2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
33 #include "transaction.h"
37 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
109 int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 if (block_group->key.objectid < cache->key.objectid) {
159 } else if (block_group->key.objectid > cache->key.objectid) {
162 spin_unlock(&info->block_group_cache_lock);
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
174 spin_unlock(&info->block_group_cache_lock);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
187 struct btrfs_block_group_cache *cache, *ret = NULL;
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
195 cache = rb_entry(n, struct btrfs_block_group_cache,
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
220 spin_unlock(&info->block_group_cache_lock);
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (cache->cached != BTRFS_CACHE_STARTED) {
315 spin_unlock(&cache->lock);
319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache->caching_ctl) {
321 spin_unlock(&cache->lock);
325 ctl = cache->caching_ctl;
326 atomic_inc(&ctl->count);
327 spin_unlock(&cache->lock);
331 static void put_caching_control(struct btrfs_caching_control *ctl)
333 if (atomic_dec_and_test(&ctl->count))
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
343 struct btrfs_fs_info *info, u64 start, u64 end)
345 u64 extent_start, extent_end, size, total_added = 0;
348 while (start < end) {
349 ret = find_first_extent_bit(info->pinned_extents, start,
350 &extent_start, &extent_end,
351 EXTENT_DIRTY | EXTENT_UPTODATE,
356 if (extent_start <= start) {
357 start = extent_end + 1;
358 } else if (extent_start > start && extent_start < end) {
359 size = extent_start - start;
361 ret = btrfs_add_free_space(block_group, start,
363 BUG_ON(ret); /* -ENOMEM or logic error */
364 start = extent_end + 1;
373 ret = btrfs_add_free_space(block_group, start, size);
374 BUG_ON(ret); /* -ENOMEM or logic error */
380 static noinline void caching_thread(struct btrfs_work *work)
382 struct btrfs_block_group_cache *block_group;
383 struct btrfs_fs_info *fs_info;
384 struct btrfs_caching_control *caching_ctl;
385 struct btrfs_root *extent_root;
386 struct btrfs_path *path;
387 struct extent_buffer *leaf;
388 struct btrfs_key key;
394 caching_ctl = container_of(work, struct btrfs_caching_control, work);
395 block_group = caching_ctl->block_group;
396 fs_info = block_group->fs_info;
397 extent_root = fs_info->extent_root;
399 path = btrfs_alloc_path();
403 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
411 path->skip_locking = 1;
412 path->search_commit_root = 1;
417 key.type = BTRFS_EXTENT_ITEM_KEY;
419 mutex_lock(&caching_ctl->mutex);
420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info->extent_commit_sem);
423 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
427 leaf = path->nodes[0];
428 nritems = btrfs_header_nritems(leaf);
431 if (btrfs_fs_closing(fs_info) > 1) {
436 if (path->slots[0] < nritems) {
437 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
439 ret = find_next_key(path, 0, &key);
443 if (need_resched()) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->extent_commit_sem);
447 mutex_unlock(&caching_ctl->mutex);
452 ret = btrfs_next_leaf(extent_root, path);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
462 if (key.objectid < block_group->key.objectid) {
467 if (key.objectid >= block_group->key.objectid +
468 block_group->key.offset)
471 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
472 key.type == BTRFS_METADATA_ITEM_KEY) {
473 total_found += add_new_free_space(block_group,
476 if (key.type == BTRFS_METADATA_ITEM_KEY)
477 last = key.objectid +
478 fs_info->tree_root->leafsize;
480 last = key.objectid + key.offset;
482 if (total_found > (1024 * 1024 * 2)) {
484 wake_up(&caching_ctl->wait);
491 total_found += add_new_free_space(block_group, fs_info, last,
492 block_group->key.objectid +
493 block_group->key.offset);
494 caching_ctl->progress = (u64)-1;
496 spin_lock(&block_group->lock);
497 block_group->caching_ctl = NULL;
498 block_group->cached = BTRFS_CACHE_FINISHED;
499 spin_unlock(&block_group->lock);
502 btrfs_free_path(path);
503 up_read(&fs_info->extent_commit_sem);
505 free_excluded_extents(extent_root, block_group);
507 mutex_unlock(&caching_ctl->mutex);
509 wake_up(&caching_ctl->wait);
511 put_caching_control(caching_ctl);
512 btrfs_put_block_group(block_group);
515 static int cache_block_group(struct btrfs_block_group_cache *cache,
519 struct btrfs_fs_info *fs_info = cache->fs_info;
520 struct btrfs_caching_control *caching_ctl;
523 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
527 INIT_LIST_HEAD(&caching_ctl->list);
528 mutex_init(&caching_ctl->mutex);
529 init_waitqueue_head(&caching_ctl->wait);
530 caching_ctl->block_group = cache;
531 caching_ctl->progress = cache->key.objectid;
532 atomic_set(&caching_ctl->count, 1);
533 caching_ctl->work.func = caching_thread;
535 spin_lock(&cache->lock);
537 * This should be a rare occasion, but this could happen I think in the
538 * case where one thread starts to load the space cache info, and then
539 * some other thread starts a transaction commit which tries to do an
540 * allocation while the other thread is still loading the space cache
541 * info. The previous loop should have kept us from choosing this block
542 * group, but if we've moved to the state where we will wait on caching
543 * block groups we need to first check if we're doing a fast load here,
544 * so we can wait for it to finish, otherwise we could end up allocating
545 * from a block group who's cache gets evicted for one reason or
548 while (cache->cached == BTRFS_CACHE_FAST) {
549 struct btrfs_caching_control *ctl;
551 ctl = cache->caching_ctl;
552 atomic_inc(&ctl->count);
553 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
554 spin_unlock(&cache->lock);
558 finish_wait(&ctl->wait, &wait);
559 put_caching_control(ctl);
560 spin_lock(&cache->lock);
563 if (cache->cached != BTRFS_CACHE_NO) {
564 spin_unlock(&cache->lock);
568 WARN_ON(cache->caching_ctl);
569 cache->caching_ctl = caching_ctl;
570 cache->cached = BTRFS_CACHE_FAST;
571 spin_unlock(&cache->lock);
573 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
574 ret = load_free_space_cache(fs_info, cache);
576 spin_lock(&cache->lock);
578 cache->caching_ctl = NULL;
579 cache->cached = BTRFS_CACHE_FINISHED;
580 cache->last_byte_to_unpin = (u64)-1;
582 if (load_cache_only) {
583 cache->caching_ctl = NULL;
584 cache->cached = BTRFS_CACHE_NO;
586 cache->cached = BTRFS_CACHE_STARTED;
589 spin_unlock(&cache->lock);
590 wake_up(&caching_ctl->wait);
592 put_caching_control(caching_ctl);
593 free_excluded_extents(fs_info->extent_root, cache);
598 * We are not going to do the fast caching, set cached to the
599 * appropriate value and wakeup any waiters.
601 spin_lock(&cache->lock);
602 if (load_cache_only) {
603 cache->caching_ctl = NULL;
604 cache->cached = BTRFS_CACHE_NO;
606 cache->cached = BTRFS_CACHE_STARTED;
608 spin_unlock(&cache->lock);
609 wake_up(&caching_ctl->wait);
612 if (load_cache_only) {
613 put_caching_control(caching_ctl);
617 down_write(&fs_info->extent_commit_sem);
618 atomic_inc(&caching_ctl->count);
619 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
620 up_write(&fs_info->extent_commit_sem);
622 btrfs_get_block_group(cache);
624 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
630 * return the block group that starts at or after bytenr
632 static struct btrfs_block_group_cache *
633 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
635 struct btrfs_block_group_cache *cache;
637 cache = block_group_cache_tree_search(info, bytenr, 0);
643 * return the block group that contains the given bytenr
645 struct btrfs_block_group_cache *btrfs_lookup_block_group(
646 struct btrfs_fs_info *info,
649 struct btrfs_block_group_cache *cache;
651 cache = block_group_cache_tree_search(info, bytenr, 1);
656 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
659 struct list_head *head = &info->space_info;
660 struct btrfs_space_info *found;
662 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
665 list_for_each_entry_rcu(found, head, list) {
666 if (found->flags & flags) {
676 * after adding space to the filesystem, we need to clear the full flags
677 * on all the space infos.
679 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
681 struct list_head *head = &info->space_info;
682 struct btrfs_space_info *found;
685 list_for_each_entry_rcu(found, head, list)
690 /* simple helper to search for an existing extent at a given offset */
691 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
694 struct btrfs_key key;
695 struct btrfs_path *path;
697 path = btrfs_alloc_path();
701 key.objectid = start;
703 key.type = BTRFS_EXTENT_ITEM_KEY;
704 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
707 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
708 if (key.objectid == start &&
709 key.type == BTRFS_METADATA_ITEM_KEY)
712 btrfs_free_path(path);
717 * helper function to lookup reference count and flags of a tree block.
719 * the head node for delayed ref is used to store the sum of all the
720 * reference count modifications queued up in the rbtree. the head
721 * node may also store the extent flags to set. This way you can check
722 * to see what the reference count and extent flags would be if all of
723 * the delayed refs are not processed.
725 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
726 struct btrfs_root *root, u64 bytenr,
727 u64 offset, int metadata, u64 *refs, u64 *flags)
729 struct btrfs_delayed_ref_head *head;
730 struct btrfs_delayed_ref_root *delayed_refs;
731 struct btrfs_path *path;
732 struct btrfs_extent_item *ei;
733 struct extent_buffer *leaf;
734 struct btrfs_key key;
741 * If we don't have skinny metadata, don't bother doing anything
744 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
745 offset = root->leafsize;
749 path = btrfs_alloc_path();
754 key.objectid = bytenr;
755 key.type = BTRFS_METADATA_ITEM_KEY;
758 key.objectid = bytenr;
759 key.type = BTRFS_EXTENT_ITEM_KEY;
764 path->skip_locking = 1;
765 path->search_commit_root = 1;
768 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
773 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
774 key.type = BTRFS_EXTENT_ITEM_KEY;
775 key.offset = root->leafsize;
776 btrfs_release_path(path);
781 leaf = path->nodes[0];
782 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
783 if (item_size >= sizeof(*ei)) {
784 ei = btrfs_item_ptr(leaf, path->slots[0],
785 struct btrfs_extent_item);
786 num_refs = btrfs_extent_refs(leaf, ei);
787 extent_flags = btrfs_extent_flags(leaf, ei);
789 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
790 struct btrfs_extent_item_v0 *ei0;
791 BUG_ON(item_size != sizeof(*ei0));
792 ei0 = btrfs_item_ptr(leaf, path->slots[0],
793 struct btrfs_extent_item_v0);
794 num_refs = btrfs_extent_refs_v0(leaf, ei0);
795 /* FIXME: this isn't correct for data */
796 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
801 BUG_ON(num_refs == 0);
811 delayed_refs = &trans->transaction->delayed_refs;
812 spin_lock(&delayed_refs->lock);
813 head = btrfs_find_delayed_ref_head(trans, bytenr);
815 if (!mutex_trylock(&head->mutex)) {
816 atomic_inc(&head->node.refs);
817 spin_unlock(&delayed_refs->lock);
819 btrfs_release_path(path);
822 * Mutex was contended, block until it's released and try
825 mutex_lock(&head->mutex);
826 mutex_unlock(&head->mutex);
827 btrfs_put_delayed_ref(&head->node);
830 if (head->extent_op && head->extent_op->update_flags)
831 extent_flags |= head->extent_op->flags_to_set;
833 BUG_ON(num_refs == 0);
835 num_refs += head->node.ref_mod;
836 mutex_unlock(&head->mutex);
838 spin_unlock(&delayed_refs->lock);
840 WARN_ON(num_refs == 0);
844 *flags = extent_flags;
846 btrfs_free_path(path);
851 * Back reference rules. Back refs have three main goals:
853 * 1) differentiate between all holders of references to an extent so that
854 * when a reference is dropped we can make sure it was a valid reference
855 * before freeing the extent.
857 * 2) Provide enough information to quickly find the holders of an extent
858 * if we notice a given block is corrupted or bad.
860 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
861 * maintenance. This is actually the same as #2, but with a slightly
862 * different use case.
864 * There are two kinds of back refs. The implicit back refs is optimized
865 * for pointers in non-shared tree blocks. For a given pointer in a block,
866 * back refs of this kind provide information about the block's owner tree
867 * and the pointer's key. These information allow us to find the block by
868 * b-tree searching. The full back refs is for pointers in tree blocks not
869 * referenced by their owner trees. The location of tree block is recorded
870 * in the back refs. Actually the full back refs is generic, and can be
871 * used in all cases the implicit back refs is used. The major shortcoming
872 * of the full back refs is its overhead. Every time a tree block gets
873 * COWed, we have to update back refs entry for all pointers in it.
875 * For a newly allocated tree block, we use implicit back refs for
876 * pointers in it. This means most tree related operations only involve
877 * implicit back refs. For a tree block created in old transaction, the
878 * only way to drop a reference to it is COW it. So we can detect the
879 * event that tree block loses its owner tree's reference and do the
880 * back refs conversion.
882 * When a tree block is COW'd through a tree, there are four cases:
884 * The reference count of the block is one and the tree is the block's
885 * owner tree. Nothing to do in this case.
887 * The reference count of the block is one and the tree is not the
888 * block's owner tree. In this case, full back refs is used for pointers
889 * in the block. Remove these full back refs, add implicit back refs for
890 * every pointers in the new block.
892 * The reference count of the block is greater than one and the tree is
893 * the block's owner tree. In this case, implicit back refs is used for
894 * pointers in the block. Add full back refs for every pointers in the
895 * block, increase lower level extents' reference counts. The original
896 * implicit back refs are entailed to the new block.
898 * The reference count of the block is greater than one and the tree is
899 * not the block's owner tree. Add implicit back refs for every pointer in
900 * the new block, increase lower level extents' reference count.
902 * Back Reference Key composing:
904 * The key objectid corresponds to the first byte in the extent,
905 * The key type is used to differentiate between types of back refs.
906 * There are different meanings of the key offset for different types
909 * File extents can be referenced by:
911 * - multiple snapshots, subvolumes, or different generations in one subvol
912 * - different files inside a single subvolume
913 * - different offsets inside a file (bookend extents in file.c)
915 * The extent ref structure for the implicit back refs has fields for:
917 * - Objectid of the subvolume root
918 * - objectid of the file holding the reference
919 * - original offset in the file
920 * - how many bookend extents
922 * The key offset for the implicit back refs is hash of the first
925 * The extent ref structure for the full back refs has field for:
927 * - number of pointers in the tree leaf
929 * The key offset for the implicit back refs is the first byte of
932 * When a file extent is allocated, The implicit back refs is used.
933 * the fields are filled in:
935 * (root_key.objectid, inode objectid, offset in file, 1)
937 * When a file extent is removed file truncation, we find the
938 * corresponding implicit back refs and check the following fields:
940 * (btrfs_header_owner(leaf), inode objectid, offset in file)
942 * Btree extents can be referenced by:
944 * - Different subvolumes
946 * Both the implicit back refs and the full back refs for tree blocks
947 * only consist of key. The key offset for the implicit back refs is
948 * objectid of block's owner tree. The key offset for the full back refs
949 * is the first byte of parent block.
951 * When implicit back refs is used, information about the lowest key and
952 * level of the tree block are required. These information are stored in
953 * tree block info structure.
956 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
957 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
958 struct btrfs_root *root,
959 struct btrfs_path *path,
960 u64 owner, u32 extra_size)
962 struct btrfs_extent_item *item;
963 struct btrfs_extent_item_v0 *ei0;
964 struct btrfs_extent_ref_v0 *ref0;
965 struct btrfs_tree_block_info *bi;
966 struct extent_buffer *leaf;
967 struct btrfs_key key;
968 struct btrfs_key found_key;
969 u32 new_size = sizeof(*item);
973 leaf = path->nodes[0];
974 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
976 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
977 ei0 = btrfs_item_ptr(leaf, path->slots[0],
978 struct btrfs_extent_item_v0);
979 refs = btrfs_extent_refs_v0(leaf, ei0);
981 if (owner == (u64)-1) {
983 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
984 ret = btrfs_next_leaf(root, path);
987 BUG_ON(ret > 0); /* Corruption */
988 leaf = path->nodes[0];
990 btrfs_item_key_to_cpu(leaf, &found_key,
992 BUG_ON(key.objectid != found_key.objectid);
993 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
997 ref0 = btrfs_item_ptr(leaf, path->slots[0],
998 struct btrfs_extent_ref_v0);
999 owner = btrfs_ref_objectid_v0(leaf, ref0);
1003 btrfs_release_path(path);
1005 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1006 new_size += sizeof(*bi);
1008 new_size -= sizeof(*ei0);
1009 ret = btrfs_search_slot(trans, root, &key, path,
1010 new_size + extra_size, 1);
1013 BUG_ON(ret); /* Corruption */
1015 btrfs_extend_item(root, path, new_size);
1017 leaf = path->nodes[0];
1018 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1019 btrfs_set_extent_refs(leaf, item, refs);
1020 /* FIXME: get real generation */
1021 btrfs_set_extent_generation(leaf, item, 0);
1022 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1023 btrfs_set_extent_flags(leaf, item,
1024 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1025 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1026 bi = (struct btrfs_tree_block_info *)(item + 1);
1027 /* FIXME: get first key of the block */
1028 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1029 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1031 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1033 btrfs_mark_buffer_dirty(leaf);
1038 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1040 u32 high_crc = ~(u32)0;
1041 u32 low_crc = ~(u32)0;
1044 lenum = cpu_to_le64(root_objectid);
1045 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1046 lenum = cpu_to_le64(owner);
1047 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048 lenum = cpu_to_le64(offset);
1049 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1051 return ((u64)high_crc << 31) ^ (u64)low_crc;
1054 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1055 struct btrfs_extent_data_ref *ref)
1057 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1058 btrfs_extent_data_ref_objectid(leaf, ref),
1059 btrfs_extent_data_ref_offset(leaf, ref));
1062 static int match_extent_data_ref(struct extent_buffer *leaf,
1063 struct btrfs_extent_data_ref *ref,
1064 u64 root_objectid, u64 owner, u64 offset)
1066 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1067 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1068 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1073 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1074 struct btrfs_root *root,
1075 struct btrfs_path *path,
1076 u64 bytenr, u64 parent,
1078 u64 owner, u64 offset)
1080 struct btrfs_key key;
1081 struct btrfs_extent_data_ref *ref;
1082 struct extent_buffer *leaf;
1088 key.objectid = bytenr;
1090 key.type = BTRFS_SHARED_DATA_REF_KEY;
1091 key.offset = parent;
1093 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1094 key.offset = hash_extent_data_ref(root_objectid,
1099 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1108 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1109 key.type = BTRFS_EXTENT_REF_V0_KEY;
1110 btrfs_release_path(path);
1111 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1122 leaf = path->nodes[0];
1123 nritems = btrfs_header_nritems(leaf);
1125 if (path->slots[0] >= nritems) {
1126 ret = btrfs_next_leaf(root, path);
1132 leaf = path->nodes[0];
1133 nritems = btrfs_header_nritems(leaf);
1137 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1138 if (key.objectid != bytenr ||
1139 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1142 ref = btrfs_item_ptr(leaf, path->slots[0],
1143 struct btrfs_extent_data_ref);
1145 if (match_extent_data_ref(leaf, ref, root_objectid,
1148 btrfs_release_path(path);
1160 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 u64 bytenr, u64 parent,
1164 u64 root_objectid, u64 owner,
1165 u64 offset, int refs_to_add)
1167 struct btrfs_key key;
1168 struct extent_buffer *leaf;
1173 key.objectid = bytenr;
1175 key.type = BTRFS_SHARED_DATA_REF_KEY;
1176 key.offset = parent;
1177 size = sizeof(struct btrfs_shared_data_ref);
1179 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1180 key.offset = hash_extent_data_ref(root_objectid,
1182 size = sizeof(struct btrfs_extent_data_ref);
1185 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1186 if (ret && ret != -EEXIST)
1189 leaf = path->nodes[0];
1191 struct btrfs_shared_data_ref *ref;
1192 ref = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_shared_data_ref);
1195 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1197 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1198 num_refs += refs_to_add;
1199 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1202 struct btrfs_extent_data_ref *ref;
1203 while (ret == -EEXIST) {
1204 ref = btrfs_item_ptr(leaf, path->slots[0],
1205 struct btrfs_extent_data_ref);
1206 if (match_extent_data_ref(leaf, ref, root_objectid,
1209 btrfs_release_path(path);
1211 ret = btrfs_insert_empty_item(trans, root, path, &key,
1213 if (ret && ret != -EEXIST)
1216 leaf = path->nodes[0];
1218 ref = btrfs_item_ptr(leaf, path->slots[0],
1219 struct btrfs_extent_data_ref);
1221 btrfs_set_extent_data_ref_root(leaf, ref,
1223 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1224 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1225 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1227 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1228 num_refs += refs_to_add;
1229 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1232 btrfs_mark_buffer_dirty(leaf);
1235 btrfs_release_path(path);
1239 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1240 struct btrfs_root *root,
1241 struct btrfs_path *path,
1244 struct btrfs_key key;
1245 struct btrfs_extent_data_ref *ref1 = NULL;
1246 struct btrfs_shared_data_ref *ref2 = NULL;
1247 struct extent_buffer *leaf;
1251 leaf = path->nodes[0];
1252 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1254 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1255 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1256 struct btrfs_extent_data_ref);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1258 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1259 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_shared_data_ref);
1261 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1262 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1263 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1264 struct btrfs_extent_ref_v0 *ref0;
1265 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1266 struct btrfs_extent_ref_v0);
1267 num_refs = btrfs_ref_count_v0(leaf, ref0);
1273 BUG_ON(num_refs < refs_to_drop);
1274 num_refs -= refs_to_drop;
1276 if (num_refs == 0) {
1277 ret = btrfs_del_item(trans, root, path);
1279 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1280 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1281 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1282 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1283 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 struct btrfs_extent_ref_v0 *ref0;
1286 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_extent_ref_v0);
1288 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1291 btrfs_mark_buffer_dirty(leaf);
1296 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1297 struct btrfs_path *path,
1298 struct btrfs_extent_inline_ref *iref)
1300 struct btrfs_key key;
1301 struct extent_buffer *leaf;
1302 struct btrfs_extent_data_ref *ref1;
1303 struct btrfs_shared_data_ref *ref2;
1306 leaf = path->nodes[0];
1307 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1309 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1310 BTRFS_EXTENT_DATA_REF_KEY) {
1311 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1312 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1314 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1315 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1317 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1318 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_extent_data_ref);
1320 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1321 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1322 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1323 struct btrfs_shared_data_ref);
1324 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1325 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1326 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1327 struct btrfs_extent_ref_v0 *ref0;
1328 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1329 struct btrfs_extent_ref_v0);
1330 num_refs = btrfs_ref_count_v0(leaf, ref0);
1338 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1339 struct btrfs_root *root,
1340 struct btrfs_path *path,
1341 u64 bytenr, u64 parent,
1344 struct btrfs_key key;
1347 key.objectid = bytenr;
1349 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1350 key.offset = parent;
1352 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1353 key.offset = root_objectid;
1356 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1359 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1360 if (ret == -ENOENT && parent) {
1361 btrfs_release_path(path);
1362 key.type = BTRFS_EXTENT_REF_V0_KEY;
1363 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1371 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1372 struct btrfs_root *root,
1373 struct btrfs_path *path,
1374 u64 bytenr, u64 parent,
1377 struct btrfs_key key;
1380 key.objectid = bytenr;
1382 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1383 key.offset = parent;
1385 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1386 key.offset = root_objectid;
1389 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1390 btrfs_release_path(path);
1394 static inline int extent_ref_type(u64 parent, u64 owner)
1397 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1399 type = BTRFS_SHARED_BLOCK_REF_KEY;
1401 type = BTRFS_TREE_BLOCK_REF_KEY;
1404 type = BTRFS_SHARED_DATA_REF_KEY;
1406 type = BTRFS_EXTENT_DATA_REF_KEY;
1411 static int find_next_key(struct btrfs_path *path, int level,
1412 struct btrfs_key *key)
1415 for (; level < BTRFS_MAX_LEVEL; level++) {
1416 if (!path->nodes[level])
1418 if (path->slots[level] + 1 >=
1419 btrfs_header_nritems(path->nodes[level]))
1422 btrfs_item_key_to_cpu(path->nodes[level], key,
1423 path->slots[level] + 1);
1425 btrfs_node_key_to_cpu(path->nodes[level], key,
1426 path->slots[level] + 1);
1433 * look for inline back ref. if back ref is found, *ref_ret is set
1434 * to the address of inline back ref, and 0 is returned.
1436 * if back ref isn't found, *ref_ret is set to the address where it
1437 * should be inserted, and -ENOENT is returned.
1439 * if insert is true and there are too many inline back refs, the path
1440 * points to the extent item, and -EAGAIN is returned.
1442 * NOTE: inline back refs are ordered in the same way that back ref
1443 * items in the tree are ordered.
1445 static noinline_for_stack
1446 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1447 struct btrfs_root *root,
1448 struct btrfs_path *path,
1449 struct btrfs_extent_inline_ref **ref_ret,
1450 u64 bytenr, u64 num_bytes,
1451 u64 parent, u64 root_objectid,
1452 u64 owner, u64 offset, int insert)
1454 struct btrfs_key key;
1455 struct extent_buffer *leaf;
1456 struct btrfs_extent_item *ei;
1457 struct btrfs_extent_inline_ref *iref;
1467 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1470 key.objectid = bytenr;
1471 key.type = BTRFS_EXTENT_ITEM_KEY;
1472 key.offset = num_bytes;
1474 want = extent_ref_type(parent, owner);
1476 extra_size = btrfs_extent_inline_ref_size(want);
1477 path->keep_locks = 1;
1482 * Owner is our parent level, so we can just add one to get the level
1483 * for the block we are interested in.
1485 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1486 key.type = BTRFS_METADATA_ITEM_KEY;
1491 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1498 * We may be a newly converted file system which still has the old fat
1499 * extent entries for metadata, so try and see if we have one of those.
1501 if (ret > 0 && skinny_metadata) {
1502 skinny_metadata = false;
1503 if (path->slots[0]) {
1505 btrfs_item_key_to_cpu(path->nodes[0], &key,
1507 if (key.objectid == bytenr &&
1508 key.type == BTRFS_EXTENT_ITEM_KEY &&
1509 key.offset == num_bytes)
1513 key.type = BTRFS_EXTENT_ITEM_KEY;
1514 key.offset = num_bytes;
1515 btrfs_release_path(path);
1520 if (ret && !insert) {
1529 leaf = path->nodes[0];
1530 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1531 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1532 if (item_size < sizeof(*ei)) {
1537 ret = convert_extent_item_v0(trans, root, path, owner,
1543 leaf = path->nodes[0];
1544 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1547 BUG_ON(item_size < sizeof(*ei));
1549 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1550 flags = btrfs_extent_flags(leaf, ei);
1552 ptr = (unsigned long)(ei + 1);
1553 end = (unsigned long)ei + item_size;
1555 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1556 ptr += sizeof(struct btrfs_tree_block_info);
1566 iref = (struct btrfs_extent_inline_ref *)ptr;
1567 type = btrfs_extent_inline_ref_type(leaf, iref);
1571 ptr += btrfs_extent_inline_ref_size(type);
1575 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1576 struct btrfs_extent_data_ref *dref;
1577 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1578 if (match_extent_data_ref(leaf, dref, root_objectid,
1583 if (hash_extent_data_ref_item(leaf, dref) <
1584 hash_extent_data_ref(root_objectid, owner, offset))
1588 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1590 if (parent == ref_offset) {
1594 if (ref_offset < parent)
1597 if (root_objectid == ref_offset) {
1601 if (ref_offset < root_objectid)
1605 ptr += btrfs_extent_inline_ref_size(type);
1607 if (err == -ENOENT && insert) {
1608 if (item_size + extra_size >=
1609 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1614 * To add new inline back ref, we have to make sure
1615 * there is no corresponding back ref item.
1616 * For simplicity, we just do not add new inline back
1617 * ref if there is any kind of item for this block
1619 if (find_next_key(path, 0, &key) == 0 &&
1620 key.objectid == bytenr &&
1621 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1626 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1629 path->keep_locks = 0;
1630 btrfs_unlock_up_safe(path, 1);
1636 * helper to add new inline back ref
1638 static noinline_for_stack
1639 void setup_inline_extent_backref(struct btrfs_root *root,
1640 struct btrfs_path *path,
1641 struct btrfs_extent_inline_ref *iref,
1642 u64 parent, u64 root_objectid,
1643 u64 owner, u64 offset, int refs_to_add,
1644 struct btrfs_delayed_extent_op *extent_op)
1646 struct extent_buffer *leaf;
1647 struct btrfs_extent_item *ei;
1650 unsigned long item_offset;
1655 leaf = path->nodes[0];
1656 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1657 item_offset = (unsigned long)iref - (unsigned long)ei;
1659 type = extent_ref_type(parent, owner);
1660 size = btrfs_extent_inline_ref_size(type);
1662 btrfs_extend_item(root, path, size);
1664 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1665 refs = btrfs_extent_refs(leaf, ei);
1666 refs += refs_to_add;
1667 btrfs_set_extent_refs(leaf, ei, refs);
1669 __run_delayed_extent_op(extent_op, leaf, ei);
1671 ptr = (unsigned long)ei + item_offset;
1672 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1673 if (ptr < end - size)
1674 memmove_extent_buffer(leaf, ptr + size, ptr,
1677 iref = (struct btrfs_extent_inline_ref *)ptr;
1678 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1679 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1680 struct btrfs_extent_data_ref *dref;
1681 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1682 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1683 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1684 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1685 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1686 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1687 struct btrfs_shared_data_ref *sref;
1688 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1689 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1690 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1691 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1692 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1694 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1696 btrfs_mark_buffer_dirty(leaf);
1699 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1700 struct btrfs_root *root,
1701 struct btrfs_path *path,
1702 struct btrfs_extent_inline_ref **ref_ret,
1703 u64 bytenr, u64 num_bytes, u64 parent,
1704 u64 root_objectid, u64 owner, u64 offset)
1708 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1709 bytenr, num_bytes, parent,
1710 root_objectid, owner, offset, 0);
1714 btrfs_release_path(path);
1717 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1718 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1721 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1722 root_objectid, owner, offset);
1728 * helper to update/remove inline back ref
1730 static noinline_for_stack
1731 void update_inline_extent_backref(struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref *iref,
1735 struct btrfs_delayed_extent_op *extent_op)
1737 struct extent_buffer *leaf;
1738 struct btrfs_extent_item *ei;
1739 struct btrfs_extent_data_ref *dref = NULL;
1740 struct btrfs_shared_data_ref *sref = NULL;
1748 leaf = path->nodes[0];
1749 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1750 refs = btrfs_extent_refs(leaf, ei);
1751 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1752 refs += refs_to_mod;
1753 btrfs_set_extent_refs(leaf, ei, refs);
1755 __run_delayed_extent_op(extent_op, leaf, ei);
1757 type = btrfs_extent_inline_ref_type(leaf, iref);
1759 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1760 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1761 refs = btrfs_extent_data_ref_count(leaf, dref);
1762 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1763 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1764 refs = btrfs_shared_data_ref_count(leaf, sref);
1767 BUG_ON(refs_to_mod != -1);
1770 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1771 refs += refs_to_mod;
1774 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1775 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1777 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1779 size = btrfs_extent_inline_ref_size(type);
1780 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1781 ptr = (unsigned long)iref;
1782 end = (unsigned long)ei + item_size;
1783 if (ptr + size < end)
1784 memmove_extent_buffer(leaf, ptr, ptr + size,
1787 btrfs_truncate_item(root, path, item_size, 1);
1789 btrfs_mark_buffer_dirty(leaf);
1792 static noinline_for_stack
1793 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1794 struct btrfs_root *root,
1795 struct btrfs_path *path,
1796 u64 bytenr, u64 num_bytes, u64 parent,
1797 u64 root_objectid, u64 owner,
1798 u64 offset, int refs_to_add,
1799 struct btrfs_delayed_extent_op *extent_op)
1801 struct btrfs_extent_inline_ref *iref;
1804 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1805 bytenr, num_bytes, parent,
1806 root_objectid, owner, offset, 1);
1808 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1809 update_inline_extent_backref(root, path, iref,
1810 refs_to_add, extent_op);
1811 } else if (ret == -ENOENT) {
1812 setup_inline_extent_backref(root, path, iref, parent,
1813 root_objectid, owner, offset,
1814 refs_to_add, extent_op);
1820 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1821 struct btrfs_root *root,
1822 struct btrfs_path *path,
1823 u64 bytenr, u64 parent, u64 root_objectid,
1824 u64 owner, u64 offset, int refs_to_add)
1827 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1828 BUG_ON(refs_to_add != 1);
1829 ret = insert_tree_block_ref(trans, root, path, bytenr,
1830 parent, root_objectid);
1832 ret = insert_extent_data_ref(trans, root, path, bytenr,
1833 parent, root_objectid,
1834 owner, offset, refs_to_add);
1839 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1840 struct btrfs_root *root,
1841 struct btrfs_path *path,
1842 struct btrfs_extent_inline_ref *iref,
1843 int refs_to_drop, int is_data)
1847 BUG_ON(!is_data && refs_to_drop != 1);
1849 update_inline_extent_backref(root, path, iref,
1850 -refs_to_drop, NULL);
1851 } else if (is_data) {
1852 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1854 ret = btrfs_del_item(trans, root, path);
1859 static int btrfs_issue_discard(struct block_device *bdev,
1862 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1865 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1866 u64 num_bytes, u64 *actual_bytes)
1869 u64 discarded_bytes = 0;
1870 struct btrfs_bio *bbio = NULL;
1873 /* Tell the block device(s) that the sectors can be discarded */
1874 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1875 bytenr, &num_bytes, &bbio, 0);
1876 /* Error condition is -ENOMEM */
1878 struct btrfs_bio_stripe *stripe = bbio->stripes;
1882 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1883 if (!stripe->dev->can_discard)
1886 ret = btrfs_issue_discard(stripe->dev->bdev,
1890 discarded_bytes += stripe->length;
1891 else if (ret != -EOPNOTSUPP)
1892 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1895 * Just in case we get back EOPNOTSUPP for some reason,
1896 * just ignore the return value so we don't screw up
1897 * people calling discard_extent.
1905 *actual_bytes = discarded_bytes;
1908 if (ret == -EOPNOTSUPP)
1913 /* Can return -ENOMEM */
1914 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1915 struct btrfs_root *root,
1916 u64 bytenr, u64 num_bytes, u64 parent,
1917 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1920 struct btrfs_fs_info *fs_info = root->fs_info;
1922 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1923 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1925 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1926 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1928 parent, root_objectid, (int)owner,
1929 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1931 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1933 parent, root_objectid, owner, offset,
1934 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1939 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1940 struct btrfs_root *root,
1941 u64 bytenr, u64 num_bytes,
1942 u64 parent, u64 root_objectid,
1943 u64 owner, u64 offset, int refs_to_add,
1944 struct btrfs_delayed_extent_op *extent_op)
1946 struct btrfs_path *path;
1947 struct extent_buffer *leaf;
1948 struct btrfs_extent_item *item;
1953 path = btrfs_alloc_path();
1958 path->leave_spinning = 1;
1959 /* this will setup the path even if it fails to insert the back ref */
1960 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1961 path, bytenr, num_bytes, parent,
1962 root_objectid, owner, offset,
1963 refs_to_add, extent_op);
1967 if (ret != -EAGAIN) {
1972 leaf = path->nodes[0];
1973 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1974 refs = btrfs_extent_refs(leaf, item);
1975 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1977 __run_delayed_extent_op(extent_op, leaf, item);
1979 btrfs_mark_buffer_dirty(leaf);
1980 btrfs_release_path(path);
1983 path->leave_spinning = 1;
1985 /* now insert the actual backref */
1986 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1987 path, bytenr, parent, root_objectid,
1988 owner, offset, refs_to_add);
1990 btrfs_abort_transaction(trans, root, ret);
1992 btrfs_free_path(path);
1996 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1997 struct btrfs_root *root,
1998 struct btrfs_delayed_ref_node *node,
1999 struct btrfs_delayed_extent_op *extent_op,
2000 int insert_reserved)
2003 struct btrfs_delayed_data_ref *ref;
2004 struct btrfs_key ins;
2009 ins.objectid = node->bytenr;
2010 ins.offset = node->num_bytes;
2011 ins.type = BTRFS_EXTENT_ITEM_KEY;
2013 ref = btrfs_delayed_node_to_data_ref(node);
2014 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2015 parent = ref->parent;
2017 ref_root = ref->root;
2019 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2021 flags |= extent_op->flags_to_set;
2022 ret = alloc_reserved_file_extent(trans, root,
2023 parent, ref_root, flags,
2024 ref->objectid, ref->offset,
2025 &ins, node->ref_mod);
2026 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2027 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2028 node->num_bytes, parent,
2029 ref_root, ref->objectid,
2030 ref->offset, node->ref_mod,
2032 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2033 ret = __btrfs_free_extent(trans, root, node->bytenr,
2034 node->num_bytes, parent,
2035 ref_root, ref->objectid,
2036 ref->offset, node->ref_mod,
2044 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2045 struct extent_buffer *leaf,
2046 struct btrfs_extent_item *ei)
2048 u64 flags = btrfs_extent_flags(leaf, ei);
2049 if (extent_op->update_flags) {
2050 flags |= extent_op->flags_to_set;
2051 btrfs_set_extent_flags(leaf, ei, flags);
2054 if (extent_op->update_key) {
2055 struct btrfs_tree_block_info *bi;
2056 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2057 bi = (struct btrfs_tree_block_info *)(ei + 1);
2058 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2062 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2063 struct btrfs_root *root,
2064 struct btrfs_delayed_ref_node *node,
2065 struct btrfs_delayed_extent_op *extent_op)
2067 struct btrfs_key key;
2068 struct btrfs_path *path;
2069 struct btrfs_extent_item *ei;
2070 struct extent_buffer *leaf;
2074 int metadata = !extent_op->is_data;
2079 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2082 path = btrfs_alloc_path();
2086 key.objectid = node->bytenr;
2089 key.type = BTRFS_METADATA_ITEM_KEY;
2090 key.offset = extent_op->level;
2092 key.type = BTRFS_EXTENT_ITEM_KEY;
2093 key.offset = node->num_bytes;
2098 path->leave_spinning = 1;
2099 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2107 btrfs_release_path(path);
2110 key.offset = node->num_bytes;
2111 key.type = BTRFS_EXTENT_ITEM_KEY;
2118 leaf = path->nodes[0];
2119 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2120 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2121 if (item_size < sizeof(*ei)) {
2122 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2128 leaf = path->nodes[0];
2129 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2132 BUG_ON(item_size < sizeof(*ei));
2133 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2134 __run_delayed_extent_op(extent_op, leaf, ei);
2136 btrfs_mark_buffer_dirty(leaf);
2138 btrfs_free_path(path);
2142 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2143 struct btrfs_root *root,
2144 struct btrfs_delayed_ref_node *node,
2145 struct btrfs_delayed_extent_op *extent_op,
2146 int insert_reserved)
2149 struct btrfs_delayed_tree_ref *ref;
2150 struct btrfs_key ins;
2153 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2156 ref = btrfs_delayed_node_to_tree_ref(node);
2157 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2158 parent = ref->parent;
2160 ref_root = ref->root;
2162 ins.objectid = node->bytenr;
2163 if (skinny_metadata) {
2164 ins.offset = ref->level;
2165 ins.type = BTRFS_METADATA_ITEM_KEY;
2167 ins.offset = node->num_bytes;
2168 ins.type = BTRFS_EXTENT_ITEM_KEY;
2171 BUG_ON(node->ref_mod != 1);
2172 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2173 BUG_ON(!extent_op || !extent_op->update_flags);
2174 ret = alloc_reserved_tree_block(trans, root,
2176 extent_op->flags_to_set,
2179 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2180 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2181 node->num_bytes, parent, ref_root,
2182 ref->level, 0, 1, extent_op);
2183 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2184 ret = __btrfs_free_extent(trans, root, node->bytenr,
2185 node->num_bytes, parent, ref_root,
2186 ref->level, 0, 1, extent_op);
2193 /* helper function to actually process a single delayed ref entry */
2194 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2195 struct btrfs_root *root,
2196 struct btrfs_delayed_ref_node *node,
2197 struct btrfs_delayed_extent_op *extent_op,
2198 int insert_reserved)
2205 if (btrfs_delayed_ref_is_head(node)) {
2206 struct btrfs_delayed_ref_head *head;
2208 * we've hit the end of the chain and we were supposed
2209 * to insert this extent into the tree. But, it got
2210 * deleted before we ever needed to insert it, so all
2211 * we have to do is clean up the accounting
2214 head = btrfs_delayed_node_to_head(node);
2215 if (insert_reserved) {
2216 btrfs_pin_extent(root, node->bytenr,
2217 node->num_bytes, 1);
2218 if (head->is_data) {
2219 ret = btrfs_del_csums(trans, root,
2227 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2228 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2229 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2231 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2232 node->type == BTRFS_SHARED_DATA_REF_KEY)
2233 ret = run_delayed_data_ref(trans, root, node, extent_op,
2240 static noinline struct btrfs_delayed_ref_node *
2241 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2243 struct rb_node *node;
2244 struct btrfs_delayed_ref_node *ref;
2245 int action = BTRFS_ADD_DELAYED_REF;
2248 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2249 * this prevents ref count from going down to zero when
2250 * there still are pending delayed ref.
2252 node = rb_prev(&head->node.rb_node);
2256 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2258 if (ref->bytenr != head->node.bytenr)
2260 if (ref->action == action)
2262 node = rb_prev(node);
2264 if (action == BTRFS_ADD_DELAYED_REF) {
2265 action = BTRFS_DROP_DELAYED_REF;
2272 * Returns 0 on success or if called with an already aborted transaction.
2273 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2275 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2276 struct btrfs_root *root,
2277 struct list_head *cluster)
2279 struct btrfs_delayed_ref_root *delayed_refs;
2280 struct btrfs_delayed_ref_node *ref;
2281 struct btrfs_delayed_ref_head *locked_ref = NULL;
2282 struct btrfs_delayed_extent_op *extent_op;
2283 struct btrfs_fs_info *fs_info = root->fs_info;
2286 int must_insert_reserved = 0;
2288 delayed_refs = &trans->transaction->delayed_refs;
2291 /* pick a new head ref from the cluster list */
2292 if (list_empty(cluster))
2295 locked_ref = list_entry(cluster->next,
2296 struct btrfs_delayed_ref_head, cluster);
2298 /* grab the lock that says we are going to process
2299 * all the refs for this head */
2300 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2303 * we may have dropped the spin lock to get the head
2304 * mutex lock, and that might have given someone else
2305 * time to free the head. If that's true, it has been
2306 * removed from our list and we can move on.
2308 if (ret == -EAGAIN) {
2316 * We need to try and merge add/drops of the same ref since we
2317 * can run into issues with relocate dropping the implicit ref
2318 * and then it being added back again before the drop can
2319 * finish. If we merged anything we need to re-loop so we can
2322 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2326 * locked_ref is the head node, so we have to go one
2327 * node back for any delayed ref updates
2329 ref = select_delayed_ref(locked_ref);
2331 if (ref && ref->seq &&
2332 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2334 * there are still refs with lower seq numbers in the
2335 * process of being added. Don't run this ref yet.
2337 list_del_init(&locked_ref->cluster);
2338 btrfs_delayed_ref_unlock(locked_ref);
2340 delayed_refs->num_heads_ready++;
2341 spin_unlock(&delayed_refs->lock);
2343 spin_lock(&delayed_refs->lock);
2348 * record the must insert reserved flag before we
2349 * drop the spin lock.
2351 must_insert_reserved = locked_ref->must_insert_reserved;
2352 locked_ref->must_insert_reserved = 0;
2354 extent_op = locked_ref->extent_op;
2355 locked_ref->extent_op = NULL;
2358 /* All delayed refs have been processed, Go ahead
2359 * and send the head node to run_one_delayed_ref,
2360 * so that any accounting fixes can happen
2362 ref = &locked_ref->node;
2364 if (extent_op && must_insert_reserved) {
2365 btrfs_free_delayed_extent_op(extent_op);
2370 spin_unlock(&delayed_refs->lock);
2372 ret = run_delayed_extent_op(trans, root,
2374 btrfs_free_delayed_extent_op(extent_op);
2377 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2378 spin_lock(&delayed_refs->lock);
2379 btrfs_delayed_ref_unlock(locked_ref);
2388 rb_erase(&ref->rb_node, &delayed_refs->root);
2389 delayed_refs->num_entries--;
2390 if (!btrfs_delayed_ref_is_head(ref)) {
2392 * when we play the delayed ref, also correct the
2395 switch (ref->action) {
2396 case BTRFS_ADD_DELAYED_REF:
2397 case BTRFS_ADD_DELAYED_EXTENT:
2398 locked_ref->node.ref_mod -= ref->ref_mod;
2400 case BTRFS_DROP_DELAYED_REF:
2401 locked_ref->node.ref_mod += ref->ref_mod;
2407 spin_unlock(&delayed_refs->lock);
2409 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2410 must_insert_reserved);
2412 btrfs_free_delayed_extent_op(extent_op);
2414 btrfs_delayed_ref_unlock(locked_ref);
2415 btrfs_put_delayed_ref(ref);
2416 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2417 spin_lock(&delayed_refs->lock);
2422 * If this node is a head, that means all the refs in this head
2423 * have been dealt with, and we will pick the next head to deal
2424 * with, so we must unlock the head and drop it from the cluster
2425 * list before we release it.
2427 if (btrfs_delayed_ref_is_head(ref)) {
2428 list_del_init(&locked_ref->cluster);
2429 btrfs_delayed_ref_unlock(locked_ref);
2432 btrfs_put_delayed_ref(ref);
2436 spin_lock(&delayed_refs->lock);
2441 #ifdef SCRAMBLE_DELAYED_REFS
2443 * Normally delayed refs get processed in ascending bytenr order. This
2444 * correlates in most cases to the order added. To expose dependencies on this
2445 * order, we start to process the tree in the middle instead of the beginning
2447 static u64 find_middle(struct rb_root *root)
2449 struct rb_node *n = root->rb_node;
2450 struct btrfs_delayed_ref_node *entry;
2453 u64 first = 0, last = 0;
2457 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2458 first = entry->bytenr;
2462 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2463 last = entry->bytenr;
2468 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2469 WARN_ON(!entry->in_tree);
2471 middle = entry->bytenr;
2484 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2485 struct btrfs_fs_info *fs_info)
2487 struct qgroup_update *qgroup_update;
2490 if (list_empty(&trans->qgroup_ref_list) !=
2491 !trans->delayed_ref_elem.seq) {
2492 /* list without seq or seq without list */
2494 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2495 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2496 (u32)(trans->delayed_ref_elem.seq >> 32),
2497 (u32)trans->delayed_ref_elem.seq);
2501 if (!trans->delayed_ref_elem.seq)
2504 while (!list_empty(&trans->qgroup_ref_list)) {
2505 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2506 struct qgroup_update, list);
2507 list_del(&qgroup_update->list);
2509 ret = btrfs_qgroup_account_ref(
2510 trans, fs_info, qgroup_update->node,
2511 qgroup_update->extent_op);
2512 kfree(qgroup_update);
2515 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2520 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2523 int val = atomic_read(&delayed_refs->ref_seq);
2525 if (val < seq || val >= seq + count)
2530 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2534 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2535 sizeof(struct btrfs_extent_inline_ref));
2536 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2537 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2540 * We don't ever fill up leaves all the way so multiply by 2 just to be
2541 * closer to what we're really going to want to ouse.
2543 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2546 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2547 struct btrfs_root *root)
2549 struct btrfs_block_rsv *global_rsv;
2550 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2554 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2555 num_heads = heads_to_leaves(root, num_heads);
2557 num_bytes += (num_heads - 1) * root->leafsize;
2559 global_rsv = &root->fs_info->global_block_rsv;
2562 * If we can't allocate any more chunks lets make sure we have _lots_ of
2563 * wiggle room since running delayed refs can create more delayed refs.
2565 if (global_rsv->space_info->full)
2568 spin_lock(&global_rsv->lock);
2569 if (global_rsv->reserved <= num_bytes)
2571 spin_unlock(&global_rsv->lock);
2576 * this starts processing the delayed reference count updates and
2577 * extent insertions we have queued up so far. count can be
2578 * 0, which means to process everything in the tree at the start
2579 * of the run (but not newly added entries), or it can be some target
2580 * number you'd like to process.
2582 * Returns 0 on success or if called with an aborted transaction
2583 * Returns <0 on error and aborts the transaction
2585 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2586 struct btrfs_root *root, unsigned long count)
2588 struct rb_node *node;
2589 struct btrfs_delayed_ref_root *delayed_refs;
2590 struct btrfs_delayed_ref_node *ref;
2591 struct list_head cluster;
2594 int run_all = count == (unsigned long)-1;
2598 /* We'll clean this up in btrfs_cleanup_transaction */
2602 if (root == root->fs_info->extent_root)
2603 root = root->fs_info->tree_root;
2605 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2607 delayed_refs = &trans->transaction->delayed_refs;
2608 INIT_LIST_HEAD(&cluster);
2610 count = delayed_refs->num_entries * 2;
2614 if (!run_all && !run_most) {
2616 int seq = atomic_read(&delayed_refs->ref_seq);
2619 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2621 DEFINE_WAIT(__wait);
2622 if (delayed_refs->flushing ||
2623 !btrfs_should_throttle_delayed_refs(trans, root))
2626 prepare_to_wait(&delayed_refs->wait, &__wait,
2627 TASK_UNINTERRUPTIBLE);
2629 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2632 finish_wait(&delayed_refs->wait, &__wait);
2634 if (!refs_newer(delayed_refs, seq, 256))
2639 finish_wait(&delayed_refs->wait, &__wait);
2645 atomic_inc(&delayed_refs->procs_running_refs);
2650 spin_lock(&delayed_refs->lock);
2652 #ifdef SCRAMBLE_DELAYED_REFS
2653 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2657 if (!(run_all || run_most) &&
2658 !btrfs_should_throttle_delayed_refs(trans, root))
2662 * go find something we can process in the rbtree. We start at
2663 * the beginning of the tree, and then build a cluster
2664 * of refs to process starting at the first one we are able to
2667 delayed_start = delayed_refs->run_delayed_start;
2668 ret = btrfs_find_ref_cluster(trans, &cluster,
2669 delayed_refs->run_delayed_start);
2673 ret = run_clustered_refs(trans, root, &cluster);
2675 btrfs_release_ref_cluster(&cluster);
2676 spin_unlock(&delayed_refs->lock);
2677 btrfs_abort_transaction(trans, root, ret);
2678 atomic_dec(&delayed_refs->procs_running_refs);
2679 wake_up(&delayed_refs->wait);
2683 atomic_add(ret, &delayed_refs->ref_seq);
2685 count -= min_t(unsigned long, ret, count);
2690 if (delayed_start >= delayed_refs->run_delayed_start) {
2693 * btrfs_find_ref_cluster looped. let's do one
2694 * more cycle. if we don't run any delayed ref
2695 * during that cycle (because we can't because
2696 * all of them are blocked), bail out.
2701 * no runnable refs left, stop trying
2708 /* refs were run, let's reset staleness detection */
2714 if (!list_empty(&trans->new_bgs)) {
2715 spin_unlock(&delayed_refs->lock);
2716 btrfs_create_pending_block_groups(trans, root);
2717 spin_lock(&delayed_refs->lock);
2720 node = rb_first(&delayed_refs->root);
2723 count = (unsigned long)-1;
2726 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2728 if (btrfs_delayed_ref_is_head(ref)) {
2729 struct btrfs_delayed_ref_head *head;
2731 head = btrfs_delayed_node_to_head(ref);
2732 atomic_inc(&ref->refs);
2734 spin_unlock(&delayed_refs->lock);
2736 * Mutex was contended, block until it's
2737 * released and try again
2739 mutex_lock(&head->mutex);
2740 mutex_unlock(&head->mutex);
2742 btrfs_put_delayed_ref(ref);
2746 node = rb_next(node);
2748 spin_unlock(&delayed_refs->lock);
2749 schedule_timeout(1);
2753 atomic_dec(&delayed_refs->procs_running_refs);
2755 if (waitqueue_active(&delayed_refs->wait))
2756 wake_up(&delayed_refs->wait);
2758 spin_unlock(&delayed_refs->lock);
2759 assert_qgroups_uptodate(trans);
2763 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 u64 bytenr, u64 num_bytes, u64 flags,
2766 int level, int is_data)
2768 struct btrfs_delayed_extent_op *extent_op;
2771 extent_op = btrfs_alloc_delayed_extent_op();
2775 extent_op->flags_to_set = flags;
2776 extent_op->update_flags = 1;
2777 extent_op->update_key = 0;
2778 extent_op->is_data = is_data ? 1 : 0;
2779 extent_op->level = level;
2781 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2782 num_bytes, extent_op);
2784 btrfs_free_delayed_extent_op(extent_op);
2788 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2789 struct btrfs_root *root,
2790 struct btrfs_path *path,
2791 u64 objectid, u64 offset, u64 bytenr)
2793 struct btrfs_delayed_ref_head *head;
2794 struct btrfs_delayed_ref_node *ref;
2795 struct btrfs_delayed_data_ref *data_ref;
2796 struct btrfs_delayed_ref_root *delayed_refs;
2797 struct rb_node *node;
2801 delayed_refs = &trans->transaction->delayed_refs;
2802 spin_lock(&delayed_refs->lock);
2803 head = btrfs_find_delayed_ref_head(trans, bytenr);
2807 if (!mutex_trylock(&head->mutex)) {
2808 atomic_inc(&head->node.refs);
2809 spin_unlock(&delayed_refs->lock);
2811 btrfs_release_path(path);
2814 * Mutex was contended, block until it's released and let
2817 mutex_lock(&head->mutex);
2818 mutex_unlock(&head->mutex);
2819 btrfs_put_delayed_ref(&head->node);
2823 node = rb_prev(&head->node.rb_node);
2827 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2829 if (ref->bytenr != bytenr)
2833 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2836 data_ref = btrfs_delayed_node_to_data_ref(ref);
2838 node = rb_prev(node);
2842 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2843 if (ref->bytenr == bytenr && ref->seq == seq)
2847 if (data_ref->root != root->root_key.objectid ||
2848 data_ref->objectid != objectid || data_ref->offset != offset)
2853 mutex_unlock(&head->mutex);
2855 spin_unlock(&delayed_refs->lock);
2859 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2860 struct btrfs_root *root,
2861 struct btrfs_path *path,
2862 u64 objectid, u64 offset, u64 bytenr)
2864 struct btrfs_root *extent_root = root->fs_info->extent_root;
2865 struct extent_buffer *leaf;
2866 struct btrfs_extent_data_ref *ref;
2867 struct btrfs_extent_inline_ref *iref;
2868 struct btrfs_extent_item *ei;
2869 struct btrfs_key key;
2873 key.objectid = bytenr;
2874 key.offset = (u64)-1;
2875 key.type = BTRFS_EXTENT_ITEM_KEY;
2877 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2880 BUG_ON(ret == 0); /* Corruption */
2883 if (path->slots[0] == 0)
2887 leaf = path->nodes[0];
2888 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2890 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2894 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2895 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2896 if (item_size < sizeof(*ei)) {
2897 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2901 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2903 if (item_size != sizeof(*ei) +
2904 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2907 if (btrfs_extent_generation(leaf, ei) <=
2908 btrfs_root_last_snapshot(&root->root_item))
2911 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2912 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2913 BTRFS_EXTENT_DATA_REF_KEY)
2916 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2917 if (btrfs_extent_refs(leaf, ei) !=
2918 btrfs_extent_data_ref_count(leaf, ref) ||
2919 btrfs_extent_data_ref_root(leaf, ref) !=
2920 root->root_key.objectid ||
2921 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2922 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2930 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2931 struct btrfs_root *root,
2932 u64 objectid, u64 offset, u64 bytenr)
2934 struct btrfs_path *path;
2938 path = btrfs_alloc_path();
2943 ret = check_committed_ref(trans, root, path, objectid,
2945 if (ret && ret != -ENOENT)
2948 ret2 = check_delayed_ref(trans, root, path, objectid,
2950 } while (ret2 == -EAGAIN);
2952 if (ret2 && ret2 != -ENOENT) {
2957 if (ret != -ENOENT || ret2 != -ENOENT)
2960 btrfs_free_path(path);
2961 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2966 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2967 struct btrfs_root *root,
2968 struct extent_buffer *buf,
2969 int full_backref, int inc, int for_cow)
2976 struct btrfs_key key;
2977 struct btrfs_file_extent_item *fi;
2981 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2982 u64, u64, u64, u64, u64, u64, int);
2984 ref_root = btrfs_header_owner(buf);
2985 nritems = btrfs_header_nritems(buf);
2986 level = btrfs_header_level(buf);
2988 if (!root->ref_cows && level == 0)
2992 process_func = btrfs_inc_extent_ref;
2994 process_func = btrfs_free_extent;
2997 parent = buf->start;
3001 for (i = 0; i < nritems; i++) {
3003 btrfs_item_key_to_cpu(buf, &key, i);
3004 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3006 fi = btrfs_item_ptr(buf, i,
3007 struct btrfs_file_extent_item);
3008 if (btrfs_file_extent_type(buf, fi) ==
3009 BTRFS_FILE_EXTENT_INLINE)
3011 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3015 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3016 key.offset -= btrfs_file_extent_offset(buf, fi);
3017 ret = process_func(trans, root, bytenr, num_bytes,
3018 parent, ref_root, key.objectid,
3019 key.offset, for_cow);
3023 bytenr = btrfs_node_blockptr(buf, i);
3024 num_bytes = btrfs_level_size(root, level - 1);
3025 ret = process_func(trans, root, bytenr, num_bytes,
3026 parent, ref_root, level - 1, 0,
3037 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3038 struct extent_buffer *buf, int full_backref, int for_cow)
3040 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3043 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3044 struct extent_buffer *buf, int full_backref, int for_cow)
3046 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3049 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3050 struct btrfs_root *root,
3051 struct btrfs_path *path,
3052 struct btrfs_block_group_cache *cache)
3055 struct btrfs_root *extent_root = root->fs_info->extent_root;
3057 struct extent_buffer *leaf;
3059 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3062 BUG_ON(ret); /* Corruption */
3064 leaf = path->nodes[0];
3065 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3066 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3067 btrfs_mark_buffer_dirty(leaf);
3068 btrfs_release_path(path);
3071 btrfs_abort_transaction(trans, root, ret);
3078 static struct btrfs_block_group_cache *
3079 next_block_group(struct btrfs_root *root,
3080 struct btrfs_block_group_cache *cache)
3082 struct rb_node *node;
3083 spin_lock(&root->fs_info->block_group_cache_lock);
3084 node = rb_next(&cache->cache_node);
3085 btrfs_put_block_group(cache);
3087 cache = rb_entry(node, struct btrfs_block_group_cache,
3089 btrfs_get_block_group(cache);
3092 spin_unlock(&root->fs_info->block_group_cache_lock);
3096 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3097 struct btrfs_trans_handle *trans,
3098 struct btrfs_path *path)
3100 struct btrfs_root *root = block_group->fs_info->tree_root;
3101 struct inode *inode = NULL;
3103 int dcs = BTRFS_DC_ERROR;
3109 * If this block group is smaller than 100 megs don't bother caching the
3112 if (block_group->key.offset < (100 * 1024 * 1024)) {
3113 spin_lock(&block_group->lock);
3114 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3115 spin_unlock(&block_group->lock);
3120 inode = lookup_free_space_inode(root, block_group, path);
3121 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3122 ret = PTR_ERR(inode);
3123 btrfs_release_path(path);
3127 if (IS_ERR(inode)) {
3131 if (block_group->ro)
3134 ret = create_free_space_inode(root, trans, block_group, path);
3140 /* We've already setup this transaction, go ahead and exit */
3141 if (block_group->cache_generation == trans->transid &&
3142 i_size_read(inode)) {
3143 dcs = BTRFS_DC_SETUP;
3148 * We want to set the generation to 0, that way if anything goes wrong
3149 * from here on out we know not to trust this cache when we load up next
3152 BTRFS_I(inode)->generation = 0;
3153 ret = btrfs_update_inode(trans, root, inode);
3156 if (i_size_read(inode) > 0) {
3157 ret = btrfs_check_trunc_cache_free_space(root,
3158 &root->fs_info->global_block_rsv);
3162 ret = btrfs_truncate_free_space_cache(root, trans, path,
3168 spin_lock(&block_group->lock);
3169 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3170 !btrfs_test_opt(root, SPACE_CACHE)) {
3172 * don't bother trying to write stuff out _if_
3173 * a) we're not cached,
3174 * b) we're with nospace_cache mount option.
3176 dcs = BTRFS_DC_WRITTEN;
3177 spin_unlock(&block_group->lock);
3180 spin_unlock(&block_group->lock);
3183 * Try to preallocate enough space based on how big the block group is.
3184 * Keep in mind this has to include any pinned space which could end up
3185 * taking up quite a bit since it's not folded into the other space
3188 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3193 num_pages *= PAGE_CACHE_SIZE;
3195 ret = btrfs_check_data_free_space(inode, num_pages);
3199 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3200 num_pages, num_pages,
3203 dcs = BTRFS_DC_SETUP;
3204 btrfs_free_reserved_data_space(inode, num_pages);
3209 btrfs_release_path(path);
3211 spin_lock(&block_group->lock);
3212 if (!ret && dcs == BTRFS_DC_SETUP)
3213 block_group->cache_generation = trans->transid;
3214 block_group->disk_cache_state = dcs;
3215 spin_unlock(&block_group->lock);
3220 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3221 struct btrfs_root *root)
3223 struct btrfs_block_group_cache *cache;
3225 struct btrfs_path *path;
3228 path = btrfs_alloc_path();
3234 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3236 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3238 cache = next_block_group(root, cache);
3246 err = cache_save_setup(cache, trans, path);
3247 last = cache->key.objectid + cache->key.offset;
3248 btrfs_put_block_group(cache);
3253 err = btrfs_run_delayed_refs(trans, root,
3255 if (err) /* File system offline */
3259 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3261 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3262 btrfs_put_block_group(cache);
3268 cache = next_block_group(root, cache);
3277 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3278 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3280 last = cache->key.objectid + cache->key.offset;
3282 err = write_one_cache_group(trans, root, path, cache);
3283 if (err) /* File system offline */
3286 btrfs_put_block_group(cache);
3291 * I don't think this is needed since we're just marking our
3292 * preallocated extent as written, but just in case it can't
3296 err = btrfs_run_delayed_refs(trans, root,
3298 if (err) /* File system offline */
3302 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3305 * Really this shouldn't happen, but it could if we
3306 * couldn't write the entire preallocated extent and
3307 * splitting the extent resulted in a new block.
3310 btrfs_put_block_group(cache);
3313 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3315 cache = next_block_group(root, cache);
3324 err = btrfs_write_out_cache(root, trans, cache, path);
3327 * If we didn't have an error then the cache state is still
3328 * NEED_WRITE, so we can set it to WRITTEN.
3330 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3331 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3332 last = cache->key.objectid + cache->key.offset;
3333 btrfs_put_block_group(cache);
3337 btrfs_free_path(path);
3341 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3343 struct btrfs_block_group_cache *block_group;
3346 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3347 if (!block_group || block_group->ro)
3350 btrfs_put_block_group(block_group);
3354 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3355 u64 total_bytes, u64 bytes_used,
3356 struct btrfs_space_info **space_info)
3358 struct btrfs_space_info *found;
3363 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3364 BTRFS_BLOCK_GROUP_RAID10))
3369 found = __find_space_info(info, flags);
3371 spin_lock(&found->lock);
3372 found->total_bytes += total_bytes;
3373 found->disk_total += total_bytes * factor;
3374 found->bytes_used += bytes_used;
3375 found->disk_used += bytes_used * factor;
3377 spin_unlock(&found->lock);
3378 *space_info = found;
3381 found = kzalloc(sizeof(*found), GFP_NOFS);
3385 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3391 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3392 INIT_LIST_HEAD(&found->block_groups[i]);
3393 init_rwsem(&found->groups_sem);
3394 spin_lock_init(&found->lock);
3395 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3396 found->total_bytes = total_bytes;
3397 found->disk_total = total_bytes * factor;
3398 found->bytes_used = bytes_used;
3399 found->disk_used = bytes_used * factor;
3400 found->bytes_pinned = 0;
3401 found->bytes_reserved = 0;
3402 found->bytes_readonly = 0;
3403 found->bytes_may_use = 0;
3405 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3406 found->chunk_alloc = 0;
3408 init_waitqueue_head(&found->wait);
3409 *space_info = found;
3410 list_add_rcu(&found->list, &info->space_info);
3411 if (flags & BTRFS_BLOCK_GROUP_DATA)
3412 info->data_sinfo = found;
3416 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3418 u64 extra_flags = chunk_to_extended(flags) &
3419 BTRFS_EXTENDED_PROFILE_MASK;
3421 write_seqlock(&fs_info->profiles_lock);
3422 if (flags & BTRFS_BLOCK_GROUP_DATA)
3423 fs_info->avail_data_alloc_bits |= extra_flags;
3424 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3425 fs_info->avail_metadata_alloc_bits |= extra_flags;
3426 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3427 fs_info->avail_system_alloc_bits |= extra_flags;
3428 write_sequnlock(&fs_info->profiles_lock);
3432 * returns target flags in extended format or 0 if restripe for this
3433 * chunk_type is not in progress
3435 * should be called with either volume_mutex or balance_lock held
3437 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3439 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3445 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3446 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3447 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3448 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3449 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3450 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3451 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3452 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3453 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3460 * @flags: available profiles in extended format (see ctree.h)
3462 * Returns reduced profile in chunk format. If profile changing is in
3463 * progress (either running or paused) picks the target profile (if it's
3464 * already available), otherwise falls back to plain reducing.
3466 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3469 * we add in the count of missing devices because we want
3470 * to make sure that any RAID levels on a degraded FS
3471 * continue to be honored.
3473 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3474 root->fs_info->fs_devices->missing_devices;
3479 * see if restripe for this chunk_type is in progress, if so
3480 * try to reduce to the target profile
3482 spin_lock(&root->fs_info->balance_lock);
3483 target = get_restripe_target(root->fs_info, flags);
3485 /* pick target profile only if it's already available */
3486 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3487 spin_unlock(&root->fs_info->balance_lock);
3488 return extended_to_chunk(target);
3491 spin_unlock(&root->fs_info->balance_lock);
3493 /* First, mask out the RAID levels which aren't possible */
3494 if (num_devices == 1)
3495 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3496 BTRFS_BLOCK_GROUP_RAID5);
3497 if (num_devices < 3)
3498 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3499 if (num_devices < 4)
3500 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3502 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3503 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3504 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3507 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3508 tmp = BTRFS_BLOCK_GROUP_RAID6;
3509 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3510 tmp = BTRFS_BLOCK_GROUP_RAID5;
3511 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3512 tmp = BTRFS_BLOCK_GROUP_RAID10;
3513 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3514 tmp = BTRFS_BLOCK_GROUP_RAID1;
3515 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3516 tmp = BTRFS_BLOCK_GROUP_RAID0;
3518 return extended_to_chunk(flags | tmp);
3521 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3526 seq = read_seqbegin(&root->fs_info->profiles_lock);
3528 if (flags & BTRFS_BLOCK_GROUP_DATA)
3529 flags |= root->fs_info->avail_data_alloc_bits;
3530 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3531 flags |= root->fs_info->avail_system_alloc_bits;
3532 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3533 flags |= root->fs_info->avail_metadata_alloc_bits;
3534 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3536 return btrfs_reduce_alloc_profile(root, flags);
3539 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3545 flags = BTRFS_BLOCK_GROUP_DATA;
3546 else if (root == root->fs_info->chunk_root)
3547 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3549 flags = BTRFS_BLOCK_GROUP_METADATA;
3551 ret = get_alloc_profile(root, flags);
3556 * This will check the space that the inode allocates from to make sure we have
3557 * enough space for bytes.
3559 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3561 struct btrfs_space_info *data_sinfo;
3562 struct btrfs_root *root = BTRFS_I(inode)->root;
3563 struct btrfs_fs_info *fs_info = root->fs_info;
3565 int ret = 0, committed = 0, alloc_chunk = 1;
3567 /* make sure bytes are sectorsize aligned */
3568 bytes = ALIGN(bytes, root->sectorsize);
3570 if (root == root->fs_info->tree_root ||
3571 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3576 data_sinfo = fs_info->data_sinfo;
3581 /* make sure we have enough space to handle the data first */
3582 spin_lock(&data_sinfo->lock);
3583 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3584 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3585 data_sinfo->bytes_may_use;
3587 if (used + bytes > data_sinfo->total_bytes) {
3588 struct btrfs_trans_handle *trans;
3591 * if we don't have enough free bytes in this space then we need
3592 * to alloc a new chunk.
3594 if (!data_sinfo->full && alloc_chunk) {
3597 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3598 spin_unlock(&data_sinfo->lock);
3600 alloc_target = btrfs_get_alloc_profile(root, 1);
3601 trans = btrfs_join_transaction(root);
3603 return PTR_ERR(trans);
3605 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3607 CHUNK_ALLOC_NO_FORCE);
3608 btrfs_end_transaction(trans, root);
3617 data_sinfo = fs_info->data_sinfo;
3623 * If we don't have enough pinned space to deal with this
3624 * allocation don't bother committing the transaction.
3626 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3629 spin_unlock(&data_sinfo->lock);
3631 /* commit the current transaction and try again */
3634 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3637 trans = btrfs_join_transaction(root);
3639 return PTR_ERR(trans);
3640 ret = btrfs_commit_transaction(trans, root);
3648 data_sinfo->bytes_may_use += bytes;
3649 trace_btrfs_space_reservation(root->fs_info, "space_info",
3650 data_sinfo->flags, bytes, 1);
3651 spin_unlock(&data_sinfo->lock);
3657 * Called if we need to clear a data reservation for this inode.
3659 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3661 struct btrfs_root *root = BTRFS_I(inode)->root;
3662 struct btrfs_space_info *data_sinfo;
3664 /* make sure bytes are sectorsize aligned */
3665 bytes = ALIGN(bytes, root->sectorsize);
3667 data_sinfo = root->fs_info->data_sinfo;
3668 spin_lock(&data_sinfo->lock);
3669 data_sinfo->bytes_may_use -= bytes;
3670 trace_btrfs_space_reservation(root->fs_info, "space_info",
3671 data_sinfo->flags, bytes, 0);
3672 spin_unlock(&data_sinfo->lock);
3675 static void force_metadata_allocation(struct btrfs_fs_info *info)
3677 struct list_head *head = &info->space_info;
3678 struct btrfs_space_info *found;
3681 list_for_each_entry_rcu(found, head, list) {
3682 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3683 found->force_alloc = CHUNK_ALLOC_FORCE;
3688 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3690 return (global->size << 1);
3693 static int should_alloc_chunk(struct btrfs_root *root,
3694 struct btrfs_space_info *sinfo, int force)
3696 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3697 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3698 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3701 if (force == CHUNK_ALLOC_FORCE)
3705 * We need to take into account the global rsv because for all intents
3706 * and purposes it's used space. Don't worry about locking the
3707 * global_rsv, it doesn't change except when the transaction commits.
3709 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3710 num_allocated += calc_global_rsv_need_space(global_rsv);
3713 * in limited mode, we want to have some free space up to
3714 * about 1% of the FS size.
3716 if (force == CHUNK_ALLOC_LIMITED) {
3717 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3718 thresh = max_t(u64, 64 * 1024 * 1024,
3719 div_factor_fine(thresh, 1));
3721 if (num_bytes - num_allocated < thresh)
3725 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3730 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3734 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3735 BTRFS_BLOCK_GROUP_RAID0 |
3736 BTRFS_BLOCK_GROUP_RAID5 |
3737 BTRFS_BLOCK_GROUP_RAID6))
3738 num_dev = root->fs_info->fs_devices->rw_devices;
3739 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3742 num_dev = 1; /* DUP or single */
3744 /* metadata for updaing devices and chunk tree */
3745 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3748 static void check_system_chunk(struct btrfs_trans_handle *trans,
3749 struct btrfs_root *root, u64 type)
3751 struct btrfs_space_info *info;
3755 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3756 spin_lock(&info->lock);
3757 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3758 info->bytes_reserved - info->bytes_readonly;
3759 spin_unlock(&info->lock);
3761 thresh = get_system_chunk_thresh(root, type);
3762 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3763 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3764 left, thresh, type);
3765 dump_space_info(info, 0, 0);
3768 if (left < thresh) {
3771 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3772 btrfs_alloc_chunk(trans, root, flags);
3776 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3777 struct btrfs_root *extent_root, u64 flags, int force)
3779 struct btrfs_space_info *space_info;
3780 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3781 int wait_for_alloc = 0;
3784 /* Don't re-enter if we're already allocating a chunk */
3785 if (trans->allocating_chunk)
3788 space_info = __find_space_info(extent_root->fs_info, flags);
3790 ret = update_space_info(extent_root->fs_info, flags,
3792 BUG_ON(ret); /* -ENOMEM */
3794 BUG_ON(!space_info); /* Logic error */
3797 spin_lock(&space_info->lock);
3798 if (force < space_info->force_alloc)
3799 force = space_info->force_alloc;
3800 if (space_info->full) {
3801 spin_unlock(&space_info->lock);
3805 if (!should_alloc_chunk(extent_root, space_info, force)) {
3806 spin_unlock(&space_info->lock);
3808 } else if (space_info->chunk_alloc) {
3811 space_info->chunk_alloc = 1;
3814 spin_unlock(&space_info->lock);
3816 mutex_lock(&fs_info->chunk_mutex);
3819 * The chunk_mutex is held throughout the entirety of a chunk
3820 * allocation, so once we've acquired the chunk_mutex we know that the
3821 * other guy is done and we need to recheck and see if we should
3824 if (wait_for_alloc) {
3825 mutex_unlock(&fs_info->chunk_mutex);
3830 trans->allocating_chunk = true;
3833 * If we have mixed data/metadata chunks we want to make sure we keep
3834 * allocating mixed chunks instead of individual chunks.
3836 if (btrfs_mixed_space_info(space_info))
3837 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3840 * if we're doing a data chunk, go ahead and make sure that
3841 * we keep a reasonable number of metadata chunks allocated in the
3844 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3845 fs_info->data_chunk_allocations++;
3846 if (!(fs_info->data_chunk_allocations %
3847 fs_info->metadata_ratio))
3848 force_metadata_allocation(fs_info);
3852 * Check if we have enough space in SYSTEM chunk because we may need
3853 * to update devices.
3855 check_system_chunk(trans, extent_root, flags);
3857 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3858 trans->allocating_chunk = false;
3860 spin_lock(&space_info->lock);
3861 if (ret < 0 && ret != -ENOSPC)
3864 space_info->full = 1;
3868 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3870 space_info->chunk_alloc = 0;
3871 spin_unlock(&space_info->lock);
3872 mutex_unlock(&fs_info->chunk_mutex);
3876 static int can_overcommit(struct btrfs_root *root,
3877 struct btrfs_space_info *space_info, u64 bytes,
3878 enum btrfs_reserve_flush_enum flush)
3880 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3881 u64 profile = btrfs_get_alloc_profile(root, 0);
3887 used = space_info->bytes_used + space_info->bytes_reserved +
3888 space_info->bytes_pinned + space_info->bytes_readonly;
3891 * We only want to allow over committing if we have lots of actual space
3892 * free, but if we don't have enough space to handle the global reserve
3893 * space then we could end up having a real enospc problem when trying
3894 * to allocate a chunk or some other such important allocation.
3896 spin_lock(&global_rsv->lock);
3897 space_size = calc_global_rsv_need_space(global_rsv);
3898 spin_unlock(&global_rsv->lock);
3899 if (used + space_size >= space_info->total_bytes)
3902 used += space_info->bytes_may_use;
3904 spin_lock(&root->fs_info->free_chunk_lock);
3905 avail = root->fs_info->free_chunk_space;
3906 spin_unlock(&root->fs_info->free_chunk_lock);
3909 * If we have dup, raid1 or raid10 then only half of the free
3910 * space is actually useable. For raid56, the space info used
3911 * doesn't include the parity drive, so we don't have to
3914 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3915 BTRFS_BLOCK_GROUP_RAID1 |
3916 BTRFS_BLOCK_GROUP_RAID10))
3919 to_add = space_info->total_bytes;
3922 * If we aren't flushing all things, let us overcommit up to
3923 * 1/2th of the space. If we can flush, don't let us overcommit
3924 * too much, let it overcommit up to 1/8 of the space.
3926 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3932 * Limit the overcommit to the amount of free space we could possibly
3933 * allocate for chunks.
3935 to_add = min(avail, to_add);
3937 if (used + bytes < space_info->total_bytes + to_add)
3942 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3943 unsigned long nr_pages)
3945 struct super_block *sb = root->fs_info->sb;
3947 if (down_read_trylock(&sb->s_umount)) {
3948 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3949 up_read(&sb->s_umount);
3952 * We needn't worry the filesystem going from r/w to r/o though
3953 * we don't acquire ->s_umount mutex, because the filesystem
3954 * should guarantee the delalloc inodes list be empty after
3955 * the filesystem is readonly(all dirty pages are written to
3958 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
3959 if (!current->journal_info)
3960 btrfs_wait_all_ordered_extents(root->fs_info, 0);
3965 * shrink metadata reservation for delalloc
3967 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3970 struct btrfs_block_rsv *block_rsv;
3971 struct btrfs_space_info *space_info;
3972 struct btrfs_trans_handle *trans;
3976 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3978 enum btrfs_reserve_flush_enum flush;
3980 trans = (struct btrfs_trans_handle *)current->journal_info;
3981 block_rsv = &root->fs_info->delalloc_block_rsv;
3982 space_info = block_rsv->space_info;
3985 delalloc_bytes = percpu_counter_sum_positive(
3986 &root->fs_info->delalloc_bytes);
3987 if (delalloc_bytes == 0) {
3990 btrfs_wait_all_ordered_extents(root->fs_info, 0);
3994 while (delalloc_bytes && loops < 3) {
3995 max_reclaim = min(delalloc_bytes, to_reclaim);
3996 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3997 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3999 * We need to wait for the async pages to actually start before
4002 wait_event(root->fs_info->async_submit_wait,
4003 !atomic_read(&root->fs_info->async_delalloc_pages));
4006 flush = BTRFS_RESERVE_FLUSH_ALL;
4008 flush = BTRFS_RESERVE_NO_FLUSH;
4009 spin_lock(&space_info->lock);
4010 if (can_overcommit(root, space_info, orig, flush)) {
4011 spin_unlock(&space_info->lock);
4014 spin_unlock(&space_info->lock);
4017 if (wait_ordered && !trans) {
4018 btrfs_wait_all_ordered_extents(root->fs_info, 0);
4020 time_left = schedule_timeout_killable(1);
4025 delalloc_bytes = percpu_counter_sum_positive(
4026 &root->fs_info->delalloc_bytes);
4031 * maybe_commit_transaction - possibly commit the transaction if its ok to
4032 * @root - the root we're allocating for
4033 * @bytes - the number of bytes we want to reserve
4034 * @force - force the commit
4036 * This will check to make sure that committing the transaction will actually
4037 * get us somewhere and then commit the transaction if it does. Otherwise it
4038 * will return -ENOSPC.
4040 static int may_commit_transaction(struct btrfs_root *root,
4041 struct btrfs_space_info *space_info,
4042 u64 bytes, int force)
4044 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4045 struct btrfs_trans_handle *trans;
4047 trans = (struct btrfs_trans_handle *)current->journal_info;
4054 /* See if there is enough pinned space to make this reservation */
4055 spin_lock(&space_info->lock);
4056 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4058 spin_unlock(&space_info->lock);
4061 spin_unlock(&space_info->lock);
4064 * See if there is some space in the delayed insertion reservation for
4067 if (space_info != delayed_rsv->space_info)
4070 spin_lock(&space_info->lock);
4071 spin_lock(&delayed_rsv->lock);
4072 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4073 bytes - delayed_rsv->size) >= 0) {
4074 spin_unlock(&delayed_rsv->lock);
4075 spin_unlock(&space_info->lock);
4078 spin_unlock(&delayed_rsv->lock);
4079 spin_unlock(&space_info->lock);
4082 trans = btrfs_join_transaction(root);
4086 return btrfs_commit_transaction(trans, root);
4090 FLUSH_DELAYED_ITEMS_NR = 1,
4091 FLUSH_DELAYED_ITEMS = 2,
4093 FLUSH_DELALLOC_WAIT = 4,
4098 static int flush_space(struct btrfs_root *root,
4099 struct btrfs_space_info *space_info, u64 num_bytes,
4100 u64 orig_bytes, int state)
4102 struct btrfs_trans_handle *trans;
4107 case FLUSH_DELAYED_ITEMS_NR:
4108 case FLUSH_DELAYED_ITEMS:
4109 if (state == FLUSH_DELAYED_ITEMS_NR) {
4110 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4112 nr = (int)div64_u64(num_bytes, bytes);
4119 trans = btrfs_join_transaction(root);
4120 if (IS_ERR(trans)) {
4121 ret = PTR_ERR(trans);
4124 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4125 btrfs_end_transaction(trans, root);
4127 case FLUSH_DELALLOC:
4128 case FLUSH_DELALLOC_WAIT:
4129 shrink_delalloc(root, num_bytes, orig_bytes,
4130 state == FLUSH_DELALLOC_WAIT);
4133 trans = btrfs_join_transaction(root);
4134 if (IS_ERR(trans)) {
4135 ret = PTR_ERR(trans);
4138 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4139 btrfs_get_alloc_profile(root, 0),
4140 CHUNK_ALLOC_NO_FORCE);
4141 btrfs_end_transaction(trans, root);
4146 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4156 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4157 * @root - the root we're allocating for
4158 * @block_rsv - the block_rsv we're allocating for
4159 * @orig_bytes - the number of bytes we want
4160 * @flush - whether or not we can flush to make our reservation
4162 * This will reserve orgi_bytes number of bytes from the space info associated
4163 * with the block_rsv. If there is not enough space it will make an attempt to
4164 * flush out space to make room. It will do this by flushing delalloc if
4165 * possible or committing the transaction. If flush is 0 then no attempts to
4166 * regain reservations will be made and this will fail if there is not enough
4169 static int reserve_metadata_bytes(struct btrfs_root *root,
4170 struct btrfs_block_rsv *block_rsv,
4172 enum btrfs_reserve_flush_enum flush)
4174 struct btrfs_space_info *space_info = block_rsv->space_info;
4176 u64 num_bytes = orig_bytes;
4177 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4179 bool flushing = false;
4183 spin_lock(&space_info->lock);
4185 * We only want to wait if somebody other than us is flushing and we
4186 * are actually allowed to flush all things.
4188 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4189 space_info->flush) {
4190 spin_unlock(&space_info->lock);
4192 * If we have a trans handle we can't wait because the flusher
4193 * may have to commit the transaction, which would mean we would
4194 * deadlock since we are waiting for the flusher to finish, but
4195 * hold the current transaction open.
4197 if (current->journal_info)
4199 ret = wait_event_killable(space_info->wait, !space_info->flush);
4200 /* Must have been killed, return */
4204 spin_lock(&space_info->lock);
4208 used = space_info->bytes_used + space_info->bytes_reserved +
4209 space_info->bytes_pinned + space_info->bytes_readonly +
4210 space_info->bytes_may_use;
4213 * The idea here is that we've not already over-reserved the block group
4214 * then we can go ahead and save our reservation first and then start
4215 * flushing if we need to. Otherwise if we've already overcommitted
4216 * lets start flushing stuff first and then come back and try to make
4219 if (used <= space_info->total_bytes) {
4220 if (used + orig_bytes <= space_info->total_bytes) {
4221 space_info->bytes_may_use += orig_bytes;
4222 trace_btrfs_space_reservation(root->fs_info,
4223 "space_info", space_info->flags, orig_bytes, 1);
4227 * Ok set num_bytes to orig_bytes since we aren't
4228 * overocmmitted, this way we only try and reclaim what
4231 num_bytes = orig_bytes;
4235 * Ok we're over committed, set num_bytes to the overcommitted
4236 * amount plus the amount of bytes that we need for this
4239 num_bytes = used - space_info->total_bytes +
4243 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4244 space_info->bytes_may_use += orig_bytes;
4245 trace_btrfs_space_reservation(root->fs_info, "space_info",
4246 space_info->flags, orig_bytes,
4252 * Couldn't make our reservation, save our place so while we're trying
4253 * to reclaim space we can actually use it instead of somebody else
4254 * stealing it from us.
4256 * We make the other tasks wait for the flush only when we can flush
4259 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4261 space_info->flush = 1;
4264 spin_unlock(&space_info->lock);
4266 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4269 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4274 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4275 * would happen. So skip delalloc flush.
4277 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4278 (flush_state == FLUSH_DELALLOC ||
4279 flush_state == FLUSH_DELALLOC_WAIT))
4280 flush_state = ALLOC_CHUNK;
4284 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4285 flush_state < COMMIT_TRANS)
4287 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4288 flush_state <= COMMIT_TRANS)
4292 if (ret == -ENOSPC &&
4293 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4294 struct btrfs_block_rsv *global_rsv =
4295 &root->fs_info->global_block_rsv;
4297 if (block_rsv != global_rsv &&
4298 !block_rsv_use_bytes(global_rsv, orig_bytes))
4302 spin_lock(&space_info->lock);
4303 space_info->flush = 0;
4304 wake_up_all(&space_info->wait);
4305 spin_unlock(&space_info->lock);
4310 static struct btrfs_block_rsv *get_block_rsv(
4311 const struct btrfs_trans_handle *trans,
4312 const struct btrfs_root *root)
4314 struct btrfs_block_rsv *block_rsv = NULL;
4317 block_rsv = trans->block_rsv;
4319 if (root == root->fs_info->csum_root && trans->adding_csums)
4320 block_rsv = trans->block_rsv;
4323 block_rsv = root->block_rsv;
4326 block_rsv = &root->fs_info->empty_block_rsv;
4331 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4335 spin_lock(&block_rsv->lock);
4336 if (block_rsv->reserved >= num_bytes) {
4337 block_rsv->reserved -= num_bytes;
4338 if (block_rsv->reserved < block_rsv->size)
4339 block_rsv->full = 0;
4342 spin_unlock(&block_rsv->lock);
4346 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4347 u64 num_bytes, int update_size)
4349 spin_lock(&block_rsv->lock);
4350 block_rsv->reserved += num_bytes;
4352 block_rsv->size += num_bytes;
4353 else if (block_rsv->reserved >= block_rsv->size)
4354 block_rsv->full = 1;
4355 spin_unlock(&block_rsv->lock);
4358 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4359 struct btrfs_block_rsv *dest, u64 num_bytes,
4362 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4365 if (global_rsv->space_info != dest->space_info)
4368 spin_lock(&global_rsv->lock);
4369 min_bytes = div_factor(global_rsv->size, min_factor);
4370 if (global_rsv->reserved < min_bytes + num_bytes) {
4371 spin_unlock(&global_rsv->lock);
4374 global_rsv->reserved -= num_bytes;
4375 if (global_rsv->reserved < global_rsv->size)
4376 global_rsv->full = 0;
4377 spin_unlock(&global_rsv->lock);
4379 block_rsv_add_bytes(dest, num_bytes, 1);
4383 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4384 struct btrfs_block_rsv *block_rsv,
4385 struct btrfs_block_rsv *dest, u64 num_bytes)
4387 struct btrfs_space_info *space_info = block_rsv->space_info;
4389 spin_lock(&block_rsv->lock);
4390 if (num_bytes == (u64)-1)
4391 num_bytes = block_rsv->size;
4392 block_rsv->size -= num_bytes;
4393 if (block_rsv->reserved >= block_rsv->size) {
4394 num_bytes = block_rsv->reserved - block_rsv->size;
4395 block_rsv->reserved = block_rsv->size;
4396 block_rsv->full = 1;
4400 spin_unlock(&block_rsv->lock);
4402 if (num_bytes > 0) {
4404 spin_lock(&dest->lock);
4408 bytes_to_add = dest->size - dest->reserved;
4409 bytes_to_add = min(num_bytes, bytes_to_add);
4410 dest->reserved += bytes_to_add;
4411 if (dest->reserved >= dest->size)
4413 num_bytes -= bytes_to_add;
4415 spin_unlock(&dest->lock);
4418 spin_lock(&space_info->lock);
4419 space_info->bytes_may_use -= num_bytes;
4420 trace_btrfs_space_reservation(fs_info, "space_info",
4421 space_info->flags, num_bytes, 0);
4422 space_info->reservation_progress++;
4423 spin_unlock(&space_info->lock);
4428 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4429 struct btrfs_block_rsv *dst, u64 num_bytes)
4433 ret = block_rsv_use_bytes(src, num_bytes);
4437 block_rsv_add_bytes(dst, num_bytes, 1);
4441 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4443 memset(rsv, 0, sizeof(*rsv));
4444 spin_lock_init(&rsv->lock);
4448 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4449 unsigned short type)
4451 struct btrfs_block_rsv *block_rsv;
4452 struct btrfs_fs_info *fs_info = root->fs_info;
4454 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4458 btrfs_init_block_rsv(block_rsv, type);
4459 block_rsv->space_info = __find_space_info(fs_info,
4460 BTRFS_BLOCK_GROUP_METADATA);
4464 void btrfs_free_block_rsv(struct btrfs_root *root,
4465 struct btrfs_block_rsv *rsv)
4469 btrfs_block_rsv_release(root, rsv, (u64)-1);
4473 int btrfs_block_rsv_add(struct btrfs_root *root,
4474 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4475 enum btrfs_reserve_flush_enum flush)
4482 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4484 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4491 int btrfs_block_rsv_check(struct btrfs_root *root,
4492 struct btrfs_block_rsv *block_rsv, int min_factor)
4500 spin_lock(&block_rsv->lock);
4501 num_bytes = div_factor(block_rsv->size, min_factor);
4502 if (block_rsv->reserved >= num_bytes)
4504 spin_unlock(&block_rsv->lock);
4509 int btrfs_block_rsv_refill(struct btrfs_root *root,
4510 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4511 enum btrfs_reserve_flush_enum flush)
4519 spin_lock(&block_rsv->lock);
4520 num_bytes = min_reserved;
4521 if (block_rsv->reserved >= num_bytes)
4524 num_bytes -= block_rsv->reserved;
4525 spin_unlock(&block_rsv->lock);
4530 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4532 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4539 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4540 struct btrfs_block_rsv *dst_rsv,
4543 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4546 void btrfs_block_rsv_release(struct btrfs_root *root,
4547 struct btrfs_block_rsv *block_rsv,
4550 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4551 if (global_rsv->full || global_rsv == block_rsv ||
4552 block_rsv->space_info != global_rsv->space_info)
4554 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4559 * helper to calculate size of global block reservation.
4560 * the desired value is sum of space used by extent tree,
4561 * checksum tree and root tree
4563 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4565 struct btrfs_space_info *sinfo;
4569 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4571 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4572 spin_lock(&sinfo->lock);
4573 data_used = sinfo->bytes_used;
4574 spin_unlock(&sinfo->lock);
4576 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4577 spin_lock(&sinfo->lock);
4578 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4580 meta_used = sinfo->bytes_used;
4581 spin_unlock(&sinfo->lock);
4583 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4585 num_bytes += div64_u64(data_used + meta_used, 50);
4587 if (num_bytes * 3 > meta_used)
4588 num_bytes = div64_u64(meta_used, 3);
4590 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4593 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4595 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4596 struct btrfs_space_info *sinfo = block_rsv->space_info;
4599 num_bytes = calc_global_metadata_size(fs_info);
4601 spin_lock(&sinfo->lock);
4602 spin_lock(&block_rsv->lock);
4604 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4606 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4607 sinfo->bytes_reserved + sinfo->bytes_readonly +
4608 sinfo->bytes_may_use;
4610 if (sinfo->total_bytes > num_bytes) {
4611 num_bytes = sinfo->total_bytes - num_bytes;
4612 block_rsv->reserved += num_bytes;
4613 sinfo->bytes_may_use += num_bytes;
4614 trace_btrfs_space_reservation(fs_info, "space_info",
4615 sinfo->flags, num_bytes, 1);
4618 if (block_rsv->reserved >= block_rsv->size) {
4619 num_bytes = block_rsv->reserved - block_rsv->size;
4620 sinfo->bytes_may_use -= num_bytes;
4621 trace_btrfs_space_reservation(fs_info, "space_info",
4622 sinfo->flags, num_bytes, 0);
4623 sinfo->reservation_progress++;
4624 block_rsv->reserved = block_rsv->size;
4625 block_rsv->full = 1;
4628 spin_unlock(&block_rsv->lock);
4629 spin_unlock(&sinfo->lock);
4632 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4634 struct btrfs_space_info *space_info;
4636 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4637 fs_info->chunk_block_rsv.space_info = space_info;
4639 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4640 fs_info->global_block_rsv.space_info = space_info;
4641 fs_info->delalloc_block_rsv.space_info = space_info;
4642 fs_info->trans_block_rsv.space_info = space_info;
4643 fs_info->empty_block_rsv.space_info = space_info;
4644 fs_info->delayed_block_rsv.space_info = space_info;
4646 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4647 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4648 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4649 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4650 if (fs_info->quota_root)
4651 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4652 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4654 update_global_block_rsv(fs_info);
4657 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4659 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4661 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4662 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4663 WARN_ON(fs_info->trans_block_rsv.size > 0);
4664 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4665 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4666 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4667 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4668 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4671 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4672 struct btrfs_root *root)
4674 if (!trans->block_rsv)
4677 if (!trans->bytes_reserved)
4680 trace_btrfs_space_reservation(root->fs_info, "transaction",
4681 trans->transid, trans->bytes_reserved, 0);
4682 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4683 trans->bytes_reserved = 0;
4686 /* Can only return 0 or -ENOSPC */
4687 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4688 struct inode *inode)
4690 struct btrfs_root *root = BTRFS_I(inode)->root;
4691 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4692 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4695 * We need to hold space in order to delete our orphan item once we've
4696 * added it, so this takes the reservation so we can release it later
4697 * when we are truly done with the orphan item.
4699 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4700 trace_btrfs_space_reservation(root->fs_info, "orphan",
4701 btrfs_ino(inode), num_bytes, 1);
4702 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4705 void btrfs_orphan_release_metadata(struct inode *inode)
4707 struct btrfs_root *root = BTRFS_I(inode)->root;
4708 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4709 trace_btrfs_space_reservation(root->fs_info, "orphan",
4710 btrfs_ino(inode), num_bytes, 0);
4711 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4715 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4716 * root: the root of the parent directory
4717 * rsv: block reservation
4718 * items: the number of items that we need do reservation
4719 * qgroup_reserved: used to return the reserved size in qgroup
4721 * This function is used to reserve the space for snapshot/subvolume
4722 * creation and deletion. Those operations are different with the
4723 * common file/directory operations, they change two fs/file trees
4724 * and root tree, the number of items that the qgroup reserves is
4725 * different with the free space reservation. So we can not use
4726 * the space reseravtion mechanism in start_transaction().
4728 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4729 struct btrfs_block_rsv *rsv,
4731 u64 *qgroup_reserved)
4736 if (root->fs_info->quota_enabled) {
4737 /* One for parent inode, two for dir entries */
4738 num_bytes = 3 * root->leafsize;
4739 ret = btrfs_qgroup_reserve(root, num_bytes);
4746 *qgroup_reserved = num_bytes;
4748 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4749 rsv->space_info = __find_space_info(root->fs_info,
4750 BTRFS_BLOCK_GROUP_METADATA);
4751 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4752 BTRFS_RESERVE_FLUSH_ALL);
4754 if (*qgroup_reserved)
4755 btrfs_qgroup_free(root, *qgroup_reserved);
4761 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4762 struct btrfs_block_rsv *rsv,
4763 u64 qgroup_reserved)
4765 btrfs_block_rsv_release(root, rsv, (u64)-1);
4766 if (qgroup_reserved)
4767 btrfs_qgroup_free(root, qgroup_reserved);
4771 * drop_outstanding_extent - drop an outstanding extent
4772 * @inode: the inode we're dropping the extent for
4774 * This is called when we are freeing up an outstanding extent, either called
4775 * after an error or after an extent is written. This will return the number of
4776 * reserved extents that need to be freed. This must be called with
4777 * BTRFS_I(inode)->lock held.
4779 static unsigned drop_outstanding_extent(struct inode *inode)
4781 unsigned drop_inode_space = 0;
4782 unsigned dropped_extents = 0;
4784 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4785 BTRFS_I(inode)->outstanding_extents--;
4787 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4788 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4789 &BTRFS_I(inode)->runtime_flags))
4790 drop_inode_space = 1;
4793 * If we have more or the same amount of outsanding extents than we have
4794 * reserved then we need to leave the reserved extents count alone.
4796 if (BTRFS_I(inode)->outstanding_extents >=
4797 BTRFS_I(inode)->reserved_extents)
4798 return drop_inode_space;
4800 dropped_extents = BTRFS_I(inode)->reserved_extents -
4801 BTRFS_I(inode)->outstanding_extents;
4802 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4803 return dropped_extents + drop_inode_space;
4807 * calc_csum_metadata_size - return the amount of metada space that must be
4808 * reserved/free'd for the given bytes.
4809 * @inode: the inode we're manipulating
4810 * @num_bytes: the number of bytes in question
4811 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4813 * This adjusts the number of csum_bytes in the inode and then returns the
4814 * correct amount of metadata that must either be reserved or freed. We
4815 * calculate how many checksums we can fit into one leaf and then divide the
4816 * number of bytes that will need to be checksumed by this value to figure out
4817 * how many checksums will be required. If we are adding bytes then the number
4818 * may go up and we will return the number of additional bytes that must be
4819 * reserved. If it is going down we will return the number of bytes that must
4822 * This must be called with BTRFS_I(inode)->lock held.
4824 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4827 struct btrfs_root *root = BTRFS_I(inode)->root;
4829 int num_csums_per_leaf;
4833 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4834 BTRFS_I(inode)->csum_bytes == 0)
4837 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4839 BTRFS_I(inode)->csum_bytes += num_bytes;
4841 BTRFS_I(inode)->csum_bytes -= num_bytes;
4842 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4843 num_csums_per_leaf = (int)div64_u64(csum_size,
4844 sizeof(struct btrfs_csum_item) +
4845 sizeof(struct btrfs_disk_key));
4846 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4847 num_csums = num_csums + num_csums_per_leaf - 1;
4848 num_csums = num_csums / num_csums_per_leaf;
4850 old_csums = old_csums + num_csums_per_leaf - 1;
4851 old_csums = old_csums / num_csums_per_leaf;
4853 /* No change, no need to reserve more */
4854 if (old_csums == num_csums)
4858 return btrfs_calc_trans_metadata_size(root,
4859 num_csums - old_csums);
4861 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4864 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4866 struct btrfs_root *root = BTRFS_I(inode)->root;
4867 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4870 unsigned nr_extents = 0;
4871 int extra_reserve = 0;
4872 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4874 bool delalloc_lock = true;
4878 /* If we are a free space inode we need to not flush since we will be in
4879 * the middle of a transaction commit. We also don't need the delalloc
4880 * mutex since we won't race with anybody. We need this mostly to make
4881 * lockdep shut its filthy mouth.
4883 if (btrfs_is_free_space_inode(inode)) {
4884 flush = BTRFS_RESERVE_NO_FLUSH;
4885 delalloc_lock = false;
4888 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4889 btrfs_transaction_in_commit(root->fs_info))
4890 schedule_timeout(1);
4893 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4895 num_bytes = ALIGN(num_bytes, root->sectorsize);
4897 spin_lock(&BTRFS_I(inode)->lock);
4898 BTRFS_I(inode)->outstanding_extents++;
4900 if (BTRFS_I(inode)->outstanding_extents >
4901 BTRFS_I(inode)->reserved_extents)
4902 nr_extents = BTRFS_I(inode)->outstanding_extents -
4903 BTRFS_I(inode)->reserved_extents;
4906 * Add an item to reserve for updating the inode when we complete the
4909 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4910 &BTRFS_I(inode)->runtime_flags)) {
4915 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4916 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4917 csum_bytes = BTRFS_I(inode)->csum_bytes;
4918 spin_unlock(&BTRFS_I(inode)->lock);
4920 if (root->fs_info->quota_enabled) {
4921 ret = btrfs_qgroup_reserve(root, num_bytes +
4922 nr_extents * root->leafsize);
4927 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4928 if (unlikely(ret)) {
4929 if (root->fs_info->quota_enabled)
4930 btrfs_qgroup_free(root, num_bytes +
4931 nr_extents * root->leafsize);
4935 spin_lock(&BTRFS_I(inode)->lock);
4936 if (extra_reserve) {
4937 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4938 &BTRFS_I(inode)->runtime_flags);
4941 BTRFS_I(inode)->reserved_extents += nr_extents;
4942 spin_unlock(&BTRFS_I(inode)->lock);
4945 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4948 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4949 btrfs_ino(inode), to_reserve, 1);
4950 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4955 spin_lock(&BTRFS_I(inode)->lock);
4956 dropped = drop_outstanding_extent(inode);
4958 * If the inodes csum_bytes is the same as the original
4959 * csum_bytes then we know we haven't raced with any free()ers
4960 * so we can just reduce our inodes csum bytes and carry on.
4962 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4963 calc_csum_metadata_size(inode, num_bytes, 0);
4965 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4969 * This is tricky, but first we need to figure out how much we
4970 * free'd from any free-ers that occured during this
4971 * reservation, so we reset ->csum_bytes to the csum_bytes
4972 * before we dropped our lock, and then call the free for the
4973 * number of bytes that were freed while we were trying our
4976 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4977 BTRFS_I(inode)->csum_bytes = csum_bytes;
4978 to_free = calc_csum_metadata_size(inode, bytes, 0);
4982 * Now we need to see how much we would have freed had we not
4983 * been making this reservation and our ->csum_bytes were not
4984 * artificially inflated.
4986 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4987 bytes = csum_bytes - orig_csum_bytes;
4988 bytes = calc_csum_metadata_size(inode, bytes, 0);
4991 * Now reset ->csum_bytes to what it should be. If bytes is
4992 * more than to_free then we would have free'd more space had we
4993 * not had an artificially high ->csum_bytes, so we need to free
4994 * the remainder. If bytes is the same or less then we don't
4995 * need to do anything, the other free-ers did the correct
4998 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4999 if (bytes > to_free)
5000 to_free = bytes - to_free;
5004 spin_unlock(&BTRFS_I(inode)->lock);
5006 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5009 btrfs_block_rsv_release(root, block_rsv, to_free);
5010 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5011 btrfs_ino(inode), to_free, 0);
5014 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5019 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5020 * @inode: the inode to release the reservation for
5021 * @num_bytes: the number of bytes we're releasing
5023 * This will release the metadata reservation for an inode. This can be called
5024 * once we complete IO for a given set of bytes to release their metadata
5027 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5029 struct btrfs_root *root = BTRFS_I(inode)->root;
5033 num_bytes = ALIGN(num_bytes, root->sectorsize);
5034 spin_lock(&BTRFS_I(inode)->lock);
5035 dropped = drop_outstanding_extent(inode);
5038 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5039 spin_unlock(&BTRFS_I(inode)->lock);
5041 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5043 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5044 btrfs_ino(inode), to_free, 0);
5045 if (root->fs_info->quota_enabled) {
5046 btrfs_qgroup_free(root, num_bytes +
5047 dropped * root->leafsize);
5050 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5055 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5056 * @inode: inode we're writing to
5057 * @num_bytes: the number of bytes we want to allocate
5059 * This will do the following things
5061 * o reserve space in the data space info for num_bytes
5062 * o reserve space in the metadata space info based on number of outstanding
5063 * extents and how much csums will be needed
5064 * o add to the inodes ->delalloc_bytes
5065 * o add it to the fs_info's delalloc inodes list.
5067 * This will return 0 for success and -ENOSPC if there is no space left.
5069 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5073 ret = btrfs_check_data_free_space(inode, num_bytes);
5077 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5079 btrfs_free_reserved_data_space(inode, num_bytes);
5087 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5088 * @inode: inode we're releasing space for
5089 * @num_bytes: the number of bytes we want to free up
5091 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5092 * called in the case that we don't need the metadata AND data reservations
5093 * anymore. So if there is an error or we insert an inline extent.
5095 * This function will release the metadata space that was not used and will
5096 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5097 * list if there are no delalloc bytes left.
5099 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5101 btrfs_delalloc_release_metadata(inode, num_bytes);
5102 btrfs_free_reserved_data_space(inode, num_bytes);
5105 static int update_block_group(struct btrfs_root *root,
5106 u64 bytenr, u64 num_bytes, int alloc)
5108 struct btrfs_block_group_cache *cache = NULL;
5109 struct btrfs_fs_info *info = root->fs_info;
5110 u64 total = num_bytes;
5115 /* block accounting for super block */
5116 spin_lock(&info->delalloc_root_lock);
5117 old_val = btrfs_super_bytes_used(info->super_copy);
5119 old_val += num_bytes;
5121 old_val -= num_bytes;
5122 btrfs_set_super_bytes_used(info->super_copy, old_val);
5123 spin_unlock(&info->delalloc_root_lock);
5126 cache = btrfs_lookup_block_group(info, bytenr);
5129 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5130 BTRFS_BLOCK_GROUP_RAID1 |
5131 BTRFS_BLOCK_GROUP_RAID10))
5136 * If this block group has free space cache written out, we
5137 * need to make sure to load it if we are removing space. This
5138 * is because we need the unpinning stage to actually add the
5139 * space back to the block group, otherwise we will leak space.
5141 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5142 cache_block_group(cache, 1);
5144 byte_in_group = bytenr - cache->key.objectid;
5145 WARN_ON(byte_in_group > cache->key.offset);
5147 spin_lock(&cache->space_info->lock);
5148 spin_lock(&cache->lock);
5150 if (btrfs_test_opt(root, SPACE_CACHE) &&
5151 cache->disk_cache_state < BTRFS_DC_CLEAR)
5152 cache->disk_cache_state = BTRFS_DC_CLEAR;
5155 old_val = btrfs_block_group_used(&cache->item);
5156 num_bytes = min(total, cache->key.offset - byte_in_group);
5158 old_val += num_bytes;
5159 btrfs_set_block_group_used(&cache->item, old_val);
5160 cache->reserved -= num_bytes;
5161 cache->space_info->bytes_reserved -= num_bytes;
5162 cache->space_info->bytes_used += num_bytes;
5163 cache->space_info->disk_used += num_bytes * factor;
5164 spin_unlock(&cache->lock);
5165 spin_unlock(&cache->space_info->lock);
5167 old_val -= num_bytes;
5168 btrfs_set_block_group_used(&cache->item, old_val);
5169 cache->pinned += num_bytes;
5170 cache->space_info->bytes_pinned += num_bytes;
5171 cache->space_info->bytes_used -= num_bytes;
5172 cache->space_info->disk_used -= num_bytes * factor;
5173 spin_unlock(&cache->lock);
5174 spin_unlock(&cache->space_info->lock);
5176 set_extent_dirty(info->pinned_extents,
5177 bytenr, bytenr + num_bytes - 1,
5178 GFP_NOFS | __GFP_NOFAIL);
5180 btrfs_put_block_group(cache);
5182 bytenr += num_bytes;
5187 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5189 struct btrfs_block_group_cache *cache;
5192 spin_lock(&root->fs_info->block_group_cache_lock);
5193 bytenr = root->fs_info->first_logical_byte;
5194 spin_unlock(&root->fs_info->block_group_cache_lock);
5196 if (bytenr < (u64)-1)
5199 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5203 bytenr = cache->key.objectid;
5204 btrfs_put_block_group(cache);
5209 static int pin_down_extent(struct btrfs_root *root,
5210 struct btrfs_block_group_cache *cache,
5211 u64 bytenr, u64 num_bytes, int reserved)
5213 spin_lock(&cache->space_info->lock);
5214 spin_lock(&cache->lock);
5215 cache->pinned += num_bytes;
5216 cache->space_info->bytes_pinned += num_bytes;
5218 cache->reserved -= num_bytes;
5219 cache->space_info->bytes_reserved -= num_bytes;
5221 spin_unlock(&cache->lock);
5222 spin_unlock(&cache->space_info->lock);
5224 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5225 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5230 * this function must be called within transaction
5232 int btrfs_pin_extent(struct btrfs_root *root,
5233 u64 bytenr, u64 num_bytes, int reserved)
5235 struct btrfs_block_group_cache *cache;
5237 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5238 BUG_ON(!cache); /* Logic error */
5240 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5242 btrfs_put_block_group(cache);
5247 * this function must be called within transaction
5249 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5250 u64 bytenr, u64 num_bytes)
5252 struct btrfs_block_group_cache *cache;
5255 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5260 * pull in the free space cache (if any) so that our pin
5261 * removes the free space from the cache. We have load_only set
5262 * to one because the slow code to read in the free extents does check
5263 * the pinned extents.
5265 cache_block_group(cache, 1);
5267 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5269 /* remove us from the free space cache (if we're there at all) */
5270 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5271 btrfs_put_block_group(cache);
5275 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5278 struct btrfs_block_group_cache *block_group;
5279 struct btrfs_caching_control *caching_ctl;
5281 block_group = btrfs_lookup_block_group(root->fs_info, start);
5285 cache_block_group(block_group, 0);
5286 caching_ctl = get_caching_control(block_group);
5290 BUG_ON(!block_group_cache_done(block_group));
5291 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5293 mutex_lock(&caching_ctl->mutex);
5295 if (start >= caching_ctl->progress) {
5296 ret = add_excluded_extent(root, start, num_bytes);
5297 } else if (start + num_bytes <= caching_ctl->progress) {
5298 ret = btrfs_remove_free_space(block_group,
5301 num_bytes = caching_ctl->progress - start;
5302 ret = btrfs_remove_free_space(block_group,
5307 num_bytes = (start + num_bytes) -
5308 caching_ctl->progress;
5309 start = caching_ctl->progress;
5310 ret = add_excluded_extent(root, start, num_bytes);
5313 mutex_unlock(&caching_ctl->mutex);
5314 put_caching_control(caching_ctl);
5316 btrfs_put_block_group(block_group);
5320 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5321 struct extent_buffer *eb)
5323 struct btrfs_file_extent_item *item;
5324 struct btrfs_key key;
5328 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5331 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5332 btrfs_item_key_to_cpu(eb, &key, i);
5333 if (key.type != BTRFS_EXTENT_DATA_KEY)
5335 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5336 found_type = btrfs_file_extent_type(eb, item);
5337 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5339 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5341 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5342 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5343 __exclude_logged_extent(log, key.objectid, key.offset);
5350 * btrfs_update_reserved_bytes - update the block_group and space info counters
5351 * @cache: The cache we are manipulating
5352 * @num_bytes: The number of bytes in question
5353 * @reserve: One of the reservation enums
5355 * This is called by the allocator when it reserves space, or by somebody who is
5356 * freeing space that was never actually used on disk. For example if you
5357 * reserve some space for a new leaf in transaction A and before transaction A
5358 * commits you free that leaf, you call this with reserve set to 0 in order to
5359 * clear the reservation.
5361 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5362 * ENOSPC accounting. For data we handle the reservation through clearing the
5363 * delalloc bits in the io_tree. We have to do this since we could end up
5364 * allocating less disk space for the amount of data we have reserved in the
5365 * case of compression.
5367 * If this is a reservation and the block group has become read only we cannot
5368 * make the reservation and return -EAGAIN, otherwise this function always
5371 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5372 u64 num_bytes, int reserve)
5374 struct btrfs_space_info *space_info = cache->space_info;
5377 spin_lock(&space_info->lock);
5378 spin_lock(&cache->lock);
5379 if (reserve != RESERVE_FREE) {
5383 cache->reserved += num_bytes;
5384 space_info->bytes_reserved += num_bytes;
5385 if (reserve == RESERVE_ALLOC) {
5386 trace_btrfs_space_reservation(cache->fs_info,
5387 "space_info", space_info->flags,
5389 space_info->bytes_may_use -= num_bytes;
5394 space_info->bytes_readonly += num_bytes;
5395 cache->reserved -= num_bytes;
5396 space_info->bytes_reserved -= num_bytes;
5397 space_info->reservation_progress++;
5399 spin_unlock(&cache->lock);
5400 spin_unlock(&space_info->lock);
5404 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5405 struct btrfs_root *root)
5407 struct btrfs_fs_info *fs_info = root->fs_info;
5408 struct btrfs_caching_control *next;
5409 struct btrfs_caching_control *caching_ctl;
5410 struct btrfs_block_group_cache *cache;
5411 struct btrfs_space_info *space_info;
5413 down_write(&fs_info->extent_commit_sem);
5415 list_for_each_entry_safe(caching_ctl, next,
5416 &fs_info->caching_block_groups, list) {
5417 cache = caching_ctl->block_group;
5418 if (block_group_cache_done(cache)) {
5419 cache->last_byte_to_unpin = (u64)-1;
5420 list_del_init(&caching_ctl->list);
5421 put_caching_control(caching_ctl);
5423 cache->last_byte_to_unpin = caching_ctl->progress;
5427 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5428 fs_info->pinned_extents = &fs_info->freed_extents[1];
5430 fs_info->pinned_extents = &fs_info->freed_extents[0];
5432 up_write(&fs_info->extent_commit_sem);
5434 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5435 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5437 update_global_block_rsv(fs_info);
5440 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5442 struct btrfs_fs_info *fs_info = root->fs_info;
5443 struct btrfs_block_group_cache *cache = NULL;
5444 struct btrfs_space_info *space_info;
5445 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5449 while (start <= end) {
5452 start >= cache->key.objectid + cache->key.offset) {
5454 btrfs_put_block_group(cache);
5455 cache = btrfs_lookup_block_group(fs_info, start);
5456 BUG_ON(!cache); /* Logic error */
5459 len = cache->key.objectid + cache->key.offset - start;
5460 len = min(len, end + 1 - start);
5462 if (start < cache->last_byte_to_unpin) {
5463 len = min(len, cache->last_byte_to_unpin - start);
5464 btrfs_add_free_space(cache, start, len);
5468 space_info = cache->space_info;
5470 spin_lock(&space_info->lock);
5471 spin_lock(&cache->lock);
5472 cache->pinned -= len;
5473 space_info->bytes_pinned -= len;
5475 space_info->bytes_readonly += len;
5478 spin_unlock(&cache->lock);
5479 if (!readonly && global_rsv->space_info == space_info) {
5480 spin_lock(&global_rsv->lock);
5481 if (!global_rsv->full) {
5482 len = min(len, global_rsv->size -
5483 global_rsv->reserved);
5484 global_rsv->reserved += len;
5485 space_info->bytes_may_use += len;
5486 if (global_rsv->reserved >= global_rsv->size)
5487 global_rsv->full = 1;
5489 spin_unlock(&global_rsv->lock);
5491 spin_unlock(&space_info->lock);
5495 btrfs_put_block_group(cache);
5499 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5500 struct btrfs_root *root)
5502 struct btrfs_fs_info *fs_info = root->fs_info;
5503 struct extent_io_tree *unpin;
5511 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5512 unpin = &fs_info->freed_extents[1];
5514 unpin = &fs_info->freed_extents[0];
5517 ret = find_first_extent_bit(unpin, 0, &start, &end,
5518 EXTENT_DIRTY, NULL);
5522 if (btrfs_test_opt(root, DISCARD))
5523 ret = btrfs_discard_extent(root, start,
5524 end + 1 - start, NULL);
5526 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5527 unpin_extent_range(root, start, end);
5534 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5535 u64 owner, u64 root_objectid)
5537 struct btrfs_space_info *space_info;
5540 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5541 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5542 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5544 flags = BTRFS_BLOCK_GROUP_METADATA;
5546 flags = BTRFS_BLOCK_GROUP_DATA;
5549 space_info = __find_space_info(fs_info, flags);
5550 BUG_ON(!space_info); /* Logic bug */
5551 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5555 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5556 struct btrfs_root *root,
5557 u64 bytenr, u64 num_bytes, u64 parent,
5558 u64 root_objectid, u64 owner_objectid,
5559 u64 owner_offset, int refs_to_drop,
5560 struct btrfs_delayed_extent_op *extent_op)
5562 struct btrfs_key key;
5563 struct btrfs_path *path;
5564 struct btrfs_fs_info *info = root->fs_info;
5565 struct btrfs_root *extent_root = info->extent_root;
5566 struct extent_buffer *leaf;
5567 struct btrfs_extent_item *ei;
5568 struct btrfs_extent_inline_ref *iref;
5571 int extent_slot = 0;
5572 int found_extent = 0;
5576 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5579 path = btrfs_alloc_path();
5584 path->leave_spinning = 1;
5586 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5587 BUG_ON(!is_data && refs_to_drop != 1);
5590 skinny_metadata = 0;
5592 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5593 bytenr, num_bytes, parent,
5594 root_objectid, owner_objectid,
5597 extent_slot = path->slots[0];
5598 while (extent_slot >= 0) {
5599 btrfs_item_key_to_cpu(path->nodes[0], &key,
5601 if (key.objectid != bytenr)
5603 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5604 key.offset == num_bytes) {
5608 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5609 key.offset == owner_objectid) {
5613 if (path->slots[0] - extent_slot > 5)
5617 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5618 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5619 if (found_extent && item_size < sizeof(*ei))
5622 if (!found_extent) {
5624 ret = remove_extent_backref(trans, extent_root, path,
5628 btrfs_abort_transaction(trans, extent_root, ret);
5631 btrfs_release_path(path);
5632 path->leave_spinning = 1;
5634 key.objectid = bytenr;
5635 key.type = BTRFS_EXTENT_ITEM_KEY;
5636 key.offset = num_bytes;
5638 if (!is_data && skinny_metadata) {
5639 key.type = BTRFS_METADATA_ITEM_KEY;
5640 key.offset = owner_objectid;
5643 ret = btrfs_search_slot(trans, extent_root,
5645 if (ret > 0 && skinny_metadata && path->slots[0]) {
5647 * Couldn't find our skinny metadata item,
5648 * see if we have ye olde extent item.
5651 btrfs_item_key_to_cpu(path->nodes[0], &key,
5653 if (key.objectid == bytenr &&
5654 key.type == BTRFS_EXTENT_ITEM_KEY &&
5655 key.offset == num_bytes)
5659 if (ret > 0 && skinny_metadata) {
5660 skinny_metadata = false;
5661 key.type = BTRFS_EXTENT_ITEM_KEY;
5662 key.offset = num_bytes;
5663 btrfs_release_path(path);
5664 ret = btrfs_search_slot(trans, extent_root,
5669 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5670 ret, (unsigned long long)bytenr);
5672 btrfs_print_leaf(extent_root,
5676 btrfs_abort_transaction(trans, extent_root, ret);
5679 extent_slot = path->slots[0];
5681 } else if (ret == -ENOENT) {
5682 btrfs_print_leaf(extent_root, path->nodes[0]);
5685 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5686 (unsigned long long)bytenr,
5687 (unsigned long long)parent,
5688 (unsigned long long)root_objectid,
5689 (unsigned long long)owner_objectid,
5690 (unsigned long long)owner_offset);
5692 btrfs_abort_transaction(trans, extent_root, ret);
5696 leaf = path->nodes[0];
5697 item_size = btrfs_item_size_nr(leaf, extent_slot);
5698 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5699 if (item_size < sizeof(*ei)) {
5700 BUG_ON(found_extent || extent_slot != path->slots[0]);
5701 ret = convert_extent_item_v0(trans, extent_root, path,
5704 btrfs_abort_transaction(trans, extent_root, ret);
5708 btrfs_release_path(path);
5709 path->leave_spinning = 1;
5711 key.objectid = bytenr;
5712 key.type = BTRFS_EXTENT_ITEM_KEY;
5713 key.offset = num_bytes;
5715 ret = btrfs_search_slot(trans, extent_root, &key, path,
5718 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5719 ret, (unsigned long long)bytenr);
5720 btrfs_print_leaf(extent_root, path->nodes[0]);
5723 btrfs_abort_transaction(trans, extent_root, ret);
5727 extent_slot = path->slots[0];
5728 leaf = path->nodes[0];
5729 item_size = btrfs_item_size_nr(leaf, extent_slot);
5732 BUG_ON(item_size < sizeof(*ei));
5733 ei = btrfs_item_ptr(leaf, extent_slot,
5734 struct btrfs_extent_item);
5735 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5736 key.type == BTRFS_EXTENT_ITEM_KEY) {
5737 struct btrfs_tree_block_info *bi;
5738 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5739 bi = (struct btrfs_tree_block_info *)(ei + 1);
5740 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5743 refs = btrfs_extent_refs(leaf, ei);
5744 if (refs < refs_to_drop) {
5745 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5746 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5748 btrfs_abort_transaction(trans, extent_root, ret);
5751 refs -= refs_to_drop;
5755 __run_delayed_extent_op(extent_op, leaf, ei);
5757 * In the case of inline back ref, reference count will
5758 * be updated by remove_extent_backref
5761 BUG_ON(!found_extent);
5763 btrfs_set_extent_refs(leaf, ei, refs);
5764 btrfs_mark_buffer_dirty(leaf);
5767 ret = remove_extent_backref(trans, extent_root, path,
5771 btrfs_abort_transaction(trans, extent_root, ret);
5775 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5779 BUG_ON(is_data && refs_to_drop !=
5780 extent_data_ref_count(root, path, iref));
5782 BUG_ON(path->slots[0] != extent_slot);
5784 BUG_ON(path->slots[0] != extent_slot + 1);
5785 path->slots[0] = extent_slot;
5790 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5793 btrfs_abort_transaction(trans, extent_root, ret);
5796 btrfs_release_path(path);
5799 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5801 btrfs_abort_transaction(trans, extent_root, ret);
5806 ret = update_block_group(root, bytenr, num_bytes, 0);
5808 btrfs_abort_transaction(trans, extent_root, ret);
5813 btrfs_free_path(path);
5818 * when we free an block, it is possible (and likely) that we free the last
5819 * delayed ref for that extent as well. This searches the delayed ref tree for
5820 * a given extent, and if there are no other delayed refs to be processed, it
5821 * removes it from the tree.
5823 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5824 struct btrfs_root *root, u64 bytenr)
5826 struct btrfs_delayed_ref_head *head;
5827 struct btrfs_delayed_ref_root *delayed_refs;
5828 struct btrfs_delayed_ref_node *ref;
5829 struct rb_node *node;
5832 delayed_refs = &trans->transaction->delayed_refs;
5833 spin_lock(&delayed_refs->lock);
5834 head = btrfs_find_delayed_ref_head(trans, bytenr);
5838 node = rb_prev(&head->node.rb_node);
5842 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5844 /* there are still entries for this ref, we can't drop it */
5845 if (ref->bytenr == bytenr)
5848 if (head->extent_op) {
5849 if (!head->must_insert_reserved)
5851 btrfs_free_delayed_extent_op(head->extent_op);
5852 head->extent_op = NULL;
5856 * waiting for the lock here would deadlock. If someone else has it
5857 * locked they are already in the process of dropping it anyway
5859 if (!mutex_trylock(&head->mutex))
5863 * at this point we have a head with no other entries. Go
5864 * ahead and process it.
5866 head->node.in_tree = 0;
5867 rb_erase(&head->node.rb_node, &delayed_refs->root);
5869 delayed_refs->num_entries--;
5872 * we don't take a ref on the node because we're removing it from the
5873 * tree, so we just steal the ref the tree was holding.
5875 delayed_refs->num_heads--;
5876 if (list_empty(&head->cluster))
5877 delayed_refs->num_heads_ready--;
5879 list_del_init(&head->cluster);
5880 spin_unlock(&delayed_refs->lock);
5882 BUG_ON(head->extent_op);
5883 if (head->must_insert_reserved)
5886 mutex_unlock(&head->mutex);
5887 btrfs_put_delayed_ref(&head->node);
5890 spin_unlock(&delayed_refs->lock);
5894 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5895 struct btrfs_root *root,
5896 struct extent_buffer *buf,
5897 u64 parent, int last_ref)
5899 struct btrfs_block_group_cache *cache = NULL;
5903 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5904 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5905 buf->start, buf->len,
5906 parent, root->root_key.objectid,
5907 btrfs_header_level(buf),
5908 BTRFS_DROP_DELAYED_REF, NULL, 0);
5909 BUG_ON(ret); /* -ENOMEM */
5915 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5917 if (btrfs_header_generation(buf) == trans->transid) {
5918 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5919 ret = check_ref_cleanup(trans, root, buf->start);
5924 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5925 pin_down_extent(root, cache, buf->start, buf->len, 1);
5929 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5931 btrfs_add_free_space(cache, buf->start, buf->len);
5932 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5937 add_pinned_bytes(root->fs_info, buf->len,
5938 btrfs_header_level(buf),
5939 root->root_key.objectid);
5942 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5945 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5946 btrfs_put_block_group(cache);
5949 /* Can return -ENOMEM */
5950 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5951 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5952 u64 owner, u64 offset, int for_cow)
5955 struct btrfs_fs_info *fs_info = root->fs_info;
5957 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
5960 * tree log blocks never actually go into the extent allocation
5961 * tree, just update pinning info and exit early.
5963 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5964 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5965 /* unlocks the pinned mutex */
5966 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5968 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5969 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5971 parent, root_objectid, (int)owner,
5972 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5974 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5976 parent, root_objectid, owner,
5977 offset, BTRFS_DROP_DELAYED_REF,
5983 static u64 stripe_align(struct btrfs_root *root,
5984 struct btrfs_block_group_cache *cache,
5985 u64 val, u64 num_bytes)
5987 u64 ret = ALIGN(val, root->stripesize);
5992 * when we wait for progress in the block group caching, its because
5993 * our allocation attempt failed at least once. So, we must sleep
5994 * and let some progress happen before we try again.
5996 * This function will sleep at least once waiting for new free space to
5997 * show up, and then it will check the block group free space numbers
5998 * for our min num_bytes. Another option is to have it go ahead
5999 * and look in the rbtree for a free extent of a given size, but this
6003 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6006 struct btrfs_caching_control *caching_ctl;
6008 caching_ctl = get_caching_control(cache);
6012 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6013 (cache->free_space_ctl->free_space >= num_bytes));
6015 put_caching_control(caching_ctl);
6020 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6022 struct btrfs_caching_control *caching_ctl;
6024 caching_ctl = get_caching_control(cache);
6028 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6030 put_caching_control(caching_ctl);
6034 int __get_raid_index(u64 flags)
6036 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6037 return BTRFS_RAID_RAID10;
6038 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6039 return BTRFS_RAID_RAID1;
6040 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6041 return BTRFS_RAID_DUP;
6042 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6043 return BTRFS_RAID_RAID0;
6044 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6045 return BTRFS_RAID_RAID5;
6046 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6047 return BTRFS_RAID_RAID6;
6049 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6052 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6054 return __get_raid_index(cache->flags);
6057 enum btrfs_loop_type {
6058 LOOP_CACHING_NOWAIT = 0,
6059 LOOP_CACHING_WAIT = 1,
6060 LOOP_ALLOC_CHUNK = 2,
6061 LOOP_NO_EMPTY_SIZE = 3,
6065 * walks the btree of allocated extents and find a hole of a given size.
6066 * The key ins is changed to record the hole:
6067 * ins->objectid == block start
6068 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6069 * ins->offset == number of blocks
6070 * Any available blocks before search_start are skipped.
6072 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
6073 struct btrfs_root *orig_root,
6074 u64 num_bytes, u64 empty_size,
6075 u64 hint_byte, struct btrfs_key *ins,
6079 struct btrfs_root *root = orig_root->fs_info->extent_root;
6080 struct btrfs_free_cluster *last_ptr = NULL;
6081 struct btrfs_block_group_cache *block_group = NULL;
6082 struct btrfs_block_group_cache *used_block_group;
6083 u64 search_start = 0;
6084 int empty_cluster = 2 * 1024 * 1024;
6085 struct btrfs_space_info *space_info;
6087 int index = __get_raid_index(flags);
6088 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6089 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6090 bool found_uncached_bg = false;
6091 bool failed_cluster_refill = false;
6092 bool failed_alloc = false;
6093 bool use_cluster = true;
6094 bool have_caching_bg = false;
6096 WARN_ON(num_bytes < root->sectorsize);
6097 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6101 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6103 space_info = __find_space_info(root->fs_info, flags);
6105 btrfs_err(root->fs_info, "No space info for %llu", flags);
6110 * If the space info is for both data and metadata it means we have a
6111 * small filesystem and we can't use the clustering stuff.
6113 if (btrfs_mixed_space_info(space_info))
6114 use_cluster = false;
6116 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6117 last_ptr = &root->fs_info->meta_alloc_cluster;
6118 if (!btrfs_test_opt(root, SSD))
6119 empty_cluster = 64 * 1024;
6122 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6123 btrfs_test_opt(root, SSD)) {
6124 last_ptr = &root->fs_info->data_alloc_cluster;
6128 spin_lock(&last_ptr->lock);
6129 if (last_ptr->block_group)
6130 hint_byte = last_ptr->window_start;
6131 spin_unlock(&last_ptr->lock);
6134 search_start = max(search_start, first_logical_byte(root, 0));
6135 search_start = max(search_start, hint_byte);
6140 if (search_start == hint_byte) {
6141 block_group = btrfs_lookup_block_group(root->fs_info,
6143 used_block_group = block_group;
6145 * we don't want to use the block group if it doesn't match our
6146 * allocation bits, or if its not cached.
6148 * However if we are re-searching with an ideal block group
6149 * picked out then we don't care that the block group is cached.
6151 if (block_group && block_group_bits(block_group, flags) &&
6152 block_group->cached != BTRFS_CACHE_NO) {
6153 down_read(&space_info->groups_sem);
6154 if (list_empty(&block_group->list) ||
6157 * someone is removing this block group,
6158 * we can't jump into the have_block_group
6159 * target because our list pointers are not
6162 btrfs_put_block_group(block_group);
6163 up_read(&space_info->groups_sem);
6165 index = get_block_group_index(block_group);
6166 goto have_block_group;
6168 } else if (block_group) {
6169 btrfs_put_block_group(block_group);
6173 have_caching_bg = false;
6174 down_read(&space_info->groups_sem);
6175 list_for_each_entry(block_group, &space_info->block_groups[index],
6180 used_block_group = block_group;
6181 btrfs_get_block_group(block_group);
6182 search_start = block_group->key.objectid;
6185 * this can happen if we end up cycling through all the
6186 * raid types, but we want to make sure we only allocate
6187 * for the proper type.
6189 if (!block_group_bits(block_group, flags)) {
6190 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6191 BTRFS_BLOCK_GROUP_RAID1 |
6192 BTRFS_BLOCK_GROUP_RAID5 |
6193 BTRFS_BLOCK_GROUP_RAID6 |
6194 BTRFS_BLOCK_GROUP_RAID10;
6197 * if they asked for extra copies and this block group
6198 * doesn't provide them, bail. This does allow us to
6199 * fill raid0 from raid1.
6201 if ((flags & extra) && !(block_group->flags & extra))
6206 cached = block_group_cache_done(block_group);
6207 if (unlikely(!cached)) {
6208 found_uncached_bg = true;
6209 ret = cache_block_group(block_group, 0);
6214 if (unlikely(block_group->ro))
6218 * Ok we want to try and use the cluster allocator, so
6222 unsigned long aligned_cluster;
6224 * the refill lock keeps out other
6225 * people trying to start a new cluster
6227 spin_lock(&last_ptr->refill_lock);
6228 used_block_group = last_ptr->block_group;
6229 if (used_block_group != block_group &&
6230 (!used_block_group ||
6231 used_block_group->ro ||
6232 !block_group_bits(used_block_group, flags))) {
6233 used_block_group = block_group;
6234 goto refill_cluster;
6237 if (used_block_group != block_group)
6238 btrfs_get_block_group(used_block_group);
6240 offset = btrfs_alloc_from_cluster(used_block_group,
6241 last_ptr, num_bytes, used_block_group->key.objectid);
6243 /* we have a block, we're done */
6244 spin_unlock(&last_ptr->refill_lock);
6245 trace_btrfs_reserve_extent_cluster(root,
6246 block_group, search_start, num_bytes);
6250 WARN_ON(last_ptr->block_group != used_block_group);
6251 if (used_block_group != block_group) {
6252 btrfs_put_block_group(used_block_group);
6253 used_block_group = block_group;
6256 BUG_ON(used_block_group != block_group);
6257 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6258 * set up a new clusters, so lets just skip it
6259 * and let the allocator find whatever block
6260 * it can find. If we reach this point, we
6261 * will have tried the cluster allocator
6262 * plenty of times and not have found
6263 * anything, so we are likely way too
6264 * fragmented for the clustering stuff to find
6267 * However, if the cluster is taken from the
6268 * current block group, release the cluster
6269 * first, so that we stand a better chance of
6270 * succeeding in the unclustered
6272 if (loop >= LOOP_NO_EMPTY_SIZE &&
6273 last_ptr->block_group != block_group) {
6274 spin_unlock(&last_ptr->refill_lock);
6275 goto unclustered_alloc;
6279 * this cluster didn't work out, free it and
6282 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6284 if (loop >= LOOP_NO_EMPTY_SIZE) {
6285 spin_unlock(&last_ptr->refill_lock);
6286 goto unclustered_alloc;
6289 aligned_cluster = max_t(unsigned long,
6290 empty_cluster + empty_size,
6291 block_group->full_stripe_len);
6293 /* allocate a cluster in this block group */
6294 ret = btrfs_find_space_cluster(trans, root,
6295 block_group, last_ptr,
6296 search_start, num_bytes,
6300 * now pull our allocation out of this
6303 offset = btrfs_alloc_from_cluster(block_group,
6304 last_ptr, num_bytes,
6307 /* we found one, proceed */
6308 spin_unlock(&last_ptr->refill_lock);
6309 trace_btrfs_reserve_extent_cluster(root,
6310 block_group, search_start,
6314 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6315 && !failed_cluster_refill) {
6316 spin_unlock(&last_ptr->refill_lock);
6318 failed_cluster_refill = true;
6319 wait_block_group_cache_progress(block_group,
6320 num_bytes + empty_cluster + empty_size);
6321 goto have_block_group;
6325 * at this point we either didn't find a cluster
6326 * or we weren't able to allocate a block from our
6327 * cluster. Free the cluster we've been trying
6328 * to use, and go to the next block group
6330 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6331 spin_unlock(&last_ptr->refill_lock);
6336 spin_lock(&block_group->free_space_ctl->tree_lock);
6338 block_group->free_space_ctl->free_space <
6339 num_bytes + empty_cluster + empty_size) {
6340 spin_unlock(&block_group->free_space_ctl->tree_lock);
6343 spin_unlock(&block_group->free_space_ctl->tree_lock);
6345 offset = btrfs_find_space_for_alloc(block_group, search_start,
6346 num_bytes, empty_size);
6348 * If we didn't find a chunk, and we haven't failed on this
6349 * block group before, and this block group is in the middle of
6350 * caching and we are ok with waiting, then go ahead and wait
6351 * for progress to be made, and set failed_alloc to true.
6353 * If failed_alloc is true then we've already waited on this
6354 * block group once and should move on to the next block group.
6356 if (!offset && !failed_alloc && !cached &&
6357 loop > LOOP_CACHING_NOWAIT) {
6358 wait_block_group_cache_progress(block_group,
6359 num_bytes + empty_size);
6360 failed_alloc = true;
6361 goto have_block_group;
6362 } else if (!offset) {
6364 have_caching_bg = true;
6368 search_start = stripe_align(root, used_block_group,
6371 /* move on to the next group */
6372 if (search_start + num_bytes >
6373 used_block_group->key.objectid + used_block_group->key.offset) {
6374 btrfs_add_free_space(used_block_group, offset, num_bytes);
6378 if (offset < search_start)
6379 btrfs_add_free_space(used_block_group, offset,
6380 search_start - offset);
6381 BUG_ON(offset > search_start);
6383 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6385 if (ret == -EAGAIN) {
6386 btrfs_add_free_space(used_block_group, offset, num_bytes);
6390 /* we are all good, lets return */
6391 ins->objectid = search_start;
6392 ins->offset = num_bytes;
6394 trace_btrfs_reserve_extent(orig_root, block_group,
6395 search_start, num_bytes);
6396 if (used_block_group != block_group)
6397 btrfs_put_block_group(used_block_group);
6398 btrfs_put_block_group(block_group);
6401 failed_cluster_refill = false;
6402 failed_alloc = false;
6403 BUG_ON(index != get_block_group_index(block_group));
6404 if (used_block_group != block_group)
6405 btrfs_put_block_group(used_block_group);
6406 btrfs_put_block_group(block_group);
6408 up_read(&space_info->groups_sem);
6410 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6413 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6417 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6418 * caching kthreads as we move along
6419 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6420 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6421 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6424 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6427 if (loop == LOOP_ALLOC_CHUNK) {
6428 ret = do_chunk_alloc(trans, root, flags,
6431 * Do not bail out on ENOSPC since we
6432 * can do more things.
6434 if (ret < 0 && ret != -ENOSPC) {
6435 btrfs_abort_transaction(trans,
6441 if (loop == LOOP_NO_EMPTY_SIZE) {
6447 } else if (!ins->objectid) {
6449 } else if (ins->objectid) {
6457 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6458 int dump_block_groups)
6460 struct btrfs_block_group_cache *cache;
6463 spin_lock(&info->lock);
6464 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6465 (unsigned long long)info->flags,
6466 (unsigned long long)(info->total_bytes - info->bytes_used -
6467 info->bytes_pinned - info->bytes_reserved -
6468 info->bytes_readonly),
6469 (info->full) ? "" : "not ");
6470 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6471 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6472 (unsigned long long)info->total_bytes,
6473 (unsigned long long)info->bytes_used,
6474 (unsigned long long)info->bytes_pinned,
6475 (unsigned long long)info->bytes_reserved,
6476 (unsigned long long)info->bytes_may_use,
6477 (unsigned long long)info->bytes_readonly);
6478 spin_unlock(&info->lock);
6480 if (!dump_block_groups)
6483 down_read(&info->groups_sem);
6485 list_for_each_entry(cache, &info->block_groups[index], list) {
6486 spin_lock(&cache->lock);
6487 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6488 (unsigned long long)cache->key.objectid,
6489 (unsigned long long)cache->key.offset,
6490 (unsigned long long)btrfs_block_group_used(&cache->item),
6491 (unsigned long long)cache->pinned,
6492 (unsigned long long)cache->reserved,
6493 cache->ro ? "[readonly]" : "");
6494 btrfs_dump_free_space(cache, bytes);
6495 spin_unlock(&cache->lock);
6497 if (++index < BTRFS_NR_RAID_TYPES)
6499 up_read(&info->groups_sem);
6502 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6503 struct btrfs_root *root,
6504 u64 num_bytes, u64 min_alloc_size,
6505 u64 empty_size, u64 hint_byte,
6506 struct btrfs_key *ins, int is_data)
6508 bool final_tried = false;
6512 flags = btrfs_get_alloc_profile(root, is_data);
6514 WARN_ON(num_bytes < root->sectorsize);
6515 ret = find_free_extent(trans, root, num_bytes, empty_size,
6516 hint_byte, ins, flags);
6518 if (ret == -ENOSPC) {
6520 num_bytes = num_bytes >> 1;
6521 num_bytes = round_down(num_bytes, root->sectorsize);
6522 num_bytes = max(num_bytes, min_alloc_size);
6523 if (num_bytes == min_alloc_size)
6526 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6527 struct btrfs_space_info *sinfo;
6529 sinfo = __find_space_info(root->fs_info, flags);
6530 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6531 (unsigned long long)flags,
6532 (unsigned long long)num_bytes);
6534 dump_space_info(sinfo, num_bytes, 1);
6538 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6543 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6544 u64 start, u64 len, int pin)
6546 struct btrfs_block_group_cache *cache;
6549 cache = btrfs_lookup_block_group(root->fs_info, start);
6551 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6552 (unsigned long long)start);
6556 if (btrfs_test_opt(root, DISCARD))
6557 ret = btrfs_discard_extent(root, start, len, NULL);
6560 pin_down_extent(root, cache, start, len, 1);
6562 btrfs_add_free_space(cache, start, len);
6563 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6565 btrfs_put_block_group(cache);
6567 trace_btrfs_reserved_extent_free(root, start, len);
6572 int btrfs_free_reserved_extent(struct btrfs_root *root,
6575 return __btrfs_free_reserved_extent(root, start, len, 0);
6578 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6581 return __btrfs_free_reserved_extent(root, start, len, 1);
6584 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6585 struct btrfs_root *root,
6586 u64 parent, u64 root_objectid,
6587 u64 flags, u64 owner, u64 offset,
6588 struct btrfs_key *ins, int ref_mod)
6591 struct btrfs_fs_info *fs_info = root->fs_info;
6592 struct btrfs_extent_item *extent_item;
6593 struct btrfs_extent_inline_ref *iref;
6594 struct btrfs_path *path;
6595 struct extent_buffer *leaf;
6600 type = BTRFS_SHARED_DATA_REF_KEY;
6602 type = BTRFS_EXTENT_DATA_REF_KEY;
6604 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6606 path = btrfs_alloc_path();
6610 path->leave_spinning = 1;
6611 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6614 btrfs_free_path(path);
6618 leaf = path->nodes[0];
6619 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6620 struct btrfs_extent_item);
6621 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6622 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6623 btrfs_set_extent_flags(leaf, extent_item,
6624 flags | BTRFS_EXTENT_FLAG_DATA);
6626 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6627 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6629 struct btrfs_shared_data_ref *ref;
6630 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6631 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6632 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6634 struct btrfs_extent_data_ref *ref;
6635 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6636 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6637 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6638 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6639 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6642 btrfs_mark_buffer_dirty(path->nodes[0]);
6643 btrfs_free_path(path);
6645 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6646 if (ret) { /* -ENOENT, logic error */
6647 btrfs_err(fs_info, "update block group failed for %llu %llu",
6648 (unsigned long long)ins->objectid,
6649 (unsigned long long)ins->offset);
6655 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6656 struct btrfs_root *root,
6657 u64 parent, u64 root_objectid,
6658 u64 flags, struct btrfs_disk_key *key,
6659 int level, struct btrfs_key *ins)
6662 struct btrfs_fs_info *fs_info = root->fs_info;
6663 struct btrfs_extent_item *extent_item;
6664 struct btrfs_tree_block_info *block_info;
6665 struct btrfs_extent_inline_ref *iref;
6666 struct btrfs_path *path;
6667 struct extent_buffer *leaf;
6668 u32 size = sizeof(*extent_item) + sizeof(*iref);
6669 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6672 if (!skinny_metadata)
6673 size += sizeof(*block_info);
6675 path = btrfs_alloc_path();
6679 path->leave_spinning = 1;
6680 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6683 btrfs_free_path(path);
6687 leaf = path->nodes[0];
6688 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6689 struct btrfs_extent_item);
6690 btrfs_set_extent_refs(leaf, extent_item, 1);
6691 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6692 btrfs_set_extent_flags(leaf, extent_item,
6693 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6695 if (skinny_metadata) {
6696 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6698 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6699 btrfs_set_tree_block_key(leaf, block_info, key);
6700 btrfs_set_tree_block_level(leaf, block_info, level);
6701 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6705 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6706 btrfs_set_extent_inline_ref_type(leaf, iref,
6707 BTRFS_SHARED_BLOCK_REF_KEY);
6708 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6710 btrfs_set_extent_inline_ref_type(leaf, iref,
6711 BTRFS_TREE_BLOCK_REF_KEY);
6712 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6715 btrfs_mark_buffer_dirty(leaf);
6716 btrfs_free_path(path);
6718 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6719 if (ret) { /* -ENOENT, logic error */
6720 btrfs_err(fs_info, "update block group failed for %llu %llu",
6721 (unsigned long long)ins->objectid,
6722 (unsigned long long)ins->offset);
6728 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6729 struct btrfs_root *root,
6730 u64 root_objectid, u64 owner,
6731 u64 offset, struct btrfs_key *ins)
6735 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6737 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6739 root_objectid, owner, offset,
6740 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6745 * this is used by the tree logging recovery code. It records that
6746 * an extent has been allocated and makes sure to clear the free
6747 * space cache bits as well
6749 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6750 struct btrfs_root *root,
6751 u64 root_objectid, u64 owner, u64 offset,
6752 struct btrfs_key *ins)
6755 struct btrfs_block_group_cache *block_group;
6758 * Mixed block groups will exclude before processing the log so we only
6759 * need to do the exlude dance if this fs isn't mixed.
6761 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6762 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6767 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6771 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6772 RESERVE_ALLOC_NO_ACCOUNT);
6773 BUG_ON(ret); /* logic error */
6774 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6775 0, owner, offset, ins, 1);
6776 btrfs_put_block_group(block_group);
6780 static struct extent_buffer *
6781 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6782 u64 bytenr, u32 blocksize, int level)
6784 struct extent_buffer *buf;
6786 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6788 return ERR_PTR(-ENOMEM);
6789 btrfs_set_header_generation(buf, trans->transid);
6790 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6791 btrfs_tree_lock(buf);
6792 clean_tree_block(trans, root, buf);
6793 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6795 btrfs_set_lock_blocking(buf);
6796 btrfs_set_buffer_uptodate(buf);
6798 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6800 * we allow two log transactions at a time, use different
6801 * EXENT bit to differentiate dirty pages.
6803 if (root->log_transid % 2 == 0)
6804 set_extent_dirty(&root->dirty_log_pages, buf->start,
6805 buf->start + buf->len - 1, GFP_NOFS);
6807 set_extent_new(&root->dirty_log_pages, buf->start,
6808 buf->start + buf->len - 1, GFP_NOFS);
6810 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6811 buf->start + buf->len - 1, GFP_NOFS);
6813 trans->blocks_used++;
6814 /* this returns a buffer locked for blocking */
6818 static struct btrfs_block_rsv *
6819 use_block_rsv(struct btrfs_trans_handle *trans,
6820 struct btrfs_root *root, u32 blocksize)
6822 struct btrfs_block_rsv *block_rsv;
6823 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6825 bool global_updated = false;
6827 block_rsv = get_block_rsv(trans, root);
6829 if (unlikely(block_rsv->size == 0))
6832 ret = block_rsv_use_bytes(block_rsv, blocksize);
6836 if (block_rsv->failfast)
6837 return ERR_PTR(ret);
6839 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6840 global_updated = true;
6841 update_global_block_rsv(root->fs_info);
6845 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6846 static DEFINE_RATELIMIT_STATE(_rs,
6847 DEFAULT_RATELIMIT_INTERVAL * 10,
6848 /*DEFAULT_RATELIMIT_BURST*/ 1);
6849 if (__ratelimit(&_rs))
6851 "btrfs: block rsv returned %d\n", ret);
6854 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6855 BTRFS_RESERVE_NO_FLUSH);
6859 * If we couldn't reserve metadata bytes try and use some from
6860 * the global reserve if its space type is the same as the global
6863 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6864 block_rsv->space_info == global_rsv->space_info) {
6865 ret = block_rsv_use_bytes(global_rsv, blocksize);
6869 return ERR_PTR(ret);
6872 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6873 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6875 block_rsv_add_bytes(block_rsv, blocksize, 0);
6876 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6880 * finds a free extent and does all the dirty work required for allocation
6881 * returns the key for the extent through ins, and a tree buffer for
6882 * the first block of the extent through buf.
6884 * returns the tree buffer or NULL.
6886 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6887 struct btrfs_root *root, u32 blocksize,
6888 u64 parent, u64 root_objectid,
6889 struct btrfs_disk_key *key, int level,
6890 u64 hint, u64 empty_size)
6892 struct btrfs_key ins;
6893 struct btrfs_block_rsv *block_rsv;
6894 struct extent_buffer *buf;
6897 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6900 block_rsv = use_block_rsv(trans, root, blocksize);
6901 if (IS_ERR(block_rsv))
6902 return ERR_CAST(block_rsv);
6904 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6905 empty_size, hint, &ins, 0);
6907 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6908 return ERR_PTR(ret);
6911 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6913 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6915 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6917 parent = ins.objectid;
6918 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6922 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6923 struct btrfs_delayed_extent_op *extent_op;
6924 extent_op = btrfs_alloc_delayed_extent_op();
6925 BUG_ON(!extent_op); /* -ENOMEM */
6927 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6929 memset(&extent_op->key, 0, sizeof(extent_op->key));
6930 extent_op->flags_to_set = flags;
6931 if (skinny_metadata)
6932 extent_op->update_key = 0;
6934 extent_op->update_key = 1;
6935 extent_op->update_flags = 1;
6936 extent_op->is_data = 0;
6937 extent_op->level = level;
6939 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6941 ins.offset, parent, root_objectid,
6942 level, BTRFS_ADD_DELAYED_EXTENT,
6944 BUG_ON(ret); /* -ENOMEM */
6949 struct walk_control {
6950 u64 refs[BTRFS_MAX_LEVEL];
6951 u64 flags[BTRFS_MAX_LEVEL];
6952 struct btrfs_key update_progress;
6963 #define DROP_REFERENCE 1
6964 #define UPDATE_BACKREF 2
6966 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6967 struct btrfs_root *root,
6968 struct walk_control *wc,
6969 struct btrfs_path *path)
6977 struct btrfs_key key;
6978 struct extent_buffer *eb;
6983 if (path->slots[wc->level] < wc->reada_slot) {
6984 wc->reada_count = wc->reada_count * 2 / 3;
6985 wc->reada_count = max(wc->reada_count, 2);
6987 wc->reada_count = wc->reada_count * 3 / 2;
6988 wc->reada_count = min_t(int, wc->reada_count,
6989 BTRFS_NODEPTRS_PER_BLOCK(root));
6992 eb = path->nodes[wc->level];
6993 nritems = btrfs_header_nritems(eb);
6994 blocksize = btrfs_level_size(root, wc->level - 1);
6996 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6997 if (nread >= wc->reada_count)
7001 bytenr = btrfs_node_blockptr(eb, slot);
7002 generation = btrfs_node_ptr_generation(eb, slot);
7004 if (slot == path->slots[wc->level])
7007 if (wc->stage == UPDATE_BACKREF &&
7008 generation <= root->root_key.offset)
7011 /* We don't lock the tree block, it's OK to be racy here */
7012 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7013 wc->level - 1, 1, &refs,
7015 /* We don't care about errors in readahead. */
7020 if (wc->stage == DROP_REFERENCE) {
7024 if (wc->level == 1 &&
7025 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7027 if (!wc->update_ref ||
7028 generation <= root->root_key.offset)
7030 btrfs_node_key_to_cpu(eb, &key, slot);
7031 ret = btrfs_comp_cpu_keys(&key,
7032 &wc->update_progress);
7036 if (wc->level == 1 &&
7037 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7041 ret = readahead_tree_block(root, bytenr, blocksize,
7047 wc->reada_slot = slot;
7051 * helper to process tree block while walking down the tree.
7053 * when wc->stage == UPDATE_BACKREF, this function updates
7054 * back refs for pointers in the block.
7056 * NOTE: return value 1 means we should stop walking down.
7058 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7059 struct btrfs_root *root,
7060 struct btrfs_path *path,
7061 struct walk_control *wc, int lookup_info)
7063 int level = wc->level;
7064 struct extent_buffer *eb = path->nodes[level];
7065 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7068 if (wc->stage == UPDATE_BACKREF &&
7069 btrfs_header_owner(eb) != root->root_key.objectid)
7073 * when reference count of tree block is 1, it won't increase
7074 * again. once full backref flag is set, we never clear it.
7077 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7078 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7079 BUG_ON(!path->locks[level]);
7080 ret = btrfs_lookup_extent_info(trans, root,
7081 eb->start, level, 1,
7084 BUG_ON(ret == -ENOMEM);
7087 BUG_ON(wc->refs[level] == 0);
7090 if (wc->stage == DROP_REFERENCE) {
7091 if (wc->refs[level] > 1)
7094 if (path->locks[level] && !wc->keep_locks) {
7095 btrfs_tree_unlock_rw(eb, path->locks[level]);
7096 path->locks[level] = 0;
7101 /* wc->stage == UPDATE_BACKREF */
7102 if (!(wc->flags[level] & flag)) {
7103 BUG_ON(!path->locks[level]);
7104 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7105 BUG_ON(ret); /* -ENOMEM */
7106 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7107 BUG_ON(ret); /* -ENOMEM */
7108 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7110 btrfs_header_level(eb), 0);
7111 BUG_ON(ret); /* -ENOMEM */
7112 wc->flags[level] |= flag;
7116 * the block is shared by multiple trees, so it's not good to
7117 * keep the tree lock
7119 if (path->locks[level] && level > 0) {
7120 btrfs_tree_unlock_rw(eb, path->locks[level]);
7121 path->locks[level] = 0;
7127 * helper to process tree block pointer.
7129 * when wc->stage == DROP_REFERENCE, this function checks
7130 * reference count of the block pointed to. if the block
7131 * is shared and we need update back refs for the subtree
7132 * rooted at the block, this function changes wc->stage to
7133 * UPDATE_BACKREF. if the block is shared and there is no
7134 * need to update back, this function drops the reference
7137 * NOTE: return value 1 means we should stop walking down.
7139 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7140 struct btrfs_root *root,
7141 struct btrfs_path *path,
7142 struct walk_control *wc, int *lookup_info)
7148 struct btrfs_key key;
7149 struct extent_buffer *next;
7150 int level = wc->level;
7154 generation = btrfs_node_ptr_generation(path->nodes[level],
7155 path->slots[level]);
7157 * if the lower level block was created before the snapshot
7158 * was created, we know there is no need to update back refs
7161 if (wc->stage == UPDATE_BACKREF &&
7162 generation <= root->root_key.offset) {
7167 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7168 blocksize = btrfs_level_size(root, level - 1);
7170 next = btrfs_find_tree_block(root, bytenr, blocksize);
7172 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7177 btrfs_tree_lock(next);
7178 btrfs_set_lock_blocking(next);
7180 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7181 &wc->refs[level - 1],
7182 &wc->flags[level - 1]);
7184 btrfs_tree_unlock(next);
7188 if (unlikely(wc->refs[level - 1] == 0)) {
7189 btrfs_err(root->fs_info, "Missing references.");
7194 if (wc->stage == DROP_REFERENCE) {
7195 if (wc->refs[level - 1] > 1) {
7197 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7200 if (!wc->update_ref ||
7201 generation <= root->root_key.offset)
7204 btrfs_node_key_to_cpu(path->nodes[level], &key,
7205 path->slots[level]);
7206 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7210 wc->stage = UPDATE_BACKREF;
7211 wc->shared_level = level - 1;
7215 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7219 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7220 btrfs_tree_unlock(next);
7221 free_extent_buffer(next);
7227 if (reada && level == 1)
7228 reada_walk_down(trans, root, wc, path);
7229 next = read_tree_block(root, bytenr, blocksize, generation);
7230 if (!next || !extent_buffer_uptodate(next)) {
7231 free_extent_buffer(next);
7234 btrfs_tree_lock(next);
7235 btrfs_set_lock_blocking(next);
7239 BUG_ON(level != btrfs_header_level(next));
7240 path->nodes[level] = next;
7241 path->slots[level] = 0;
7242 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7248 wc->refs[level - 1] = 0;
7249 wc->flags[level - 1] = 0;
7250 if (wc->stage == DROP_REFERENCE) {
7251 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7252 parent = path->nodes[level]->start;
7254 BUG_ON(root->root_key.objectid !=
7255 btrfs_header_owner(path->nodes[level]));
7259 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7260 root->root_key.objectid, level - 1, 0, 0);
7261 BUG_ON(ret); /* -ENOMEM */
7263 btrfs_tree_unlock(next);
7264 free_extent_buffer(next);
7270 * helper to process tree block while walking up the tree.
7272 * when wc->stage == DROP_REFERENCE, this function drops
7273 * reference count on the block.
7275 * when wc->stage == UPDATE_BACKREF, this function changes
7276 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7277 * to UPDATE_BACKREF previously while processing the block.
7279 * NOTE: return value 1 means we should stop walking up.
7281 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7282 struct btrfs_root *root,
7283 struct btrfs_path *path,
7284 struct walk_control *wc)
7287 int level = wc->level;
7288 struct extent_buffer *eb = path->nodes[level];
7291 if (wc->stage == UPDATE_BACKREF) {
7292 BUG_ON(wc->shared_level < level);
7293 if (level < wc->shared_level)
7296 ret = find_next_key(path, level + 1, &wc->update_progress);
7300 wc->stage = DROP_REFERENCE;
7301 wc->shared_level = -1;
7302 path->slots[level] = 0;
7305 * check reference count again if the block isn't locked.
7306 * we should start walking down the tree again if reference
7309 if (!path->locks[level]) {
7311 btrfs_tree_lock(eb);
7312 btrfs_set_lock_blocking(eb);
7313 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7315 ret = btrfs_lookup_extent_info(trans, root,
7316 eb->start, level, 1,
7320 btrfs_tree_unlock_rw(eb, path->locks[level]);
7321 path->locks[level] = 0;
7324 BUG_ON(wc->refs[level] == 0);
7325 if (wc->refs[level] == 1) {
7326 btrfs_tree_unlock_rw(eb, path->locks[level]);
7327 path->locks[level] = 0;
7333 /* wc->stage == DROP_REFERENCE */
7334 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7336 if (wc->refs[level] == 1) {
7338 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7339 ret = btrfs_dec_ref(trans, root, eb, 1,
7342 ret = btrfs_dec_ref(trans, root, eb, 0,
7344 BUG_ON(ret); /* -ENOMEM */
7346 /* make block locked assertion in clean_tree_block happy */
7347 if (!path->locks[level] &&
7348 btrfs_header_generation(eb) == trans->transid) {
7349 btrfs_tree_lock(eb);
7350 btrfs_set_lock_blocking(eb);
7351 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7353 clean_tree_block(trans, root, eb);
7356 if (eb == root->node) {
7357 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7360 BUG_ON(root->root_key.objectid !=
7361 btrfs_header_owner(eb));
7363 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7364 parent = path->nodes[level + 1]->start;
7366 BUG_ON(root->root_key.objectid !=
7367 btrfs_header_owner(path->nodes[level + 1]));
7370 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7372 wc->refs[level] = 0;
7373 wc->flags[level] = 0;
7377 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7378 struct btrfs_root *root,
7379 struct btrfs_path *path,
7380 struct walk_control *wc)
7382 int level = wc->level;
7383 int lookup_info = 1;
7386 while (level >= 0) {
7387 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7394 if (path->slots[level] >=
7395 btrfs_header_nritems(path->nodes[level]))
7398 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7400 path->slots[level]++;
7409 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7410 struct btrfs_root *root,
7411 struct btrfs_path *path,
7412 struct walk_control *wc, int max_level)
7414 int level = wc->level;
7417 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7418 while (level < max_level && path->nodes[level]) {
7420 if (path->slots[level] + 1 <
7421 btrfs_header_nritems(path->nodes[level])) {
7422 path->slots[level]++;
7425 ret = walk_up_proc(trans, root, path, wc);
7429 if (path->locks[level]) {
7430 btrfs_tree_unlock_rw(path->nodes[level],
7431 path->locks[level]);
7432 path->locks[level] = 0;
7434 free_extent_buffer(path->nodes[level]);
7435 path->nodes[level] = NULL;
7443 * drop a subvolume tree.
7445 * this function traverses the tree freeing any blocks that only
7446 * referenced by the tree.
7448 * when a shared tree block is found. this function decreases its
7449 * reference count by one. if update_ref is true, this function
7450 * also make sure backrefs for the shared block and all lower level
7451 * blocks are properly updated.
7453 * If called with for_reloc == 0, may exit early with -EAGAIN
7455 int btrfs_drop_snapshot(struct btrfs_root *root,
7456 struct btrfs_block_rsv *block_rsv, int update_ref,
7459 struct btrfs_path *path;
7460 struct btrfs_trans_handle *trans;
7461 struct btrfs_root *tree_root = root->fs_info->tree_root;
7462 struct btrfs_root_item *root_item = &root->root_item;
7463 struct walk_control *wc;
7464 struct btrfs_key key;
7469 path = btrfs_alloc_path();
7475 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7477 btrfs_free_path(path);
7482 trans = btrfs_start_transaction(tree_root, 0);
7483 if (IS_ERR(trans)) {
7484 err = PTR_ERR(trans);
7489 trans->block_rsv = block_rsv;
7491 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7492 level = btrfs_header_level(root->node);
7493 path->nodes[level] = btrfs_lock_root_node(root);
7494 btrfs_set_lock_blocking(path->nodes[level]);
7495 path->slots[level] = 0;
7496 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7497 memset(&wc->update_progress, 0,
7498 sizeof(wc->update_progress));
7500 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7501 memcpy(&wc->update_progress, &key,
7502 sizeof(wc->update_progress));
7504 level = root_item->drop_level;
7506 path->lowest_level = level;
7507 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7508 path->lowest_level = 0;
7516 * unlock our path, this is safe because only this
7517 * function is allowed to delete this snapshot
7519 btrfs_unlock_up_safe(path, 0);
7521 level = btrfs_header_level(root->node);
7523 btrfs_tree_lock(path->nodes[level]);
7524 btrfs_set_lock_blocking(path->nodes[level]);
7526 ret = btrfs_lookup_extent_info(trans, root,
7527 path->nodes[level]->start,
7528 level, 1, &wc->refs[level],
7534 BUG_ON(wc->refs[level] == 0);
7536 if (level == root_item->drop_level)
7539 btrfs_tree_unlock(path->nodes[level]);
7540 WARN_ON(wc->refs[level] != 1);
7546 wc->shared_level = -1;
7547 wc->stage = DROP_REFERENCE;
7548 wc->update_ref = update_ref;
7550 wc->for_reloc = for_reloc;
7551 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7554 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7555 pr_debug("btrfs: drop snapshot early exit\n");
7560 ret = walk_down_tree(trans, root, path, wc);
7566 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7573 BUG_ON(wc->stage != DROP_REFERENCE);
7577 if (wc->stage == DROP_REFERENCE) {
7579 btrfs_node_key(path->nodes[level],
7580 &root_item->drop_progress,
7581 path->slots[level]);
7582 root_item->drop_level = level;
7585 BUG_ON(wc->level == 0);
7586 if (btrfs_should_end_transaction(trans, tree_root)) {
7587 ret = btrfs_update_root(trans, tree_root,
7591 btrfs_abort_transaction(trans, tree_root, ret);
7596 btrfs_end_transaction_throttle(trans, tree_root);
7597 trans = btrfs_start_transaction(tree_root, 0);
7598 if (IS_ERR(trans)) {
7599 err = PTR_ERR(trans);
7603 trans->block_rsv = block_rsv;
7606 btrfs_release_path(path);
7610 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7612 btrfs_abort_transaction(trans, tree_root, ret);
7616 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7617 ret = btrfs_find_root(tree_root, &root->root_key, path,
7620 btrfs_abort_transaction(trans, tree_root, ret);
7623 } else if (ret > 0) {
7624 /* if we fail to delete the orphan item this time
7625 * around, it'll get picked up the next time.
7627 * The most common failure here is just -ENOENT.
7629 btrfs_del_orphan_item(trans, tree_root,
7630 root->root_key.objectid);
7634 if (root->in_radix) {
7635 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7637 free_extent_buffer(root->node);
7638 free_extent_buffer(root->commit_root);
7639 btrfs_put_fs_root(root);
7642 btrfs_end_transaction_throttle(trans, tree_root);
7645 btrfs_free_path(path);
7648 btrfs_std_error(root->fs_info, err);
7653 * drop subtree rooted at tree block 'node'.
7655 * NOTE: this function will unlock and release tree block 'node'
7656 * only used by relocation code
7658 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7659 struct btrfs_root *root,
7660 struct extent_buffer *node,
7661 struct extent_buffer *parent)
7663 struct btrfs_path *path;
7664 struct walk_control *wc;
7670 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7672 path = btrfs_alloc_path();
7676 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7678 btrfs_free_path(path);
7682 btrfs_assert_tree_locked(parent);
7683 parent_level = btrfs_header_level(parent);
7684 extent_buffer_get(parent);
7685 path->nodes[parent_level] = parent;
7686 path->slots[parent_level] = btrfs_header_nritems(parent);
7688 btrfs_assert_tree_locked(node);
7689 level = btrfs_header_level(node);
7690 path->nodes[level] = node;
7691 path->slots[level] = 0;
7692 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7694 wc->refs[parent_level] = 1;
7695 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7697 wc->shared_level = -1;
7698 wc->stage = DROP_REFERENCE;
7702 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7705 wret = walk_down_tree(trans, root, path, wc);
7711 wret = walk_up_tree(trans, root, path, wc, parent_level);
7719 btrfs_free_path(path);
7723 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7729 * if restripe for this chunk_type is on pick target profile and
7730 * return, otherwise do the usual balance
7732 stripped = get_restripe_target(root->fs_info, flags);
7734 return extended_to_chunk(stripped);
7737 * we add in the count of missing devices because we want
7738 * to make sure that any RAID levels on a degraded FS
7739 * continue to be honored.
7741 num_devices = root->fs_info->fs_devices->rw_devices +
7742 root->fs_info->fs_devices->missing_devices;
7744 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7745 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7746 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7748 if (num_devices == 1) {
7749 stripped |= BTRFS_BLOCK_GROUP_DUP;
7750 stripped = flags & ~stripped;
7752 /* turn raid0 into single device chunks */
7753 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7756 /* turn mirroring into duplication */
7757 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7758 BTRFS_BLOCK_GROUP_RAID10))
7759 return stripped | BTRFS_BLOCK_GROUP_DUP;
7761 /* they already had raid on here, just return */
7762 if (flags & stripped)
7765 stripped |= BTRFS_BLOCK_GROUP_DUP;
7766 stripped = flags & ~stripped;
7768 /* switch duplicated blocks with raid1 */
7769 if (flags & BTRFS_BLOCK_GROUP_DUP)
7770 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7772 /* this is drive concat, leave it alone */
7778 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7780 struct btrfs_space_info *sinfo = cache->space_info;
7782 u64 min_allocable_bytes;
7787 * We need some metadata space and system metadata space for
7788 * allocating chunks in some corner cases until we force to set
7789 * it to be readonly.
7792 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7794 min_allocable_bytes = 1 * 1024 * 1024;
7796 min_allocable_bytes = 0;
7798 spin_lock(&sinfo->lock);
7799 spin_lock(&cache->lock);
7806 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7807 cache->bytes_super - btrfs_block_group_used(&cache->item);
7809 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7810 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7811 min_allocable_bytes <= sinfo->total_bytes) {
7812 sinfo->bytes_readonly += num_bytes;
7817 spin_unlock(&cache->lock);
7818 spin_unlock(&sinfo->lock);
7822 int btrfs_set_block_group_ro(struct btrfs_root *root,
7823 struct btrfs_block_group_cache *cache)
7826 struct btrfs_trans_handle *trans;
7832 trans = btrfs_join_transaction(root);
7834 return PTR_ERR(trans);
7836 alloc_flags = update_block_group_flags(root, cache->flags);
7837 if (alloc_flags != cache->flags) {
7838 ret = do_chunk_alloc(trans, root, alloc_flags,
7844 ret = set_block_group_ro(cache, 0);
7847 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7848 ret = do_chunk_alloc(trans, root, alloc_flags,
7852 ret = set_block_group_ro(cache, 0);
7854 btrfs_end_transaction(trans, root);
7858 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7859 struct btrfs_root *root, u64 type)
7861 u64 alloc_flags = get_alloc_profile(root, type);
7862 return do_chunk_alloc(trans, root, alloc_flags,
7867 * helper to account the unused space of all the readonly block group in the
7868 * list. takes mirrors into account.
7870 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7872 struct btrfs_block_group_cache *block_group;
7876 list_for_each_entry(block_group, groups_list, list) {
7877 spin_lock(&block_group->lock);
7879 if (!block_group->ro) {
7880 spin_unlock(&block_group->lock);
7884 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7885 BTRFS_BLOCK_GROUP_RAID10 |
7886 BTRFS_BLOCK_GROUP_DUP))
7891 free_bytes += (block_group->key.offset -
7892 btrfs_block_group_used(&block_group->item)) *
7895 spin_unlock(&block_group->lock);
7902 * helper to account the unused space of all the readonly block group in the
7903 * space_info. takes mirrors into account.
7905 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7910 spin_lock(&sinfo->lock);
7912 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7913 if (!list_empty(&sinfo->block_groups[i]))
7914 free_bytes += __btrfs_get_ro_block_group_free_space(
7915 &sinfo->block_groups[i]);
7917 spin_unlock(&sinfo->lock);
7922 void btrfs_set_block_group_rw(struct btrfs_root *root,
7923 struct btrfs_block_group_cache *cache)
7925 struct btrfs_space_info *sinfo = cache->space_info;
7930 spin_lock(&sinfo->lock);
7931 spin_lock(&cache->lock);
7932 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7933 cache->bytes_super - btrfs_block_group_used(&cache->item);
7934 sinfo->bytes_readonly -= num_bytes;
7936 spin_unlock(&cache->lock);
7937 spin_unlock(&sinfo->lock);
7941 * checks to see if its even possible to relocate this block group.
7943 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7944 * ok to go ahead and try.
7946 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7948 struct btrfs_block_group_cache *block_group;
7949 struct btrfs_space_info *space_info;
7950 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7951 struct btrfs_device *device;
7960 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7962 /* odd, couldn't find the block group, leave it alone */
7966 min_free = btrfs_block_group_used(&block_group->item);
7968 /* no bytes used, we're good */
7972 space_info = block_group->space_info;
7973 spin_lock(&space_info->lock);
7975 full = space_info->full;
7978 * if this is the last block group we have in this space, we can't
7979 * relocate it unless we're able to allocate a new chunk below.
7981 * Otherwise, we need to make sure we have room in the space to handle
7982 * all of the extents from this block group. If we can, we're good
7984 if ((space_info->total_bytes != block_group->key.offset) &&
7985 (space_info->bytes_used + space_info->bytes_reserved +
7986 space_info->bytes_pinned + space_info->bytes_readonly +
7987 min_free < space_info->total_bytes)) {
7988 spin_unlock(&space_info->lock);
7991 spin_unlock(&space_info->lock);
7994 * ok we don't have enough space, but maybe we have free space on our
7995 * devices to allocate new chunks for relocation, so loop through our
7996 * alloc devices and guess if we have enough space. if this block
7997 * group is going to be restriped, run checks against the target
7998 * profile instead of the current one.
8010 target = get_restripe_target(root->fs_info, block_group->flags);
8012 index = __get_raid_index(extended_to_chunk(target));
8015 * this is just a balance, so if we were marked as full
8016 * we know there is no space for a new chunk
8021 index = get_block_group_index(block_group);
8024 if (index == BTRFS_RAID_RAID10) {
8028 } else if (index == BTRFS_RAID_RAID1) {
8030 } else if (index == BTRFS_RAID_DUP) {
8033 } else if (index == BTRFS_RAID_RAID0) {
8034 dev_min = fs_devices->rw_devices;
8035 do_div(min_free, dev_min);
8038 mutex_lock(&root->fs_info->chunk_mutex);
8039 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8043 * check to make sure we can actually find a chunk with enough
8044 * space to fit our block group in.
8046 if (device->total_bytes > device->bytes_used + min_free &&
8047 !device->is_tgtdev_for_dev_replace) {
8048 ret = find_free_dev_extent(device, min_free,
8053 if (dev_nr >= dev_min)
8059 mutex_unlock(&root->fs_info->chunk_mutex);
8061 btrfs_put_block_group(block_group);
8065 static int find_first_block_group(struct btrfs_root *root,
8066 struct btrfs_path *path, struct btrfs_key *key)
8069 struct btrfs_key found_key;
8070 struct extent_buffer *leaf;
8073 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8078 slot = path->slots[0];
8079 leaf = path->nodes[0];
8080 if (slot >= btrfs_header_nritems(leaf)) {
8081 ret = btrfs_next_leaf(root, path);
8088 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8090 if (found_key.objectid >= key->objectid &&
8091 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8101 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8103 struct btrfs_block_group_cache *block_group;
8107 struct inode *inode;
8109 block_group = btrfs_lookup_first_block_group(info, last);
8110 while (block_group) {
8111 spin_lock(&block_group->lock);
8112 if (block_group->iref)
8114 spin_unlock(&block_group->lock);
8115 block_group = next_block_group(info->tree_root,
8125 inode = block_group->inode;
8126 block_group->iref = 0;
8127 block_group->inode = NULL;
8128 spin_unlock(&block_group->lock);
8130 last = block_group->key.objectid + block_group->key.offset;
8131 btrfs_put_block_group(block_group);
8135 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8137 struct btrfs_block_group_cache *block_group;
8138 struct btrfs_space_info *space_info;
8139 struct btrfs_caching_control *caching_ctl;
8142 down_write(&info->extent_commit_sem);
8143 while (!list_empty(&info->caching_block_groups)) {
8144 caching_ctl = list_entry(info->caching_block_groups.next,
8145 struct btrfs_caching_control, list);
8146 list_del(&caching_ctl->list);
8147 put_caching_control(caching_ctl);
8149 up_write(&info->extent_commit_sem);
8151 spin_lock(&info->block_group_cache_lock);
8152 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8153 block_group = rb_entry(n, struct btrfs_block_group_cache,
8155 rb_erase(&block_group->cache_node,
8156 &info->block_group_cache_tree);
8157 spin_unlock(&info->block_group_cache_lock);
8159 down_write(&block_group->space_info->groups_sem);
8160 list_del(&block_group->list);
8161 up_write(&block_group->space_info->groups_sem);
8163 if (block_group->cached == BTRFS_CACHE_STARTED)
8164 wait_block_group_cache_done(block_group);
8167 * We haven't cached this block group, which means we could
8168 * possibly have excluded extents on this block group.
8170 if (block_group->cached == BTRFS_CACHE_NO)
8171 free_excluded_extents(info->extent_root, block_group);
8173 btrfs_remove_free_space_cache(block_group);
8174 btrfs_put_block_group(block_group);
8176 spin_lock(&info->block_group_cache_lock);
8178 spin_unlock(&info->block_group_cache_lock);
8180 /* now that all the block groups are freed, go through and
8181 * free all the space_info structs. This is only called during
8182 * the final stages of unmount, and so we know nobody is
8183 * using them. We call synchronize_rcu() once before we start,
8184 * just to be on the safe side.
8188 release_global_block_rsv(info);
8190 while(!list_empty(&info->space_info)) {
8191 space_info = list_entry(info->space_info.next,
8192 struct btrfs_space_info,
8194 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8195 if (space_info->bytes_pinned > 0 ||
8196 space_info->bytes_reserved > 0 ||
8197 space_info->bytes_may_use > 0) {
8199 dump_space_info(space_info, 0, 0);
8202 percpu_counter_destroy(&space_info->total_bytes_pinned);
8203 list_del(&space_info->list);
8209 static void __link_block_group(struct btrfs_space_info *space_info,
8210 struct btrfs_block_group_cache *cache)
8212 int index = get_block_group_index(cache);
8214 down_write(&space_info->groups_sem);
8215 list_add_tail(&cache->list, &space_info->block_groups[index]);
8216 up_write(&space_info->groups_sem);
8219 int btrfs_read_block_groups(struct btrfs_root *root)
8221 struct btrfs_path *path;
8223 struct btrfs_block_group_cache *cache;
8224 struct btrfs_fs_info *info = root->fs_info;
8225 struct btrfs_space_info *space_info;
8226 struct btrfs_key key;
8227 struct btrfs_key found_key;
8228 struct extent_buffer *leaf;
8232 root = info->extent_root;
8235 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8236 path = btrfs_alloc_path();
8241 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8242 if (btrfs_test_opt(root, SPACE_CACHE) &&
8243 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8245 if (btrfs_test_opt(root, CLEAR_CACHE))
8249 ret = find_first_block_group(root, path, &key);
8254 leaf = path->nodes[0];
8255 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8256 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8261 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8263 if (!cache->free_space_ctl) {
8269 atomic_set(&cache->count, 1);
8270 spin_lock_init(&cache->lock);
8271 cache->fs_info = info;
8272 INIT_LIST_HEAD(&cache->list);
8273 INIT_LIST_HEAD(&cache->cluster_list);
8277 * When we mount with old space cache, we need to
8278 * set BTRFS_DC_CLEAR and set dirty flag.
8280 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8281 * truncate the old free space cache inode and
8283 * b) Setting 'dirty flag' makes sure that we flush
8284 * the new space cache info onto disk.
8286 cache->disk_cache_state = BTRFS_DC_CLEAR;
8287 if (btrfs_test_opt(root, SPACE_CACHE))
8291 read_extent_buffer(leaf, &cache->item,
8292 btrfs_item_ptr_offset(leaf, path->slots[0]),
8293 sizeof(cache->item));
8294 memcpy(&cache->key, &found_key, sizeof(found_key));
8296 key.objectid = found_key.objectid + found_key.offset;
8297 btrfs_release_path(path);
8298 cache->flags = btrfs_block_group_flags(&cache->item);
8299 cache->sectorsize = root->sectorsize;
8300 cache->full_stripe_len = btrfs_full_stripe_len(root,
8301 &root->fs_info->mapping_tree,
8302 found_key.objectid);
8303 btrfs_init_free_space_ctl(cache);
8306 * We need to exclude the super stripes now so that the space
8307 * info has super bytes accounted for, otherwise we'll think
8308 * we have more space than we actually do.
8310 ret = exclude_super_stripes(root, cache);
8313 * We may have excluded something, so call this just in
8316 free_excluded_extents(root, cache);
8317 kfree(cache->free_space_ctl);
8323 * check for two cases, either we are full, and therefore
8324 * don't need to bother with the caching work since we won't
8325 * find any space, or we are empty, and we can just add all
8326 * the space in and be done with it. This saves us _alot_ of
8327 * time, particularly in the full case.
8329 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8330 cache->last_byte_to_unpin = (u64)-1;
8331 cache->cached = BTRFS_CACHE_FINISHED;
8332 free_excluded_extents(root, cache);
8333 } else if (btrfs_block_group_used(&cache->item) == 0) {
8334 cache->last_byte_to_unpin = (u64)-1;
8335 cache->cached = BTRFS_CACHE_FINISHED;
8336 add_new_free_space(cache, root->fs_info,
8338 found_key.objectid +
8340 free_excluded_extents(root, cache);
8343 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8345 btrfs_remove_free_space_cache(cache);
8346 btrfs_put_block_group(cache);
8350 ret = update_space_info(info, cache->flags, found_key.offset,
8351 btrfs_block_group_used(&cache->item),
8354 btrfs_remove_free_space_cache(cache);
8355 spin_lock(&info->block_group_cache_lock);
8356 rb_erase(&cache->cache_node,
8357 &info->block_group_cache_tree);
8358 spin_unlock(&info->block_group_cache_lock);
8359 btrfs_put_block_group(cache);
8363 cache->space_info = space_info;
8364 spin_lock(&cache->space_info->lock);
8365 cache->space_info->bytes_readonly += cache->bytes_super;
8366 spin_unlock(&cache->space_info->lock);
8368 __link_block_group(space_info, cache);
8370 set_avail_alloc_bits(root->fs_info, cache->flags);
8371 if (btrfs_chunk_readonly(root, cache->key.objectid))
8372 set_block_group_ro(cache, 1);
8375 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8376 if (!(get_alloc_profile(root, space_info->flags) &
8377 (BTRFS_BLOCK_GROUP_RAID10 |
8378 BTRFS_BLOCK_GROUP_RAID1 |
8379 BTRFS_BLOCK_GROUP_RAID5 |
8380 BTRFS_BLOCK_GROUP_RAID6 |
8381 BTRFS_BLOCK_GROUP_DUP)))
8384 * avoid allocating from un-mirrored block group if there are
8385 * mirrored block groups.
8387 list_for_each_entry(cache, &space_info->block_groups[3], list)
8388 set_block_group_ro(cache, 1);
8389 list_for_each_entry(cache, &space_info->block_groups[4], list)
8390 set_block_group_ro(cache, 1);
8393 init_global_block_rsv(info);
8396 btrfs_free_path(path);
8400 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8401 struct btrfs_root *root)
8403 struct btrfs_block_group_cache *block_group, *tmp;
8404 struct btrfs_root *extent_root = root->fs_info->extent_root;
8405 struct btrfs_block_group_item item;
8406 struct btrfs_key key;
8409 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8411 list_del_init(&block_group->new_bg_list);
8416 spin_lock(&block_group->lock);
8417 memcpy(&item, &block_group->item, sizeof(item));
8418 memcpy(&key, &block_group->key, sizeof(key));
8419 spin_unlock(&block_group->lock);
8421 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8424 btrfs_abort_transaction(trans, extent_root, ret);
8428 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8429 struct btrfs_root *root, u64 bytes_used,
8430 u64 type, u64 chunk_objectid, u64 chunk_offset,
8434 struct btrfs_root *extent_root;
8435 struct btrfs_block_group_cache *cache;
8437 extent_root = root->fs_info->extent_root;
8439 root->fs_info->last_trans_log_full_commit = trans->transid;
8441 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8444 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8446 if (!cache->free_space_ctl) {
8451 cache->key.objectid = chunk_offset;
8452 cache->key.offset = size;
8453 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8454 cache->sectorsize = root->sectorsize;
8455 cache->fs_info = root->fs_info;
8456 cache->full_stripe_len = btrfs_full_stripe_len(root,
8457 &root->fs_info->mapping_tree,
8460 atomic_set(&cache->count, 1);
8461 spin_lock_init(&cache->lock);
8462 INIT_LIST_HEAD(&cache->list);
8463 INIT_LIST_HEAD(&cache->cluster_list);
8464 INIT_LIST_HEAD(&cache->new_bg_list);
8466 btrfs_init_free_space_ctl(cache);
8468 btrfs_set_block_group_used(&cache->item, bytes_used);
8469 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8470 cache->flags = type;
8471 btrfs_set_block_group_flags(&cache->item, type);
8473 cache->last_byte_to_unpin = (u64)-1;
8474 cache->cached = BTRFS_CACHE_FINISHED;
8475 ret = exclude_super_stripes(root, cache);
8478 * We may have excluded something, so call this just in
8481 free_excluded_extents(root, cache);
8482 kfree(cache->free_space_ctl);
8487 add_new_free_space(cache, root->fs_info, chunk_offset,
8488 chunk_offset + size);
8490 free_excluded_extents(root, cache);
8492 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8494 btrfs_remove_free_space_cache(cache);
8495 btrfs_put_block_group(cache);
8499 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8500 &cache->space_info);
8502 btrfs_remove_free_space_cache(cache);
8503 spin_lock(&root->fs_info->block_group_cache_lock);
8504 rb_erase(&cache->cache_node,
8505 &root->fs_info->block_group_cache_tree);
8506 spin_unlock(&root->fs_info->block_group_cache_lock);
8507 btrfs_put_block_group(cache);
8510 update_global_block_rsv(root->fs_info);
8512 spin_lock(&cache->space_info->lock);
8513 cache->space_info->bytes_readonly += cache->bytes_super;
8514 spin_unlock(&cache->space_info->lock);
8516 __link_block_group(cache->space_info, cache);
8518 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8520 set_avail_alloc_bits(extent_root->fs_info, type);
8525 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8527 u64 extra_flags = chunk_to_extended(flags) &
8528 BTRFS_EXTENDED_PROFILE_MASK;
8530 write_seqlock(&fs_info->profiles_lock);
8531 if (flags & BTRFS_BLOCK_GROUP_DATA)
8532 fs_info->avail_data_alloc_bits &= ~extra_flags;
8533 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8534 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8535 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8536 fs_info->avail_system_alloc_bits &= ~extra_flags;
8537 write_sequnlock(&fs_info->profiles_lock);
8540 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8541 struct btrfs_root *root, u64 group_start)
8543 struct btrfs_path *path;
8544 struct btrfs_block_group_cache *block_group;
8545 struct btrfs_free_cluster *cluster;
8546 struct btrfs_root *tree_root = root->fs_info->tree_root;
8547 struct btrfs_key key;
8548 struct inode *inode;
8553 root = root->fs_info->extent_root;
8555 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8556 BUG_ON(!block_group);
8557 BUG_ON(!block_group->ro);
8560 * Free the reserved super bytes from this block group before
8563 free_excluded_extents(root, block_group);
8565 memcpy(&key, &block_group->key, sizeof(key));
8566 index = get_block_group_index(block_group);
8567 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8568 BTRFS_BLOCK_GROUP_RAID1 |
8569 BTRFS_BLOCK_GROUP_RAID10))
8574 /* make sure this block group isn't part of an allocation cluster */
8575 cluster = &root->fs_info->data_alloc_cluster;
8576 spin_lock(&cluster->refill_lock);
8577 btrfs_return_cluster_to_free_space(block_group, cluster);
8578 spin_unlock(&cluster->refill_lock);
8581 * make sure this block group isn't part of a metadata
8582 * allocation cluster
8584 cluster = &root->fs_info->meta_alloc_cluster;
8585 spin_lock(&cluster->refill_lock);
8586 btrfs_return_cluster_to_free_space(block_group, cluster);
8587 spin_unlock(&cluster->refill_lock);
8589 path = btrfs_alloc_path();
8595 inode = lookup_free_space_inode(tree_root, block_group, path);
8596 if (!IS_ERR(inode)) {
8597 ret = btrfs_orphan_add(trans, inode);
8599 btrfs_add_delayed_iput(inode);
8603 /* One for the block groups ref */
8604 spin_lock(&block_group->lock);
8605 if (block_group->iref) {
8606 block_group->iref = 0;
8607 block_group->inode = NULL;
8608 spin_unlock(&block_group->lock);
8611 spin_unlock(&block_group->lock);
8613 /* One for our lookup ref */
8614 btrfs_add_delayed_iput(inode);
8617 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8618 key.offset = block_group->key.objectid;
8621 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8625 btrfs_release_path(path);
8627 ret = btrfs_del_item(trans, tree_root, path);
8630 btrfs_release_path(path);
8633 spin_lock(&root->fs_info->block_group_cache_lock);
8634 rb_erase(&block_group->cache_node,
8635 &root->fs_info->block_group_cache_tree);
8637 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8638 root->fs_info->first_logical_byte = (u64)-1;
8639 spin_unlock(&root->fs_info->block_group_cache_lock);
8641 down_write(&block_group->space_info->groups_sem);
8643 * we must use list_del_init so people can check to see if they
8644 * are still on the list after taking the semaphore
8646 list_del_init(&block_group->list);
8647 if (list_empty(&block_group->space_info->block_groups[index]))
8648 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8649 up_write(&block_group->space_info->groups_sem);
8651 if (block_group->cached == BTRFS_CACHE_STARTED)
8652 wait_block_group_cache_done(block_group);
8654 btrfs_remove_free_space_cache(block_group);
8656 spin_lock(&block_group->space_info->lock);
8657 block_group->space_info->total_bytes -= block_group->key.offset;
8658 block_group->space_info->bytes_readonly -= block_group->key.offset;
8659 block_group->space_info->disk_total -= block_group->key.offset * factor;
8660 spin_unlock(&block_group->space_info->lock);
8662 memcpy(&key, &block_group->key, sizeof(key));
8664 btrfs_clear_space_info_full(root->fs_info);
8666 btrfs_put_block_group(block_group);
8667 btrfs_put_block_group(block_group);
8669 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8675 ret = btrfs_del_item(trans, root, path);
8677 btrfs_free_path(path);
8681 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8683 struct btrfs_space_info *space_info;
8684 struct btrfs_super_block *disk_super;
8690 disk_super = fs_info->super_copy;
8691 if (!btrfs_super_root(disk_super))
8694 features = btrfs_super_incompat_flags(disk_super);
8695 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8698 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8699 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8704 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8705 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8707 flags = BTRFS_BLOCK_GROUP_METADATA;
8708 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8712 flags = BTRFS_BLOCK_GROUP_DATA;
8713 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8719 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8721 return unpin_extent_range(root, start, end);
8724 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8725 u64 num_bytes, u64 *actual_bytes)
8727 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8730 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8732 struct btrfs_fs_info *fs_info = root->fs_info;
8733 struct btrfs_block_group_cache *cache = NULL;
8738 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8742 * try to trim all FS space, our block group may start from non-zero.
8744 if (range->len == total_bytes)
8745 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8747 cache = btrfs_lookup_block_group(fs_info, range->start);
8750 if (cache->key.objectid >= (range->start + range->len)) {
8751 btrfs_put_block_group(cache);
8755 start = max(range->start, cache->key.objectid);
8756 end = min(range->start + range->len,
8757 cache->key.objectid + cache->key.offset);
8759 if (end - start >= range->minlen) {
8760 if (!block_group_cache_done(cache)) {
8761 ret = cache_block_group(cache, 0);
8763 btrfs_put_block_group(cache);
8766 ret = wait_block_group_cache_done(cache);
8768 btrfs_put_block_group(cache);
8772 ret = btrfs_trim_block_group(cache,
8778 trimmed += group_trimmed;
8780 btrfs_put_block_group(cache);
8785 cache = next_block_group(fs_info->tree_root, cache);
8788 range->len = trimmed;