes_pinned > 0 ||
8341 space_info->bytes_reserved > 0 ||
8342 space_info->bytes_may_use > 0)) {
8343 dump_space_info(space_info, 0, 0);
8346 percpu_counter_destroy(&space_info->total_bytes_pinned);
8347 list_del(&space_info->list);
8353 static void __link_block_group(struct btrfs_space_info *space_info,
8354 struct btrfs_block_group_cache *cache)
8356 int index = get_block_group_index(cache);
8358 down_write(&space_info->groups_sem);
8359 list_add_tail(&cache->list, &space_info->block_groups[index]);
8360 up_write(&space_info->groups_sem);
8363 int btrfs_read_block_groups(struct btrfs_root *root)
8365 struct btrfs_path *path;
8367 struct btrfs_block_group_cache *cache;
8368 struct btrfs_fs_info *info = root->fs_info;
8369 struct btrfs_space_info *space_info;
8370 struct btrfs_key key;
8371 struct btrfs_key found_key;
8372 struct extent_buffer *leaf;
8376 root = info->extent_root;
8379 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8380 path = btrfs_alloc_path();
8385 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8386 if (btrfs_test_opt(root, SPACE_CACHE) &&
8387 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8389 if (btrfs_test_opt(root, CLEAR_CACHE))
8393 ret = find_first_block_group(root, path, &key);
8398 leaf = path->nodes[0];
8399 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8400 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8405 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8407 if (!cache->free_space_ctl) {
8413 atomic_set(&cache->count, 1);
8414 spin_lock_init(&cache->lock);
8415 cache->fs_info = info;
8416 INIT_LIST_HEAD(&cache->list);
8417 INIT_LIST_HEAD(&cache->cluster_list);
8421 * When we mount with old space cache, we need to
8422 * set BTRFS_DC_CLEAR and set dirty flag.
8424 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8425 * truncate the old free space cache inode and
8427 * b) Setting 'dirty flag' makes sure that we flush
8428 * the new space cache info onto disk.
8430 cache->disk_cache_state = BTRFS_DC_CLEAR;
8431 if (btrfs_test_opt(root, SPACE_CACHE))
8435 read_extent_buffer(leaf, &cache->item,
8436 btrfs_item_ptr_offset(leaf, path->slots[0]),
8437 sizeof(cache->item));
8438 memcpy(&cache->key, &found_key, sizeof(found_key));
8440 key.objectid = found_key.objectid + found_key.offset;
8441 btrfs_release_path(path);
8442 cache->flags = btrfs_block_group_flags(&cache->item);
8443 cache->sectorsize = root->sectorsize;
8444 cache->full_stripe_len = btrfs_full_stripe_len(root,
8445 &root->fs_info->mapping_tree,
8446 found_key.objectid);
8447 btrfs_init_free_space_ctl(cache);
8450 * We need to exclude the super stripes now so that the space
8451 * info has super bytes accounted for, otherwise we'll think
8452 * we have more space than we actually do.
8454 ret = exclude_super_stripes(root, cache);
8457 * We may have excluded something, so call this just in
8460 free_excluded_extents(root, cache);
8461 kfree(cache->free_space_ctl);
8467 * check for two cases, either we are full, and therefore
8468 * don't need to bother with the caching work since we won't
8469 * find any space, or we are empty, and we can just add all
8470 * the space in and be done with it. This saves us _alot_ of
8471 * time, particularly in the full case.
8473 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8474 cache->last_byte_to_unpin = (u64)-1;
8475 cache->cached = BTRFS_CACHE_FINISHED;
8476 free_excluded_extents(root, cache);
8477 } else if (btrfs_block_group_used(&cache->item) == 0) {
8478 cache->last_byte_to_unpin = (u64)-1;
8479 cache->cached = BTRFS_CACHE_FINISHED;
8480 add_new_free_space(cache, root->fs_info,
8482 found_key.objectid +
8484 free_excluded_extents(root, cache);
8487 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8489 btrfs_remove_free_space_cache(cache);
8490 btrfs_put_block_group(cache);
8494 ret = update_space_info(info, cache->flags, found_key.offset,
8495 btrfs_block_group_used(&cache->item),
8498 btrfs_remove_free_space_cache(cache);
8499 spin_lock(&info->block_group_cache_lock);
8500 rb_erase(&cache->cache_node,
8501 &info->block_group_cache_tree);
8502 spin_unlock(&info->block_group_cache_lock);
8503 btrfs_put_block_group(cache);
8507 cache->space_info = space_info;
8508 spin_lock(&cache->space_info->lock);
8509 cache->space_info->bytes_readonly += cache->bytes_super;
8510 spin_unlock(&cache->space_info->lock);
8512 __link_block_group(space_info, cache);
8514 set_avail_alloc_bits(root->fs_info, cache->flags);
8515 if (btrfs_chunk_readonly(root, cache->key.objectid))
8516 set_block_group_ro(cache, 1);
8519 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8520 if (!(get_alloc_profile(root, space_info->flags) &
8521 (BTRFS_BLOCK_GROUP_RAID10 |
8522 BTRFS_BLOCK_GROUP_RAID1 |
8523 BTRFS_BLOCK_GROUP_RAID5 |
8524 BTRFS_BLOCK_GROUP_RAID6 |
8525 BTRFS_BLOCK_GROUP_DUP)))
8528 * avoid allocating from un-mirrored block group if there are
8529 * mirrored block groups.
8531 list_for_each_entry(cache,
8532 &space_info->block_groups[BTRFS_RAID_RAID0],
8534 set_block_group_ro(cache, 1);
8535 list_for_each_entry(cache,
8536 &space_info->block_groups[BTRFS_RAID_SINGLE],
8538 set_block_group_ro(cache, 1);
8541 init_global_block_rsv(info);
8544 btrfs_free_path(path);
8548 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8549 struct btrfs_root *root)
8551 struct btrfs_block_group_cache *block_group, *tmp;
8552 struct btrfs_root *extent_root = root->fs_info->extent_root;
8553 struct btrfs_block_group_item item;
8554 struct btrfs_key key;
8557 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8559 list_del_init(&block_group->new_bg_list);
8564 spin_lock(&block_group->lock);
8565 memcpy(&item, &block_group->item, sizeof(item));
8566 memcpy(&key, &block_group->key, sizeof(key));
8567 spin_unlock(&block_group->lock);
8569 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8572 btrfs_abort_transaction(trans, extent_root, ret);
8573 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8574 key.objectid, key.offset);
8576 btrfs_abort_transaction(trans, extent_root, ret);
8580 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8581 struct btrfs_root *root, u64 bytes_used,
8582 u64 type, u64 chunk_objectid, u64 chunk_offset,
8586 struct btrfs_root *extent_root;
8587 struct btrfs_block_group_cache *cache;
8589 extent_root = root->fs_info->extent_root;
8591 root->fs_info->last_trans_log_full_commit = trans->transid;
8593 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8596 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8598 if (!cache->free_space_ctl) {
8603 cache->key.objectid = chunk_offset;
8604 cache->key.offset = size;
8605 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8606 cache->sectorsize = root->sectorsize;
8607 cache->fs_info = root->fs_info;
8608 cache->full_stripe_len = btrfs_full_stripe_len(root,
8609 &root->fs_info->mapping_tree,
8612 atomic_set(&cache->count, 1);
8613 spin_lock_init(&cache->lock);
8614 INIT_LIST_HEAD(&cache->list);
8615 INIT_LIST_HEAD(&cache->cluster_list);
8616 INIT_LIST_HEAD(&cache->new_bg_list);
8618 btrfs_init_free_space_ctl(cache);
8620 btrfs_set_block_group_used(&cache->item, bytes_used);
8621 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8622 cache->flags = type;
8623 btrfs_set_block_group_flags(&cache->item, type);
8625 cache->last_byte_to_unpin = (u64)-1;
8626 cache->cached = BTRFS_CACHE_FINISHED;
8627 ret = exclude_super_stripes(root, cache);
8630 * We may have excluded something, so call this just in
8633 free_excluded_extents(root, cache);
8634 kfree(cache->free_space_ctl);
8639 add_new_free_space(cache, root->fs_info, chunk_offset,
8640 chunk_offset + size);
8642 free_excluded_extents(root, cache);
8644 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8646 btrfs_remove_free_space_cache(cache);
8647 btrfs_put_block_group(cache);
8651 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8652 &cache->space_info);
8654 btrfs_remove_free_space_cache(cache);
8655 spin_lock(&root->fs_info->block_group_cache_lock);
8656 rb_erase(&cache->cache_node,
8657 &root->fs_info->block_group_cache_tree);
8658 spin_unlock(&root->fs_info->block_group_cache_lock);
8659 btrfs_put_block_group(cache);
8662 update_global_block_rsv(root->fs_info);
8664 spin_lock(&cache->space_info->lock);
8665 cache->space_info->bytes_readonly += cache->bytes_super;
8666 spin_unlock(&cache->space_info->lock);
8668 __link_block_group(cache->space_info, cache);
8670 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8672 set_avail_alloc_bits(extent_root->fs_info, type);
8677 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8679 u64 extra_flags = chunk_to_extended(flags) &
8680 BTRFS_EXTENDED_PROFILE_MASK;
8682 write_seqlock(&fs_info->profiles_lock);
8683 if (flags & BTRFS_BLOCK_GROUP_DATA)
8684 fs_info->avail_data_alloc_bits &= ~extra_flags;
8685 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8686 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8687 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8688 fs_info->avail_system_alloc_bits &= ~extra_flags;
8689 write_sequnlock(&fs_info->profiles_lock);
8692 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8693 struct btrfs_root *root, u64 group_start)
8695 struct btrfs_path *path;
8696 struct btrfs_block_group_cache *block_group;
8697 struct btrfs_free_cluster *cluster;
8698 struct btrfs_root *tree_root = root->fs_info->tree_root;
8699 struct btrfs_key key;
8700 struct inode *inode;
8705 root = root->fs_info->extent_root;
8707 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8708 BUG_ON(!block_group);
8709 BUG_ON(!block_group->ro);
8712 * Free the reserved super bytes from this block group before
8715 free_excluded_extents(root, block_group);
8717 memcpy(&key, &block_group->key, sizeof(key));
8718 index = get_block_group_index(block_group);
8719 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8720 BTRFS_BLOCK_GROUP_RAID1 |
8721 BTRFS_BLOCK_GROUP_RAID10))
8726 /* make sure this block group isn't part of an allocation cluster */
8727 cluster = &root->fs_info->data_alloc_cluster;
8728 spin_lock(&cluster->refill_lock);
8729 btrfs_return_cluster_to_free_space(block_group, cluster);
8730 spin_unlock(&cluster->refill_lock);
8733 * make sure this block group isn't part of a metadata
8734 * allocation cluster
8736 cluster = &root->fs_info->meta_alloc_cluster;
8737 spin_lock(&cluster->refill_lock);
8738 btrfs_return_cluster_to_free_space(block_group, cluster);
8739 spin_unlock(&cluster->refill_lock);
8741 path = btrfs_alloc_path();
8747 inode = lookup_free_space_inode(tree_root, block_group, path);
8748 if (!IS_ERR(inode)) {
8749 ret = btrfs_orphan_add(trans, inode);
8751 btrfs_add_delayed_iput(inode);
8755 /* One for the block groups ref */
8756 spin_lock(&block_group->lock);
8757 if (block_group->iref) {
8758 block_group->iref = 0;
8759 block_group->inode = NULL;
8760 spin_unlock(&block_group->lock);
8763 spin_unlock(&block_group->lock);
8765 /* One for our lookup ref */
8766 btrfs_add_delayed_iput(inode);
8769 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8770 key.offset = block_group->key.objectid;
8773 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8777 btrfs_release_path(path);
8779 ret = btrfs_del_item(trans, tree_root, path);
8782 btrfs_release_path(path);
8785 spin_lock(&root->fs_info->block_group_cache_lock);
8786 rb_erase(&block_group->cache_node,
8787 &root->fs_info->block_group_cache_tree);
8789 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8790 root->fs_info->first_logical_byte = (u64)-1;
8791 spin_unlock(&root->fs_info->block_group_cache_lock);
8793 down_write(&block_group->space_info->groups_sem);
8795 * we must use list_del_init so people can check to see if they
8796 * are still on the list after taking the semaphore
8798 list_del_init(&block_group->list);
8799 if (list_empty(&block_group->space_info->block_groups[index]))
8800 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8801 up_write(&block_group->space_info->groups_sem);
8803 if (block_group->cached == BTRFS_CACHE_STARTED)
8804 wait_block_group_cache_done(block_group);
8806 btrfs_remove_free_space_cache(block_group);
8808 spin_lock(&block_group->space_info->lock);
8809 block_group->space_info->total_bytes -= block_group->key.offset;
8810 block_group->space_info->bytes_readonly -= block_group->key.offset;
8811 block_group->space_info->disk_total -= block_group->key.offset * factor;
8812 spin_unlock(&block_group->space_info->lock);
8814 memcpy(&key, &block_group->key, sizeof(key));
8816 btrfs_clear_space_info_full(root->fs_info);
8818 btrfs_put_block_group(block_group);
8819 btrfs_put_block_group(block_group);
8821 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8827 ret = btrfs_del_item(trans, root, path);
8829 btrfs_free_path(path);
8833 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8835 struct btrfs_space_info *space_info;
8836 struct btrfs_super_block *disk_super;
8842 disk_super = fs_info->super_copy;
8843 if (!btrfs_super_root(disk_super))
8846 features = btrfs_super_incompat_flags(disk_super);
8847 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8850 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8851 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8856 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8857 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8859 flags = BTRFS_BLOCK_GROUP_METADATA;
8860 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8864 flags = BTRFS_BLOCK_GROUP_DATA;
8865 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8871 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8873 return unpin_extent_range(root, start, end);
8876 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8877 u64 num_bytes, u64 *actual_bytes)
8879 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8882 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8884 struct btrfs_fs_info *fs_info = root->fs_info;
8885 struct btrfs_block_group_cache *cache = NULL;
8890 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8894 * try to trim all FS space, our block group may start from non-zero.
8896 if (range->len == total_bytes)
8897 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8899 cache = btrfs_lookup_block_group(fs_info, range->start);
8902 if (cache->key.objectid >= (range->start + range->len)) {
8903 btrfs_put_block_group(cache);
8907 start = max(range->start, cache->key.objectid);
8908 end = min(range->start + range->len,
8909 cache->key.objectid + cache->key.offset);
8911 if (end - start >= range->minlen) {
8912 if (!block_group_cache_done(cache)) {
8913 ret = cache_block_group(cache, 0);
8915 btrfs_put_block_group(cache);
8918 ret = wait_block_group_cache_done(cache);
8920 btrfs_put_block_group(cache);
8924 ret = btrfs_trim_block_group(cache,
8930 trimmed += group_trimmed;
8932 btrfs_put_block_group(cache);
8937 cache = next_block_group(fs_info->tree_root, cache);
8940 range->len = trimmed;