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>
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,
110 block_group_cache_done(struct btrfs_block_group_cache *cache)
113 return cache->cached == BTRFS_CACHE_FINISHED;
116 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
118 return (cache->flags & bits) == bits;
121 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
123 atomic_inc(&cache->count);
126 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
128 if (atomic_dec_and_test(&cache->count)) {
129 WARN_ON(cache->pinned > 0);
130 WARN_ON(cache->reserved > 0);
131 kfree(cache->free_space_ctl);
137 * this adds the block group to the fs_info rb tree for the block group
140 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
141 struct btrfs_block_group_cache *block_group)
144 struct rb_node *parent = NULL;
145 struct btrfs_block_group_cache *cache;
147 spin_lock(&info->block_group_cache_lock);
148 p = &info->block_group_cache_tree.rb_node;
152 cache = rb_entry(parent, struct btrfs_block_group_cache,
154 if (block_group->key.objectid < cache->key.objectid) {
156 } else if (block_group->key.objectid > cache->key.objectid) {
159 spin_unlock(&info->block_group_cache_lock);
164 rb_link_node(&block_group->cache_node, parent, p);
165 rb_insert_color(&block_group->cache_node,
166 &info->block_group_cache_tree);
168 if (info->first_logical_byte > block_group->key.objectid)
169 info->first_logical_byte = block_group->key.objectid;
171 spin_unlock(&info->block_group_cache_lock);
177 * This will return the block group at or after bytenr if contains is 0, else
178 * it will return the block group that contains the bytenr
180 static struct btrfs_block_group_cache *
181 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
184 struct btrfs_block_group_cache *cache, *ret = NULL;
188 spin_lock(&info->block_group_cache_lock);
189 n = info->block_group_cache_tree.rb_node;
192 cache = rb_entry(n, struct btrfs_block_group_cache,
194 end = cache->key.objectid + cache->key.offset - 1;
195 start = cache->key.objectid;
197 if (bytenr < start) {
198 if (!contains && (!ret || start < ret->key.objectid))
201 } else if (bytenr > start) {
202 if (contains && bytenr <= end) {
213 btrfs_get_block_group(ret);
214 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
215 info->first_logical_byte = ret->key.objectid;
217 spin_unlock(&info->block_group_cache_lock);
222 static int add_excluded_extent(struct btrfs_root *root,
223 u64 start, u64 num_bytes)
225 u64 end = start + num_bytes - 1;
226 set_extent_bits(&root->fs_info->freed_extents[0],
227 start, end, EXTENT_UPTODATE, GFP_NOFS);
228 set_extent_bits(&root->fs_info->freed_extents[1],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 static void free_excluded_extents(struct btrfs_root *root,
234 struct btrfs_block_group_cache *cache)
238 start = cache->key.objectid;
239 end = start + cache->key.offset - 1;
241 clear_extent_bits(&root->fs_info->freed_extents[0],
242 start, end, EXTENT_UPTODATE, GFP_NOFS);
243 clear_extent_bits(&root->fs_info->freed_extents[1],
244 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 static int exclude_super_stripes(struct btrfs_root *root,
248 struct btrfs_block_group_cache *cache)
255 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
256 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
257 cache->bytes_super += stripe_len;
258 ret = add_excluded_extent(root, cache->key.objectid,
264 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
265 bytenr = btrfs_sb_offset(i);
266 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
267 cache->key.objectid, bytenr,
268 0, &logical, &nr, &stripe_len);
273 cache->bytes_super += stripe_len;
274 ret = add_excluded_extent(root, logical[nr],
287 static struct btrfs_caching_control *
288 get_caching_control(struct btrfs_block_group_cache *cache)
290 struct btrfs_caching_control *ctl;
292 spin_lock(&cache->lock);
293 if (cache->cached != BTRFS_CACHE_STARTED) {
294 spin_unlock(&cache->lock);
298 /* We're loading it the fast way, so we don't have a caching_ctl. */
299 if (!cache->caching_ctl) {
300 spin_unlock(&cache->lock);
304 ctl = cache->caching_ctl;
305 atomic_inc(&ctl->count);
306 spin_unlock(&cache->lock);
310 static void put_caching_control(struct btrfs_caching_control *ctl)
312 if (atomic_dec_and_test(&ctl->count))
317 * this is only called by cache_block_group, since we could have freed extents
318 * we need to check the pinned_extents for any extents that can't be used yet
319 * since their free space will be released as soon as the transaction commits.
321 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
322 struct btrfs_fs_info *info, u64 start, u64 end)
324 u64 extent_start, extent_end, size, total_added = 0;
327 while (start < end) {
328 ret = find_first_extent_bit(info->pinned_extents, start,
329 &extent_start, &extent_end,
330 EXTENT_DIRTY | EXTENT_UPTODATE,
335 if (extent_start <= start) {
336 start = extent_end + 1;
337 } else if (extent_start > start && extent_start < end) {
338 size = extent_start - start;
340 ret = btrfs_add_free_space(block_group, start,
342 BUG_ON(ret); /* -ENOMEM or logic error */
343 start = extent_end + 1;
352 ret = btrfs_add_free_space(block_group, start, size);
353 BUG_ON(ret); /* -ENOMEM or logic error */
359 static noinline void caching_thread(struct btrfs_work *work)
361 struct btrfs_block_group_cache *block_group;
362 struct btrfs_fs_info *fs_info;
363 struct btrfs_caching_control *caching_ctl;
364 struct btrfs_root *extent_root;
365 struct btrfs_path *path;
366 struct extent_buffer *leaf;
367 struct btrfs_key key;
373 caching_ctl = container_of(work, struct btrfs_caching_control, work);
374 block_group = caching_ctl->block_group;
375 fs_info = block_group->fs_info;
376 extent_root = fs_info->extent_root;
378 path = btrfs_alloc_path();
382 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
385 * We don't want to deadlock with somebody trying to allocate a new
386 * extent for the extent root while also trying to search the extent
387 * root to add free space. So we skip locking and search the commit
388 * root, since its read-only
390 path->skip_locking = 1;
391 path->search_commit_root = 1;
396 key.type = BTRFS_EXTENT_ITEM_KEY;
398 mutex_lock(&caching_ctl->mutex);
399 /* need to make sure the commit_root doesn't disappear */
400 down_read(&fs_info->extent_commit_sem);
402 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
406 leaf = path->nodes[0];
407 nritems = btrfs_header_nritems(leaf);
410 if (btrfs_fs_closing(fs_info) > 1) {
415 if (path->slots[0] < nritems) {
416 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
418 ret = find_next_key(path, 0, &key);
422 if (need_resched()) {
423 caching_ctl->progress = last;
424 btrfs_release_path(path);
425 up_read(&fs_info->extent_commit_sem);
426 mutex_unlock(&caching_ctl->mutex);
431 ret = btrfs_next_leaf(extent_root, path);
436 leaf = path->nodes[0];
437 nritems = btrfs_header_nritems(leaf);
441 if (key.objectid < block_group->key.objectid) {
446 if (key.objectid >= block_group->key.objectid +
447 block_group->key.offset)
450 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
451 key.type == BTRFS_METADATA_ITEM_KEY) {
452 total_found += add_new_free_space(block_group,
455 if (key.type == BTRFS_METADATA_ITEM_KEY)
456 last = key.objectid +
457 fs_info->tree_root->leafsize;
459 last = key.objectid + key.offset;
461 if (total_found > (1024 * 1024 * 2)) {
463 wake_up(&caching_ctl->wait);
470 total_found += add_new_free_space(block_group, fs_info, last,
471 block_group->key.objectid +
472 block_group->key.offset);
473 caching_ctl->progress = (u64)-1;
475 spin_lock(&block_group->lock);
476 block_group->caching_ctl = NULL;
477 block_group->cached = BTRFS_CACHE_FINISHED;
478 spin_unlock(&block_group->lock);
481 btrfs_free_path(path);
482 up_read(&fs_info->extent_commit_sem);
484 free_excluded_extents(extent_root, block_group);
486 mutex_unlock(&caching_ctl->mutex);
488 wake_up(&caching_ctl->wait);
490 put_caching_control(caching_ctl);
491 btrfs_put_block_group(block_group);
494 static int cache_block_group(struct btrfs_block_group_cache *cache,
498 struct btrfs_fs_info *fs_info = cache->fs_info;
499 struct btrfs_caching_control *caching_ctl;
502 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
506 INIT_LIST_HEAD(&caching_ctl->list);
507 mutex_init(&caching_ctl->mutex);
508 init_waitqueue_head(&caching_ctl->wait);
509 caching_ctl->block_group = cache;
510 caching_ctl->progress = cache->key.objectid;
511 atomic_set(&caching_ctl->count, 1);
512 caching_ctl->work.func = caching_thread;
514 spin_lock(&cache->lock);
516 * This should be a rare occasion, but this could happen I think in the
517 * case where one thread starts to load the space cache info, and then
518 * some other thread starts a transaction commit which tries to do an
519 * allocation while the other thread is still loading the space cache
520 * info. The previous loop should have kept us from choosing this block
521 * group, but if we've moved to the state where we will wait on caching
522 * block groups we need to first check if we're doing a fast load here,
523 * so we can wait for it to finish, otherwise we could end up allocating
524 * from a block group who's cache gets evicted for one reason or
527 while (cache->cached == BTRFS_CACHE_FAST) {
528 struct btrfs_caching_control *ctl;
530 ctl = cache->caching_ctl;
531 atomic_inc(&ctl->count);
532 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
533 spin_unlock(&cache->lock);
537 finish_wait(&ctl->wait, &wait);
538 put_caching_control(ctl);
539 spin_lock(&cache->lock);
542 if (cache->cached != BTRFS_CACHE_NO) {
543 spin_unlock(&cache->lock);
547 WARN_ON(cache->caching_ctl);
548 cache->caching_ctl = caching_ctl;
549 cache->cached = BTRFS_CACHE_FAST;
550 spin_unlock(&cache->lock);
552 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
553 ret = load_free_space_cache(fs_info, cache);
555 spin_lock(&cache->lock);
557 cache->caching_ctl = NULL;
558 cache->cached = BTRFS_CACHE_FINISHED;
559 cache->last_byte_to_unpin = (u64)-1;
561 if (load_cache_only) {
562 cache->caching_ctl = NULL;
563 cache->cached = BTRFS_CACHE_NO;
565 cache->cached = BTRFS_CACHE_STARTED;
568 spin_unlock(&cache->lock);
569 wake_up(&caching_ctl->wait);
571 put_caching_control(caching_ctl);
572 free_excluded_extents(fs_info->extent_root, cache);
577 * We are not going to do the fast caching, set cached to the
578 * appropriate value and wakeup any waiters.
580 spin_lock(&cache->lock);
581 if (load_cache_only) {
582 cache->caching_ctl = NULL;
583 cache->cached = BTRFS_CACHE_NO;
585 cache->cached = BTRFS_CACHE_STARTED;
587 spin_unlock(&cache->lock);
588 wake_up(&caching_ctl->wait);
591 if (load_cache_only) {
592 put_caching_control(caching_ctl);
596 down_write(&fs_info->extent_commit_sem);
597 atomic_inc(&caching_ctl->count);
598 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
599 up_write(&fs_info->extent_commit_sem);
601 btrfs_get_block_group(cache);
603 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
609 * return the block group that starts at or after bytenr
611 static struct btrfs_block_group_cache *
612 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
614 struct btrfs_block_group_cache *cache;
616 cache = block_group_cache_tree_search(info, bytenr, 0);
622 * return the block group that contains the given bytenr
624 struct btrfs_block_group_cache *btrfs_lookup_block_group(
625 struct btrfs_fs_info *info,
628 struct btrfs_block_group_cache *cache;
630 cache = block_group_cache_tree_search(info, bytenr, 1);
635 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
638 struct list_head *head = &info->space_info;
639 struct btrfs_space_info *found;
641 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
644 list_for_each_entry_rcu(found, head, list) {
645 if (found->flags & flags) {
655 * after adding space to the filesystem, we need to clear the full flags
656 * on all the space infos.
658 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
660 struct list_head *head = &info->space_info;
661 struct btrfs_space_info *found;
664 list_for_each_entry_rcu(found, head, list)
669 u64 btrfs_find_block_group(struct btrfs_root *root,
670 u64 search_start, u64 search_hint, int owner)
672 struct btrfs_block_group_cache *cache;
674 u64 last = max(search_hint, search_start);
681 cache = btrfs_lookup_first_block_group(root->fs_info, last);
685 spin_lock(&cache->lock);
686 last = cache->key.objectid + cache->key.offset;
687 used = btrfs_block_group_used(&cache->item);
689 if ((full_search || !cache->ro) &&
690 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
691 if (used + cache->pinned + cache->reserved <
692 div_factor(cache->key.offset, factor)) {
693 group_start = cache->key.objectid;
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
699 spin_unlock(&cache->lock);
700 btrfs_put_block_group(cache);
708 if (!full_search && factor < 10) {
718 /* simple helper to search for an existing extent at a given offset */
719 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
722 struct btrfs_key key;
723 struct btrfs_path *path;
725 path = btrfs_alloc_path();
729 key.objectid = start;
731 key.type = BTRFS_EXTENT_ITEM_KEY;
732 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
735 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
736 if (key.objectid == start &&
737 key.type == BTRFS_METADATA_ITEM_KEY)
740 btrfs_free_path(path);
745 * helper function to lookup reference count and flags of a tree block.
747 * the head node for delayed ref is used to store the sum of all the
748 * reference count modifications queued up in the rbtree. the head
749 * node may also store the extent flags to set. This way you can check
750 * to see what the reference count and extent flags would be if all of
751 * the delayed refs are not processed.
753 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
754 struct btrfs_root *root, u64 bytenr,
755 u64 offset, int metadata, u64 *refs, u64 *flags)
757 struct btrfs_delayed_ref_head *head;
758 struct btrfs_delayed_ref_root *delayed_refs;
759 struct btrfs_path *path;
760 struct btrfs_extent_item *ei;
761 struct extent_buffer *leaf;
762 struct btrfs_key key;
769 * If we don't have skinny metadata, don't bother doing anything
772 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
773 offset = root->leafsize;
777 path = btrfs_alloc_path();
782 key.objectid = bytenr;
783 key.type = BTRFS_METADATA_ITEM_KEY;
786 key.objectid = bytenr;
787 key.type = BTRFS_EXTENT_ITEM_KEY;
792 path->skip_locking = 1;
793 path->search_commit_root = 1;
796 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
801 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
802 key.type = BTRFS_EXTENT_ITEM_KEY;
803 key.offset = root->leafsize;
804 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) {
1557 leaf = path->nodes[0];
1558 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1559 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1560 if (item_size < sizeof(*ei)) {
1565 ret = convert_extent_item_v0(trans, root, path, owner,
1571 leaf = path->nodes[0];
1572 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1575 BUG_ON(item_size < sizeof(*ei));
1577 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1578 flags = btrfs_extent_flags(leaf, ei);
1580 ptr = (unsigned long)(ei + 1);
1581 end = (unsigned long)ei + item_size;
1583 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1584 ptr += sizeof(struct btrfs_tree_block_info);
1594 iref = (struct btrfs_extent_inline_ref *)ptr;
1595 type = btrfs_extent_inline_ref_type(leaf, iref);
1599 ptr += btrfs_extent_inline_ref_size(type);
1603 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1604 struct btrfs_extent_data_ref *dref;
1605 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1606 if (match_extent_data_ref(leaf, dref, root_objectid,
1611 if (hash_extent_data_ref_item(leaf, dref) <
1612 hash_extent_data_ref(root_objectid, owner, offset))
1616 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1618 if (parent == ref_offset) {
1622 if (ref_offset < parent)
1625 if (root_objectid == ref_offset) {
1629 if (ref_offset < root_objectid)
1633 ptr += btrfs_extent_inline_ref_size(type);
1635 if (err == -ENOENT && insert) {
1636 if (item_size + extra_size >=
1637 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1642 * To add new inline back ref, we have to make sure
1643 * there is no corresponding back ref item.
1644 * For simplicity, we just do not add new inline back
1645 * ref if there is any kind of item for this block
1647 if (find_next_key(path, 0, &key) == 0 &&
1648 key.objectid == bytenr &&
1649 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1654 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1657 path->keep_locks = 0;
1658 btrfs_unlock_up_safe(path, 1);
1664 * helper to add new inline back ref
1666 static noinline_for_stack
1667 void setup_inline_extent_backref(struct btrfs_root *root,
1668 struct btrfs_path *path,
1669 struct btrfs_extent_inline_ref *iref,
1670 u64 parent, u64 root_objectid,
1671 u64 owner, u64 offset, int refs_to_add,
1672 struct btrfs_delayed_extent_op *extent_op)
1674 struct extent_buffer *leaf;
1675 struct btrfs_extent_item *ei;
1678 unsigned long item_offset;
1683 leaf = path->nodes[0];
1684 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1685 item_offset = (unsigned long)iref - (unsigned long)ei;
1687 type = extent_ref_type(parent, owner);
1688 size = btrfs_extent_inline_ref_size(type);
1690 btrfs_extend_item(root, path, size);
1692 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1693 refs = btrfs_extent_refs(leaf, ei);
1694 refs += refs_to_add;
1695 btrfs_set_extent_refs(leaf, ei, refs);
1697 __run_delayed_extent_op(extent_op, leaf, ei);
1699 ptr = (unsigned long)ei + item_offset;
1700 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1701 if (ptr < end - size)
1702 memmove_extent_buffer(leaf, ptr + size, ptr,
1705 iref = (struct btrfs_extent_inline_ref *)ptr;
1706 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1707 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1708 struct btrfs_extent_data_ref *dref;
1709 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1710 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1711 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1712 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1713 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1714 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1715 struct btrfs_shared_data_ref *sref;
1716 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1717 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1719 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1720 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1724 btrfs_mark_buffer_dirty(leaf);
1727 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1728 struct btrfs_root *root,
1729 struct btrfs_path *path,
1730 struct btrfs_extent_inline_ref **ref_ret,
1731 u64 bytenr, u64 num_bytes, u64 parent,
1732 u64 root_objectid, u64 owner, u64 offset)
1736 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1737 bytenr, num_bytes, parent,
1738 root_objectid, owner, offset, 0);
1742 btrfs_release_path(path);
1745 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1746 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1749 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1750 root_objectid, owner, offset);
1756 * helper to update/remove inline back ref
1758 static noinline_for_stack
1759 void update_inline_extent_backref(struct btrfs_root *root,
1760 struct btrfs_path *path,
1761 struct btrfs_extent_inline_ref *iref,
1763 struct btrfs_delayed_extent_op *extent_op)
1765 struct extent_buffer *leaf;
1766 struct btrfs_extent_item *ei;
1767 struct btrfs_extent_data_ref *dref = NULL;
1768 struct btrfs_shared_data_ref *sref = NULL;
1776 leaf = path->nodes[0];
1777 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1778 refs = btrfs_extent_refs(leaf, ei);
1779 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1780 refs += refs_to_mod;
1781 btrfs_set_extent_refs(leaf, ei, refs);
1783 __run_delayed_extent_op(extent_op, leaf, ei);
1785 type = btrfs_extent_inline_ref_type(leaf, iref);
1787 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1788 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1789 refs = btrfs_extent_data_ref_count(leaf, dref);
1790 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1791 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1792 refs = btrfs_shared_data_ref_count(leaf, sref);
1795 BUG_ON(refs_to_mod != -1);
1798 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1799 refs += refs_to_mod;
1802 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1803 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1805 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1807 size = btrfs_extent_inline_ref_size(type);
1808 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1809 ptr = (unsigned long)iref;
1810 end = (unsigned long)ei + item_size;
1811 if (ptr + size < end)
1812 memmove_extent_buffer(leaf, ptr, ptr + size,
1815 btrfs_truncate_item(root, path, item_size, 1);
1817 btrfs_mark_buffer_dirty(leaf);
1820 static noinline_for_stack
1821 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1822 struct btrfs_root *root,
1823 struct btrfs_path *path,
1824 u64 bytenr, u64 num_bytes, u64 parent,
1825 u64 root_objectid, u64 owner,
1826 u64 offset, int refs_to_add,
1827 struct btrfs_delayed_extent_op *extent_op)
1829 struct btrfs_extent_inline_ref *iref;
1832 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1833 bytenr, num_bytes, parent,
1834 root_objectid, owner, offset, 1);
1836 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1837 update_inline_extent_backref(root, path, iref,
1838 refs_to_add, extent_op);
1839 } else if (ret == -ENOENT) {
1840 setup_inline_extent_backref(root, path, iref, parent,
1841 root_objectid, owner, offset,
1842 refs_to_add, extent_op);
1848 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1849 struct btrfs_root *root,
1850 struct btrfs_path *path,
1851 u64 bytenr, u64 parent, u64 root_objectid,
1852 u64 owner, u64 offset, int refs_to_add)
1855 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1856 BUG_ON(refs_to_add != 1);
1857 ret = insert_tree_block_ref(trans, root, path, bytenr,
1858 parent, root_objectid);
1860 ret = insert_extent_data_ref(trans, root, path, bytenr,
1861 parent, root_objectid,
1862 owner, offset, refs_to_add);
1867 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1868 struct btrfs_root *root,
1869 struct btrfs_path *path,
1870 struct btrfs_extent_inline_ref *iref,
1871 int refs_to_drop, int is_data)
1875 BUG_ON(!is_data && refs_to_drop != 1);
1877 update_inline_extent_backref(root, path, iref,
1878 -refs_to_drop, NULL);
1879 } else if (is_data) {
1880 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1882 ret = btrfs_del_item(trans, root, path);
1887 static int btrfs_issue_discard(struct block_device *bdev,
1890 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1893 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894 u64 num_bytes, u64 *actual_bytes)
1897 u64 discarded_bytes = 0;
1898 struct btrfs_bio *bbio = NULL;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903 bytenr, &num_bytes, &bbio, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911 if (!stripe->dev->can_discard)
1914 ret = btrfs_issue_discard(stripe->dev->bdev,
1918 discarded_bytes += stripe->length;
1919 else if (ret != -EOPNOTSUPP)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1933 *actual_bytes = discarded_bytes;
1936 if (ret == -EOPNOTSUPP)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 u64 bytenr, u64 num_bytes, u64 parent,
1945 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1948 struct btrfs_fs_info *fs_info = root->fs_info;
1950 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956 parent, root_objectid, (int)owner,
1957 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1959 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961 parent, root_objectid, owner, offset,
1962 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root,
1969 u64 bytenr, u64 num_bytes,
1970 u64 parent, u64 root_objectid,
1971 u64 owner, u64 offset, int refs_to_add,
1972 struct btrfs_delayed_extent_op *extent_op)
1974 struct btrfs_path *path;
1975 struct extent_buffer *leaf;
1976 struct btrfs_extent_item *item;
1981 path = btrfs_alloc_path();
1986 path->leave_spinning = 1;
1987 /* this will setup the path even if it fails to insert the back ref */
1988 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1989 path, bytenr, num_bytes, parent,
1990 root_objectid, owner, offset,
1991 refs_to_add, extent_op);
1995 if (ret != -EAGAIN) {
2000 leaf = path->nodes[0];
2001 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2002 refs = btrfs_extent_refs(leaf, item);
2003 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2005 __run_delayed_extent_op(extent_op, leaf, item);
2007 btrfs_mark_buffer_dirty(leaf);
2008 btrfs_release_path(path);
2011 path->leave_spinning = 1;
2013 /* now insert the actual backref */
2014 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2015 path, bytenr, parent, root_objectid,
2016 owner, offset, refs_to_add);
2018 btrfs_abort_transaction(trans, root, ret);
2020 btrfs_free_path(path);
2024 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2025 struct btrfs_root *root,
2026 struct btrfs_delayed_ref_node *node,
2027 struct btrfs_delayed_extent_op *extent_op,
2028 int insert_reserved)
2031 struct btrfs_delayed_data_ref *ref;
2032 struct btrfs_key ins;
2037 ins.objectid = node->bytenr;
2038 ins.offset = node->num_bytes;
2039 ins.type = BTRFS_EXTENT_ITEM_KEY;
2041 ref = btrfs_delayed_node_to_data_ref(node);
2042 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2043 parent = ref->parent;
2045 ref_root = ref->root;
2047 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2049 flags |= extent_op->flags_to_set;
2050 ret = alloc_reserved_file_extent(trans, root,
2051 parent, ref_root, flags,
2052 ref->objectid, ref->offset,
2053 &ins, node->ref_mod);
2054 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2055 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2056 node->num_bytes, parent,
2057 ref_root, ref->objectid,
2058 ref->offset, node->ref_mod,
2060 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2061 ret = __btrfs_free_extent(trans, root, node->bytenr,
2062 node->num_bytes, parent,
2063 ref_root, ref->objectid,
2064 ref->offset, node->ref_mod,
2072 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2073 struct extent_buffer *leaf,
2074 struct btrfs_extent_item *ei)
2076 u64 flags = btrfs_extent_flags(leaf, ei);
2077 if (extent_op->update_flags) {
2078 flags |= extent_op->flags_to_set;
2079 btrfs_set_extent_flags(leaf, ei, flags);
2082 if (extent_op->update_key) {
2083 struct btrfs_tree_block_info *bi;
2084 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2085 bi = (struct btrfs_tree_block_info *)(ei + 1);
2086 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2090 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2091 struct btrfs_root *root,
2092 struct btrfs_delayed_ref_node *node,
2093 struct btrfs_delayed_extent_op *extent_op)
2095 struct btrfs_key key;
2096 struct btrfs_path *path;
2097 struct btrfs_extent_item *ei;
2098 struct extent_buffer *leaf;
2102 int metadata = (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2103 node->type == BTRFS_SHARED_BLOCK_REF_KEY);
2108 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2111 path = btrfs_alloc_path();
2115 key.objectid = node->bytenr;
2118 struct btrfs_delayed_tree_ref *tree_ref;
2120 tree_ref = btrfs_delayed_node_to_tree_ref(node);
2121 key.type = BTRFS_METADATA_ITEM_KEY;
2122 key.offset = tree_ref->level;
2124 key.type = BTRFS_EXTENT_ITEM_KEY;
2125 key.offset = node->num_bytes;
2130 path->leave_spinning = 1;
2131 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2139 btrfs_release_path(path);
2142 key.offset = node->num_bytes;
2143 key.type = BTRFS_EXTENT_ITEM_KEY;
2150 leaf = path->nodes[0];
2151 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2152 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2153 if (item_size < sizeof(*ei)) {
2154 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2160 leaf = path->nodes[0];
2161 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2164 BUG_ON(item_size < sizeof(*ei));
2165 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2166 __run_delayed_extent_op(extent_op, leaf, ei);
2168 btrfs_mark_buffer_dirty(leaf);
2170 btrfs_free_path(path);
2174 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2175 struct btrfs_root *root,
2176 struct btrfs_delayed_ref_node *node,
2177 struct btrfs_delayed_extent_op *extent_op,
2178 int insert_reserved)
2181 struct btrfs_delayed_tree_ref *ref;
2182 struct btrfs_key ins;
2185 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2188 ref = btrfs_delayed_node_to_tree_ref(node);
2189 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2190 parent = ref->parent;
2192 ref_root = ref->root;
2194 ins.objectid = node->bytenr;
2195 if (skinny_metadata) {
2196 ins.offset = ref->level;
2197 ins.type = BTRFS_METADATA_ITEM_KEY;
2199 ins.offset = node->num_bytes;
2200 ins.type = BTRFS_EXTENT_ITEM_KEY;
2203 BUG_ON(node->ref_mod != 1);
2204 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2205 BUG_ON(!extent_op || !extent_op->update_flags);
2206 ret = alloc_reserved_tree_block(trans, root,
2208 extent_op->flags_to_set,
2211 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2212 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2213 node->num_bytes, parent, ref_root,
2214 ref->level, 0, 1, extent_op);
2215 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2216 ret = __btrfs_free_extent(trans, root, node->bytenr,
2217 node->num_bytes, parent, ref_root,
2218 ref->level, 0, 1, extent_op);
2225 /* helper function to actually process a single delayed ref entry */
2226 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2227 struct btrfs_root *root,
2228 struct btrfs_delayed_ref_node *node,
2229 struct btrfs_delayed_extent_op *extent_op,
2230 int insert_reserved)
2237 if (btrfs_delayed_ref_is_head(node)) {
2238 struct btrfs_delayed_ref_head *head;
2240 * we've hit the end of the chain and we were supposed
2241 * to insert this extent into the tree. But, it got
2242 * deleted before we ever needed to insert it, so all
2243 * we have to do is clean up the accounting
2246 head = btrfs_delayed_node_to_head(node);
2247 if (insert_reserved) {
2248 btrfs_pin_extent(root, node->bytenr,
2249 node->num_bytes, 1);
2250 if (head->is_data) {
2251 ret = btrfs_del_csums(trans, root,
2259 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2260 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2261 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2263 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2264 node->type == BTRFS_SHARED_DATA_REF_KEY)
2265 ret = run_delayed_data_ref(trans, root, node, extent_op,
2272 static noinline struct btrfs_delayed_ref_node *
2273 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2275 struct rb_node *node;
2276 struct btrfs_delayed_ref_node *ref;
2277 int action = BTRFS_ADD_DELAYED_REF;
2280 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2281 * this prevents ref count from going down to zero when
2282 * there still are pending delayed ref.
2284 node = rb_prev(&head->node.rb_node);
2288 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2290 if (ref->bytenr != head->node.bytenr)
2292 if (ref->action == action)
2294 node = rb_prev(node);
2296 if (action == BTRFS_ADD_DELAYED_REF) {
2297 action = BTRFS_DROP_DELAYED_REF;
2304 * Returns 0 on success or if called with an already aborted transaction.
2305 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2307 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2308 struct btrfs_root *root,
2309 struct list_head *cluster)
2311 struct btrfs_delayed_ref_root *delayed_refs;
2312 struct btrfs_delayed_ref_node *ref;
2313 struct btrfs_delayed_ref_head *locked_ref = NULL;
2314 struct btrfs_delayed_extent_op *extent_op;
2315 struct btrfs_fs_info *fs_info = root->fs_info;
2318 int must_insert_reserved = 0;
2320 delayed_refs = &trans->transaction->delayed_refs;
2323 /* pick a new head ref from the cluster list */
2324 if (list_empty(cluster))
2327 locked_ref = list_entry(cluster->next,
2328 struct btrfs_delayed_ref_head, cluster);
2330 /* grab the lock that says we are going to process
2331 * all the refs for this head */
2332 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2335 * we may have dropped the spin lock to get the head
2336 * mutex lock, and that might have given someone else
2337 * time to free the head. If that's true, it has been
2338 * removed from our list and we can move on.
2340 if (ret == -EAGAIN) {
2348 * We need to try and merge add/drops of the same ref since we
2349 * can run into issues with relocate dropping the implicit ref
2350 * and then it being added back again before the drop can
2351 * finish. If we merged anything we need to re-loop so we can
2354 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2358 * locked_ref is the head node, so we have to go one
2359 * node back for any delayed ref updates
2361 ref = select_delayed_ref(locked_ref);
2363 if (ref && ref->seq &&
2364 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2366 * there are still refs with lower seq numbers in the
2367 * process of being added. Don't run this ref yet.
2369 list_del_init(&locked_ref->cluster);
2370 btrfs_delayed_ref_unlock(locked_ref);
2372 delayed_refs->num_heads_ready++;
2373 spin_unlock(&delayed_refs->lock);
2375 spin_lock(&delayed_refs->lock);
2380 * record the must insert reserved flag before we
2381 * drop the spin lock.
2383 must_insert_reserved = locked_ref->must_insert_reserved;
2384 locked_ref->must_insert_reserved = 0;
2386 extent_op = locked_ref->extent_op;
2387 locked_ref->extent_op = NULL;
2390 /* All delayed refs have been processed, Go ahead
2391 * and send the head node to run_one_delayed_ref,
2392 * so that any accounting fixes can happen
2394 ref = &locked_ref->node;
2396 if (extent_op && must_insert_reserved) {
2397 btrfs_free_delayed_extent_op(extent_op);
2402 spin_unlock(&delayed_refs->lock);
2404 ret = run_delayed_extent_op(trans, root,
2406 btrfs_free_delayed_extent_op(extent_op);
2409 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2410 spin_lock(&delayed_refs->lock);
2411 btrfs_delayed_ref_unlock(locked_ref);
2420 rb_erase(&ref->rb_node, &delayed_refs->root);
2421 delayed_refs->num_entries--;
2422 if (!btrfs_delayed_ref_is_head(ref)) {
2424 * when we play the delayed ref, also correct the
2427 switch (ref->action) {
2428 case BTRFS_ADD_DELAYED_REF:
2429 case BTRFS_ADD_DELAYED_EXTENT:
2430 locked_ref->node.ref_mod -= ref->ref_mod;
2432 case BTRFS_DROP_DELAYED_REF:
2433 locked_ref->node.ref_mod += ref->ref_mod;
2439 spin_unlock(&delayed_refs->lock);
2441 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2442 must_insert_reserved);
2444 btrfs_free_delayed_extent_op(extent_op);
2446 btrfs_delayed_ref_unlock(locked_ref);
2447 btrfs_put_delayed_ref(ref);
2448 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2449 spin_lock(&delayed_refs->lock);
2454 * If this node is a head, that means all the refs in this head
2455 * have been dealt with, and we will pick the next head to deal
2456 * with, so we must unlock the head and drop it from the cluster
2457 * list before we release it.
2459 if (btrfs_delayed_ref_is_head(ref)) {
2460 list_del_init(&locked_ref->cluster);
2461 btrfs_delayed_ref_unlock(locked_ref);
2464 btrfs_put_delayed_ref(ref);
2468 spin_lock(&delayed_refs->lock);
2473 #ifdef SCRAMBLE_DELAYED_REFS
2475 * Normally delayed refs get processed in ascending bytenr order. This
2476 * correlates in most cases to the order added. To expose dependencies on this
2477 * order, we start to process the tree in the middle instead of the beginning
2479 static u64 find_middle(struct rb_root *root)
2481 struct rb_node *n = root->rb_node;
2482 struct btrfs_delayed_ref_node *entry;
2485 u64 first = 0, last = 0;
2489 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2490 first = entry->bytenr;
2494 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2495 last = entry->bytenr;
2500 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501 WARN_ON(!entry->in_tree);
2503 middle = entry->bytenr;
2516 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2517 struct btrfs_fs_info *fs_info)
2519 struct qgroup_update *qgroup_update;
2522 if (list_empty(&trans->qgroup_ref_list) !=
2523 !trans->delayed_ref_elem.seq) {
2524 /* list without seq or seq without list */
2526 "qgroup accounting update error, list is%s empty, seq is %llu",
2527 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2528 trans->delayed_ref_elem.seq);
2532 if (!trans->delayed_ref_elem.seq)
2535 while (!list_empty(&trans->qgroup_ref_list)) {
2536 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2537 struct qgroup_update, list);
2538 list_del(&qgroup_update->list);
2540 ret = btrfs_qgroup_account_ref(
2541 trans, fs_info, qgroup_update->node,
2542 qgroup_update->extent_op);
2543 kfree(qgroup_update);
2546 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2551 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2554 int val = atomic_read(&delayed_refs->ref_seq);
2556 if (val < seq || val >= seq + count)
2562 * this starts processing the delayed reference count updates and
2563 * extent insertions we have queued up so far. count can be
2564 * 0, which means to process everything in the tree at the start
2565 * of the run (but not newly added entries), or it can be some target
2566 * number you'd like to process.
2568 * Returns 0 on success or if called with an aborted transaction
2569 * Returns <0 on error and aborts the transaction
2571 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2572 struct btrfs_root *root, unsigned long count)
2574 struct rb_node *node;
2575 struct btrfs_delayed_ref_root *delayed_refs;
2576 struct btrfs_delayed_ref_node *ref;
2577 struct list_head cluster;
2580 int run_all = count == (unsigned long)-1;
2584 /* We'll clean this up in btrfs_cleanup_transaction */
2588 if (root == root->fs_info->extent_root)
2589 root = root->fs_info->tree_root;
2591 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2593 delayed_refs = &trans->transaction->delayed_refs;
2594 INIT_LIST_HEAD(&cluster);
2596 count = delayed_refs->num_entries * 2;
2600 if (!run_all && !run_most) {
2602 int seq = atomic_read(&delayed_refs->ref_seq);
2605 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2607 DEFINE_WAIT(__wait);
2608 if (delayed_refs->num_entries < 16348)
2611 prepare_to_wait(&delayed_refs->wait, &__wait,
2612 TASK_UNINTERRUPTIBLE);
2614 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2617 finish_wait(&delayed_refs->wait, &__wait);
2619 if (!refs_newer(delayed_refs, seq, 256))
2624 finish_wait(&delayed_refs->wait, &__wait);
2630 atomic_inc(&delayed_refs->procs_running_refs);
2635 spin_lock(&delayed_refs->lock);
2637 #ifdef SCRAMBLE_DELAYED_REFS
2638 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2642 if (!(run_all || run_most) &&
2643 delayed_refs->num_heads_ready < 64)
2647 * go find something we can process in the rbtree. We start at
2648 * the beginning of the tree, and then build a cluster
2649 * of refs to process starting at the first one we are able to
2652 delayed_start = delayed_refs->run_delayed_start;
2653 ret = btrfs_find_ref_cluster(trans, &cluster,
2654 delayed_refs->run_delayed_start);
2658 ret = run_clustered_refs(trans, root, &cluster);
2660 btrfs_release_ref_cluster(&cluster);
2661 spin_unlock(&delayed_refs->lock);
2662 btrfs_abort_transaction(trans, root, ret);
2663 atomic_dec(&delayed_refs->procs_running_refs);
2667 atomic_add(ret, &delayed_refs->ref_seq);
2669 count -= min_t(unsigned long, ret, count);
2674 if (delayed_start >= delayed_refs->run_delayed_start) {
2677 * btrfs_find_ref_cluster looped. let's do one
2678 * more cycle. if we don't run any delayed ref
2679 * during that cycle (because we can't because
2680 * all of them are blocked), bail out.
2685 * no runnable refs left, stop trying
2692 /* refs were run, let's reset staleness detection */
2698 if (!list_empty(&trans->new_bgs)) {
2699 spin_unlock(&delayed_refs->lock);
2700 btrfs_create_pending_block_groups(trans, root);
2701 spin_lock(&delayed_refs->lock);
2704 node = rb_first(&delayed_refs->root);
2707 count = (unsigned long)-1;
2710 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2712 if (btrfs_delayed_ref_is_head(ref)) {
2713 struct btrfs_delayed_ref_head *head;
2715 head = btrfs_delayed_node_to_head(ref);
2716 atomic_inc(&ref->refs);
2718 spin_unlock(&delayed_refs->lock);
2720 * Mutex was contended, block until it's
2721 * released and try again
2723 mutex_lock(&head->mutex);
2724 mutex_unlock(&head->mutex);
2726 btrfs_put_delayed_ref(ref);
2730 node = rb_next(node);
2732 spin_unlock(&delayed_refs->lock);
2733 schedule_timeout(1);
2737 atomic_dec(&delayed_refs->procs_running_refs);
2739 if (waitqueue_active(&delayed_refs->wait))
2740 wake_up(&delayed_refs->wait);
2742 spin_unlock(&delayed_refs->lock);
2743 assert_qgroups_uptodate(trans);
2747 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root,
2749 u64 bytenr, u64 num_bytes, u64 flags,
2752 struct btrfs_delayed_extent_op *extent_op;
2755 extent_op = btrfs_alloc_delayed_extent_op();
2759 extent_op->flags_to_set = flags;
2760 extent_op->update_flags = 1;
2761 extent_op->update_key = 0;
2762 extent_op->is_data = is_data ? 1 : 0;
2764 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2765 num_bytes, extent_op);
2767 btrfs_free_delayed_extent_op(extent_op);
2771 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2772 struct btrfs_root *root,
2773 struct btrfs_path *path,
2774 u64 objectid, u64 offset, u64 bytenr)
2776 struct btrfs_delayed_ref_head *head;
2777 struct btrfs_delayed_ref_node *ref;
2778 struct btrfs_delayed_data_ref *data_ref;
2779 struct btrfs_delayed_ref_root *delayed_refs;
2780 struct rb_node *node;
2784 delayed_refs = &trans->transaction->delayed_refs;
2785 spin_lock(&delayed_refs->lock);
2786 head = btrfs_find_delayed_ref_head(trans, bytenr);
2790 if (!mutex_trylock(&head->mutex)) {
2791 atomic_inc(&head->node.refs);
2792 spin_unlock(&delayed_refs->lock);
2794 btrfs_release_path(path);
2797 * Mutex was contended, block until it's released and let
2800 mutex_lock(&head->mutex);
2801 mutex_unlock(&head->mutex);
2802 btrfs_put_delayed_ref(&head->node);
2806 node = rb_prev(&head->node.rb_node);
2810 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2812 if (ref->bytenr != bytenr)
2816 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2819 data_ref = btrfs_delayed_node_to_data_ref(ref);
2821 node = rb_prev(node);
2825 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2826 if (ref->bytenr == bytenr && ref->seq == seq)
2830 if (data_ref->root != root->root_key.objectid ||
2831 data_ref->objectid != objectid || data_ref->offset != offset)
2836 mutex_unlock(&head->mutex);
2838 spin_unlock(&delayed_refs->lock);
2842 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2843 struct btrfs_root *root,
2844 struct btrfs_path *path,
2845 u64 objectid, u64 offset, u64 bytenr)
2847 struct btrfs_root *extent_root = root->fs_info->extent_root;
2848 struct extent_buffer *leaf;
2849 struct btrfs_extent_data_ref *ref;
2850 struct btrfs_extent_inline_ref *iref;
2851 struct btrfs_extent_item *ei;
2852 struct btrfs_key key;
2856 key.objectid = bytenr;
2857 key.offset = (u64)-1;
2858 key.type = BTRFS_EXTENT_ITEM_KEY;
2860 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2863 BUG_ON(ret == 0); /* Corruption */
2866 if (path->slots[0] == 0)
2870 leaf = path->nodes[0];
2871 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2873 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2877 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2878 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2879 if (item_size < sizeof(*ei)) {
2880 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2884 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2886 if (item_size != sizeof(*ei) +
2887 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2890 if (btrfs_extent_generation(leaf, ei) <=
2891 btrfs_root_last_snapshot(&root->root_item))
2894 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2895 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2896 BTRFS_EXTENT_DATA_REF_KEY)
2899 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2900 if (btrfs_extent_refs(leaf, ei) !=
2901 btrfs_extent_data_ref_count(leaf, ref) ||
2902 btrfs_extent_data_ref_root(leaf, ref) !=
2903 root->root_key.objectid ||
2904 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2905 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2913 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2914 struct btrfs_root *root,
2915 u64 objectid, u64 offset, u64 bytenr)
2917 struct btrfs_path *path;
2921 path = btrfs_alloc_path();
2926 ret = check_committed_ref(trans, root, path, objectid,
2928 if (ret && ret != -ENOENT)
2931 ret2 = check_delayed_ref(trans, root, path, objectid,
2933 } while (ret2 == -EAGAIN);
2935 if (ret2 && ret2 != -ENOENT) {
2940 if (ret != -ENOENT || ret2 != -ENOENT)
2943 btrfs_free_path(path);
2944 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2949 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2950 struct btrfs_root *root,
2951 struct extent_buffer *buf,
2952 int full_backref, int inc, int for_cow)
2959 struct btrfs_key key;
2960 struct btrfs_file_extent_item *fi;
2964 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2965 u64, u64, u64, u64, u64, u64, int);
2967 ref_root = btrfs_header_owner(buf);
2968 nritems = btrfs_header_nritems(buf);
2969 level = btrfs_header_level(buf);
2971 if (!root->ref_cows && level == 0)
2975 process_func = btrfs_inc_extent_ref;
2977 process_func = btrfs_free_extent;
2980 parent = buf->start;
2984 for (i = 0; i < nritems; i++) {
2986 btrfs_item_key_to_cpu(buf, &key, i);
2987 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2989 fi = btrfs_item_ptr(buf, i,
2990 struct btrfs_file_extent_item);
2991 if (btrfs_file_extent_type(buf, fi) ==
2992 BTRFS_FILE_EXTENT_INLINE)
2994 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2998 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2999 key.offset -= btrfs_file_extent_offset(buf, fi);
3000 ret = process_func(trans, root, bytenr, num_bytes,
3001 parent, ref_root, key.objectid,
3002 key.offset, for_cow);
3006 bytenr = btrfs_node_blockptr(buf, i);
3007 num_bytes = btrfs_level_size(root, level - 1);
3008 ret = process_func(trans, root, bytenr, num_bytes,
3009 parent, ref_root, level - 1, 0,
3020 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3021 struct extent_buffer *buf, int full_backref, int for_cow)
3023 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3026 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3027 struct extent_buffer *buf, int full_backref, int for_cow)
3029 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3032 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3033 struct btrfs_root *root,
3034 struct btrfs_path *path,
3035 struct btrfs_block_group_cache *cache)
3038 struct btrfs_root *extent_root = root->fs_info->extent_root;
3040 struct extent_buffer *leaf;
3042 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3045 BUG_ON(ret); /* Corruption */
3047 leaf = path->nodes[0];
3048 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3049 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3050 btrfs_mark_buffer_dirty(leaf);
3051 btrfs_release_path(path);
3054 btrfs_abort_transaction(trans, root, ret);
3061 static struct btrfs_block_group_cache *
3062 next_block_group(struct btrfs_root *root,
3063 struct btrfs_block_group_cache *cache)
3065 struct rb_node *node;
3066 spin_lock(&root->fs_info->block_group_cache_lock);
3067 node = rb_next(&cache->cache_node);
3068 btrfs_put_block_group(cache);
3070 cache = rb_entry(node, struct btrfs_block_group_cache,
3072 btrfs_get_block_group(cache);
3075 spin_unlock(&root->fs_info->block_group_cache_lock);
3079 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3080 struct btrfs_trans_handle *trans,
3081 struct btrfs_path *path)
3083 struct btrfs_root *root = block_group->fs_info->tree_root;
3084 struct inode *inode = NULL;
3086 int dcs = BTRFS_DC_ERROR;
3092 * If this block group is smaller than 100 megs don't bother caching the
3095 if (block_group->key.offset < (100 * 1024 * 1024)) {
3096 spin_lock(&block_group->lock);
3097 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3098 spin_unlock(&block_group->lock);
3103 inode = lookup_free_space_inode(root, block_group, path);
3104 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3105 ret = PTR_ERR(inode);
3106 btrfs_release_path(path);
3110 if (IS_ERR(inode)) {
3114 if (block_group->ro)
3117 ret = create_free_space_inode(root, trans, block_group, path);
3123 /* We've already setup this transaction, go ahead and exit */
3124 if (block_group->cache_generation == trans->transid &&
3125 i_size_read(inode)) {
3126 dcs = BTRFS_DC_SETUP;
3131 * We want to set the generation to 0, that way if anything goes wrong
3132 * from here on out we know not to trust this cache when we load up next
3135 BTRFS_I(inode)->generation = 0;
3136 ret = btrfs_update_inode(trans, root, inode);
3139 if (i_size_read(inode) > 0) {
3140 ret = btrfs_truncate_free_space_cache(root, trans, path,
3146 spin_lock(&block_group->lock);
3147 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3148 !btrfs_test_opt(root, SPACE_CACHE)) {
3150 * don't bother trying to write stuff out _if_
3151 * a) we're not cached,
3152 * b) we're with nospace_cache mount option.
3154 dcs = BTRFS_DC_WRITTEN;
3155 spin_unlock(&block_group->lock);
3158 spin_unlock(&block_group->lock);
3161 * Try to preallocate enough space based on how big the block group is.
3162 * Keep in mind this has to include any pinned space which could end up
3163 * taking up quite a bit since it's not folded into the other space
3166 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3171 num_pages *= PAGE_CACHE_SIZE;
3173 ret = btrfs_check_data_free_space(inode, num_pages);
3177 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3178 num_pages, num_pages,
3181 dcs = BTRFS_DC_SETUP;
3182 btrfs_free_reserved_data_space(inode, num_pages);
3187 btrfs_release_path(path);
3189 spin_lock(&block_group->lock);
3190 if (!ret && dcs == BTRFS_DC_SETUP)
3191 block_group->cache_generation = trans->transid;
3192 block_group->disk_cache_state = dcs;
3193 spin_unlock(&block_group->lock);
3198 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3199 struct btrfs_root *root)
3201 struct btrfs_block_group_cache *cache;
3203 struct btrfs_path *path;
3206 path = btrfs_alloc_path();
3212 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3214 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3216 cache = next_block_group(root, cache);
3224 err = cache_save_setup(cache, trans, path);
3225 last = cache->key.objectid + cache->key.offset;
3226 btrfs_put_block_group(cache);
3231 err = btrfs_run_delayed_refs(trans, root,
3233 if (err) /* File system offline */
3237 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3239 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3240 btrfs_put_block_group(cache);
3246 cache = next_block_group(root, cache);
3255 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3256 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3258 last = cache->key.objectid + cache->key.offset;
3260 err = write_one_cache_group(trans, root, path, cache);
3261 if (err) /* File system offline */
3264 btrfs_put_block_group(cache);
3269 * I don't think this is needed since we're just marking our
3270 * preallocated extent as written, but just in case it can't
3274 err = btrfs_run_delayed_refs(trans, root,
3276 if (err) /* File system offline */
3280 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3283 * Really this shouldn't happen, but it could if we
3284 * couldn't write the entire preallocated extent and
3285 * splitting the extent resulted in a new block.
3288 btrfs_put_block_group(cache);
3291 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3293 cache = next_block_group(root, cache);
3302 err = btrfs_write_out_cache(root, trans, cache, path);
3305 * If we didn't have an error then the cache state is still
3306 * NEED_WRITE, so we can set it to WRITTEN.
3308 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3309 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3310 last = cache->key.objectid + cache->key.offset;
3311 btrfs_put_block_group(cache);
3315 btrfs_free_path(path);
3319 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3321 struct btrfs_block_group_cache *block_group;
3324 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3325 if (!block_group || block_group->ro)
3328 btrfs_put_block_group(block_group);
3332 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3333 u64 total_bytes, u64 bytes_used,
3334 struct btrfs_space_info **space_info)
3336 struct btrfs_space_info *found;
3340 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3341 BTRFS_BLOCK_GROUP_RAID10))
3346 found = __find_space_info(info, flags);
3348 spin_lock(&found->lock);
3349 found->total_bytes += total_bytes;
3350 found->disk_total += total_bytes * factor;
3351 found->bytes_used += bytes_used;
3352 found->disk_used += bytes_used * factor;
3354 spin_unlock(&found->lock);
3355 *space_info = found;
3358 found = kzalloc(sizeof(*found), GFP_NOFS);
3362 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3363 INIT_LIST_HEAD(&found->block_groups[i]);
3364 init_rwsem(&found->groups_sem);
3365 spin_lock_init(&found->lock);
3366 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3367 found->total_bytes = total_bytes;
3368 found->disk_total = total_bytes * factor;
3369 found->bytes_used = bytes_used;
3370 found->disk_used = bytes_used * factor;
3371 found->bytes_pinned = 0;
3372 found->bytes_reserved = 0;
3373 found->bytes_readonly = 0;
3374 found->bytes_may_use = 0;
3376 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3377 found->chunk_alloc = 0;
3379 init_waitqueue_head(&found->wait);
3380 *space_info = found;
3381 list_add_rcu(&found->list, &info->space_info);
3382 if (flags & BTRFS_BLOCK_GROUP_DATA)
3383 info->data_sinfo = found;
3387 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3389 u64 extra_flags = chunk_to_extended(flags) &
3390 BTRFS_EXTENDED_PROFILE_MASK;
3392 write_seqlock(&fs_info->profiles_lock);
3393 if (flags & BTRFS_BLOCK_GROUP_DATA)
3394 fs_info->avail_data_alloc_bits |= extra_flags;
3395 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3396 fs_info->avail_metadata_alloc_bits |= extra_flags;
3397 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3398 fs_info->avail_system_alloc_bits |= extra_flags;
3399 write_sequnlock(&fs_info->profiles_lock);
3403 * returns target flags in extended format or 0 if restripe for this
3404 * chunk_type is not in progress
3406 * should be called with either volume_mutex or balance_lock held
3408 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3410 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3416 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3417 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3418 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3419 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3420 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3421 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3422 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3423 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3424 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3431 * @flags: available profiles in extended format (see ctree.h)
3433 * Returns reduced profile in chunk format. If profile changing is in
3434 * progress (either running or paused) picks the target profile (if it's
3435 * already available), otherwise falls back to plain reducing.
3437 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3440 * we add in the count of missing devices because we want
3441 * to make sure that any RAID levels on a degraded FS
3442 * continue to be honored.
3444 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3445 root->fs_info->fs_devices->missing_devices;
3450 * see if restripe for this chunk_type is in progress, if so
3451 * try to reduce to the target profile
3453 spin_lock(&root->fs_info->balance_lock);
3454 target = get_restripe_target(root->fs_info, flags);
3456 /* pick target profile only if it's already available */
3457 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3458 spin_unlock(&root->fs_info->balance_lock);
3459 return extended_to_chunk(target);
3462 spin_unlock(&root->fs_info->balance_lock);
3464 /* First, mask out the RAID levels which aren't possible */
3465 if (num_devices == 1)
3466 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3467 BTRFS_BLOCK_GROUP_RAID5);
3468 if (num_devices < 3)
3469 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3470 if (num_devices < 4)
3471 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3473 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3474 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3475 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3478 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3479 tmp = BTRFS_BLOCK_GROUP_RAID6;
3480 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3481 tmp = BTRFS_BLOCK_GROUP_RAID5;
3482 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3483 tmp = BTRFS_BLOCK_GROUP_RAID10;
3484 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3485 tmp = BTRFS_BLOCK_GROUP_RAID1;
3486 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3487 tmp = BTRFS_BLOCK_GROUP_RAID0;
3489 return extended_to_chunk(flags | tmp);
3492 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3497 seq = read_seqbegin(&root->fs_info->profiles_lock);
3499 if (flags & BTRFS_BLOCK_GROUP_DATA)
3500 flags |= root->fs_info->avail_data_alloc_bits;
3501 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3502 flags |= root->fs_info->avail_system_alloc_bits;
3503 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3504 flags |= root->fs_info->avail_metadata_alloc_bits;
3505 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3507 return btrfs_reduce_alloc_profile(root, flags);
3510 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3516 flags = BTRFS_BLOCK_GROUP_DATA;
3517 else if (root == root->fs_info->chunk_root)
3518 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3520 flags = BTRFS_BLOCK_GROUP_METADATA;
3522 ret = get_alloc_profile(root, flags);
3527 * This will check the space that the inode allocates from to make sure we have
3528 * enough space for bytes.
3530 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3532 struct btrfs_space_info *data_sinfo;
3533 struct btrfs_root *root = BTRFS_I(inode)->root;
3534 struct btrfs_fs_info *fs_info = root->fs_info;
3536 int ret = 0, committed = 0, alloc_chunk = 1;
3538 /* make sure bytes are sectorsize aligned */
3539 bytes = ALIGN(bytes, root->sectorsize);
3541 if (root == root->fs_info->tree_root ||
3542 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3547 data_sinfo = fs_info->data_sinfo;
3552 /* make sure we have enough space to handle the data first */
3553 spin_lock(&data_sinfo->lock);
3554 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3555 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3556 data_sinfo->bytes_may_use;
3558 if (used + bytes > data_sinfo->total_bytes) {
3559 struct btrfs_trans_handle *trans;
3562 * if we don't have enough free bytes in this space then we need
3563 * to alloc a new chunk.
3565 if (!data_sinfo->full && alloc_chunk) {
3568 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3569 spin_unlock(&data_sinfo->lock);
3571 alloc_target = btrfs_get_alloc_profile(root, 1);
3572 trans = btrfs_join_transaction(root);
3574 return PTR_ERR(trans);
3576 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3578 CHUNK_ALLOC_NO_FORCE);
3579 btrfs_end_transaction(trans, root);
3588 data_sinfo = fs_info->data_sinfo;
3594 * If we have less pinned bytes than we want to allocate then
3595 * don't bother committing the transaction, it won't help us.
3597 if (data_sinfo->bytes_pinned < bytes)
3599 spin_unlock(&data_sinfo->lock);
3601 /* commit the current transaction and try again */
3604 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3606 trans = btrfs_join_transaction(root);
3608 return PTR_ERR(trans);
3609 ret = btrfs_commit_transaction(trans, root);
3617 data_sinfo->bytes_may_use += bytes;
3618 trace_btrfs_space_reservation(root->fs_info, "space_info",
3619 data_sinfo->flags, bytes, 1);
3620 spin_unlock(&data_sinfo->lock);
3626 * Called if we need to clear a data reservation for this inode.
3628 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3630 struct btrfs_root *root = BTRFS_I(inode)->root;
3631 struct btrfs_space_info *data_sinfo;
3633 /* make sure bytes are sectorsize aligned */
3634 bytes = ALIGN(bytes, root->sectorsize);
3636 data_sinfo = root->fs_info->data_sinfo;
3637 spin_lock(&data_sinfo->lock);
3638 data_sinfo->bytes_may_use -= bytes;
3639 trace_btrfs_space_reservation(root->fs_info, "space_info",
3640 data_sinfo->flags, bytes, 0);
3641 spin_unlock(&data_sinfo->lock);
3644 static void force_metadata_allocation(struct btrfs_fs_info *info)
3646 struct list_head *head = &info->space_info;
3647 struct btrfs_space_info *found;
3650 list_for_each_entry_rcu(found, head, list) {
3651 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3652 found->force_alloc = CHUNK_ALLOC_FORCE;
3657 static int should_alloc_chunk(struct btrfs_root *root,
3658 struct btrfs_space_info *sinfo, int force)
3660 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3661 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3662 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3665 if (force == CHUNK_ALLOC_FORCE)
3669 * We need to take into account the global rsv because for all intents
3670 * and purposes it's used space. Don't worry about locking the
3671 * global_rsv, it doesn't change except when the transaction commits.
3673 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3674 num_allocated += global_rsv->size;
3677 * in limited mode, we want to have some free space up to
3678 * about 1% of the FS size.
3680 if (force == CHUNK_ALLOC_LIMITED) {
3681 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3682 thresh = max_t(u64, 64 * 1024 * 1024,
3683 div_factor_fine(thresh, 1));
3685 if (num_bytes - num_allocated < thresh)
3689 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3694 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3698 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3699 BTRFS_BLOCK_GROUP_RAID0 |
3700 BTRFS_BLOCK_GROUP_RAID5 |
3701 BTRFS_BLOCK_GROUP_RAID6))
3702 num_dev = root->fs_info->fs_devices->rw_devices;
3703 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3706 num_dev = 1; /* DUP or single */
3708 /* metadata for updaing devices and chunk tree */
3709 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3712 static void check_system_chunk(struct btrfs_trans_handle *trans,
3713 struct btrfs_root *root, u64 type)
3715 struct btrfs_space_info *info;
3719 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3720 spin_lock(&info->lock);
3721 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3722 info->bytes_reserved - info->bytes_readonly;
3723 spin_unlock(&info->lock);
3725 thresh = get_system_chunk_thresh(root, type);
3726 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3727 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3728 left, thresh, type);
3729 dump_space_info(info, 0, 0);
3732 if (left < thresh) {
3735 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3736 btrfs_alloc_chunk(trans, root, flags);
3740 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3741 struct btrfs_root *extent_root, u64 flags, int force)
3743 struct btrfs_space_info *space_info;
3744 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3745 int wait_for_alloc = 0;
3748 /* Don't re-enter if we're already allocating a chunk */
3749 if (trans->allocating_chunk)
3752 space_info = __find_space_info(extent_root->fs_info, flags);
3754 ret = update_space_info(extent_root->fs_info, flags,
3756 BUG_ON(ret); /* -ENOMEM */
3758 BUG_ON(!space_info); /* Logic error */
3761 spin_lock(&space_info->lock);
3762 if (force < space_info->force_alloc)
3763 force = space_info->force_alloc;
3764 if (space_info->full) {
3765 spin_unlock(&space_info->lock);
3769 if (!should_alloc_chunk(extent_root, space_info, force)) {
3770 spin_unlock(&space_info->lock);
3772 } else if (space_info->chunk_alloc) {
3775 space_info->chunk_alloc = 1;
3778 spin_unlock(&space_info->lock);
3780 mutex_lock(&fs_info->chunk_mutex);
3783 * The chunk_mutex is held throughout the entirety of a chunk
3784 * allocation, so once we've acquired the chunk_mutex we know that the
3785 * other guy is done and we need to recheck and see if we should
3788 if (wait_for_alloc) {
3789 mutex_unlock(&fs_info->chunk_mutex);
3794 trans->allocating_chunk = true;
3797 * If we have mixed data/metadata chunks we want to make sure we keep
3798 * allocating mixed chunks instead of individual chunks.
3800 if (btrfs_mixed_space_info(space_info))
3801 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3804 * if we're doing a data chunk, go ahead and make sure that
3805 * we keep a reasonable number of metadata chunks allocated in the
3808 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3809 fs_info->data_chunk_allocations++;
3810 if (!(fs_info->data_chunk_allocations %
3811 fs_info->metadata_ratio))
3812 force_metadata_allocation(fs_info);
3816 * Check if we have enough space in SYSTEM chunk because we may need
3817 * to update devices.
3819 check_system_chunk(trans, extent_root, flags);
3821 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3822 trans->allocating_chunk = false;
3824 spin_lock(&space_info->lock);
3825 if (ret < 0 && ret != -ENOSPC)
3828 space_info->full = 1;
3832 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3834 space_info->chunk_alloc = 0;
3835 spin_unlock(&space_info->lock);
3836 mutex_unlock(&fs_info->chunk_mutex);
3840 static int can_overcommit(struct btrfs_root *root,
3841 struct btrfs_space_info *space_info, u64 bytes,
3842 enum btrfs_reserve_flush_enum flush)
3844 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3845 u64 profile = btrfs_get_alloc_profile(root, 0);
3851 used = space_info->bytes_used + space_info->bytes_reserved +
3852 space_info->bytes_pinned + space_info->bytes_readonly;
3854 spin_lock(&global_rsv->lock);
3855 rsv_size = global_rsv->size;
3856 spin_unlock(&global_rsv->lock);
3859 * We only want to allow over committing if we have lots of actual space
3860 * free, but if we don't have enough space to handle the global reserve
3861 * space then we could end up having a real enospc problem when trying
3862 * to allocate a chunk or some other such important allocation.
3865 if (used + rsv_size >= space_info->total_bytes)
3868 used += space_info->bytes_may_use;
3870 spin_lock(&root->fs_info->free_chunk_lock);
3871 avail = root->fs_info->free_chunk_space;
3872 spin_unlock(&root->fs_info->free_chunk_lock);
3875 * If we have dup, raid1 or raid10 then only half of the free
3876 * space is actually useable. For raid56, the space info used
3877 * doesn't include the parity drive, so we don't have to
3880 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3881 BTRFS_BLOCK_GROUP_RAID1 |
3882 BTRFS_BLOCK_GROUP_RAID10))
3885 to_add = space_info->total_bytes;
3888 * If we aren't flushing all things, let us overcommit up to
3889 * 1/2th of the space. If we can flush, don't let us overcommit
3890 * too much, let it overcommit up to 1/8 of the space.
3892 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3898 * Limit the overcommit to the amount of free space we could possibly
3899 * allocate for chunks.
3901 to_add = min(avail, to_add);
3903 if (used + bytes < space_info->total_bytes + to_add)
3908 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3909 unsigned long nr_pages)
3911 struct super_block *sb = root->fs_info->sb;
3914 /* If we can not start writeback, just sync all the delalloc file. */
3915 started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
3916 WB_REASON_FS_FREE_SPACE);
3919 * We needn't worry the filesystem going from r/w to r/o though
3920 * we don't acquire ->s_umount mutex, because the filesystem
3921 * should guarantee the delalloc inodes list be empty after
3922 * the filesystem is readonly(all dirty pages are written to
3925 btrfs_start_delalloc_inodes(root, 0);
3926 if (!current->journal_info)
3927 btrfs_wait_ordered_extents(root, 0);
3932 * shrink metadata reservation for delalloc
3934 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3937 struct btrfs_block_rsv *block_rsv;
3938 struct btrfs_space_info *space_info;
3939 struct btrfs_trans_handle *trans;
3943 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3945 enum btrfs_reserve_flush_enum flush;
3947 trans = (struct btrfs_trans_handle *)current->journal_info;
3948 block_rsv = &root->fs_info->delalloc_block_rsv;
3949 space_info = block_rsv->space_info;
3952 delalloc_bytes = percpu_counter_sum_positive(
3953 &root->fs_info->delalloc_bytes);
3954 if (delalloc_bytes == 0) {
3957 btrfs_wait_ordered_extents(root, 0);
3961 while (delalloc_bytes && loops < 3) {
3962 max_reclaim = min(delalloc_bytes, to_reclaim);
3963 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3964 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3966 * We need to wait for the async pages to actually start before
3969 wait_event(root->fs_info->async_submit_wait,
3970 !atomic_read(&root->fs_info->async_delalloc_pages));
3973 flush = BTRFS_RESERVE_FLUSH_ALL;
3975 flush = BTRFS_RESERVE_NO_FLUSH;
3976 spin_lock(&space_info->lock);
3977 if (can_overcommit(root, space_info, orig, flush)) {
3978 spin_unlock(&space_info->lock);
3981 spin_unlock(&space_info->lock);
3984 if (wait_ordered && !trans) {
3985 btrfs_wait_ordered_extents(root, 0);
3987 time_left = schedule_timeout_killable(1);
3992 delalloc_bytes = percpu_counter_sum_positive(
3993 &root->fs_info->delalloc_bytes);
3998 * maybe_commit_transaction - possibly commit the transaction if its ok to
3999 * @root - the root we're allocating for
4000 * @bytes - the number of bytes we want to reserve
4001 * @force - force the commit
4003 * This will check to make sure that committing the transaction will actually
4004 * get us somewhere and then commit the transaction if it does. Otherwise it
4005 * will return -ENOSPC.
4007 static int may_commit_transaction(struct btrfs_root *root,
4008 struct btrfs_space_info *space_info,
4009 u64 bytes, int force)
4011 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4012 struct btrfs_trans_handle *trans;
4014 trans = (struct btrfs_trans_handle *)current->journal_info;
4021 /* See if there is enough pinned space to make this reservation */
4022 spin_lock(&space_info->lock);
4023 if (space_info->bytes_pinned >= bytes) {
4024 spin_unlock(&space_info->lock);
4027 spin_unlock(&space_info->lock);
4030 * See if there is some space in the delayed insertion reservation for
4033 if (space_info != delayed_rsv->space_info)
4036 spin_lock(&space_info->lock);
4037 spin_lock(&delayed_rsv->lock);
4038 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
4039 spin_unlock(&delayed_rsv->lock);
4040 spin_unlock(&space_info->lock);
4043 spin_unlock(&delayed_rsv->lock);
4044 spin_unlock(&space_info->lock);
4047 trans = btrfs_join_transaction(root);
4051 return btrfs_commit_transaction(trans, root);
4055 FLUSH_DELAYED_ITEMS_NR = 1,
4056 FLUSH_DELAYED_ITEMS = 2,
4058 FLUSH_DELALLOC_WAIT = 4,
4063 static int flush_space(struct btrfs_root *root,
4064 struct btrfs_space_info *space_info, u64 num_bytes,
4065 u64 orig_bytes, int state)
4067 struct btrfs_trans_handle *trans;
4072 case FLUSH_DELAYED_ITEMS_NR:
4073 case FLUSH_DELAYED_ITEMS:
4074 if (state == FLUSH_DELAYED_ITEMS_NR) {
4075 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4077 nr = (int)div64_u64(num_bytes, bytes);
4084 trans = btrfs_join_transaction(root);
4085 if (IS_ERR(trans)) {
4086 ret = PTR_ERR(trans);
4089 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4090 btrfs_end_transaction(trans, root);
4092 case FLUSH_DELALLOC:
4093 case FLUSH_DELALLOC_WAIT:
4094 shrink_delalloc(root, num_bytes, orig_bytes,
4095 state == FLUSH_DELALLOC_WAIT);
4098 trans = btrfs_join_transaction(root);
4099 if (IS_ERR(trans)) {
4100 ret = PTR_ERR(trans);
4103 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4104 btrfs_get_alloc_profile(root, 0),
4105 CHUNK_ALLOC_NO_FORCE);
4106 btrfs_end_transaction(trans, root);
4111 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4121 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4122 * @root - the root we're allocating for
4123 * @block_rsv - the block_rsv we're allocating for
4124 * @orig_bytes - the number of bytes we want
4125 * @flush - whether or not we can flush to make our reservation
4127 * This will reserve orgi_bytes number of bytes from the space info associated
4128 * with the block_rsv. If there is not enough space it will make an attempt to
4129 * flush out space to make room. It will do this by flushing delalloc if
4130 * possible or committing the transaction. If flush is 0 then no attempts to
4131 * regain reservations will be made and this will fail if there is not enough
4134 static int reserve_metadata_bytes(struct btrfs_root *root,
4135 struct btrfs_block_rsv *block_rsv,
4137 enum btrfs_reserve_flush_enum flush)
4139 struct btrfs_space_info *space_info = block_rsv->space_info;
4141 u64 num_bytes = orig_bytes;
4142 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4144 bool flushing = false;
4148 spin_lock(&space_info->lock);
4150 * We only want to wait if somebody other than us is flushing and we
4151 * are actually allowed to flush all things.
4153 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4154 space_info->flush) {
4155 spin_unlock(&space_info->lock);
4157 * If we have a trans handle we can't wait because the flusher
4158 * may have to commit the transaction, which would mean we would
4159 * deadlock since we are waiting for the flusher to finish, but
4160 * hold the current transaction open.
4162 if (current->journal_info)
4164 ret = wait_event_killable(space_info->wait, !space_info->flush);
4165 /* Must have been killed, return */
4169 spin_lock(&space_info->lock);
4173 used = space_info->bytes_used + space_info->bytes_reserved +
4174 space_info->bytes_pinned + space_info->bytes_readonly +
4175 space_info->bytes_may_use;
4178 * The idea here is that we've not already over-reserved the block group
4179 * then we can go ahead and save our reservation first and then start
4180 * flushing if we need to. Otherwise if we've already overcommitted
4181 * lets start flushing stuff first and then come back and try to make
4184 if (used <= space_info->total_bytes) {
4185 if (used + orig_bytes <= space_info->total_bytes) {
4186 space_info->bytes_may_use += orig_bytes;
4187 trace_btrfs_space_reservation(root->fs_info,
4188 "space_info", space_info->flags, orig_bytes, 1);
4192 * Ok set num_bytes to orig_bytes since we aren't
4193 * overocmmitted, this way we only try and reclaim what
4196 num_bytes = orig_bytes;
4200 * Ok we're over committed, set num_bytes to the overcommitted
4201 * amount plus the amount of bytes that we need for this
4204 num_bytes = used - space_info->total_bytes +
4208 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4209 space_info->bytes_may_use += orig_bytes;
4210 trace_btrfs_space_reservation(root->fs_info, "space_info",
4211 space_info->flags, orig_bytes,
4217 * Couldn't make our reservation, save our place so while we're trying
4218 * to reclaim space we can actually use it instead of somebody else
4219 * stealing it from us.
4221 * We make the other tasks wait for the flush only when we can flush
4224 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4226 space_info->flush = 1;
4229 spin_unlock(&space_info->lock);
4231 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4234 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4239 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4240 * would happen. So skip delalloc flush.
4242 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4243 (flush_state == FLUSH_DELALLOC ||
4244 flush_state == FLUSH_DELALLOC_WAIT))
4245 flush_state = ALLOC_CHUNK;
4249 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4250 flush_state < COMMIT_TRANS)
4252 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4253 flush_state <= COMMIT_TRANS)
4257 if (ret == -ENOSPC &&
4258 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4259 struct btrfs_block_rsv *global_rsv =
4260 &root->fs_info->global_block_rsv;
4262 if (block_rsv != global_rsv &&
4263 !block_rsv_use_bytes(global_rsv, orig_bytes))
4267 spin_lock(&space_info->lock);
4268 space_info->flush = 0;
4269 wake_up_all(&space_info->wait);
4270 spin_unlock(&space_info->lock);
4275 static struct btrfs_block_rsv *get_block_rsv(
4276 const struct btrfs_trans_handle *trans,
4277 const struct btrfs_root *root)
4279 struct btrfs_block_rsv *block_rsv = NULL;
4282 block_rsv = trans->block_rsv;
4284 if (root == root->fs_info->csum_root && trans->adding_csums)
4285 block_rsv = trans->block_rsv;
4288 block_rsv = root->block_rsv;
4291 block_rsv = &root->fs_info->empty_block_rsv;
4296 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4300 spin_lock(&block_rsv->lock);
4301 if (block_rsv->reserved >= num_bytes) {
4302 block_rsv->reserved -= num_bytes;
4303 if (block_rsv->reserved < block_rsv->size)
4304 block_rsv->full = 0;
4307 spin_unlock(&block_rsv->lock);
4311 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4312 u64 num_bytes, int update_size)
4314 spin_lock(&block_rsv->lock);
4315 block_rsv->reserved += num_bytes;
4317 block_rsv->size += num_bytes;
4318 else if (block_rsv->reserved >= block_rsv->size)
4319 block_rsv->full = 1;
4320 spin_unlock(&block_rsv->lock);
4323 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4324 struct btrfs_block_rsv *block_rsv,
4325 struct btrfs_block_rsv *dest, u64 num_bytes)
4327 struct btrfs_space_info *space_info = block_rsv->space_info;
4329 spin_lock(&block_rsv->lock);
4330 if (num_bytes == (u64)-1)
4331 num_bytes = block_rsv->size;
4332 block_rsv->size -= num_bytes;
4333 if (block_rsv->reserved >= block_rsv->size) {
4334 num_bytes = block_rsv->reserved - block_rsv->size;
4335 block_rsv->reserved = block_rsv->size;
4336 block_rsv->full = 1;
4340 spin_unlock(&block_rsv->lock);
4342 if (num_bytes > 0) {
4344 spin_lock(&dest->lock);
4348 bytes_to_add = dest->size - dest->reserved;
4349 bytes_to_add = min(num_bytes, bytes_to_add);
4350 dest->reserved += bytes_to_add;
4351 if (dest->reserved >= dest->size)
4353 num_bytes -= bytes_to_add;
4355 spin_unlock(&dest->lock);
4358 spin_lock(&space_info->lock);
4359 space_info->bytes_may_use -= num_bytes;
4360 trace_btrfs_space_reservation(fs_info, "space_info",
4361 space_info->flags, num_bytes, 0);
4362 space_info->reservation_progress++;
4363 spin_unlock(&space_info->lock);
4368 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4369 struct btrfs_block_rsv *dst, u64 num_bytes)
4373 ret = block_rsv_use_bytes(src, num_bytes);
4377 block_rsv_add_bytes(dst, num_bytes, 1);
4381 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4383 memset(rsv, 0, sizeof(*rsv));
4384 spin_lock_init(&rsv->lock);
4388 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4389 unsigned short type)
4391 struct btrfs_block_rsv *block_rsv;
4392 struct btrfs_fs_info *fs_info = root->fs_info;
4394 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4398 btrfs_init_block_rsv(block_rsv, type);
4399 block_rsv->space_info = __find_space_info(fs_info,
4400 BTRFS_BLOCK_GROUP_METADATA);
4404 void btrfs_free_block_rsv(struct btrfs_root *root,
4405 struct btrfs_block_rsv *rsv)
4409 btrfs_block_rsv_release(root, rsv, (u64)-1);
4413 int btrfs_block_rsv_add(struct btrfs_root *root,
4414 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4415 enum btrfs_reserve_flush_enum flush)
4422 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4424 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4431 int btrfs_block_rsv_check(struct btrfs_root *root,
4432 struct btrfs_block_rsv *block_rsv, int min_factor)
4440 spin_lock(&block_rsv->lock);
4441 num_bytes = div_factor(block_rsv->size, min_factor);
4442 if (block_rsv->reserved >= num_bytes)
4444 spin_unlock(&block_rsv->lock);
4449 int btrfs_block_rsv_refill(struct btrfs_root *root,
4450 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4451 enum btrfs_reserve_flush_enum flush)
4459 spin_lock(&block_rsv->lock);
4460 num_bytes = min_reserved;
4461 if (block_rsv->reserved >= num_bytes)
4464 num_bytes -= block_rsv->reserved;
4465 spin_unlock(&block_rsv->lock);
4470 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4472 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4479 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4480 struct btrfs_block_rsv *dst_rsv,
4483 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4486 void btrfs_block_rsv_release(struct btrfs_root *root,
4487 struct btrfs_block_rsv *block_rsv,
4490 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4491 if (global_rsv->full || global_rsv == block_rsv ||
4492 block_rsv->space_info != global_rsv->space_info)
4494 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4499 * helper to calculate size of global block reservation.
4500 * the desired value is sum of space used by extent tree,
4501 * checksum tree and root tree
4503 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4505 struct btrfs_space_info *sinfo;
4509 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4511 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4512 spin_lock(&sinfo->lock);
4513 data_used = sinfo->bytes_used;
4514 spin_unlock(&sinfo->lock);
4516 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4517 spin_lock(&sinfo->lock);
4518 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4520 meta_used = sinfo->bytes_used;
4521 spin_unlock(&sinfo->lock);
4523 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4525 num_bytes += div64_u64(data_used + meta_used, 50);
4527 if (num_bytes * 3 > meta_used)
4528 num_bytes = div64_u64(meta_used, 3);
4530 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4533 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4535 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4536 struct btrfs_space_info *sinfo = block_rsv->space_info;
4539 num_bytes = calc_global_metadata_size(fs_info);
4541 spin_lock(&sinfo->lock);
4542 spin_lock(&block_rsv->lock);
4544 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4546 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4547 sinfo->bytes_reserved + sinfo->bytes_readonly +
4548 sinfo->bytes_may_use;
4550 if (sinfo->total_bytes > num_bytes) {
4551 num_bytes = sinfo->total_bytes - num_bytes;
4552 block_rsv->reserved += num_bytes;
4553 sinfo->bytes_may_use += num_bytes;
4554 trace_btrfs_space_reservation(fs_info, "space_info",
4555 sinfo->flags, num_bytes, 1);
4558 if (block_rsv->reserved >= block_rsv->size) {
4559 num_bytes = block_rsv->reserved - block_rsv->size;
4560 sinfo->bytes_may_use -= num_bytes;
4561 trace_btrfs_space_reservation(fs_info, "space_info",
4562 sinfo->flags, num_bytes, 0);
4563 sinfo->reservation_progress++;
4564 block_rsv->reserved = block_rsv->size;
4565 block_rsv->full = 1;
4568 spin_unlock(&block_rsv->lock);
4569 spin_unlock(&sinfo->lock);
4572 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4574 struct btrfs_space_info *space_info;
4576 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4577 fs_info->chunk_block_rsv.space_info = space_info;
4579 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4580 fs_info->global_block_rsv.space_info = space_info;
4581 fs_info->delalloc_block_rsv.space_info = space_info;
4582 fs_info->trans_block_rsv.space_info = space_info;
4583 fs_info->empty_block_rsv.space_info = space_info;
4584 fs_info->delayed_block_rsv.space_info = space_info;
4586 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4587 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4588 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4589 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4590 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4592 update_global_block_rsv(fs_info);
4595 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4597 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4599 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4600 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4601 WARN_ON(fs_info->trans_block_rsv.size > 0);
4602 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4603 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4604 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4605 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4606 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4609 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4610 struct btrfs_root *root)
4612 if (!trans->block_rsv)
4615 if (!trans->bytes_reserved)
4618 trace_btrfs_space_reservation(root->fs_info, "transaction",
4619 trans->transid, trans->bytes_reserved, 0);
4620 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4621 trans->bytes_reserved = 0;
4624 /* Can only return 0 or -ENOSPC */
4625 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4626 struct inode *inode)
4628 struct btrfs_root *root = BTRFS_I(inode)->root;
4629 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4630 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4633 * We need to hold space in order to delete our orphan item once we've
4634 * added it, so this takes the reservation so we can release it later
4635 * when we are truly done with the orphan item.
4637 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4638 trace_btrfs_space_reservation(root->fs_info, "orphan",
4639 btrfs_ino(inode), num_bytes, 1);
4640 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4643 void btrfs_orphan_release_metadata(struct inode *inode)
4645 struct btrfs_root *root = BTRFS_I(inode)->root;
4646 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4647 trace_btrfs_space_reservation(root->fs_info, "orphan",
4648 btrfs_ino(inode), num_bytes, 0);
4649 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4653 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4654 * root: the root of the parent directory
4655 * rsv: block reservation
4656 * items: the number of items that we need do reservation
4657 * qgroup_reserved: used to return the reserved size in qgroup
4659 * This function is used to reserve the space for snapshot/subvolume
4660 * creation and deletion. Those operations are different with the
4661 * common file/directory operations, they change two fs/file trees
4662 * and root tree, the number of items that the qgroup reserves is
4663 * different with the free space reservation. So we can not use
4664 * the space reseravtion mechanism in start_transaction().
4666 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4667 struct btrfs_block_rsv *rsv,
4669 u64 *qgroup_reserved)
4674 if (root->fs_info->quota_enabled) {
4675 /* One for parent inode, two for dir entries */
4676 num_bytes = 3 * root->leafsize;
4677 ret = btrfs_qgroup_reserve(root, num_bytes);
4684 *qgroup_reserved = num_bytes;
4686 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4687 rsv->space_info = __find_space_info(root->fs_info,
4688 BTRFS_BLOCK_GROUP_METADATA);
4689 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4690 BTRFS_RESERVE_FLUSH_ALL);
4692 if (*qgroup_reserved)
4693 btrfs_qgroup_free(root, *qgroup_reserved);
4699 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4700 struct btrfs_block_rsv *rsv,
4701 u64 qgroup_reserved)
4703 btrfs_block_rsv_release(root, rsv, (u64)-1);
4704 if (qgroup_reserved)
4705 btrfs_qgroup_free(root, qgroup_reserved);
4709 * drop_outstanding_extent - drop an outstanding extent
4710 * @inode: the inode we're dropping the extent for
4712 * This is called when we are freeing up an outstanding extent, either called
4713 * after an error or after an extent is written. This will return the number of
4714 * reserved extents that need to be freed. This must be called with
4715 * BTRFS_I(inode)->lock held.
4717 static unsigned drop_outstanding_extent(struct inode *inode)
4719 unsigned drop_inode_space = 0;
4720 unsigned dropped_extents = 0;
4722 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4723 BTRFS_I(inode)->outstanding_extents--;
4725 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4726 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4727 &BTRFS_I(inode)->runtime_flags))
4728 drop_inode_space = 1;
4731 * If we have more or the same amount of outsanding extents than we have
4732 * reserved then we need to leave the reserved extents count alone.
4734 if (BTRFS_I(inode)->outstanding_extents >=
4735 BTRFS_I(inode)->reserved_extents)
4736 return drop_inode_space;
4738 dropped_extents = BTRFS_I(inode)->reserved_extents -
4739 BTRFS_I(inode)->outstanding_extents;
4740 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4741 return dropped_extents + drop_inode_space;
4745 * calc_csum_metadata_size - return the amount of metada space that must be
4746 * reserved/free'd for the given bytes.
4747 * @inode: the inode we're manipulating
4748 * @num_bytes: the number of bytes in question
4749 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4751 * This adjusts the number of csum_bytes in the inode and then returns the
4752 * correct amount of metadata that must either be reserved or freed. We
4753 * calculate how many checksums we can fit into one leaf and then divide the
4754 * number of bytes that will need to be checksumed by this value to figure out
4755 * how many checksums will be required. If we are adding bytes then the number
4756 * may go up and we will return the number of additional bytes that must be
4757 * reserved. If it is going down we will return the number of bytes that must
4760 * This must be called with BTRFS_I(inode)->lock held.
4762 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4765 struct btrfs_root *root = BTRFS_I(inode)->root;
4767 int num_csums_per_leaf;
4771 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4772 BTRFS_I(inode)->csum_bytes == 0)
4775 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4777 BTRFS_I(inode)->csum_bytes += num_bytes;
4779 BTRFS_I(inode)->csum_bytes -= num_bytes;
4780 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4781 num_csums_per_leaf = (int)div64_u64(csum_size,
4782 sizeof(struct btrfs_csum_item) +
4783 sizeof(struct btrfs_disk_key));
4784 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4785 num_csums = num_csums + num_csums_per_leaf - 1;
4786 num_csums = num_csums / num_csums_per_leaf;
4788 old_csums = old_csums + num_csums_per_leaf - 1;
4789 old_csums = old_csums / num_csums_per_leaf;
4791 /* No change, no need to reserve more */
4792 if (old_csums == num_csums)
4796 return btrfs_calc_trans_metadata_size(root,
4797 num_csums - old_csums);
4799 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4802 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4804 struct btrfs_root *root = BTRFS_I(inode)->root;
4805 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4808 unsigned nr_extents = 0;
4809 int extra_reserve = 0;
4810 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4812 bool delalloc_lock = true;
4816 /* If we are a free space inode we need to not flush since we will be in
4817 * the middle of a transaction commit. We also don't need the delalloc
4818 * mutex since we won't race with anybody. We need this mostly to make
4819 * lockdep shut its filthy mouth.
4821 if (btrfs_is_free_space_inode(inode)) {
4822 flush = BTRFS_RESERVE_NO_FLUSH;
4823 delalloc_lock = false;
4826 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4827 btrfs_transaction_in_commit(root->fs_info))
4828 schedule_timeout(1);
4831 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4833 num_bytes = ALIGN(num_bytes, root->sectorsize);
4835 spin_lock(&BTRFS_I(inode)->lock);
4836 BTRFS_I(inode)->outstanding_extents++;
4838 if (BTRFS_I(inode)->outstanding_extents >
4839 BTRFS_I(inode)->reserved_extents)
4840 nr_extents = BTRFS_I(inode)->outstanding_extents -
4841 BTRFS_I(inode)->reserved_extents;
4844 * Add an item to reserve for updating the inode when we complete the
4847 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4848 &BTRFS_I(inode)->runtime_flags)) {
4853 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4854 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4855 csum_bytes = BTRFS_I(inode)->csum_bytes;
4856 spin_unlock(&BTRFS_I(inode)->lock);
4858 if (root->fs_info->quota_enabled) {
4859 ret = btrfs_qgroup_reserve(root, num_bytes +
4860 nr_extents * root->leafsize);
4865 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4866 if (unlikely(ret)) {
4867 if (root->fs_info->quota_enabled)
4868 btrfs_qgroup_free(root, num_bytes +
4869 nr_extents * root->leafsize);
4873 spin_lock(&BTRFS_I(inode)->lock);
4874 if (extra_reserve) {
4875 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4876 &BTRFS_I(inode)->runtime_flags);
4879 BTRFS_I(inode)->reserved_extents += nr_extents;
4880 spin_unlock(&BTRFS_I(inode)->lock);
4883 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4886 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4887 btrfs_ino(inode), to_reserve, 1);
4888 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4893 spin_lock(&BTRFS_I(inode)->lock);
4894 dropped = drop_outstanding_extent(inode);
4896 * If the inodes csum_bytes is the same as the original
4897 * csum_bytes then we know we haven't raced with any free()ers
4898 * so we can just reduce our inodes csum bytes and carry on.
4900 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4901 calc_csum_metadata_size(inode, num_bytes, 0);
4903 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4907 * This is tricky, but first we need to figure out how much we
4908 * free'd from any free-ers that occured during this
4909 * reservation, so we reset ->csum_bytes to the csum_bytes
4910 * before we dropped our lock, and then call the free for the
4911 * number of bytes that were freed while we were trying our
4914 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4915 BTRFS_I(inode)->csum_bytes = csum_bytes;
4916 to_free = calc_csum_metadata_size(inode, bytes, 0);
4920 * Now we need to see how much we would have freed had we not
4921 * been making this reservation and our ->csum_bytes were not
4922 * artificially inflated.
4924 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4925 bytes = csum_bytes - orig_csum_bytes;
4926 bytes = calc_csum_metadata_size(inode, bytes, 0);
4929 * Now reset ->csum_bytes to what it should be. If bytes is
4930 * more than to_free then we would have free'd more space had we
4931 * not had an artificially high ->csum_bytes, so we need to free
4932 * the remainder. If bytes is the same or less then we don't
4933 * need to do anything, the other free-ers did the correct
4936 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4937 if (bytes > to_free)
4938 to_free = bytes - to_free;
4942 spin_unlock(&BTRFS_I(inode)->lock);
4944 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4947 btrfs_block_rsv_release(root, block_rsv, to_free);
4948 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4949 btrfs_ino(inode), to_free, 0);
4952 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4957 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4958 * @inode: the inode to release the reservation for
4959 * @num_bytes: the number of bytes we're releasing
4961 * This will release the metadata reservation for an inode. This can be called
4962 * once we complete IO for a given set of bytes to release their metadata
4965 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4967 struct btrfs_root *root = BTRFS_I(inode)->root;
4971 num_bytes = ALIGN(num_bytes, root->sectorsize);
4972 spin_lock(&BTRFS_I(inode)->lock);
4973 dropped = drop_outstanding_extent(inode);
4976 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4977 spin_unlock(&BTRFS_I(inode)->lock);
4979 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4981 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4982 btrfs_ino(inode), to_free, 0);
4983 if (root->fs_info->quota_enabled) {
4984 btrfs_qgroup_free(root, num_bytes +
4985 dropped * root->leafsize);
4988 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4993 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4994 * @inode: inode we're writing to
4995 * @num_bytes: the number of bytes we want to allocate
4997 * This will do the following things
4999 * o reserve space in the data space info for num_bytes
5000 * o reserve space in the metadata space info based on number of outstanding
5001 * extents and how much csums will be needed
5002 * o add to the inodes ->delalloc_bytes
5003 * o add it to the fs_info's delalloc inodes list.
5005 * This will return 0 for success and -ENOSPC if there is no space left.
5007 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5011 ret = btrfs_check_data_free_space(inode, num_bytes);
5015 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5017 btrfs_free_reserved_data_space(inode, num_bytes);
5025 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5026 * @inode: inode we're releasing space for
5027 * @num_bytes: the number of bytes we want to free up
5029 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5030 * called in the case that we don't need the metadata AND data reservations
5031 * anymore. So if there is an error or we insert an inline extent.
5033 * This function will release the metadata space that was not used and will
5034 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5035 * list if there are no delalloc bytes left.
5037 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5039 btrfs_delalloc_release_metadata(inode, num_bytes);
5040 btrfs_free_reserved_data_space(inode, num_bytes);
5043 static int update_block_group(struct btrfs_root *root,
5044 u64 bytenr, u64 num_bytes, int alloc)
5046 struct btrfs_block_group_cache *cache = NULL;
5047 struct btrfs_fs_info *info = root->fs_info;
5048 u64 total = num_bytes;
5053 /* block accounting for super block */
5054 spin_lock(&info->delalloc_lock);
5055 old_val = btrfs_super_bytes_used(info->super_copy);
5057 old_val += num_bytes;
5059 old_val -= num_bytes;
5060 btrfs_set_super_bytes_used(info->super_copy, old_val);
5061 spin_unlock(&info->delalloc_lock);
5064 cache = btrfs_lookup_block_group(info, bytenr);
5067 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5068 BTRFS_BLOCK_GROUP_RAID1 |
5069 BTRFS_BLOCK_GROUP_RAID10))
5074 * If this block group has free space cache written out, we
5075 * need to make sure to load it if we are removing space. This
5076 * is because we need the unpinning stage to actually add the
5077 * space back to the block group, otherwise we will leak space.
5079 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5080 cache_block_group(cache, 1);
5082 byte_in_group = bytenr - cache->key.objectid;
5083 WARN_ON(byte_in_group > cache->key.offset);
5085 spin_lock(&cache->space_info->lock);
5086 spin_lock(&cache->lock);
5088 if (btrfs_test_opt(root, SPACE_CACHE) &&
5089 cache->disk_cache_state < BTRFS_DC_CLEAR)
5090 cache->disk_cache_state = BTRFS_DC_CLEAR;
5093 old_val = btrfs_block_group_used(&cache->item);
5094 num_bytes = min(total, cache->key.offset - byte_in_group);
5096 old_val += num_bytes;
5097 btrfs_set_block_group_used(&cache->item, old_val);
5098 cache->reserved -= num_bytes;
5099 cache->space_info->bytes_reserved -= num_bytes;
5100 cache->space_info->bytes_used += num_bytes;
5101 cache->space_info->disk_used += num_bytes * factor;
5102 spin_unlock(&cache->lock);
5103 spin_unlock(&cache->space_info->lock);
5105 old_val -= num_bytes;
5106 btrfs_set_block_group_used(&cache->item, old_val);
5107 cache->pinned += num_bytes;
5108 cache->space_info->bytes_pinned += num_bytes;
5109 cache->space_info->bytes_used -= num_bytes;
5110 cache->space_info->disk_used -= num_bytes * factor;
5111 spin_unlock(&cache->lock);
5112 spin_unlock(&cache->space_info->lock);
5114 set_extent_dirty(info->pinned_extents,
5115 bytenr, bytenr + num_bytes - 1,
5116 GFP_NOFS | __GFP_NOFAIL);
5118 btrfs_put_block_group(cache);
5120 bytenr += num_bytes;
5125 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5127 struct btrfs_block_group_cache *cache;
5130 spin_lock(&root->fs_info->block_group_cache_lock);
5131 bytenr = root->fs_info->first_logical_byte;
5132 spin_unlock(&root->fs_info->block_group_cache_lock);
5134 if (bytenr < (u64)-1)
5137 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5141 bytenr = cache->key.objectid;
5142 btrfs_put_block_group(cache);
5147 static int pin_down_extent(struct btrfs_root *root,
5148 struct btrfs_block_group_cache *cache,
5149 u64 bytenr, u64 num_bytes, int reserved)
5151 spin_lock(&cache->space_info->lock);
5152 spin_lock(&cache->lock);
5153 cache->pinned += num_bytes;
5154 cache->space_info->bytes_pinned += num_bytes;
5156 cache->reserved -= num_bytes;
5157 cache->space_info->bytes_reserved -= num_bytes;
5159 spin_unlock(&cache->lock);
5160 spin_unlock(&cache->space_info->lock);
5162 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5163 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5168 * this function must be called within transaction
5170 int btrfs_pin_extent(struct btrfs_root *root,
5171 u64 bytenr, u64 num_bytes, int reserved)
5173 struct btrfs_block_group_cache *cache;
5175 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5176 BUG_ON(!cache); /* Logic error */
5178 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5180 btrfs_put_block_group(cache);
5185 * this function must be called within transaction
5187 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5188 u64 bytenr, u64 num_bytes)
5190 struct btrfs_block_group_cache *cache;
5192 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5193 BUG_ON(!cache); /* Logic error */
5196 * pull in the free space cache (if any) so that our pin
5197 * removes the free space from the cache. We have load_only set
5198 * to one because the slow code to read in the free extents does check
5199 * the pinned extents.
5201 cache_block_group(cache, 1);
5203 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5205 /* remove us from the free space cache (if we're there at all) */
5206 btrfs_remove_free_space(cache, bytenr, num_bytes);
5207 btrfs_put_block_group(cache);
5212 * btrfs_update_reserved_bytes - update the block_group and space info counters
5213 * @cache: The cache we are manipulating
5214 * @num_bytes: The number of bytes in question
5215 * @reserve: One of the reservation enums
5217 * This is called by the allocator when it reserves space, or by somebody who is
5218 * freeing space that was never actually used on disk. For example if you
5219 * reserve some space for a new leaf in transaction A and before transaction A
5220 * commits you free that leaf, you call this with reserve set to 0 in order to
5221 * clear the reservation.
5223 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5224 * ENOSPC accounting. For data we handle the reservation through clearing the
5225 * delalloc bits in the io_tree. We have to do this since we could end up
5226 * allocating less disk space for the amount of data we have reserved in the
5227 * case of compression.
5229 * If this is a reservation and the block group has become read only we cannot
5230 * make the reservation and return -EAGAIN, otherwise this function always
5233 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5234 u64 num_bytes, int reserve)
5236 struct btrfs_space_info *space_info = cache->space_info;
5239 spin_lock(&space_info->lock);
5240 spin_lock(&cache->lock);
5241 if (reserve != RESERVE_FREE) {
5245 cache->reserved += num_bytes;
5246 space_info->bytes_reserved += num_bytes;
5247 if (reserve == RESERVE_ALLOC) {
5248 trace_btrfs_space_reservation(cache->fs_info,
5249 "space_info", space_info->flags,
5251 space_info->bytes_may_use -= num_bytes;
5256 space_info->bytes_readonly += num_bytes;
5257 cache->reserved -= num_bytes;
5258 space_info->bytes_reserved -= num_bytes;
5259 space_info->reservation_progress++;
5261 spin_unlock(&cache->lock);
5262 spin_unlock(&space_info->lock);
5266 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5267 struct btrfs_root *root)
5269 struct btrfs_fs_info *fs_info = root->fs_info;
5270 struct btrfs_caching_control *next;
5271 struct btrfs_caching_control *caching_ctl;
5272 struct btrfs_block_group_cache *cache;
5274 down_write(&fs_info->extent_commit_sem);
5276 list_for_each_entry_safe(caching_ctl, next,
5277 &fs_info->caching_block_groups, list) {
5278 cache = caching_ctl->block_group;
5279 if (block_group_cache_done(cache)) {
5280 cache->last_byte_to_unpin = (u64)-1;
5281 list_del_init(&caching_ctl->list);
5282 put_caching_control(caching_ctl);
5284 cache->last_byte_to_unpin = caching_ctl->progress;
5288 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5289 fs_info->pinned_extents = &fs_info->freed_extents[1];
5291 fs_info->pinned_extents = &fs_info->freed_extents[0];
5293 up_write(&fs_info->extent_commit_sem);
5295 update_global_block_rsv(fs_info);
5298 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5300 struct btrfs_fs_info *fs_info = root->fs_info;
5301 struct btrfs_block_group_cache *cache = NULL;
5302 struct btrfs_space_info *space_info;
5303 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5307 while (start <= end) {
5310 start >= cache->key.objectid + cache->key.offset) {
5312 btrfs_put_block_group(cache);
5313 cache = btrfs_lookup_block_group(fs_info, start);
5314 BUG_ON(!cache); /* Logic error */
5317 len = cache->key.objectid + cache->key.offset - start;
5318 len = min(len, end + 1 - start);
5320 if (start < cache->last_byte_to_unpin) {
5321 len = min(len, cache->last_byte_to_unpin - start);
5322 btrfs_add_free_space(cache, start, len);
5326 space_info = cache->space_info;
5328 spin_lock(&space_info->lock);
5329 spin_lock(&cache->lock);
5330 cache->pinned -= len;
5331 space_info->bytes_pinned -= len;
5333 space_info->bytes_readonly += len;
5336 spin_unlock(&cache->lock);
5337 if (!readonly && global_rsv->space_info == space_info) {
5338 spin_lock(&global_rsv->lock);
5339 if (!global_rsv->full) {
5340 len = min(len, global_rsv->size -
5341 global_rsv->reserved);
5342 global_rsv->reserved += len;
5343 space_info->bytes_may_use += len;
5344 if (global_rsv->reserved >= global_rsv->size)
5345 global_rsv->full = 1;
5347 spin_unlock(&global_rsv->lock);
5349 spin_unlock(&space_info->lock);
5353 btrfs_put_block_group(cache);
5357 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5358 struct btrfs_root *root)
5360 struct btrfs_fs_info *fs_info = root->fs_info;
5361 struct extent_io_tree *unpin;
5369 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5370 unpin = &fs_info->freed_extents[1];
5372 unpin = &fs_info->freed_extents[0];
5375 ret = find_first_extent_bit(unpin, 0, &start, &end,
5376 EXTENT_DIRTY, NULL);
5380 if (btrfs_test_opt(root, DISCARD))
5381 ret = btrfs_discard_extent(root, start,
5382 end + 1 - start, NULL);
5384 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5385 unpin_extent_range(root, start, end);
5392 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5393 struct btrfs_root *root,
5394 u64 bytenr, u64 num_bytes, u64 parent,
5395 u64 root_objectid, u64 owner_objectid,
5396 u64 owner_offset, int refs_to_drop,
5397 struct btrfs_delayed_extent_op *extent_op)
5399 struct btrfs_key key;
5400 struct btrfs_path *path;
5401 struct btrfs_fs_info *info = root->fs_info;
5402 struct btrfs_root *extent_root = info->extent_root;
5403 struct extent_buffer *leaf;
5404 struct btrfs_extent_item *ei;
5405 struct btrfs_extent_inline_ref *iref;
5408 int extent_slot = 0;
5409 int found_extent = 0;
5413 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5416 path = btrfs_alloc_path();
5421 path->leave_spinning = 1;
5423 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5424 BUG_ON(!is_data && refs_to_drop != 1);
5427 skinny_metadata = 0;
5429 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5430 bytenr, num_bytes, parent,
5431 root_objectid, owner_objectid,
5434 extent_slot = path->slots[0];
5435 while (extent_slot >= 0) {
5436 btrfs_item_key_to_cpu(path->nodes[0], &key,
5438 if (key.objectid != bytenr)
5440 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5441 key.offset == num_bytes) {
5445 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5446 key.offset == owner_objectid) {
5450 if (path->slots[0] - extent_slot > 5)
5454 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5455 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5456 if (found_extent && item_size < sizeof(*ei))
5459 if (!found_extent) {
5461 ret = remove_extent_backref(trans, extent_root, path,
5465 btrfs_abort_transaction(trans, extent_root, ret);
5468 btrfs_release_path(path);
5469 path->leave_spinning = 1;
5471 key.objectid = bytenr;
5472 key.type = BTRFS_EXTENT_ITEM_KEY;
5473 key.offset = num_bytes;
5475 if (!is_data && skinny_metadata) {
5476 key.type = BTRFS_METADATA_ITEM_KEY;
5477 key.offset = owner_objectid;
5480 ret = btrfs_search_slot(trans, extent_root,
5482 if (ret > 0 && skinny_metadata && path->slots[0]) {
5484 * Couldn't find our skinny metadata item,
5485 * see if we have ye olde extent item.
5488 btrfs_item_key_to_cpu(path->nodes[0], &key,
5490 if (key.objectid == bytenr &&
5491 key.type == BTRFS_EXTENT_ITEM_KEY &&
5492 key.offset == num_bytes)
5496 if (ret > 0 && skinny_metadata) {
5497 skinny_metadata = false;
5498 key.type = BTRFS_EXTENT_ITEM_KEY;
5499 key.offset = num_bytes;
5500 btrfs_release_path(path);
5501 ret = btrfs_search_slot(trans, extent_root,
5506 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5507 ret, (unsigned long long)bytenr);
5509 btrfs_print_leaf(extent_root,
5513 btrfs_abort_transaction(trans, extent_root, ret);
5516 extent_slot = path->slots[0];
5518 } else if (ret == -ENOENT) {
5519 btrfs_print_leaf(extent_root, path->nodes[0]);
5522 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5523 (unsigned long long)bytenr,
5524 (unsigned long long)parent,
5525 (unsigned long long)root_objectid,
5526 (unsigned long long)owner_objectid,
5527 (unsigned long long)owner_offset);
5529 btrfs_abort_transaction(trans, extent_root, ret);
5533 leaf = path->nodes[0];
5534 item_size = btrfs_item_size_nr(leaf, extent_slot);
5535 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5536 if (item_size < sizeof(*ei)) {
5537 BUG_ON(found_extent || extent_slot != path->slots[0]);
5538 ret = convert_extent_item_v0(trans, extent_root, path,
5541 btrfs_abort_transaction(trans, extent_root, ret);
5545 btrfs_release_path(path);
5546 path->leave_spinning = 1;
5548 key.objectid = bytenr;
5549 key.type = BTRFS_EXTENT_ITEM_KEY;
5550 key.offset = num_bytes;
5552 ret = btrfs_search_slot(trans, extent_root, &key, path,
5555 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5556 ret, (unsigned long long)bytenr);
5557 btrfs_print_leaf(extent_root, path->nodes[0]);
5560 btrfs_abort_transaction(trans, extent_root, ret);
5564 extent_slot = path->slots[0];
5565 leaf = path->nodes[0];
5566 item_size = btrfs_item_size_nr(leaf, extent_slot);
5569 BUG_ON(item_size < sizeof(*ei));
5570 ei = btrfs_item_ptr(leaf, extent_slot,
5571 struct btrfs_extent_item);
5572 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5573 key.type == BTRFS_EXTENT_ITEM_KEY) {
5574 struct btrfs_tree_block_info *bi;
5575 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5576 bi = (struct btrfs_tree_block_info *)(ei + 1);
5577 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5580 refs = btrfs_extent_refs(leaf, ei);
5581 BUG_ON(refs < refs_to_drop);
5582 refs -= refs_to_drop;
5586 __run_delayed_extent_op(extent_op, leaf, ei);
5588 * In the case of inline back ref, reference count will
5589 * be updated by remove_extent_backref
5592 BUG_ON(!found_extent);
5594 btrfs_set_extent_refs(leaf, ei, refs);
5595 btrfs_mark_buffer_dirty(leaf);
5598 ret = remove_extent_backref(trans, extent_root, path,
5602 btrfs_abort_transaction(trans, extent_root, ret);
5608 BUG_ON(is_data && refs_to_drop !=
5609 extent_data_ref_count(root, path, iref));
5611 BUG_ON(path->slots[0] != extent_slot);
5613 BUG_ON(path->slots[0] != extent_slot + 1);
5614 path->slots[0] = extent_slot;
5619 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5622 btrfs_abort_transaction(trans, extent_root, ret);
5625 btrfs_release_path(path);
5628 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5630 btrfs_abort_transaction(trans, extent_root, ret);
5635 ret = update_block_group(root, bytenr, num_bytes, 0);
5637 btrfs_abort_transaction(trans, extent_root, ret);
5642 btrfs_free_path(path);
5647 * when we free an block, it is possible (and likely) that we free the last
5648 * delayed ref for that extent as well. This searches the delayed ref tree for
5649 * a given extent, and if there are no other delayed refs to be processed, it
5650 * removes it from the tree.
5652 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5653 struct btrfs_root *root, u64 bytenr)
5655 struct btrfs_delayed_ref_head *head;
5656 struct btrfs_delayed_ref_root *delayed_refs;
5657 struct btrfs_delayed_ref_node *ref;
5658 struct rb_node *node;
5661 delayed_refs = &trans->transaction->delayed_refs;
5662 spin_lock(&delayed_refs->lock);
5663 head = btrfs_find_delayed_ref_head(trans, bytenr);
5667 node = rb_prev(&head->node.rb_node);
5671 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5673 /* there are still entries for this ref, we can't drop it */
5674 if (ref->bytenr == bytenr)
5677 if (head->extent_op) {
5678 if (!head->must_insert_reserved)
5680 btrfs_free_delayed_extent_op(head->extent_op);
5681 head->extent_op = NULL;
5685 * waiting for the lock here would deadlock. If someone else has it
5686 * locked they are already in the process of dropping it anyway
5688 if (!mutex_trylock(&head->mutex))
5692 * at this point we have a head with no other entries. Go
5693 * ahead and process it.
5695 head->node.in_tree = 0;
5696 rb_erase(&head->node.rb_node, &delayed_refs->root);
5698 delayed_refs->num_entries--;
5701 * we don't take a ref on the node because we're removing it from the
5702 * tree, so we just steal the ref the tree was holding.
5704 delayed_refs->num_heads--;
5705 if (list_empty(&head->cluster))
5706 delayed_refs->num_heads_ready--;
5708 list_del_init(&head->cluster);
5709 spin_unlock(&delayed_refs->lock);
5711 BUG_ON(head->extent_op);
5712 if (head->must_insert_reserved)
5715 mutex_unlock(&head->mutex);
5716 btrfs_put_delayed_ref(&head->node);
5719 spin_unlock(&delayed_refs->lock);
5723 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5724 struct btrfs_root *root,
5725 struct extent_buffer *buf,
5726 u64 parent, int last_ref)
5728 struct btrfs_block_group_cache *cache = NULL;
5731 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5732 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5733 buf->start, buf->len,
5734 parent, root->root_key.objectid,
5735 btrfs_header_level(buf),
5736 BTRFS_DROP_DELAYED_REF, NULL, 0);
5737 BUG_ON(ret); /* -ENOMEM */
5743 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5745 if (btrfs_header_generation(buf) == trans->transid) {
5746 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5747 ret = check_ref_cleanup(trans, root, buf->start);
5752 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5753 pin_down_extent(root, cache, buf->start, buf->len, 1);
5757 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5759 btrfs_add_free_space(cache, buf->start, buf->len);
5760 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5764 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5767 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5768 btrfs_put_block_group(cache);
5771 /* Can return -ENOMEM */
5772 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5773 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5774 u64 owner, u64 offset, int for_cow)
5777 struct btrfs_fs_info *fs_info = root->fs_info;
5780 * tree log blocks never actually go into the extent allocation
5781 * tree, just update pinning info and exit early.
5783 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5784 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5785 /* unlocks the pinned mutex */
5786 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5788 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5789 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5791 parent, root_objectid, (int)owner,
5792 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5794 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5796 parent, root_objectid, owner,
5797 offset, BTRFS_DROP_DELAYED_REF,
5803 static u64 stripe_align(struct btrfs_root *root,
5804 struct btrfs_block_group_cache *cache,
5805 u64 val, u64 num_bytes)
5807 u64 ret = ALIGN(val, root->stripesize);
5812 * when we wait for progress in the block group caching, its because
5813 * our allocation attempt failed at least once. So, we must sleep
5814 * and let some progress happen before we try again.
5816 * This function will sleep at least once waiting for new free space to
5817 * show up, and then it will check the block group free space numbers
5818 * for our min num_bytes. Another option is to have it go ahead
5819 * and look in the rbtree for a free extent of a given size, but this
5823 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5826 struct btrfs_caching_control *caching_ctl;
5828 caching_ctl = get_caching_control(cache);
5832 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5833 (cache->free_space_ctl->free_space >= num_bytes));
5835 put_caching_control(caching_ctl);
5840 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5842 struct btrfs_caching_control *caching_ctl;
5844 caching_ctl = get_caching_control(cache);
5848 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5850 put_caching_control(caching_ctl);
5854 int __get_raid_index(u64 flags)
5856 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5857 return BTRFS_RAID_RAID10;
5858 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5859 return BTRFS_RAID_RAID1;
5860 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5861 return BTRFS_RAID_DUP;
5862 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5863 return BTRFS_RAID_RAID0;
5864 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5865 return BTRFS_RAID_RAID5;
5866 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5867 return BTRFS_RAID_RAID6;
5869 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
5872 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5874 return __get_raid_index(cache->flags);
5877 enum btrfs_loop_type {
5878 LOOP_CACHING_NOWAIT = 0,
5879 LOOP_CACHING_WAIT = 1,
5880 LOOP_ALLOC_CHUNK = 2,
5881 LOOP_NO_EMPTY_SIZE = 3,
5885 * walks the btree of allocated extents and find a hole of a given size.
5886 * The key ins is changed to record the hole:
5887 * ins->objectid == block start
5888 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5889 * ins->offset == number of blocks
5890 * Any available blocks before search_start are skipped.
5892 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5893 struct btrfs_root *orig_root,
5894 u64 num_bytes, u64 empty_size,
5895 u64 hint_byte, struct btrfs_key *ins,
5899 struct btrfs_root *root = orig_root->fs_info->extent_root;
5900 struct btrfs_free_cluster *last_ptr = NULL;
5901 struct btrfs_block_group_cache *block_group = NULL;
5902 struct btrfs_block_group_cache *used_block_group;
5903 u64 search_start = 0;
5904 int empty_cluster = 2 * 1024 * 1024;
5905 struct btrfs_space_info *space_info;
5907 int index = __get_raid_index(data);
5908 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5909 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5910 bool found_uncached_bg = false;
5911 bool failed_cluster_refill = false;
5912 bool failed_alloc = false;
5913 bool use_cluster = true;
5914 bool have_caching_bg = false;
5916 WARN_ON(num_bytes < root->sectorsize);
5917 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5921 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5923 space_info = __find_space_info(root->fs_info, data);
5925 btrfs_err(root->fs_info, "No space info for %llu", data);
5930 * If the space info is for both data and metadata it means we have a
5931 * small filesystem and we can't use the clustering stuff.
5933 if (btrfs_mixed_space_info(space_info))
5934 use_cluster = false;
5936 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5937 last_ptr = &root->fs_info->meta_alloc_cluster;
5938 if (!btrfs_test_opt(root, SSD))
5939 empty_cluster = 64 * 1024;
5942 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5943 btrfs_test_opt(root, SSD)) {
5944 last_ptr = &root->fs_info->data_alloc_cluster;
5948 spin_lock(&last_ptr->lock);
5949 if (last_ptr->block_group)
5950 hint_byte = last_ptr->window_start;
5951 spin_unlock(&last_ptr->lock);
5954 search_start = max(search_start, first_logical_byte(root, 0));
5955 search_start = max(search_start, hint_byte);
5960 if (search_start == hint_byte) {
5961 block_group = btrfs_lookup_block_group(root->fs_info,
5963 used_block_group = block_group;
5965 * we don't want to use the block group if it doesn't match our
5966 * allocation bits, or if its not cached.
5968 * However if we are re-searching with an ideal block group
5969 * picked out then we don't care that the block group is cached.
5971 if (block_group && block_group_bits(block_group, data) &&
5972 block_group->cached != BTRFS_CACHE_NO) {
5973 down_read(&space_info->groups_sem);
5974 if (list_empty(&block_group->list) ||
5977 * someone is removing this block group,
5978 * we can't jump into the have_block_group
5979 * target because our list pointers are not
5982 btrfs_put_block_group(block_group);
5983 up_read(&space_info->groups_sem);
5985 index = get_block_group_index(block_group);
5986 goto have_block_group;
5988 } else if (block_group) {
5989 btrfs_put_block_group(block_group);
5993 have_caching_bg = false;
5994 down_read(&space_info->groups_sem);
5995 list_for_each_entry(block_group, &space_info->block_groups[index],
6000 used_block_group = block_group;
6001 btrfs_get_block_group(block_group);
6002 search_start = block_group->key.objectid;
6005 * this can happen if we end up cycling through all the
6006 * raid types, but we want to make sure we only allocate
6007 * for the proper type.
6009 if (!block_group_bits(block_group, data)) {
6010 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6011 BTRFS_BLOCK_GROUP_RAID1 |
6012 BTRFS_BLOCK_GROUP_RAID5 |
6013 BTRFS_BLOCK_GROUP_RAID6 |
6014 BTRFS_BLOCK_GROUP_RAID10;
6017 * if they asked for extra copies and this block group
6018 * doesn't provide them, bail. This does allow us to
6019 * fill raid0 from raid1.
6021 if ((data & extra) && !(block_group->flags & extra))
6026 cached = block_group_cache_done(block_group);
6027 if (unlikely(!cached)) {
6028 found_uncached_bg = true;
6029 ret = cache_block_group(block_group, 0);
6034 if (unlikely(block_group->ro))
6038 * Ok we want to try and use the cluster allocator, so
6042 unsigned long aligned_cluster;
6044 * the refill lock keeps out other
6045 * people trying to start a new cluster
6047 spin_lock(&last_ptr->refill_lock);
6048 used_block_group = last_ptr->block_group;
6049 if (used_block_group != block_group &&
6050 (!used_block_group ||
6051 used_block_group->ro ||
6052 !block_group_bits(used_block_group, data))) {
6053 used_block_group = block_group;
6054 goto refill_cluster;
6057 if (used_block_group != block_group)
6058 btrfs_get_block_group(used_block_group);
6060 offset = btrfs_alloc_from_cluster(used_block_group,
6061 last_ptr, num_bytes, used_block_group->key.objectid);
6063 /* we have a block, we're done */
6064 spin_unlock(&last_ptr->refill_lock);
6065 trace_btrfs_reserve_extent_cluster(root,
6066 block_group, search_start, num_bytes);
6070 WARN_ON(last_ptr->block_group != used_block_group);
6071 if (used_block_group != block_group) {
6072 btrfs_put_block_group(used_block_group);
6073 used_block_group = block_group;
6076 BUG_ON(used_block_group != block_group);
6077 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6078 * set up a new clusters, so lets just skip it
6079 * and let the allocator find whatever block
6080 * it can find. If we reach this point, we
6081 * will have tried the cluster allocator
6082 * plenty of times and not have found
6083 * anything, so we are likely way too
6084 * fragmented for the clustering stuff to find
6087 * However, if the cluster is taken from the
6088 * current block group, release the cluster
6089 * first, so that we stand a better chance of
6090 * succeeding in the unclustered
6092 if (loop >= LOOP_NO_EMPTY_SIZE &&
6093 last_ptr->block_group != block_group) {
6094 spin_unlock(&last_ptr->refill_lock);
6095 goto unclustered_alloc;
6099 * this cluster didn't work out, free it and
6102 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6104 if (loop >= LOOP_NO_EMPTY_SIZE) {
6105 spin_unlock(&last_ptr->refill_lock);
6106 goto unclustered_alloc;
6109 aligned_cluster = max_t(unsigned long,
6110 empty_cluster + empty_size,
6111 block_group->full_stripe_len);
6113 /* allocate a cluster in this block group */
6114 ret = btrfs_find_space_cluster(trans, root,
6115 block_group, last_ptr,
6116 search_start, num_bytes,
6120 * now pull our allocation out of this
6123 offset = btrfs_alloc_from_cluster(block_group,
6124 last_ptr, num_bytes,
6127 /* we found one, proceed */
6128 spin_unlock(&last_ptr->refill_lock);
6129 trace_btrfs_reserve_extent_cluster(root,
6130 block_group, search_start,
6134 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6135 && !failed_cluster_refill) {
6136 spin_unlock(&last_ptr->refill_lock);
6138 failed_cluster_refill = true;
6139 wait_block_group_cache_progress(block_group,
6140 num_bytes + empty_cluster + empty_size);
6141 goto have_block_group;
6145 * at this point we either didn't find a cluster
6146 * or we weren't able to allocate a block from our
6147 * cluster. Free the cluster we've been trying
6148 * to use, and go to the next block group
6150 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6151 spin_unlock(&last_ptr->refill_lock);
6156 spin_lock(&block_group->free_space_ctl->tree_lock);
6158 block_group->free_space_ctl->free_space <
6159 num_bytes + empty_cluster + empty_size) {
6160 spin_unlock(&block_group->free_space_ctl->tree_lock);
6163 spin_unlock(&block_group->free_space_ctl->tree_lock);
6165 offset = btrfs_find_space_for_alloc(block_group, search_start,
6166 num_bytes, empty_size);
6168 * If we didn't find a chunk, and we haven't failed on this
6169 * block group before, and this block group is in the middle of
6170 * caching and we are ok with waiting, then go ahead and wait
6171 * for progress to be made, and set failed_alloc to true.
6173 * If failed_alloc is true then we've already waited on this
6174 * block group once and should move on to the next block group.
6176 if (!offset && !failed_alloc && !cached &&
6177 loop > LOOP_CACHING_NOWAIT) {
6178 wait_block_group_cache_progress(block_group,
6179 num_bytes + empty_size);
6180 failed_alloc = true;
6181 goto have_block_group;
6182 } else if (!offset) {
6184 have_caching_bg = true;
6188 search_start = stripe_align(root, used_block_group,
6191 /* move on to the next group */
6192 if (search_start + num_bytes >
6193 used_block_group->key.objectid + used_block_group->key.offset) {
6194 btrfs_add_free_space(used_block_group, offset, num_bytes);
6198 if (offset < search_start)
6199 btrfs_add_free_space(used_block_group, offset,
6200 search_start - offset);
6201 BUG_ON(offset > search_start);
6203 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6205 if (ret == -EAGAIN) {
6206 btrfs_add_free_space(used_block_group, offset, num_bytes);
6210 /* we are all good, lets return */
6211 ins->objectid = search_start;
6212 ins->offset = num_bytes;
6214 trace_btrfs_reserve_extent(orig_root, block_group,
6215 search_start, num_bytes);
6216 if (used_block_group != block_group)
6217 btrfs_put_block_group(used_block_group);
6218 btrfs_put_block_group(block_group);
6221 failed_cluster_refill = false;
6222 failed_alloc = false;
6223 BUG_ON(index != get_block_group_index(block_group));
6224 if (used_block_group != block_group)
6225 btrfs_put_block_group(used_block_group);
6226 btrfs_put_block_group(block_group);
6228 up_read(&space_info->groups_sem);
6230 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6233 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6237 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6238 * caching kthreads as we move along
6239 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6240 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6241 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6244 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6247 if (loop == LOOP_ALLOC_CHUNK) {
6248 ret = do_chunk_alloc(trans, root, data,
6251 * Do not bail out on ENOSPC since we
6252 * can do more things.
6254 if (ret < 0 && ret != -ENOSPC) {
6255 btrfs_abort_transaction(trans,
6261 if (loop == LOOP_NO_EMPTY_SIZE) {
6267 } else if (!ins->objectid) {
6269 } else if (ins->objectid) {
6277 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6278 int dump_block_groups)
6280 struct btrfs_block_group_cache *cache;
6283 spin_lock(&info->lock);
6284 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6285 (unsigned long long)info->flags,
6286 (unsigned long long)(info->total_bytes - info->bytes_used -
6287 info->bytes_pinned - info->bytes_reserved -
6288 info->bytes_readonly),
6289 (info->full) ? "" : "not ");
6290 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6291 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6292 (unsigned long long)info->total_bytes,
6293 (unsigned long long)info->bytes_used,
6294 (unsigned long long)info->bytes_pinned,
6295 (unsigned long long)info->bytes_reserved,
6296 (unsigned long long)info->bytes_may_use,
6297 (unsigned long long)info->bytes_readonly);
6298 spin_unlock(&info->lock);
6300 if (!dump_block_groups)
6303 down_read(&info->groups_sem);
6305 list_for_each_entry(cache, &info->block_groups[index], list) {
6306 spin_lock(&cache->lock);
6307 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6308 (unsigned long long)cache->key.objectid,
6309 (unsigned long long)cache->key.offset,
6310 (unsigned long long)btrfs_block_group_used(&cache->item),
6311 (unsigned long long)cache->pinned,
6312 (unsigned long long)cache->reserved,
6313 cache->ro ? "[readonly]" : "");
6314 btrfs_dump_free_space(cache, bytes);
6315 spin_unlock(&cache->lock);
6317 if (++index < BTRFS_NR_RAID_TYPES)
6319 up_read(&info->groups_sem);
6322 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6323 struct btrfs_root *root,
6324 u64 num_bytes, u64 min_alloc_size,
6325 u64 empty_size, u64 hint_byte,
6326 struct btrfs_key *ins, u64 data)
6328 bool final_tried = false;
6331 data = btrfs_get_alloc_profile(root, data);
6333 WARN_ON(num_bytes < root->sectorsize);
6334 ret = find_free_extent(trans, root, num_bytes, empty_size,
6335 hint_byte, ins, data);
6337 if (ret == -ENOSPC) {
6339 num_bytes = num_bytes >> 1;
6340 num_bytes = round_down(num_bytes, root->sectorsize);
6341 num_bytes = max(num_bytes, min_alloc_size);
6342 if (num_bytes == min_alloc_size)
6345 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6346 struct btrfs_space_info *sinfo;
6348 sinfo = __find_space_info(root->fs_info, data);
6349 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6350 (unsigned long long)data,
6351 (unsigned long long)num_bytes);
6353 dump_space_info(sinfo, num_bytes, 1);
6357 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6362 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6363 u64 start, u64 len, int pin)
6365 struct btrfs_block_group_cache *cache;
6368 cache = btrfs_lookup_block_group(root->fs_info, start);
6370 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6371 (unsigned long long)start);
6375 if (btrfs_test_opt(root, DISCARD))
6376 ret = btrfs_discard_extent(root, start, len, NULL);
6379 pin_down_extent(root, cache, start, len, 1);
6381 btrfs_add_free_space(cache, start, len);
6382 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6384 btrfs_put_block_group(cache);
6386 trace_btrfs_reserved_extent_free(root, start, len);
6391 int btrfs_free_reserved_extent(struct btrfs_root *root,
6394 return __btrfs_free_reserved_extent(root, start, len, 0);
6397 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6400 return __btrfs_free_reserved_extent(root, start, len, 1);
6403 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6404 struct btrfs_root *root,
6405 u64 parent, u64 root_objectid,
6406 u64 flags, u64 owner, u64 offset,
6407 struct btrfs_key *ins, int ref_mod)
6410 struct btrfs_fs_info *fs_info = root->fs_info;
6411 struct btrfs_extent_item *extent_item;
6412 struct btrfs_extent_inline_ref *iref;
6413 struct btrfs_path *path;
6414 struct extent_buffer *leaf;
6419 type = BTRFS_SHARED_DATA_REF_KEY;
6421 type = BTRFS_EXTENT_DATA_REF_KEY;
6423 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6425 path = btrfs_alloc_path();
6429 path->leave_spinning = 1;
6430 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6433 btrfs_free_path(path);
6437 leaf = path->nodes[0];
6438 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6439 struct btrfs_extent_item);
6440 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6441 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6442 btrfs_set_extent_flags(leaf, extent_item,
6443 flags | BTRFS_EXTENT_FLAG_DATA);
6445 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6446 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6448 struct btrfs_shared_data_ref *ref;
6449 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6450 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6451 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6453 struct btrfs_extent_data_ref *ref;
6454 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6455 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6456 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6457 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6458 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6461 btrfs_mark_buffer_dirty(path->nodes[0]);
6462 btrfs_free_path(path);
6464 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6465 if (ret) { /* -ENOENT, logic error */
6466 btrfs_err(fs_info, "update block group failed for %llu %llu",
6467 (unsigned long long)ins->objectid,
6468 (unsigned long long)ins->offset);
6474 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6475 struct btrfs_root *root,
6476 u64 parent, u64 root_objectid,
6477 u64 flags, struct btrfs_disk_key *key,
6478 int level, struct btrfs_key *ins)
6481 struct btrfs_fs_info *fs_info = root->fs_info;
6482 struct btrfs_extent_item *extent_item;
6483 struct btrfs_tree_block_info *block_info;
6484 struct btrfs_extent_inline_ref *iref;
6485 struct btrfs_path *path;
6486 struct extent_buffer *leaf;
6487 u32 size = sizeof(*extent_item) + sizeof(*iref);
6488 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6491 if (!skinny_metadata)
6492 size += sizeof(*block_info);
6494 path = btrfs_alloc_path();
6498 path->leave_spinning = 1;
6499 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6502 btrfs_free_path(path);
6506 leaf = path->nodes[0];
6507 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6508 struct btrfs_extent_item);
6509 btrfs_set_extent_refs(leaf, extent_item, 1);
6510 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6511 btrfs_set_extent_flags(leaf, extent_item,
6512 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6514 if (skinny_metadata) {
6515 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6517 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6518 btrfs_set_tree_block_key(leaf, block_info, key);
6519 btrfs_set_tree_block_level(leaf, block_info, level);
6520 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6524 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6525 btrfs_set_extent_inline_ref_type(leaf, iref,
6526 BTRFS_SHARED_BLOCK_REF_KEY);
6527 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6529 btrfs_set_extent_inline_ref_type(leaf, iref,
6530 BTRFS_TREE_BLOCK_REF_KEY);
6531 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6534 btrfs_mark_buffer_dirty(leaf);
6535 btrfs_free_path(path);
6537 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6538 if (ret) { /* -ENOENT, logic error */
6539 btrfs_err(fs_info, "update block group failed for %llu %llu",
6540 (unsigned long long)ins->objectid,
6541 (unsigned long long)ins->offset);
6547 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6548 struct btrfs_root *root,
6549 u64 root_objectid, u64 owner,
6550 u64 offset, struct btrfs_key *ins)
6554 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6556 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6558 root_objectid, owner, offset,
6559 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6564 * this is used by the tree logging recovery code. It records that
6565 * an extent has been allocated and makes sure to clear the free
6566 * space cache bits as well
6568 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6569 struct btrfs_root *root,
6570 u64 root_objectid, u64 owner, u64 offset,
6571 struct btrfs_key *ins)
6574 struct btrfs_block_group_cache *block_group;
6575 struct btrfs_caching_control *caching_ctl;
6576 u64 start = ins->objectid;
6577 u64 num_bytes = ins->offset;
6579 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6580 cache_block_group(block_group, 0);
6581 caching_ctl = get_caching_control(block_group);
6584 BUG_ON(!block_group_cache_done(block_group));
6585 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6586 BUG_ON(ret); /* -ENOMEM */
6588 mutex_lock(&caching_ctl->mutex);
6590 if (start >= caching_ctl->progress) {
6591 ret = add_excluded_extent(root, start, num_bytes);
6592 BUG_ON(ret); /* -ENOMEM */
6593 } else if (start + num_bytes <= caching_ctl->progress) {
6594 ret = btrfs_remove_free_space(block_group,
6596 BUG_ON(ret); /* -ENOMEM */
6598 num_bytes = caching_ctl->progress - start;
6599 ret = btrfs_remove_free_space(block_group,
6601 BUG_ON(ret); /* -ENOMEM */
6603 start = caching_ctl->progress;
6604 num_bytes = ins->objectid + ins->offset -
6605 caching_ctl->progress;
6606 ret = add_excluded_extent(root, start, num_bytes);
6607 BUG_ON(ret); /* -ENOMEM */
6610 mutex_unlock(&caching_ctl->mutex);
6611 put_caching_control(caching_ctl);
6614 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6615 RESERVE_ALLOC_NO_ACCOUNT);
6616 BUG_ON(ret); /* logic error */
6617 btrfs_put_block_group(block_group);
6618 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6619 0, owner, offset, ins, 1);
6623 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6624 struct btrfs_root *root,
6625 u64 bytenr, u32 blocksize,
6628 struct extent_buffer *buf;
6630 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6632 return ERR_PTR(-ENOMEM);
6633 btrfs_set_header_generation(buf, trans->transid);
6634 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6635 btrfs_tree_lock(buf);
6636 clean_tree_block(trans, root, buf);
6637 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6639 btrfs_set_lock_blocking(buf);
6640 btrfs_set_buffer_uptodate(buf);
6642 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6644 * we allow two log transactions at a time, use different
6645 * EXENT bit to differentiate dirty pages.
6647 if (root->log_transid % 2 == 0)
6648 set_extent_dirty(&root->dirty_log_pages, buf->start,
6649 buf->start + buf->len - 1, GFP_NOFS);
6651 set_extent_new(&root->dirty_log_pages, buf->start,
6652 buf->start + buf->len - 1, GFP_NOFS);
6654 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6655 buf->start + buf->len - 1, GFP_NOFS);
6657 trans->blocks_used++;
6658 /* this returns a buffer locked for blocking */
6662 static struct btrfs_block_rsv *
6663 use_block_rsv(struct btrfs_trans_handle *trans,
6664 struct btrfs_root *root, u32 blocksize)
6666 struct btrfs_block_rsv *block_rsv;
6667 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6670 block_rsv = get_block_rsv(trans, root);
6672 if (block_rsv->size == 0) {
6673 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6674 BTRFS_RESERVE_NO_FLUSH);
6676 * If we couldn't reserve metadata bytes try and use some from
6677 * the global reserve.
6679 if (ret && block_rsv != global_rsv) {
6680 ret = block_rsv_use_bytes(global_rsv, blocksize);
6683 return ERR_PTR(ret);
6685 return ERR_PTR(ret);
6690 ret = block_rsv_use_bytes(block_rsv, blocksize);
6693 if (ret && !block_rsv->failfast) {
6694 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6695 static DEFINE_RATELIMIT_STATE(_rs,
6696 DEFAULT_RATELIMIT_INTERVAL * 10,
6697 /*DEFAULT_RATELIMIT_BURST*/ 1);
6698 if (__ratelimit(&_rs))
6700 "btrfs: block rsv returned %d\n", ret);
6702 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6703 BTRFS_RESERVE_NO_FLUSH);
6706 } else if (ret && block_rsv != global_rsv) {
6707 ret = block_rsv_use_bytes(global_rsv, blocksize);
6713 return ERR_PTR(-ENOSPC);
6716 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6717 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6719 block_rsv_add_bytes(block_rsv, blocksize, 0);
6720 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6724 * finds a free extent and does all the dirty work required for allocation
6725 * returns the key for the extent through ins, and a tree buffer for
6726 * the first block of the extent through buf.
6728 * returns the tree buffer or NULL.
6730 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6731 struct btrfs_root *root, u32 blocksize,
6732 u64 parent, u64 root_objectid,
6733 struct btrfs_disk_key *key, int level,
6734 u64 hint, u64 empty_size)
6736 struct btrfs_key ins;
6737 struct btrfs_block_rsv *block_rsv;
6738 struct extent_buffer *buf;
6741 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6744 block_rsv = use_block_rsv(trans, root, blocksize);
6745 if (IS_ERR(block_rsv))
6746 return ERR_CAST(block_rsv);
6748 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6749 empty_size, hint, &ins, 0);
6751 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6752 return ERR_PTR(ret);
6755 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6757 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6759 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6761 parent = ins.objectid;
6762 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6766 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6767 struct btrfs_delayed_extent_op *extent_op;
6768 extent_op = btrfs_alloc_delayed_extent_op();
6769 BUG_ON(!extent_op); /* -ENOMEM */
6771 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6773 memset(&extent_op->key, 0, sizeof(extent_op->key));
6774 extent_op->flags_to_set = flags;
6775 if (skinny_metadata)
6776 extent_op->update_key = 0;
6778 extent_op->update_key = 1;
6779 extent_op->update_flags = 1;
6780 extent_op->is_data = 0;
6782 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6784 ins.offset, parent, root_objectid,
6785 level, BTRFS_ADD_DELAYED_EXTENT,
6787 BUG_ON(ret); /* -ENOMEM */
6792 struct walk_control {
6793 u64 refs[BTRFS_MAX_LEVEL];
6794 u64 flags[BTRFS_MAX_LEVEL];
6795 struct btrfs_key update_progress;
6806 #define DROP_REFERENCE 1
6807 #define UPDATE_BACKREF 2
6809 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6810 struct btrfs_root *root,
6811 struct walk_control *wc,
6812 struct btrfs_path *path)
6820 struct btrfs_key key;
6821 struct extent_buffer *eb;
6826 if (path->slots[wc->level] < wc->reada_slot) {
6827 wc->reada_count = wc->reada_count * 2 / 3;
6828 wc->reada_count = max(wc->reada_count, 2);
6830 wc->reada_count = wc->reada_count * 3 / 2;
6831 wc->reada_count = min_t(int, wc->reada_count,
6832 BTRFS_NODEPTRS_PER_BLOCK(root));
6835 eb = path->nodes[wc->level];
6836 nritems = btrfs_header_nritems(eb);
6837 blocksize = btrfs_level_size(root, wc->level - 1);
6839 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6840 if (nread >= wc->reada_count)
6844 bytenr = btrfs_node_blockptr(eb, slot);
6845 generation = btrfs_node_ptr_generation(eb, slot);
6847 if (slot == path->slots[wc->level])
6850 if (wc->stage == UPDATE_BACKREF &&
6851 generation <= root->root_key.offset)
6854 /* We don't lock the tree block, it's OK to be racy here */
6855 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6856 wc->level - 1, 1, &refs,
6858 /* We don't care about errors in readahead. */
6863 if (wc->stage == DROP_REFERENCE) {
6867 if (wc->level == 1 &&
6868 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6870 if (!wc->update_ref ||
6871 generation <= root->root_key.offset)
6873 btrfs_node_key_to_cpu(eb, &key, slot);
6874 ret = btrfs_comp_cpu_keys(&key,
6875 &wc->update_progress);
6879 if (wc->level == 1 &&
6880 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6884 ret = readahead_tree_block(root, bytenr, blocksize,
6890 wc->reada_slot = slot;
6894 * helper to process tree block while walking down the tree.
6896 * when wc->stage == UPDATE_BACKREF, this function updates
6897 * back refs for pointers in the block.
6899 * NOTE: return value 1 means we should stop walking down.
6901 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6902 struct btrfs_root *root,
6903 struct btrfs_path *path,
6904 struct walk_control *wc, int lookup_info)
6906 int level = wc->level;
6907 struct extent_buffer *eb = path->nodes[level];
6908 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6911 if (wc->stage == UPDATE_BACKREF &&
6912 btrfs_header_owner(eb) != root->root_key.objectid)
6916 * when reference count of tree block is 1, it won't increase
6917 * again. once full backref flag is set, we never clear it.
6920 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6921 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6922 BUG_ON(!path->locks[level]);
6923 ret = btrfs_lookup_extent_info(trans, root,
6924 eb->start, level, 1,
6927 BUG_ON(ret == -ENOMEM);
6930 BUG_ON(wc->refs[level] == 0);
6933 if (wc->stage == DROP_REFERENCE) {
6934 if (wc->refs[level] > 1)
6937 if (path->locks[level] && !wc->keep_locks) {
6938 btrfs_tree_unlock_rw(eb, path->locks[level]);
6939 path->locks[level] = 0;
6944 /* wc->stage == UPDATE_BACKREF */
6945 if (!(wc->flags[level] & flag)) {
6946 BUG_ON(!path->locks[level]);
6947 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6948 BUG_ON(ret); /* -ENOMEM */
6949 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6950 BUG_ON(ret); /* -ENOMEM */
6951 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6953 BUG_ON(ret); /* -ENOMEM */
6954 wc->flags[level] |= flag;
6958 * the block is shared by multiple trees, so it's not good to
6959 * keep the tree lock
6961 if (path->locks[level] && level > 0) {
6962 btrfs_tree_unlock_rw(eb, path->locks[level]);
6963 path->locks[level] = 0;
6969 * helper to process tree block pointer.
6971 * when wc->stage == DROP_REFERENCE, this function checks
6972 * reference count of the block pointed to. if the block
6973 * is shared and we need update back refs for the subtree
6974 * rooted at the block, this function changes wc->stage to
6975 * UPDATE_BACKREF. if the block is shared and there is no
6976 * need to update back, this function drops the reference
6979 * NOTE: return value 1 means we should stop walking down.
6981 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6982 struct btrfs_root *root,
6983 struct btrfs_path *path,
6984 struct walk_control *wc, int *lookup_info)
6990 struct btrfs_key key;
6991 struct extent_buffer *next;
6992 int level = wc->level;
6996 generation = btrfs_node_ptr_generation(path->nodes[level],
6997 path->slots[level]);
6999 * if the lower level block was created before the snapshot
7000 * was created, we know there is no need to update back refs
7003 if (wc->stage == UPDATE_BACKREF &&
7004 generation <= root->root_key.offset) {
7009 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7010 blocksize = btrfs_level_size(root, level - 1);
7012 next = btrfs_find_tree_block(root, bytenr, blocksize);
7014 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7019 btrfs_tree_lock(next);
7020 btrfs_set_lock_blocking(next);
7022 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7023 &wc->refs[level - 1],
7024 &wc->flags[level - 1]);
7026 btrfs_tree_unlock(next);
7030 if (unlikely(wc->refs[level - 1] == 0)) {
7031 btrfs_err(root->fs_info, "Missing references.");
7036 if (wc->stage == DROP_REFERENCE) {
7037 if (wc->refs[level - 1] > 1) {
7039 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7042 if (!wc->update_ref ||
7043 generation <= root->root_key.offset)
7046 btrfs_node_key_to_cpu(path->nodes[level], &key,
7047 path->slots[level]);
7048 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7052 wc->stage = UPDATE_BACKREF;
7053 wc->shared_level = level - 1;
7057 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7061 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7062 btrfs_tree_unlock(next);
7063 free_extent_buffer(next);
7069 if (reada && level == 1)
7070 reada_walk_down(trans, root, wc, path);
7071 next = read_tree_block(root, bytenr, blocksize, generation);
7074 btrfs_tree_lock(next);
7075 btrfs_set_lock_blocking(next);
7079 BUG_ON(level != btrfs_header_level(next));
7080 path->nodes[level] = next;
7081 path->slots[level] = 0;
7082 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7088 wc->refs[level - 1] = 0;
7089 wc->flags[level - 1] = 0;
7090 if (wc->stage == DROP_REFERENCE) {
7091 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7092 parent = path->nodes[level]->start;
7094 BUG_ON(root->root_key.objectid !=
7095 btrfs_header_owner(path->nodes[level]));
7099 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7100 root->root_key.objectid, level - 1, 0, 0);
7101 BUG_ON(ret); /* -ENOMEM */
7103 btrfs_tree_unlock(next);
7104 free_extent_buffer(next);
7110 * helper to process tree block while walking up the tree.
7112 * when wc->stage == DROP_REFERENCE, this function drops
7113 * reference count on the block.
7115 * when wc->stage == UPDATE_BACKREF, this function changes
7116 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7117 * to UPDATE_BACKREF previously while processing the block.
7119 * NOTE: return value 1 means we should stop walking up.
7121 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7122 struct btrfs_root *root,
7123 struct btrfs_path *path,
7124 struct walk_control *wc)
7127 int level = wc->level;
7128 struct extent_buffer *eb = path->nodes[level];
7131 if (wc->stage == UPDATE_BACKREF) {
7132 BUG_ON(wc->shared_level < level);
7133 if (level < wc->shared_level)
7136 ret = find_next_key(path, level + 1, &wc->update_progress);
7140 wc->stage = DROP_REFERENCE;
7141 wc->shared_level = -1;
7142 path->slots[level] = 0;
7145 * check reference count again if the block isn't locked.
7146 * we should start walking down the tree again if reference
7149 if (!path->locks[level]) {
7151 btrfs_tree_lock(eb);
7152 btrfs_set_lock_blocking(eb);
7153 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7155 ret = btrfs_lookup_extent_info(trans, root,
7156 eb->start, level, 1,
7160 btrfs_tree_unlock_rw(eb, path->locks[level]);
7161 path->locks[level] = 0;
7164 BUG_ON(wc->refs[level] == 0);
7165 if (wc->refs[level] == 1) {
7166 btrfs_tree_unlock_rw(eb, path->locks[level]);
7167 path->locks[level] = 0;
7173 /* wc->stage == DROP_REFERENCE */
7174 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7176 if (wc->refs[level] == 1) {
7178 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7179 ret = btrfs_dec_ref(trans, root, eb, 1,
7182 ret = btrfs_dec_ref(trans, root, eb, 0,
7184 BUG_ON(ret); /* -ENOMEM */
7186 /* make block locked assertion in clean_tree_block happy */
7187 if (!path->locks[level] &&
7188 btrfs_header_generation(eb) == trans->transid) {
7189 btrfs_tree_lock(eb);
7190 btrfs_set_lock_blocking(eb);
7191 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7193 clean_tree_block(trans, root, eb);
7196 if (eb == root->node) {
7197 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7200 BUG_ON(root->root_key.objectid !=
7201 btrfs_header_owner(eb));
7203 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7204 parent = path->nodes[level + 1]->start;
7206 BUG_ON(root->root_key.objectid !=
7207 btrfs_header_owner(path->nodes[level + 1]));
7210 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7212 wc->refs[level] = 0;
7213 wc->flags[level] = 0;
7217 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7218 struct btrfs_root *root,
7219 struct btrfs_path *path,
7220 struct walk_control *wc)
7222 int level = wc->level;
7223 int lookup_info = 1;
7226 while (level >= 0) {
7227 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7234 if (path->slots[level] >=
7235 btrfs_header_nritems(path->nodes[level]))
7238 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7240 path->slots[level]++;
7249 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7250 struct btrfs_root *root,
7251 struct btrfs_path *path,
7252 struct walk_control *wc, int max_level)
7254 int level = wc->level;
7257 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7258 while (level < max_level && path->nodes[level]) {
7260 if (path->slots[level] + 1 <
7261 btrfs_header_nritems(path->nodes[level])) {
7262 path->slots[level]++;
7265 ret = walk_up_proc(trans, root, path, wc);
7269 if (path->locks[level]) {
7270 btrfs_tree_unlock_rw(path->nodes[level],
7271 path->locks[level]);
7272 path->locks[level] = 0;
7274 free_extent_buffer(path->nodes[level]);
7275 path->nodes[level] = NULL;
7283 * drop a subvolume tree.
7285 * this function traverses the tree freeing any blocks that only
7286 * referenced by the tree.
7288 * when a shared tree block is found. this function decreases its
7289 * reference count by one. if update_ref is true, this function
7290 * also make sure backrefs for the shared block and all lower level
7291 * blocks are properly updated.
7293 * If called with for_reloc == 0, may exit early with -EAGAIN
7295 int btrfs_drop_snapshot(struct btrfs_root *root,
7296 struct btrfs_block_rsv *block_rsv, int update_ref,
7299 struct btrfs_path *path;
7300 struct btrfs_trans_handle *trans;
7301 struct btrfs_root *tree_root = root->fs_info->tree_root;
7302 struct btrfs_root_item *root_item = &root->root_item;
7303 struct walk_control *wc;
7304 struct btrfs_key key;
7309 path = btrfs_alloc_path();
7315 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7317 btrfs_free_path(path);
7322 trans = btrfs_start_transaction(tree_root, 0);
7323 if (IS_ERR(trans)) {
7324 err = PTR_ERR(trans);
7329 trans->block_rsv = block_rsv;
7331 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7332 level = btrfs_header_level(root->node);
7333 path->nodes[level] = btrfs_lock_root_node(root);
7334 btrfs_set_lock_blocking(path->nodes[level]);
7335 path->slots[level] = 0;
7336 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7337 memset(&wc->update_progress, 0,
7338 sizeof(wc->update_progress));
7340 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7341 memcpy(&wc->update_progress, &key,
7342 sizeof(wc->update_progress));
7344 level = root_item->drop_level;
7346 path->lowest_level = level;
7347 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7348 path->lowest_level = 0;
7356 * unlock our path, this is safe because only this
7357 * function is allowed to delete this snapshot
7359 btrfs_unlock_up_safe(path, 0);
7361 level = btrfs_header_level(root->node);
7363 btrfs_tree_lock(path->nodes[level]);
7364 btrfs_set_lock_blocking(path->nodes[level]);
7366 ret = btrfs_lookup_extent_info(trans, root,
7367 path->nodes[level]->start,
7368 level, 1, &wc->refs[level],
7374 BUG_ON(wc->refs[level] == 0);
7376 if (level == root_item->drop_level)
7379 btrfs_tree_unlock(path->nodes[level]);
7380 WARN_ON(wc->refs[level] != 1);
7386 wc->shared_level = -1;
7387 wc->stage = DROP_REFERENCE;
7388 wc->update_ref = update_ref;
7390 wc->for_reloc = for_reloc;
7391 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7394 if (!for_reloc && btrfs_fs_closing(root->fs_info)) {
7395 pr_debug("btrfs: drop snapshot early exit\n");
7400 ret = walk_down_tree(trans, root, path, wc);
7406 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7413 BUG_ON(wc->stage != DROP_REFERENCE);
7417 if (wc->stage == DROP_REFERENCE) {
7419 btrfs_node_key(path->nodes[level],
7420 &root_item->drop_progress,
7421 path->slots[level]);
7422 root_item->drop_level = level;
7425 BUG_ON(wc->level == 0);
7426 if (btrfs_should_end_transaction(trans, tree_root)) {
7427 ret = btrfs_update_root(trans, tree_root,
7431 btrfs_abort_transaction(trans, tree_root, ret);
7436 btrfs_end_transaction_throttle(trans, tree_root);
7437 trans = btrfs_start_transaction(tree_root, 0);
7438 if (IS_ERR(trans)) {
7439 err = PTR_ERR(trans);
7443 trans->block_rsv = block_rsv;
7446 btrfs_release_path(path);
7450 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7452 btrfs_abort_transaction(trans, tree_root, ret);
7456 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7457 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7460 btrfs_abort_transaction(trans, tree_root, ret);
7463 } else if (ret > 0) {
7464 /* if we fail to delete the orphan item this time
7465 * around, it'll get picked up the next time.
7467 * The most common failure here is just -ENOENT.
7469 btrfs_del_orphan_item(trans, tree_root,
7470 root->root_key.objectid);
7474 if (root->in_radix) {
7475 btrfs_free_fs_root(tree_root->fs_info, root);
7477 free_extent_buffer(root->node);
7478 free_extent_buffer(root->commit_root);
7482 btrfs_end_transaction_throttle(trans, tree_root);
7485 btrfs_free_path(path);
7488 btrfs_std_error(root->fs_info, err);
7493 * drop subtree rooted at tree block 'node'.
7495 * NOTE: this function will unlock and release tree block 'node'
7496 * only used by relocation code
7498 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7499 struct btrfs_root *root,
7500 struct extent_buffer *node,
7501 struct extent_buffer *parent)
7503 struct btrfs_path *path;
7504 struct walk_control *wc;
7510 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7512 path = btrfs_alloc_path();
7516 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7518 btrfs_free_path(path);
7522 btrfs_assert_tree_locked(parent);
7523 parent_level = btrfs_header_level(parent);
7524 extent_buffer_get(parent);
7525 path->nodes[parent_level] = parent;
7526 path->slots[parent_level] = btrfs_header_nritems(parent);
7528 btrfs_assert_tree_locked(node);
7529 level = btrfs_header_level(node);
7530 path->nodes[level] = node;
7531 path->slots[level] = 0;
7532 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7534 wc->refs[parent_level] = 1;
7535 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7537 wc->shared_level = -1;
7538 wc->stage = DROP_REFERENCE;
7542 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7545 wret = walk_down_tree(trans, root, path, wc);
7551 wret = walk_up_tree(trans, root, path, wc, parent_level);
7559 btrfs_free_path(path);
7563 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7569 * if restripe for this chunk_type is on pick target profile and
7570 * return, otherwise do the usual balance
7572 stripped = get_restripe_target(root->fs_info, flags);
7574 return extended_to_chunk(stripped);
7577 * we add in the count of missing devices because we want
7578 * to make sure that any RAID levels on a degraded FS
7579 * continue to be honored.
7581 num_devices = root->fs_info->fs_devices->rw_devices +
7582 root->fs_info->fs_devices->missing_devices;
7584 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7585 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7586 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7588 if (num_devices == 1) {
7589 stripped |= BTRFS_BLOCK_GROUP_DUP;
7590 stripped = flags & ~stripped;
7592 /* turn raid0 into single device chunks */
7593 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7596 /* turn mirroring into duplication */
7597 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7598 BTRFS_BLOCK_GROUP_RAID10))
7599 return stripped | BTRFS_BLOCK_GROUP_DUP;
7601 /* they already had raid on here, just return */
7602 if (flags & stripped)
7605 stripped |= BTRFS_BLOCK_GROUP_DUP;
7606 stripped = flags & ~stripped;
7608 /* switch duplicated blocks with raid1 */
7609 if (flags & BTRFS_BLOCK_GROUP_DUP)
7610 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7612 /* this is drive concat, leave it alone */
7618 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7620 struct btrfs_space_info *sinfo = cache->space_info;
7622 u64 min_allocable_bytes;
7627 * We need some metadata space and system metadata space for
7628 * allocating chunks in some corner cases until we force to set
7629 * it to be readonly.
7632 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7634 min_allocable_bytes = 1 * 1024 * 1024;
7636 min_allocable_bytes = 0;
7638 spin_lock(&sinfo->lock);
7639 spin_lock(&cache->lock);
7646 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7647 cache->bytes_super - btrfs_block_group_used(&cache->item);
7649 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7650 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7651 min_allocable_bytes <= sinfo->total_bytes) {
7652 sinfo->bytes_readonly += num_bytes;
7657 spin_unlock(&cache->lock);
7658 spin_unlock(&sinfo->lock);
7662 int btrfs_set_block_group_ro(struct btrfs_root *root,
7663 struct btrfs_block_group_cache *cache)
7666 struct btrfs_trans_handle *trans;
7672 trans = btrfs_join_transaction(root);
7674 return PTR_ERR(trans);
7676 alloc_flags = update_block_group_flags(root, cache->flags);
7677 if (alloc_flags != cache->flags) {
7678 ret = do_chunk_alloc(trans, root, alloc_flags,
7684 ret = set_block_group_ro(cache, 0);
7687 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7688 ret = do_chunk_alloc(trans, root, alloc_flags,
7692 ret = set_block_group_ro(cache, 0);
7694 btrfs_end_transaction(trans, root);
7698 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7699 struct btrfs_root *root, u64 type)
7701 u64 alloc_flags = get_alloc_profile(root, type);
7702 return do_chunk_alloc(trans, root, alloc_flags,
7707 * helper to account the unused space of all the readonly block group in the
7708 * list. takes mirrors into account.
7710 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7712 struct btrfs_block_group_cache *block_group;
7716 list_for_each_entry(block_group, groups_list, list) {
7717 spin_lock(&block_group->lock);
7719 if (!block_group->ro) {
7720 spin_unlock(&block_group->lock);
7724 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7725 BTRFS_BLOCK_GROUP_RAID10 |
7726 BTRFS_BLOCK_GROUP_DUP))
7731 free_bytes += (block_group->key.offset -
7732 btrfs_block_group_used(&block_group->item)) *
7735 spin_unlock(&block_group->lock);
7742 * helper to account the unused space of all the readonly block group in the
7743 * space_info. takes mirrors into account.
7745 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7750 spin_lock(&sinfo->lock);
7752 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7753 if (!list_empty(&sinfo->block_groups[i]))
7754 free_bytes += __btrfs_get_ro_block_group_free_space(
7755 &sinfo->block_groups[i]);
7757 spin_unlock(&sinfo->lock);
7762 void btrfs_set_block_group_rw(struct btrfs_root *root,
7763 struct btrfs_block_group_cache *cache)
7765 struct btrfs_space_info *sinfo = cache->space_info;
7770 spin_lock(&sinfo->lock);
7771 spin_lock(&cache->lock);
7772 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7773 cache->bytes_super - btrfs_block_group_used(&cache->item);
7774 sinfo->bytes_readonly -= num_bytes;
7776 spin_unlock(&cache->lock);
7777 spin_unlock(&sinfo->lock);
7781 * checks to see if its even possible to relocate this block group.
7783 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7784 * ok to go ahead and try.
7786 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7788 struct btrfs_block_group_cache *block_group;
7789 struct btrfs_space_info *space_info;
7790 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7791 struct btrfs_device *device;
7800 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7802 /* odd, couldn't find the block group, leave it alone */
7806 min_free = btrfs_block_group_used(&block_group->item);
7808 /* no bytes used, we're good */
7812 space_info = block_group->space_info;
7813 spin_lock(&space_info->lock);
7815 full = space_info->full;
7818 * if this is the last block group we have in this space, we can't
7819 * relocate it unless we're able to allocate a new chunk below.
7821 * Otherwise, we need to make sure we have room in the space to handle
7822 * all of the extents from this block group. If we can, we're good
7824 if ((space_info->total_bytes != block_group->key.offset) &&
7825 (space_info->bytes_used + space_info->bytes_reserved +
7826 space_info->bytes_pinned + space_info->bytes_readonly +
7827 min_free < space_info->total_bytes)) {
7828 spin_unlock(&space_info->lock);
7831 spin_unlock(&space_info->lock);
7834 * ok we don't have enough space, but maybe we have free space on our
7835 * devices to allocate new chunks for relocation, so loop through our
7836 * alloc devices and guess if we have enough space. if this block
7837 * group is going to be restriped, run checks against the target
7838 * profile instead of the current one.
7850 target = get_restripe_target(root->fs_info, block_group->flags);
7852 index = __get_raid_index(extended_to_chunk(target));
7855 * this is just a balance, so if we were marked as full
7856 * we know there is no space for a new chunk
7861 index = get_block_group_index(block_group);
7864 if (index == BTRFS_RAID_RAID10) {
7868 } else if (index == BTRFS_RAID_RAID1) {
7870 } else if (index == BTRFS_RAID_DUP) {
7873 } else if (index == BTRFS_RAID_RAID0) {
7874 dev_min = fs_devices->rw_devices;
7875 do_div(min_free, dev_min);
7878 mutex_lock(&root->fs_info->chunk_mutex);
7879 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7883 * check to make sure we can actually find a chunk with enough
7884 * space to fit our block group in.
7886 if (device->total_bytes > device->bytes_used + min_free &&
7887 !device->is_tgtdev_for_dev_replace) {
7888 ret = find_free_dev_extent(device, min_free,
7893 if (dev_nr >= dev_min)
7899 mutex_unlock(&root->fs_info->chunk_mutex);
7901 btrfs_put_block_group(block_group);
7905 static int find_first_block_group(struct btrfs_root *root,
7906 struct btrfs_path *path, struct btrfs_key *key)
7909 struct btrfs_key found_key;
7910 struct extent_buffer *leaf;
7913 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7918 slot = path->slots[0];
7919 leaf = path->nodes[0];
7920 if (slot >= btrfs_header_nritems(leaf)) {
7921 ret = btrfs_next_leaf(root, path);
7928 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7930 if (found_key.objectid >= key->objectid &&
7931 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7941 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7943 struct btrfs_block_group_cache *block_group;
7947 struct inode *inode;
7949 block_group = btrfs_lookup_first_block_group(info, last);
7950 while (block_group) {
7951 spin_lock(&block_group->lock);
7952 if (block_group->iref)
7954 spin_unlock(&block_group->lock);
7955 block_group = next_block_group(info->tree_root,
7965 inode = block_group->inode;
7966 block_group->iref = 0;
7967 block_group->inode = NULL;
7968 spin_unlock(&block_group->lock);
7970 last = block_group->key.objectid + block_group->key.offset;
7971 btrfs_put_block_group(block_group);
7975 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7977 struct btrfs_block_group_cache *block_group;
7978 struct btrfs_space_info *space_info;
7979 struct btrfs_caching_control *caching_ctl;
7982 down_write(&info->extent_commit_sem);
7983 while (!list_empty(&info->caching_block_groups)) {
7984 caching_ctl = list_entry(info->caching_block_groups.next,
7985 struct btrfs_caching_control, list);
7986 list_del(&caching_ctl->list);
7987 put_caching_control(caching_ctl);
7989 up_write(&info->extent_commit_sem);
7991 spin_lock(&info->block_group_cache_lock);
7992 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7993 block_group = rb_entry(n, struct btrfs_block_group_cache,
7995 rb_erase(&block_group->cache_node,
7996 &info->block_group_cache_tree);
7997 spin_unlock(&info->block_group_cache_lock);
7999 down_write(&block_group->space_info->groups_sem);
8000 list_del(&block_group->list);
8001 up_write(&block_group->space_info->groups_sem);
8003 if (block_group->cached == BTRFS_CACHE_STARTED)
8004 wait_block_group_cache_done(block_group);
8007 * We haven't cached this block group, which means we could
8008 * possibly have excluded extents on this block group.
8010 if (block_group->cached == BTRFS_CACHE_NO)
8011 free_excluded_extents(info->extent_root, block_group);
8013 btrfs_remove_free_space_cache(block_group);
8014 btrfs_put_block_group(block_group);
8016 spin_lock(&info->block_group_cache_lock);
8018 spin_unlock(&info->block_group_cache_lock);
8020 /* now that all the block groups are freed, go through and
8021 * free all the space_info structs. This is only called during
8022 * the final stages of unmount, and so we know nobody is
8023 * using them. We call synchronize_rcu() once before we start,
8024 * just to be on the safe side.
8028 release_global_block_rsv(info);
8030 while(!list_empty(&info->space_info)) {
8031 space_info = list_entry(info->space_info.next,
8032 struct btrfs_space_info,
8034 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8035 if (space_info->bytes_pinned > 0 ||
8036 space_info->bytes_reserved > 0 ||
8037 space_info->bytes_may_use > 0) {
8039 dump_space_info(space_info, 0, 0);
8042 list_del(&space_info->list);
8048 static void __link_block_group(struct btrfs_space_info *space_info,
8049 struct btrfs_block_group_cache *cache)
8051 int index = get_block_group_index(cache);
8053 down_write(&space_info->groups_sem);
8054 list_add_tail(&cache->list, &space_info->block_groups[index]);
8055 up_write(&space_info->groups_sem);
8058 int btrfs_read_block_groups(struct btrfs_root *root)
8060 struct btrfs_path *path;
8062 struct btrfs_block_group_cache *cache;
8063 struct btrfs_fs_info *info = root->fs_info;
8064 struct btrfs_space_info *space_info;
8065 struct btrfs_key key;
8066 struct btrfs_key found_key;
8067 struct extent_buffer *leaf;
8071 root = info->extent_root;
8074 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8075 path = btrfs_alloc_path();
8080 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8081 if (btrfs_test_opt(root, SPACE_CACHE) &&
8082 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8084 if (btrfs_test_opt(root, CLEAR_CACHE))
8088 ret = find_first_block_group(root, path, &key);
8093 leaf = path->nodes[0];
8094 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8095 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8100 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8102 if (!cache->free_space_ctl) {
8108 atomic_set(&cache->count, 1);
8109 spin_lock_init(&cache->lock);
8110 cache->fs_info = info;
8111 INIT_LIST_HEAD(&cache->list);
8112 INIT_LIST_HEAD(&cache->cluster_list);
8116 * When we mount with old space cache, we need to
8117 * set BTRFS_DC_CLEAR and set dirty flag.
8119 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8120 * truncate the old free space cache inode and
8122 * b) Setting 'dirty flag' makes sure that we flush
8123 * the new space cache info onto disk.
8125 cache->disk_cache_state = BTRFS_DC_CLEAR;
8126 if (btrfs_test_opt(root, SPACE_CACHE))
8130 read_extent_buffer(leaf, &cache->item,
8131 btrfs_item_ptr_offset(leaf, path->slots[0]),
8132 sizeof(cache->item));
8133 memcpy(&cache->key, &found_key, sizeof(found_key));
8135 key.objectid = found_key.objectid + found_key.offset;
8136 btrfs_release_path(path);
8137 cache->flags = btrfs_block_group_flags(&cache->item);
8138 cache->sectorsize = root->sectorsize;
8139 cache->full_stripe_len = btrfs_full_stripe_len(root,
8140 &root->fs_info->mapping_tree,
8141 found_key.objectid);
8142 btrfs_init_free_space_ctl(cache);
8145 * We need to exclude the super stripes now so that the space
8146 * info has super bytes accounted for, otherwise we'll think
8147 * we have more space than we actually do.
8149 ret = exclude_super_stripes(root, cache);
8152 * We may have excluded something, so call this just in
8155 free_excluded_extents(root, cache);
8156 kfree(cache->free_space_ctl);
8162 * check for two cases, either we are full, and therefore
8163 * don't need to bother with the caching work since we won't
8164 * find any space, or we are empty, and we can just add all
8165 * the space in and be done with it. This saves us _alot_ of
8166 * time, particularly in the full case.
8168 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8169 cache->last_byte_to_unpin = (u64)-1;
8170 cache->cached = BTRFS_CACHE_FINISHED;
8171 free_excluded_extents(root, cache);
8172 } else if (btrfs_block_group_used(&cache->item) == 0) {
8173 cache->last_byte_to_unpin = (u64)-1;
8174 cache->cached = BTRFS_CACHE_FINISHED;
8175 add_new_free_space(cache, root->fs_info,
8177 found_key.objectid +
8179 free_excluded_extents(root, cache);
8182 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8184 btrfs_remove_free_space_cache(cache);
8185 btrfs_put_block_group(cache);
8189 ret = update_space_info(info, cache->flags, found_key.offset,
8190 btrfs_block_group_used(&cache->item),
8193 btrfs_remove_free_space_cache(cache);
8194 spin_lock(&info->block_group_cache_lock);
8195 rb_erase(&cache->cache_node,
8196 &info->block_group_cache_tree);
8197 spin_unlock(&info->block_group_cache_lock);
8198 btrfs_put_block_group(cache);
8202 cache->space_info = space_info;
8203 spin_lock(&cache->space_info->lock);
8204 cache->space_info->bytes_readonly += cache->bytes_super;
8205 spin_unlock(&cache->space_info->lock);
8207 __link_block_group(space_info, cache);
8209 set_avail_alloc_bits(root->fs_info, cache->flags);
8210 if (btrfs_chunk_readonly(root, cache->key.objectid))
8211 set_block_group_ro(cache, 1);
8214 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8215 if (!(get_alloc_profile(root, space_info->flags) &
8216 (BTRFS_BLOCK_GROUP_RAID10 |
8217 BTRFS_BLOCK_GROUP_RAID1 |
8218 BTRFS_BLOCK_GROUP_RAID5 |
8219 BTRFS_BLOCK_GROUP_RAID6 |
8220 BTRFS_BLOCK_GROUP_DUP)))
8223 * avoid allocating from un-mirrored block group if there are
8224 * mirrored block groups.
8226 list_for_each_entry(cache, &space_info->block_groups[3], list)
8227 set_block_group_ro(cache, 1);
8228 list_for_each_entry(cache, &space_info->block_groups[4], list)
8229 set_block_group_ro(cache, 1);
8232 init_global_block_rsv(info);
8235 btrfs_free_path(path);
8239 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8240 struct btrfs_root *root)
8242 struct btrfs_block_group_cache *block_group, *tmp;
8243 struct btrfs_root *extent_root = root->fs_info->extent_root;
8244 struct btrfs_block_group_item item;
8245 struct btrfs_key key;
8248 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8250 list_del_init(&block_group->new_bg_list);
8255 spin_lock(&block_group->lock);
8256 memcpy(&item, &block_group->item, sizeof(item));
8257 memcpy(&key, &block_group->key, sizeof(key));
8258 spin_unlock(&block_group->lock);
8260 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8263 btrfs_abort_transaction(trans, extent_root, ret);
8267 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8268 struct btrfs_root *root, u64 bytes_used,
8269 u64 type, u64 chunk_objectid, u64 chunk_offset,
8273 struct btrfs_root *extent_root;
8274 struct btrfs_block_group_cache *cache;
8276 extent_root = root->fs_info->extent_root;
8278 root->fs_info->last_trans_log_full_commit = trans->transid;
8280 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8283 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8285 if (!cache->free_space_ctl) {
8290 cache->key.objectid = chunk_offset;
8291 cache->key.offset = size;
8292 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8293 cache->sectorsize = root->sectorsize;
8294 cache->fs_info = root->fs_info;
8295 cache->full_stripe_len = btrfs_full_stripe_len(root,
8296 &root->fs_info->mapping_tree,
8299 atomic_set(&cache->count, 1);
8300 spin_lock_init(&cache->lock);
8301 INIT_LIST_HEAD(&cache->list);
8302 INIT_LIST_HEAD(&cache->cluster_list);
8303 INIT_LIST_HEAD(&cache->new_bg_list);
8305 btrfs_init_free_space_ctl(cache);
8307 btrfs_set_block_group_used(&cache->item, bytes_used);
8308 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8309 cache->flags = type;
8310 btrfs_set_block_group_flags(&cache->item, type);
8312 cache->last_byte_to_unpin = (u64)-1;
8313 cache->cached = BTRFS_CACHE_FINISHED;
8314 ret = exclude_super_stripes(root, cache);
8317 * We may have excluded something, so call this just in
8320 free_excluded_extents(root, cache);
8321 kfree(cache->free_space_ctl);
8326 add_new_free_space(cache, root->fs_info, chunk_offset,
8327 chunk_offset + size);
8329 free_excluded_extents(root, cache);
8331 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8333 btrfs_remove_free_space_cache(cache);
8334 btrfs_put_block_group(cache);
8338 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8339 &cache->space_info);
8341 btrfs_remove_free_space_cache(cache);
8342 spin_lock(&root->fs_info->block_group_cache_lock);
8343 rb_erase(&cache->cache_node,
8344 &root->fs_info->block_group_cache_tree);
8345 spin_unlock(&root->fs_info->block_group_cache_lock);
8346 btrfs_put_block_group(cache);
8349 update_global_block_rsv(root->fs_info);
8351 spin_lock(&cache->space_info->lock);
8352 cache->space_info->bytes_readonly += cache->bytes_super;
8353 spin_unlock(&cache->space_info->lock);
8355 __link_block_group(cache->space_info, cache);
8357 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8359 set_avail_alloc_bits(extent_root->fs_info, type);
8364 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8366 u64 extra_flags = chunk_to_extended(flags) &
8367 BTRFS_EXTENDED_PROFILE_MASK;
8369 write_seqlock(&fs_info->profiles_lock);
8370 if (flags & BTRFS_BLOCK_GROUP_DATA)
8371 fs_info->avail_data_alloc_bits &= ~extra_flags;
8372 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8373 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8374 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8375 fs_info->avail_system_alloc_bits &= ~extra_flags;
8376 write_sequnlock(&fs_info->profiles_lock);
8379 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8380 struct btrfs_root *root, u64 group_start)
8382 struct btrfs_path *path;
8383 struct btrfs_block_group_cache *block_group;
8384 struct btrfs_free_cluster *cluster;
8385 struct btrfs_root *tree_root = root->fs_info->tree_root;
8386 struct btrfs_key key;
8387 struct inode *inode;
8392 root = root->fs_info->extent_root;
8394 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8395 BUG_ON(!block_group);
8396 BUG_ON(!block_group->ro);
8399 * Free the reserved super bytes from this block group before
8402 free_excluded_extents(root, block_group);
8404 memcpy(&key, &block_group->key, sizeof(key));
8405 index = get_block_group_index(block_group);
8406 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8407 BTRFS_BLOCK_GROUP_RAID1 |
8408 BTRFS_BLOCK_GROUP_RAID10))
8413 /* make sure this block group isn't part of an allocation cluster */
8414 cluster = &root->fs_info->data_alloc_cluster;
8415 spin_lock(&cluster->refill_lock);
8416 btrfs_return_cluster_to_free_space(block_group, cluster);
8417 spin_unlock(&cluster->refill_lock);
8420 * make sure this block group isn't part of a metadata
8421 * allocation cluster
8423 cluster = &root->fs_info->meta_alloc_cluster;
8424 spin_lock(&cluster->refill_lock);
8425 btrfs_return_cluster_to_free_space(block_group, cluster);
8426 spin_unlock(&cluster->refill_lock);
8428 path = btrfs_alloc_path();
8434 inode = lookup_free_space_inode(tree_root, block_group, path);
8435 if (!IS_ERR(inode)) {
8436 ret = btrfs_orphan_add(trans, inode);
8438 btrfs_add_delayed_iput(inode);
8442 /* One for the block groups ref */
8443 spin_lock(&block_group->lock);
8444 if (block_group->iref) {
8445 block_group->iref = 0;
8446 block_group->inode = NULL;
8447 spin_unlock(&block_group->lock);
8450 spin_unlock(&block_group->lock);
8452 /* One for our lookup ref */
8453 btrfs_add_delayed_iput(inode);
8456 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8457 key.offset = block_group->key.objectid;
8460 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8464 btrfs_release_path(path);
8466 ret = btrfs_del_item(trans, tree_root, path);
8469 btrfs_release_path(path);
8472 spin_lock(&root->fs_info->block_group_cache_lock);
8473 rb_erase(&block_group->cache_node,
8474 &root->fs_info->block_group_cache_tree);
8476 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8477 root->fs_info->first_logical_byte = (u64)-1;
8478 spin_unlock(&root->fs_info->block_group_cache_lock);
8480 down_write(&block_group->space_info->groups_sem);
8482 * we must use list_del_init so people can check to see if they
8483 * are still on the list after taking the semaphore
8485 list_del_init(&block_group->list);
8486 if (list_empty(&block_group->space_info->block_groups[index]))
8487 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8488 up_write(&block_group->space_info->groups_sem);
8490 if (block_group->cached == BTRFS_CACHE_STARTED)
8491 wait_block_group_cache_done(block_group);
8493 btrfs_remove_free_space_cache(block_group);
8495 spin_lock(&block_group->space_info->lock);
8496 block_group->space_info->total_bytes -= block_group->key.offset;
8497 block_group->space_info->bytes_readonly -= block_group->key.offset;
8498 block_group->space_info->disk_total -= block_group->key.offset * factor;
8499 spin_unlock(&block_group->space_info->lock);
8501 memcpy(&key, &block_group->key, sizeof(key));
8503 btrfs_clear_space_info_full(root->fs_info);
8505 btrfs_put_block_group(block_group);
8506 btrfs_put_block_group(block_group);
8508 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8514 ret = btrfs_del_item(trans, root, path);
8516 btrfs_free_path(path);
8520 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8522 struct btrfs_space_info *space_info;
8523 struct btrfs_super_block *disk_super;
8529 disk_super = fs_info->super_copy;
8530 if (!btrfs_super_root(disk_super))
8533 features = btrfs_super_incompat_flags(disk_super);
8534 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8537 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8538 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8543 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8544 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8546 flags = BTRFS_BLOCK_GROUP_METADATA;
8547 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8551 flags = BTRFS_BLOCK_GROUP_DATA;
8552 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8558 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8560 return unpin_extent_range(root, start, end);
8563 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8564 u64 num_bytes, u64 *actual_bytes)
8566 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8569 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8571 struct btrfs_fs_info *fs_info = root->fs_info;
8572 struct btrfs_block_group_cache *cache = NULL;
8577 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8581 * try to trim all FS space, our block group may start from non-zero.
8583 if (range->len == total_bytes)
8584 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8586 cache = btrfs_lookup_block_group(fs_info, range->start);
8589 if (cache->key.objectid >= (range->start + range->len)) {
8590 btrfs_put_block_group(cache);
8594 start = max(range->start, cache->key.objectid);
8595 end = min(range->start + range->len,
8596 cache->key.objectid + cache->key.offset);
8598 if (end - start >= range->minlen) {
8599 if (!block_group_cache_done(cache)) {
8600 ret = cache_block_group(cache, 0);
8602 wait_block_group_cache_done(cache);
8604 ret = btrfs_trim_block_group(cache,
8610 trimmed += group_trimmed;
8612 btrfs_put_block_group(cache);
8617 cache = next_block_group(fs_info->tree_root, cache);
8620 range->len = trimmed;