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>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 static int update_block_group(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 u64 bytenr, u64 num_bytes, int alloc);
39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve, int sinfo);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61 struct btrfs_root *extent_root, u64 alloc_bytes,
62 u64 flags, int force);
63 static int find_next_key(struct btrfs_path *path, int level,
64 struct btrfs_key *key);
65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
66 int dump_block_groups);
69 block_group_cache_done(struct btrfs_block_group_cache *cache)
72 return cache->cached == BTRFS_CACHE_FINISHED;
75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
77 return (cache->flags & bits) == bits;
80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
82 atomic_inc(&cache->count);
85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
87 if (atomic_dec_and_test(&cache->count)) {
88 WARN_ON(cache->pinned > 0);
89 WARN_ON(cache->reserved > 0);
90 WARN_ON(cache->reserved_pinned > 0);
96 * this adds the block group to the fs_info rb tree for the block group
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
100 struct btrfs_block_group_cache *block_group)
103 struct rb_node *parent = NULL;
104 struct btrfs_block_group_cache *cache;
106 spin_lock(&info->block_group_cache_lock);
107 p = &info->block_group_cache_tree.rb_node;
111 cache = rb_entry(parent, struct btrfs_block_group_cache,
113 if (block_group->key.objectid < cache->key.objectid) {
115 } else if (block_group->key.objectid > cache->key.objectid) {
118 spin_unlock(&info->block_group_cache_lock);
123 rb_link_node(&block_group->cache_node, parent, p);
124 rb_insert_color(&block_group->cache_node,
125 &info->block_group_cache_tree);
126 spin_unlock(&info->block_group_cache_lock);
132 * This will return the block group at or after bytenr if contains is 0, else
133 * it will return the block group that contains the bytenr
135 static struct btrfs_block_group_cache *
136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
139 struct btrfs_block_group_cache *cache, *ret = NULL;
143 spin_lock(&info->block_group_cache_lock);
144 n = info->block_group_cache_tree.rb_node;
147 cache = rb_entry(n, struct btrfs_block_group_cache,
149 end = cache->key.objectid + cache->key.offset - 1;
150 start = cache->key.objectid;
152 if (bytenr < start) {
153 if (!contains && (!ret || start < ret->key.objectid))
156 } else if (bytenr > start) {
157 if (contains && bytenr <= end) {
168 btrfs_get_block_group(ret);
169 spin_unlock(&info->block_group_cache_lock);
174 static int add_excluded_extent(struct btrfs_root *root,
175 u64 start, u64 num_bytes)
177 u64 end = start + num_bytes - 1;
178 set_extent_bits(&root->fs_info->freed_extents[0],
179 start, end, EXTENT_UPTODATE, GFP_NOFS);
180 set_extent_bits(&root->fs_info->freed_extents[1],
181 start, end, EXTENT_UPTODATE, GFP_NOFS);
185 static void free_excluded_extents(struct btrfs_root *root,
186 struct btrfs_block_group_cache *cache)
190 start = cache->key.objectid;
191 end = start + cache->key.offset - 1;
193 clear_extent_bits(&root->fs_info->freed_extents[0],
194 start, end, EXTENT_UPTODATE, GFP_NOFS);
195 clear_extent_bits(&root->fs_info->freed_extents[1],
196 start, end, EXTENT_UPTODATE, GFP_NOFS);
199 static int exclude_super_stripes(struct btrfs_root *root,
200 struct btrfs_block_group_cache *cache)
207 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
208 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
209 cache->bytes_super += stripe_len;
210 ret = add_excluded_extent(root, cache->key.objectid,
215 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
216 bytenr = btrfs_sb_offset(i);
217 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
218 cache->key.objectid, bytenr,
219 0, &logical, &nr, &stripe_len);
223 cache->bytes_super += stripe_len;
224 ret = add_excluded_extent(root, logical[nr],
234 static struct btrfs_caching_control *
235 get_caching_control(struct btrfs_block_group_cache *cache)
237 struct btrfs_caching_control *ctl;
239 spin_lock(&cache->lock);
240 if (cache->cached != BTRFS_CACHE_STARTED) {
241 spin_unlock(&cache->lock);
245 ctl = cache->caching_ctl;
246 atomic_inc(&ctl->count);
247 spin_unlock(&cache->lock);
251 static void put_caching_control(struct btrfs_caching_control *ctl)
253 if (atomic_dec_and_test(&ctl->count))
258 * this is only called by cache_block_group, since we could have freed extents
259 * we need to check the pinned_extents for any extents that can't be used yet
260 * since their free space will be released as soon as the transaction commits.
262 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
263 struct btrfs_fs_info *info, u64 start, u64 end)
265 u64 extent_start, extent_end, size, total_added = 0;
268 while (start < end) {
269 ret = find_first_extent_bit(info->pinned_extents, start,
270 &extent_start, &extent_end,
271 EXTENT_DIRTY | EXTENT_UPTODATE);
275 if (extent_start <= start) {
276 start = extent_end + 1;
277 } else if (extent_start > start && extent_start < end) {
278 size = extent_start - start;
280 ret = btrfs_add_free_space(block_group, start,
283 start = extent_end + 1;
292 ret = btrfs_add_free_space(block_group, start, size);
299 static int caching_kthread(void *data)
301 struct btrfs_block_group_cache *block_group = data;
302 struct btrfs_fs_info *fs_info = block_group->fs_info;
303 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
304 struct btrfs_root *extent_root = fs_info->extent_root;
305 struct btrfs_path *path;
306 struct extent_buffer *leaf;
307 struct btrfs_key key;
313 path = btrfs_alloc_path();
317 exclude_super_stripes(extent_root, block_group);
318 spin_lock(&block_group->space_info->lock);
319 block_group->space_info->bytes_readonly += block_group->bytes_super;
320 spin_unlock(&block_group->space_info->lock);
322 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
325 * We don't want to deadlock with somebody trying to allocate a new
326 * extent for the extent root while also trying to search the extent
327 * root to add free space. So we skip locking and search the commit
328 * root, since its read-only
330 path->skip_locking = 1;
331 path->search_commit_root = 1;
336 key.type = BTRFS_EXTENT_ITEM_KEY;
338 mutex_lock(&caching_ctl->mutex);
339 /* need to make sure the commit_root doesn't disappear */
340 down_read(&fs_info->extent_commit_sem);
342 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
346 leaf = path->nodes[0];
347 nritems = btrfs_header_nritems(leaf);
351 if (fs_info->closing > 1) {
356 if (path->slots[0] < nritems) {
357 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
359 ret = find_next_key(path, 0, &key);
363 caching_ctl->progress = last;
364 btrfs_release_path(extent_root, path);
365 up_read(&fs_info->extent_commit_sem);
366 mutex_unlock(&caching_ctl->mutex);
367 if (btrfs_transaction_in_commit(fs_info))
374 if (key.objectid < block_group->key.objectid) {
379 if (key.objectid >= block_group->key.objectid +
380 block_group->key.offset)
383 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
384 total_found += add_new_free_space(block_group,
387 last = key.objectid + key.offset;
389 if (total_found > (1024 * 1024 * 2)) {
391 wake_up(&caching_ctl->wait);
398 total_found += add_new_free_space(block_group, fs_info, last,
399 block_group->key.objectid +
400 block_group->key.offset);
401 caching_ctl->progress = (u64)-1;
403 spin_lock(&block_group->lock);
404 block_group->caching_ctl = NULL;
405 block_group->cached = BTRFS_CACHE_FINISHED;
406 spin_unlock(&block_group->lock);
409 btrfs_free_path(path);
410 up_read(&fs_info->extent_commit_sem);
412 free_excluded_extents(extent_root, block_group);
414 mutex_unlock(&caching_ctl->mutex);
415 wake_up(&caching_ctl->wait);
417 put_caching_control(caching_ctl);
418 atomic_dec(&block_group->space_info->caching_threads);
419 btrfs_put_block_group(block_group);
424 static int cache_block_group(struct btrfs_block_group_cache *cache)
426 struct btrfs_fs_info *fs_info = cache->fs_info;
427 struct btrfs_caching_control *caching_ctl;
428 struct task_struct *tsk;
432 if (cache->cached != BTRFS_CACHE_NO)
435 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
436 BUG_ON(!caching_ctl);
438 INIT_LIST_HEAD(&caching_ctl->list);
439 mutex_init(&caching_ctl->mutex);
440 init_waitqueue_head(&caching_ctl->wait);
441 caching_ctl->block_group = cache;
442 caching_ctl->progress = cache->key.objectid;
443 /* one for caching kthread, one for caching block group list */
444 atomic_set(&caching_ctl->count, 2);
446 spin_lock(&cache->lock);
447 if (cache->cached != BTRFS_CACHE_NO) {
448 spin_unlock(&cache->lock);
452 cache->caching_ctl = caching_ctl;
453 cache->cached = BTRFS_CACHE_STARTED;
454 spin_unlock(&cache->lock);
456 down_write(&fs_info->extent_commit_sem);
457 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
458 up_write(&fs_info->extent_commit_sem);
460 atomic_inc(&cache->space_info->caching_threads);
461 btrfs_get_block_group(cache);
463 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
464 cache->key.objectid);
467 printk(KERN_ERR "error running thread %d\n", ret);
475 * return the block group that starts at or after bytenr
477 static struct btrfs_block_group_cache *
478 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
480 struct btrfs_block_group_cache *cache;
482 cache = block_group_cache_tree_search(info, bytenr, 0);
488 * return the block group that contains the given bytenr
490 struct btrfs_block_group_cache *btrfs_lookup_block_group(
491 struct btrfs_fs_info *info,
494 struct btrfs_block_group_cache *cache;
496 cache = block_group_cache_tree_search(info, bytenr, 1);
501 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
504 struct list_head *head = &info->space_info;
505 struct btrfs_space_info *found;
507 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
508 BTRFS_BLOCK_GROUP_METADATA;
511 list_for_each_entry_rcu(found, head, list) {
512 if (found->flags == flags) {
522 * after adding space to the filesystem, we need to clear the full flags
523 * on all the space infos.
525 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
527 struct list_head *head = &info->space_info;
528 struct btrfs_space_info *found;
531 list_for_each_entry_rcu(found, head, list)
536 static u64 div_factor(u64 num, int factor)
545 u64 btrfs_find_block_group(struct btrfs_root *root,
546 u64 search_start, u64 search_hint, int owner)
548 struct btrfs_block_group_cache *cache;
550 u64 last = max(search_hint, search_start);
557 cache = btrfs_lookup_first_block_group(root->fs_info, last);
561 spin_lock(&cache->lock);
562 last = cache->key.objectid + cache->key.offset;
563 used = btrfs_block_group_used(&cache->item);
565 if ((full_search || !cache->ro) &&
566 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
567 if (used + cache->pinned + cache->reserved <
568 div_factor(cache->key.offset, factor)) {
569 group_start = cache->key.objectid;
570 spin_unlock(&cache->lock);
571 btrfs_put_block_group(cache);
575 spin_unlock(&cache->lock);
576 btrfs_put_block_group(cache);
584 if (!full_search && factor < 10) {
594 /* simple helper to search for an existing extent at a given offset */
595 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
598 struct btrfs_key key;
599 struct btrfs_path *path;
601 path = btrfs_alloc_path();
603 key.objectid = start;
605 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
606 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
608 btrfs_free_path(path);
613 * helper function to lookup reference count and flags of extent.
615 * the head node for delayed ref is used to store the sum of all the
616 * reference count modifications queued up in the rbtree. the head
617 * node may also store the extent flags to set. This way you can check
618 * to see what the reference count and extent flags would be if all of
619 * the delayed refs are not processed.
621 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root, u64 bytenr,
623 u64 num_bytes, u64 *refs, u64 *flags)
625 struct btrfs_delayed_ref_head *head;
626 struct btrfs_delayed_ref_root *delayed_refs;
627 struct btrfs_path *path;
628 struct btrfs_extent_item *ei;
629 struct extent_buffer *leaf;
630 struct btrfs_key key;
636 path = btrfs_alloc_path();
640 key.objectid = bytenr;
641 key.type = BTRFS_EXTENT_ITEM_KEY;
642 key.offset = num_bytes;
644 path->skip_locking = 1;
645 path->search_commit_root = 1;
648 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
654 leaf = path->nodes[0];
655 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
656 if (item_size >= sizeof(*ei)) {
657 ei = btrfs_item_ptr(leaf, path->slots[0],
658 struct btrfs_extent_item);
659 num_refs = btrfs_extent_refs(leaf, ei);
660 extent_flags = btrfs_extent_flags(leaf, ei);
662 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
663 struct btrfs_extent_item_v0 *ei0;
664 BUG_ON(item_size != sizeof(*ei0));
665 ei0 = btrfs_item_ptr(leaf, path->slots[0],
666 struct btrfs_extent_item_v0);
667 num_refs = btrfs_extent_refs_v0(leaf, ei0);
668 /* FIXME: this isn't correct for data */
669 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
674 BUG_ON(num_refs == 0);
684 delayed_refs = &trans->transaction->delayed_refs;
685 spin_lock(&delayed_refs->lock);
686 head = btrfs_find_delayed_ref_head(trans, bytenr);
688 if (!mutex_trylock(&head->mutex)) {
689 atomic_inc(&head->node.refs);
690 spin_unlock(&delayed_refs->lock);
692 btrfs_release_path(root->fs_info->extent_root, path);
694 mutex_lock(&head->mutex);
695 mutex_unlock(&head->mutex);
696 btrfs_put_delayed_ref(&head->node);
699 if (head->extent_op && head->extent_op->update_flags)
700 extent_flags |= head->extent_op->flags_to_set;
702 BUG_ON(num_refs == 0);
704 num_refs += head->node.ref_mod;
705 mutex_unlock(&head->mutex);
707 spin_unlock(&delayed_refs->lock);
709 WARN_ON(num_refs == 0);
713 *flags = extent_flags;
715 btrfs_free_path(path);
720 * Back reference rules. Back refs have three main goals:
722 * 1) differentiate between all holders of references to an extent so that
723 * when a reference is dropped we can make sure it was a valid reference
724 * before freeing the extent.
726 * 2) Provide enough information to quickly find the holders of an extent
727 * if we notice a given block is corrupted or bad.
729 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
730 * maintenance. This is actually the same as #2, but with a slightly
731 * different use case.
733 * There are two kinds of back refs. The implicit back refs is optimized
734 * for pointers in non-shared tree blocks. For a given pointer in a block,
735 * back refs of this kind provide information about the block's owner tree
736 * and the pointer's key. These information allow us to find the block by
737 * b-tree searching. The full back refs is for pointers in tree blocks not
738 * referenced by their owner trees. The location of tree block is recorded
739 * in the back refs. Actually the full back refs is generic, and can be
740 * used in all cases the implicit back refs is used. The major shortcoming
741 * of the full back refs is its overhead. Every time a tree block gets
742 * COWed, we have to update back refs entry for all pointers in it.
744 * For a newly allocated tree block, we use implicit back refs for
745 * pointers in it. This means most tree related operations only involve
746 * implicit back refs. For a tree block created in old transaction, the
747 * only way to drop a reference to it is COW it. So we can detect the
748 * event that tree block loses its owner tree's reference and do the
749 * back refs conversion.
751 * When a tree block is COW'd through a tree, there are four cases:
753 * The reference count of the block is one and the tree is the block's
754 * owner tree. Nothing to do in this case.
756 * The reference count of the block is one and the tree is not the
757 * block's owner tree. In this case, full back refs is used for pointers
758 * in the block. Remove these full back refs, add implicit back refs for
759 * every pointers in the new block.
761 * The reference count of the block is greater than one and the tree is
762 * the block's owner tree. In this case, implicit back refs is used for
763 * pointers in the block. Add full back refs for every pointers in the
764 * block, increase lower level extents' reference counts. The original
765 * implicit back refs are entailed to the new block.
767 * The reference count of the block is greater than one and the tree is
768 * not the block's owner tree. Add implicit back refs for every pointer in
769 * the new block, increase lower level extents' reference count.
771 * Back Reference Key composing:
773 * The key objectid corresponds to the first byte in the extent,
774 * The key type is used to differentiate between types of back refs.
775 * There are different meanings of the key offset for different types
778 * File extents can be referenced by:
780 * - multiple snapshots, subvolumes, or different generations in one subvol
781 * - different files inside a single subvolume
782 * - different offsets inside a file (bookend extents in file.c)
784 * The extent ref structure for the implicit back refs has fields for:
786 * - Objectid of the subvolume root
787 * - objectid of the file holding the reference
788 * - original offset in the file
789 * - how many bookend extents
791 * The key offset for the implicit back refs is hash of the first
794 * The extent ref structure for the full back refs has field for:
796 * - number of pointers in the tree leaf
798 * The key offset for the implicit back refs is the first byte of
801 * When a file extent is allocated, The implicit back refs is used.
802 * the fields are filled in:
804 * (root_key.objectid, inode objectid, offset in file, 1)
806 * When a file extent is removed file truncation, we find the
807 * corresponding implicit back refs and check the following fields:
809 * (btrfs_header_owner(leaf), inode objectid, offset in file)
811 * Btree extents can be referenced by:
813 * - Different subvolumes
815 * Both the implicit back refs and the full back refs for tree blocks
816 * only consist of key. The key offset for the implicit back refs is
817 * objectid of block's owner tree. The key offset for the full back refs
818 * is the first byte of parent block.
820 * When implicit back refs is used, information about the lowest key and
821 * level of the tree block are required. These information are stored in
822 * tree block info structure.
825 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
826 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct btrfs_path *path,
829 u64 owner, u32 extra_size)
831 struct btrfs_extent_item *item;
832 struct btrfs_extent_item_v0 *ei0;
833 struct btrfs_extent_ref_v0 *ref0;
834 struct btrfs_tree_block_info *bi;
835 struct extent_buffer *leaf;
836 struct btrfs_key key;
837 struct btrfs_key found_key;
838 u32 new_size = sizeof(*item);
842 leaf = path->nodes[0];
843 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
845 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
846 ei0 = btrfs_item_ptr(leaf, path->slots[0],
847 struct btrfs_extent_item_v0);
848 refs = btrfs_extent_refs_v0(leaf, ei0);
850 if (owner == (u64)-1) {
852 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
853 ret = btrfs_next_leaf(root, path);
857 leaf = path->nodes[0];
859 btrfs_item_key_to_cpu(leaf, &found_key,
861 BUG_ON(key.objectid != found_key.objectid);
862 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
866 ref0 = btrfs_item_ptr(leaf, path->slots[0],
867 struct btrfs_extent_ref_v0);
868 owner = btrfs_ref_objectid_v0(leaf, ref0);
872 btrfs_release_path(root, path);
874 if (owner < BTRFS_FIRST_FREE_OBJECTID)
875 new_size += sizeof(*bi);
877 new_size -= sizeof(*ei0);
878 ret = btrfs_search_slot(trans, root, &key, path,
879 new_size + extra_size, 1);
884 ret = btrfs_extend_item(trans, root, path, new_size);
887 leaf = path->nodes[0];
888 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
889 btrfs_set_extent_refs(leaf, item, refs);
890 /* FIXME: get real generation */
891 btrfs_set_extent_generation(leaf, item, 0);
892 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
893 btrfs_set_extent_flags(leaf, item,
894 BTRFS_EXTENT_FLAG_TREE_BLOCK |
895 BTRFS_BLOCK_FLAG_FULL_BACKREF);
896 bi = (struct btrfs_tree_block_info *)(item + 1);
897 /* FIXME: get first key of the block */
898 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
899 btrfs_set_tree_block_level(leaf, bi, (int)owner);
901 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
903 btrfs_mark_buffer_dirty(leaf);
908 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
910 u32 high_crc = ~(u32)0;
911 u32 low_crc = ~(u32)0;
914 lenum = cpu_to_le64(root_objectid);
915 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
916 lenum = cpu_to_le64(owner);
917 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
918 lenum = cpu_to_le64(offset);
919 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
921 return ((u64)high_crc << 31) ^ (u64)low_crc;
924 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
925 struct btrfs_extent_data_ref *ref)
927 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
928 btrfs_extent_data_ref_objectid(leaf, ref),
929 btrfs_extent_data_ref_offset(leaf, ref));
932 static int match_extent_data_ref(struct extent_buffer *leaf,
933 struct btrfs_extent_data_ref *ref,
934 u64 root_objectid, u64 owner, u64 offset)
936 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
937 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
938 btrfs_extent_data_ref_offset(leaf, ref) != offset)
943 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
944 struct btrfs_root *root,
945 struct btrfs_path *path,
946 u64 bytenr, u64 parent,
948 u64 owner, u64 offset)
950 struct btrfs_key key;
951 struct btrfs_extent_data_ref *ref;
952 struct extent_buffer *leaf;
958 key.objectid = bytenr;
960 key.type = BTRFS_SHARED_DATA_REF_KEY;
963 key.type = BTRFS_EXTENT_DATA_REF_KEY;
964 key.offset = hash_extent_data_ref(root_objectid,
969 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 key.type = BTRFS_EXTENT_REF_V0_KEY;
980 btrfs_release_path(root, path);
981 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
992 leaf = path->nodes[0];
993 nritems = btrfs_header_nritems(leaf);
995 if (path->slots[0] >= nritems) {
996 ret = btrfs_next_leaf(root, path);
1002 leaf = path->nodes[0];
1003 nritems = btrfs_header_nritems(leaf);
1007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1008 if (key.objectid != bytenr ||
1009 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1012 ref = btrfs_item_ptr(leaf, path->slots[0],
1013 struct btrfs_extent_data_ref);
1015 if (match_extent_data_ref(leaf, ref, root_objectid,
1018 btrfs_release_path(root, path);
1030 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1031 struct btrfs_root *root,
1032 struct btrfs_path *path,
1033 u64 bytenr, u64 parent,
1034 u64 root_objectid, u64 owner,
1035 u64 offset, int refs_to_add)
1037 struct btrfs_key key;
1038 struct extent_buffer *leaf;
1043 key.objectid = bytenr;
1045 key.type = BTRFS_SHARED_DATA_REF_KEY;
1046 key.offset = parent;
1047 size = sizeof(struct btrfs_shared_data_ref);
1049 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1050 key.offset = hash_extent_data_ref(root_objectid,
1052 size = sizeof(struct btrfs_extent_data_ref);
1055 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1056 if (ret && ret != -EEXIST)
1059 leaf = path->nodes[0];
1061 struct btrfs_shared_data_ref *ref;
1062 ref = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_shared_data_ref);
1065 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1067 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1068 num_refs += refs_to_add;
1069 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1072 struct btrfs_extent_data_ref *ref;
1073 while (ret == -EEXIST) {
1074 ref = btrfs_item_ptr(leaf, path->slots[0],
1075 struct btrfs_extent_data_ref);
1076 if (match_extent_data_ref(leaf, ref, root_objectid,
1079 btrfs_release_path(root, path);
1081 ret = btrfs_insert_empty_item(trans, root, path, &key,
1083 if (ret && ret != -EEXIST)
1086 leaf = path->nodes[0];
1088 ref = btrfs_item_ptr(leaf, path->slots[0],
1089 struct btrfs_extent_data_ref);
1091 btrfs_set_extent_data_ref_root(leaf, ref,
1093 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1094 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1095 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1097 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1098 num_refs += refs_to_add;
1099 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1102 btrfs_mark_buffer_dirty(leaf);
1105 btrfs_release_path(root, path);
1109 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root,
1111 struct btrfs_path *path,
1114 struct btrfs_key key;
1115 struct btrfs_extent_data_ref *ref1 = NULL;
1116 struct btrfs_shared_data_ref *ref2 = NULL;
1117 struct extent_buffer *leaf;
1121 leaf = path->nodes[0];
1122 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1124 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1125 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_extent_data_ref);
1127 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1128 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1129 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1130 struct btrfs_shared_data_ref);
1131 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1132 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1133 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1134 struct btrfs_extent_ref_v0 *ref0;
1135 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1136 struct btrfs_extent_ref_v0);
1137 num_refs = btrfs_ref_count_v0(leaf, ref0);
1143 BUG_ON(num_refs < refs_to_drop);
1144 num_refs -= refs_to_drop;
1146 if (num_refs == 0) {
1147 ret = btrfs_del_item(trans, root, path);
1149 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1150 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1151 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1152 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1155 struct btrfs_extent_ref_v0 *ref0;
1156 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1157 struct btrfs_extent_ref_v0);
1158 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1161 btrfs_mark_buffer_dirty(leaf);
1166 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1167 struct btrfs_path *path,
1168 struct btrfs_extent_inline_ref *iref)
1170 struct btrfs_key key;
1171 struct extent_buffer *leaf;
1172 struct btrfs_extent_data_ref *ref1;
1173 struct btrfs_shared_data_ref *ref2;
1176 leaf = path->nodes[0];
1177 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1179 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1180 BTRFS_EXTENT_DATA_REF_KEY) {
1181 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1182 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1184 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1185 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1187 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1188 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1189 struct btrfs_extent_data_ref);
1190 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1191 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1192 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_shared_data_ref);
1194 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1196 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1197 struct btrfs_extent_ref_v0 *ref0;
1198 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1199 struct btrfs_extent_ref_v0);
1200 num_refs = btrfs_ref_count_v0(leaf, ref0);
1208 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root,
1210 struct btrfs_path *path,
1211 u64 bytenr, u64 parent,
1214 struct btrfs_key key;
1217 key.objectid = bytenr;
1219 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1220 key.offset = parent;
1222 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1223 key.offset = root_objectid;
1226 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1229 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1230 if (ret == -ENOENT && parent) {
1231 btrfs_release_path(root, path);
1232 key.type = BTRFS_EXTENT_REF_V0_KEY;
1233 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1241 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1242 struct btrfs_root *root,
1243 struct btrfs_path *path,
1244 u64 bytenr, u64 parent,
1247 struct btrfs_key key;
1250 key.objectid = bytenr;
1252 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1253 key.offset = parent;
1255 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1256 key.offset = root_objectid;
1259 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1260 btrfs_release_path(root, path);
1264 static inline int extent_ref_type(u64 parent, u64 owner)
1267 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1269 type = BTRFS_SHARED_BLOCK_REF_KEY;
1271 type = BTRFS_TREE_BLOCK_REF_KEY;
1274 type = BTRFS_SHARED_DATA_REF_KEY;
1276 type = BTRFS_EXTENT_DATA_REF_KEY;
1281 static int find_next_key(struct btrfs_path *path, int level,
1282 struct btrfs_key *key)
1285 for (; level < BTRFS_MAX_LEVEL; level++) {
1286 if (!path->nodes[level])
1288 if (path->slots[level] + 1 >=
1289 btrfs_header_nritems(path->nodes[level]))
1292 btrfs_item_key_to_cpu(path->nodes[level], key,
1293 path->slots[level] + 1);
1295 btrfs_node_key_to_cpu(path->nodes[level], key,
1296 path->slots[level] + 1);
1303 * look for inline back ref. if back ref is found, *ref_ret is set
1304 * to the address of inline back ref, and 0 is returned.
1306 * if back ref isn't found, *ref_ret is set to the address where it
1307 * should be inserted, and -ENOENT is returned.
1309 * if insert is true and there are too many inline back refs, the path
1310 * points to the extent item, and -EAGAIN is returned.
1312 * NOTE: inline back refs are ordered in the same way that back ref
1313 * items in the tree are ordered.
1315 static noinline_for_stack
1316 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1317 struct btrfs_root *root,
1318 struct btrfs_path *path,
1319 struct btrfs_extent_inline_ref **ref_ret,
1320 u64 bytenr, u64 num_bytes,
1321 u64 parent, u64 root_objectid,
1322 u64 owner, u64 offset, int insert)
1324 struct btrfs_key key;
1325 struct extent_buffer *leaf;
1326 struct btrfs_extent_item *ei;
1327 struct btrfs_extent_inline_ref *iref;
1338 key.objectid = bytenr;
1339 key.type = BTRFS_EXTENT_ITEM_KEY;
1340 key.offset = num_bytes;
1342 want = extent_ref_type(parent, owner);
1344 extra_size = btrfs_extent_inline_ref_size(want);
1345 path->keep_locks = 1;
1348 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1355 leaf = path->nodes[0];
1356 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (item_size < sizeof(*ei)) {
1363 ret = convert_extent_item_v0(trans, root, path, owner,
1369 leaf = path->nodes[0];
1370 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1373 BUG_ON(item_size < sizeof(*ei));
1375 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1376 flags = btrfs_extent_flags(leaf, ei);
1378 ptr = (unsigned long)(ei + 1);
1379 end = (unsigned long)ei + item_size;
1381 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1382 ptr += sizeof(struct btrfs_tree_block_info);
1385 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1394 iref = (struct btrfs_extent_inline_ref *)ptr;
1395 type = btrfs_extent_inline_ref_type(leaf, iref);
1399 ptr += btrfs_extent_inline_ref_size(type);
1403 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 struct btrfs_extent_data_ref *dref;
1405 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1406 if (match_extent_data_ref(leaf, dref, root_objectid,
1411 if (hash_extent_data_ref_item(leaf, dref) <
1412 hash_extent_data_ref(root_objectid, owner, offset))
1416 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1418 if (parent == ref_offset) {
1422 if (ref_offset < parent)
1425 if (root_objectid == ref_offset) {
1429 if (ref_offset < root_objectid)
1433 ptr += btrfs_extent_inline_ref_size(type);
1435 if (err == -ENOENT && insert) {
1436 if (item_size + extra_size >=
1437 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1442 * To add new inline back ref, we have to make sure
1443 * there is no corresponding back ref item.
1444 * For simplicity, we just do not add new inline back
1445 * ref if there is any kind of item for this block
1447 if (find_next_key(path, 0, &key) == 0 &&
1448 key.objectid == bytenr &&
1449 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1454 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1457 path->keep_locks = 0;
1458 btrfs_unlock_up_safe(path, 1);
1464 * helper to add new inline back ref
1466 static noinline_for_stack
1467 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root,
1469 struct btrfs_path *path,
1470 struct btrfs_extent_inline_ref *iref,
1471 u64 parent, u64 root_objectid,
1472 u64 owner, u64 offset, int refs_to_add,
1473 struct btrfs_delayed_extent_op *extent_op)
1475 struct extent_buffer *leaf;
1476 struct btrfs_extent_item *ei;
1479 unsigned long item_offset;
1485 leaf = path->nodes[0];
1486 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1487 item_offset = (unsigned long)iref - (unsigned long)ei;
1489 type = extent_ref_type(parent, owner);
1490 size = btrfs_extent_inline_ref_size(type);
1492 ret = btrfs_extend_item(trans, root, path, size);
1495 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1496 refs = btrfs_extent_refs(leaf, ei);
1497 refs += refs_to_add;
1498 btrfs_set_extent_refs(leaf, ei, refs);
1500 __run_delayed_extent_op(extent_op, leaf, ei);
1502 ptr = (unsigned long)ei + item_offset;
1503 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1504 if (ptr < end - size)
1505 memmove_extent_buffer(leaf, ptr + size, ptr,
1508 iref = (struct btrfs_extent_inline_ref *)ptr;
1509 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1510 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1511 struct btrfs_extent_data_ref *dref;
1512 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1513 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1514 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1515 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1516 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1517 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1518 struct btrfs_shared_data_ref *sref;
1519 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1520 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1521 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1522 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1523 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1525 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1527 btrfs_mark_buffer_dirty(leaf);
1531 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1532 struct btrfs_root *root,
1533 struct btrfs_path *path,
1534 struct btrfs_extent_inline_ref **ref_ret,
1535 u64 bytenr, u64 num_bytes, u64 parent,
1536 u64 root_objectid, u64 owner, u64 offset)
1540 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1541 bytenr, num_bytes, parent,
1542 root_objectid, owner, offset, 0);
1546 btrfs_release_path(root, path);
1549 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1550 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1553 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1554 root_objectid, owner, offset);
1560 * helper to update/remove inline back ref
1562 static noinline_for_stack
1563 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1564 struct btrfs_root *root,
1565 struct btrfs_path *path,
1566 struct btrfs_extent_inline_ref *iref,
1568 struct btrfs_delayed_extent_op *extent_op)
1570 struct extent_buffer *leaf;
1571 struct btrfs_extent_item *ei;
1572 struct btrfs_extent_data_ref *dref = NULL;
1573 struct btrfs_shared_data_ref *sref = NULL;
1582 leaf = path->nodes[0];
1583 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1584 refs = btrfs_extent_refs(leaf, ei);
1585 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1586 refs += refs_to_mod;
1587 btrfs_set_extent_refs(leaf, ei, refs);
1589 __run_delayed_extent_op(extent_op, leaf, ei);
1591 type = btrfs_extent_inline_ref_type(leaf, iref);
1593 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1594 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1595 refs = btrfs_extent_data_ref_count(leaf, dref);
1596 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1597 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1598 refs = btrfs_shared_data_ref_count(leaf, sref);
1601 BUG_ON(refs_to_mod != -1);
1604 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1605 refs += refs_to_mod;
1608 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1609 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1611 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1613 size = btrfs_extent_inline_ref_size(type);
1614 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1615 ptr = (unsigned long)iref;
1616 end = (unsigned long)ei + item_size;
1617 if (ptr + size < end)
1618 memmove_extent_buffer(leaf, ptr, ptr + size,
1621 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1624 btrfs_mark_buffer_dirty(leaf);
1628 static noinline_for_stack
1629 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1630 struct btrfs_root *root,
1631 struct btrfs_path *path,
1632 u64 bytenr, u64 num_bytes, u64 parent,
1633 u64 root_objectid, u64 owner,
1634 u64 offset, int refs_to_add,
1635 struct btrfs_delayed_extent_op *extent_op)
1637 struct btrfs_extent_inline_ref *iref;
1640 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1641 bytenr, num_bytes, parent,
1642 root_objectid, owner, offset, 1);
1644 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1645 ret = update_inline_extent_backref(trans, root, path, iref,
1646 refs_to_add, extent_op);
1647 } else if (ret == -ENOENT) {
1648 ret = setup_inline_extent_backref(trans, root, path, iref,
1649 parent, root_objectid,
1650 owner, offset, refs_to_add,
1656 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1657 struct btrfs_root *root,
1658 struct btrfs_path *path,
1659 u64 bytenr, u64 parent, u64 root_objectid,
1660 u64 owner, u64 offset, int refs_to_add)
1663 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1664 BUG_ON(refs_to_add != 1);
1665 ret = insert_tree_block_ref(trans, root, path, bytenr,
1666 parent, root_objectid);
1668 ret = insert_extent_data_ref(trans, root, path, bytenr,
1669 parent, root_objectid,
1670 owner, offset, refs_to_add);
1675 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1676 struct btrfs_root *root,
1677 struct btrfs_path *path,
1678 struct btrfs_extent_inline_ref *iref,
1679 int refs_to_drop, int is_data)
1683 BUG_ON(!is_data && refs_to_drop != 1);
1685 ret = update_inline_extent_backref(trans, root, path, iref,
1686 -refs_to_drop, NULL);
1687 } else if (is_data) {
1688 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1690 ret = btrfs_del_item(trans, root, path);
1695 static void btrfs_issue_discard(struct block_device *bdev,
1698 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1699 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1702 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1706 u64 map_length = num_bytes;
1707 struct btrfs_multi_bio *multi = NULL;
1709 if (!btrfs_test_opt(root, DISCARD))
1712 /* Tell the block device(s) that the sectors can be discarded */
1713 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1714 bytenr, &map_length, &multi, 0);
1716 struct btrfs_bio_stripe *stripe = multi->stripes;
1719 if (map_length > num_bytes)
1720 map_length = num_bytes;
1722 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1723 btrfs_issue_discard(stripe->dev->bdev,
1733 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1734 struct btrfs_root *root,
1735 u64 bytenr, u64 num_bytes, u64 parent,
1736 u64 root_objectid, u64 owner, u64 offset)
1739 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1740 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1742 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1743 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1744 parent, root_objectid, (int)owner,
1745 BTRFS_ADD_DELAYED_REF, NULL);
1747 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1748 parent, root_objectid, owner, offset,
1749 BTRFS_ADD_DELAYED_REF, NULL);
1754 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1755 struct btrfs_root *root,
1756 u64 bytenr, u64 num_bytes,
1757 u64 parent, u64 root_objectid,
1758 u64 owner, u64 offset, int refs_to_add,
1759 struct btrfs_delayed_extent_op *extent_op)
1761 struct btrfs_path *path;
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *item;
1768 path = btrfs_alloc_path();
1773 path->leave_spinning = 1;
1774 /* this will setup the path even if it fails to insert the back ref */
1775 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1776 path, bytenr, num_bytes, parent,
1777 root_objectid, owner, offset,
1778 refs_to_add, extent_op);
1782 if (ret != -EAGAIN) {
1787 leaf = path->nodes[0];
1788 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1789 refs = btrfs_extent_refs(leaf, item);
1790 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1792 __run_delayed_extent_op(extent_op, leaf, item);
1794 btrfs_mark_buffer_dirty(leaf);
1795 btrfs_release_path(root->fs_info->extent_root, path);
1798 path->leave_spinning = 1;
1800 /* now insert the actual backref */
1801 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1802 path, bytenr, parent, root_objectid,
1803 owner, offset, refs_to_add);
1806 btrfs_free_path(path);
1810 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1811 struct btrfs_root *root,
1812 struct btrfs_delayed_ref_node *node,
1813 struct btrfs_delayed_extent_op *extent_op,
1814 int insert_reserved)
1817 struct btrfs_delayed_data_ref *ref;
1818 struct btrfs_key ins;
1823 ins.objectid = node->bytenr;
1824 ins.offset = node->num_bytes;
1825 ins.type = BTRFS_EXTENT_ITEM_KEY;
1827 ref = btrfs_delayed_node_to_data_ref(node);
1828 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1829 parent = ref->parent;
1831 ref_root = ref->root;
1833 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1835 BUG_ON(extent_op->update_key);
1836 flags |= extent_op->flags_to_set;
1838 ret = alloc_reserved_file_extent(trans, root,
1839 parent, ref_root, flags,
1840 ref->objectid, ref->offset,
1841 &ins, node->ref_mod);
1842 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1843 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1844 node->num_bytes, parent,
1845 ref_root, ref->objectid,
1846 ref->offset, node->ref_mod,
1848 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1849 ret = __btrfs_free_extent(trans, root, node->bytenr,
1850 node->num_bytes, parent,
1851 ref_root, ref->objectid,
1852 ref->offset, node->ref_mod,
1860 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1861 struct extent_buffer *leaf,
1862 struct btrfs_extent_item *ei)
1864 u64 flags = btrfs_extent_flags(leaf, ei);
1865 if (extent_op->update_flags) {
1866 flags |= extent_op->flags_to_set;
1867 btrfs_set_extent_flags(leaf, ei, flags);
1870 if (extent_op->update_key) {
1871 struct btrfs_tree_block_info *bi;
1872 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1873 bi = (struct btrfs_tree_block_info *)(ei + 1);
1874 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1878 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1879 struct btrfs_root *root,
1880 struct btrfs_delayed_ref_node *node,
1881 struct btrfs_delayed_extent_op *extent_op)
1883 struct btrfs_key key;
1884 struct btrfs_path *path;
1885 struct btrfs_extent_item *ei;
1886 struct extent_buffer *leaf;
1891 path = btrfs_alloc_path();
1895 key.objectid = node->bytenr;
1896 key.type = BTRFS_EXTENT_ITEM_KEY;
1897 key.offset = node->num_bytes;
1900 path->leave_spinning = 1;
1901 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1912 leaf = path->nodes[0];
1913 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1914 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1915 if (item_size < sizeof(*ei)) {
1916 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1922 leaf = path->nodes[0];
1923 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1926 BUG_ON(item_size < sizeof(*ei));
1927 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1928 __run_delayed_extent_op(extent_op, leaf, ei);
1930 btrfs_mark_buffer_dirty(leaf);
1932 btrfs_free_path(path);
1936 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1937 struct btrfs_root *root,
1938 struct btrfs_delayed_ref_node *node,
1939 struct btrfs_delayed_extent_op *extent_op,
1940 int insert_reserved)
1943 struct btrfs_delayed_tree_ref *ref;
1944 struct btrfs_key ins;
1948 ins.objectid = node->bytenr;
1949 ins.offset = node->num_bytes;
1950 ins.type = BTRFS_EXTENT_ITEM_KEY;
1952 ref = btrfs_delayed_node_to_tree_ref(node);
1953 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1954 parent = ref->parent;
1956 ref_root = ref->root;
1958 BUG_ON(node->ref_mod != 1);
1959 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1960 BUG_ON(!extent_op || !extent_op->update_flags ||
1961 !extent_op->update_key);
1962 ret = alloc_reserved_tree_block(trans, root,
1964 extent_op->flags_to_set,
1967 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1968 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1969 node->num_bytes, parent, ref_root,
1970 ref->level, 0, 1, extent_op);
1971 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1972 ret = __btrfs_free_extent(trans, root, node->bytenr,
1973 node->num_bytes, parent, ref_root,
1974 ref->level, 0, 1, extent_op);
1981 /* helper function to actually process a single delayed ref entry */
1982 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1983 struct btrfs_root *root,
1984 struct btrfs_delayed_ref_node *node,
1985 struct btrfs_delayed_extent_op *extent_op,
1986 int insert_reserved)
1989 if (btrfs_delayed_ref_is_head(node)) {
1990 struct btrfs_delayed_ref_head *head;
1992 * we've hit the end of the chain and we were supposed
1993 * to insert this extent into the tree. But, it got
1994 * deleted before we ever needed to insert it, so all
1995 * we have to do is clean up the accounting
1998 head = btrfs_delayed_node_to_head(node);
1999 if (insert_reserved) {
2000 btrfs_pin_extent(root, node->bytenr,
2001 node->num_bytes, 1);
2002 if (head->is_data) {
2003 ret = btrfs_del_csums(trans, root,
2009 mutex_unlock(&head->mutex);
2013 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2014 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2015 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2017 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2018 node->type == BTRFS_SHARED_DATA_REF_KEY)
2019 ret = run_delayed_data_ref(trans, root, node, extent_op,
2026 static noinline struct btrfs_delayed_ref_node *
2027 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2029 struct rb_node *node;
2030 struct btrfs_delayed_ref_node *ref;
2031 int action = BTRFS_ADD_DELAYED_REF;
2034 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2035 * this prevents ref count from going down to zero when
2036 * there still are pending delayed ref.
2038 node = rb_prev(&head->node.rb_node);
2042 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2044 if (ref->bytenr != head->node.bytenr)
2046 if (ref->action == action)
2048 node = rb_prev(node);
2050 if (action == BTRFS_ADD_DELAYED_REF) {
2051 action = BTRFS_DROP_DELAYED_REF;
2057 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2058 struct btrfs_root *root,
2059 struct list_head *cluster)
2061 struct btrfs_delayed_ref_root *delayed_refs;
2062 struct btrfs_delayed_ref_node *ref;
2063 struct btrfs_delayed_ref_head *locked_ref = NULL;
2064 struct btrfs_delayed_extent_op *extent_op;
2067 int must_insert_reserved = 0;
2069 delayed_refs = &trans->transaction->delayed_refs;
2072 /* pick a new head ref from the cluster list */
2073 if (list_empty(cluster))
2076 locked_ref = list_entry(cluster->next,
2077 struct btrfs_delayed_ref_head, cluster);
2079 /* grab the lock that says we are going to process
2080 * all the refs for this head */
2081 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2084 * we may have dropped the spin lock to get the head
2085 * mutex lock, and that might have given someone else
2086 * time to free the head. If that's true, it has been
2087 * removed from our list and we can move on.
2089 if (ret == -EAGAIN) {
2097 * record the must insert reserved flag before we
2098 * drop the spin lock.
2100 must_insert_reserved = locked_ref->must_insert_reserved;
2101 locked_ref->must_insert_reserved = 0;
2103 extent_op = locked_ref->extent_op;
2104 locked_ref->extent_op = NULL;
2107 * locked_ref is the head node, so we have to go one
2108 * node back for any delayed ref updates
2110 ref = select_delayed_ref(locked_ref);
2112 /* All delayed refs have been processed, Go ahead
2113 * and send the head node to run_one_delayed_ref,
2114 * so that any accounting fixes can happen
2116 ref = &locked_ref->node;
2118 if (extent_op && must_insert_reserved) {
2124 spin_unlock(&delayed_refs->lock);
2126 ret = run_delayed_extent_op(trans, root,
2132 spin_lock(&delayed_refs->lock);
2136 list_del_init(&locked_ref->cluster);
2141 rb_erase(&ref->rb_node, &delayed_refs->root);
2142 delayed_refs->num_entries--;
2144 spin_unlock(&delayed_refs->lock);
2146 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2147 must_insert_reserved);
2150 btrfs_put_delayed_ref(ref);
2155 spin_lock(&delayed_refs->lock);
2161 * this starts processing the delayed reference count updates and
2162 * extent insertions we have queued up so far. count can be
2163 * 0, which means to process everything in the tree at the start
2164 * of the run (but not newly added entries), or it can be some target
2165 * number you'd like to process.
2167 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2168 struct btrfs_root *root, unsigned long count)
2170 struct rb_node *node;
2171 struct btrfs_delayed_ref_root *delayed_refs;
2172 struct btrfs_delayed_ref_node *ref;
2173 struct list_head cluster;
2175 int run_all = count == (unsigned long)-1;
2178 if (root == root->fs_info->extent_root)
2179 root = root->fs_info->tree_root;
2181 delayed_refs = &trans->transaction->delayed_refs;
2182 INIT_LIST_HEAD(&cluster);
2184 spin_lock(&delayed_refs->lock);
2186 count = delayed_refs->num_entries * 2;
2190 if (!(run_all || run_most) &&
2191 delayed_refs->num_heads_ready < 64)
2195 * go find something we can process in the rbtree. We start at
2196 * the beginning of the tree, and then build a cluster
2197 * of refs to process starting at the first one we are able to
2200 ret = btrfs_find_ref_cluster(trans, &cluster,
2201 delayed_refs->run_delayed_start);
2205 ret = run_clustered_refs(trans, root, &cluster);
2208 count -= min_t(unsigned long, ret, count);
2215 node = rb_first(&delayed_refs->root);
2218 count = (unsigned long)-1;
2221 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2223 if (btrfs_delayed_ref_is_head(ref)) {
2224 struct btrfs_delayed_ref_head *head;
2226 head = btrfs_delayed_node_to_head(ref);
2227 atomic_inc(&ref->refs);
2229 spin_unlock(&delayed_refs->lock);
2230 mutex_lock(&head->mutex);
2231 mutex_unlock(&head->mutex);
2233 btrfs_put_delayed_ref(ref);
2237 node = rb_next(node);
2239 spin_unlock(&delayed_refs->lock);
2240 schedule_timeout(1);
2244 spin_unlock(&delayed_refs->lock);
2248 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2249 struct btrfs_root *root,
2250 u64 bytenr, u64 num_bytes, u64 flags,
2253 struct btrfs_delayed_extent_op *extent_op;
2256 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2260 extent_op->flags_to_set = flags;
2261 extent_op->update_flags = 1;
2262 extent_op->update_key = 0;
2263 extent_op->is_data = is_data ? 1 : 0;
2265 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2271 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2272 struct btrfs_root *root,
2273 struct btrfs_path *path,
2274 u64 objectid, u64 offset, u64 bytenr)
2276 struct btrfs_delayed_ref_head *head;
2277 struct btrfs_delayed_ref_node *ref;
2278 struct btrfs_delayed_data_ref *data_ref;
2279 struct btrfs_delayed_ref_root *delayed_refs;
2280 struct rb_node *node;
2284 delayed_refs = &trans->transaction->delayed_refs;
2285 spin_lock(&delayed_refs->lock);
2286 head = btrfs_find_delayed_ref_head(trans, bytenr);
2290 if (!mutex_trylock(&head->mutex)) {
2291 atomic_inc(&head->node.refs);
2292 spin_unlock(&delayed_refs->lock);
2294 btrfs_release_path(root->fs_info->extent_root, path);
2296 mutex_lock(&head->mutex);
2297 mutex_unlock(&head->mutex);
2298 btrfs_put_delayed_ref(&head->node);
2302 node = rb_prev(&head->node.rb_node);
2306 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2308 if (ref->bytenr != bytenr)
2312 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2315 data_ref = btrfs_delayed_node_to_data_ref(ref);
2317 node = rb_prev(node);
2319 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2320 if (ref->bytenr == bytenr)
2324 if (data_ref->root != root->root_key.objectid ||
2325 data_ref->objectid != objectid || data_ref->offset != offset)
2330 mutex_unlock(&head->mutex);
2332 spin_unlock(&delayed_refs->lock);
2336 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2337 struct btrfs_root *root,
2338 struct btrfs_path *path,
2339 u64 objectid, u64 offset, u64 bytenr)
2341 struct btrfs_root *extent_root = root->fs_info->extent_root;
2342 struct extent_buffer *leaf;
2343 struct btrfs_extent_data_ref *ref;
2344 struct btrfs_extent_inline_ref *iref;
2345 struct btrfs_extent_item *ei;
2346 struct btrfs_key key;
2350 key.objectid = bytenr;
2351 key.offset = (u64)-1;
2352 key.type = BTRFS_EXTENT_ITEM_KEY;
2354 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2360 if (path->slots[0] == 0)
2364 leaf = path->nodes[0];
2365 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2367 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2371 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2372 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2373 if (item_size < sizeof(*ei)) {
2374 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2378 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2380 if (item_size != sizeof(*ei) +
2381 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2384 if (btrfs_extent_generation(leaf, ei) <=
2385 btrfs_root_last_snapshot(&root->root_item))
2388 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2389 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2390 BTRFS_EXTENT_DATA_REF_KEY)
2393 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2394 if (btrfs_extent_refs(leaf, ei) !=
2395 btrfs_extent_data_ref_count(leaf, ref) ||
2396 btrfs_extent_data_ref_root(leaf, ref) !=
2397 root->root_key.objectid ||
2398 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2399 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2407 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2408 struct btrfs_root *root,
2409 u64 objectid, u64 offset, u64 bytenr)
2411 struct btrfs_path *path;
2415 path = btrfs_alloc_path();
2420 ret = check_committed_ref(trans, root, path, objectid,
2422 if (ret && ret != -ENOENT)
2425 ret2 = check_delayed_ref(trans, root, path, objectid,
2427 } while (ret2 == -EAGAIN);
2429 if (ret2 && ret2 != -ENOENT) {
2434 if (ret != -ENOENT || ret2 != -ENOENT)
2437 btrfs_free_path(path);
2438 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2444 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2445 struct extent_buffer *buf, u32 nr_extents)
2447 struct btrfs_key key;
2448 struct btrfs_file_extent_item *fi;
2456 if (!root->ref_cows)
2459 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2461 root_gen = root->root_key.offset;
2464 root_gen = trans->transid - 1;
2467 level = btrfs_header_level(buf);
2468 nritems = btrfs_header_nritems(buf);
2471 struct btrfs_leaf_ref *ref;
2472 struct btrfs_extent_info *info;
2474 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2480 ref->root_gen = root_gen;
2481 ref->bytenr = buf->start;
2482 ref->owner = btrfs_header_owner(buf);
2483 ref->generation = btrfs_header_generation(buf);
2484 ref->nritems = nr_extents;
2485 info = ref->extents;
2487 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2489 btrfs_item_key_to_cpu(buf, &key, i);
2490 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2492 fi = btrfs_item_ptr(buf, i,
2493 struct btrfs_file_extent_item);
2494 if (btrfs_file_extent_type(buf, fi) ==
2495 BTRFS_FILE_EXTENT_INLINE)
2497 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2498 if (disk_bytenr == 0)
2501 info->bytenr = disk_bytenr;
2503 btrfs_file_extent_disk_num_bytes(buf, fi);
2504 info->objectid = key.objectid;
2505 info->offset = key.offset;
2509 ret = btrfs_add_leaf_ref(root, ref, shared);
2510 if (ret == -EEXIST && shared) {
2511 struct btrfs_leaf_ref *old;
2512 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2514 btrfs_remove_leaf_ref(root, old);
2515 btrfs_free_leaf_ref(root, old);
2516 ret = btrfs_add_leaf_ref(root, ref, shared);
2519 btrfs_free_leaf_ref(root, ref);
2525 /* when a block goes through cow, we update the reference counts of
2526 * everything that block points to. The internal pointers of the block
2527 * can be in just about any order, and it is likely to have clusters of
2528 * things that are close together and clusters of things that are not.
2530 * To help reduce the seeks that come with updating all of these reference
2531 * counts, sort them by byte number before actual updates are done.
2533 * struct refsort is used to match byte number to slot in the btree block.
2534 * we sort based on the byte number and then use the slot to actually
2537 * struct refsort is smaller than strcut btrfs_item and smaller than
2538 * struct btrfs_key_ptr. Since we're currently limited to the page size
2539 * for a btree block, there's no way for a kmalloc of refsorts for a
2540 * single node to be bigger than a page.
2548 * for passing into sort()
2550 static int refsort_cmp(const void *a_void, const void *b_void)
2552 const struct refsort *a = a_void;
2553 const struct refsort *b = b_void;
2555 if (a->bytenr < b->bytenr)
2557 if (a->bytenr > b->bytenr)
2563 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2564 struct btrfs_root *root,
2565 struct extent_buffer *buf,
2566 int full_backref, int inc)
2573 struct btrfs_key key;
2574 struct btrfs_file_extent_item *fi;
2578 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2579 u64, u64, u64, u64, u64, u64);
2581 ref_root = btrfs_header_owner(buf);
2582 nritems = btrfs_header_nritems(buf);
2583 level = btrfs_header_level(buf);
2585 if (!root->ref_cows && level == 0)
2589 process_func = btrfs_inc_extent_ref;
2591 process_func = btrfs_free_extent;
2594 parent = buf->start;
2598 for (i = 0; i < nritems; i++) {
2600 btrfs_item_key_to_cpu(buf, &key, i);
2601 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2603 fi = btrfs_item_ptr(buf, i,
2604 struct btrfs_file_extent_item);
2605 if (btrfs_file_extent_type(buf, fi) ==
2606 BTRFS_FILE_EXTENT_INLINE)
2608 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2612 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2613 key.offset -= btrfs_file_extent_offset(buf, fi);
2614 ret = process_func(trans, root, bytenr, num_bytes,
2615 parent, ref_root, key.objectid,
2620 bytenr = btrfs_node_blockptr(buf, i);
2621 num_bytes = btrfs_level_size(root, level - 1);
2622 ret = process_func(trans, root, bytenr, num_bytes,
2623 parent, ref_root, level - 1, 0);
2634 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2635 struct extent_buffer *buf, int full_backref)
2637 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2640 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2641 struct extent_buffer *buf, int full_backref)
2643 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2646 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2647 struct btrfs_root *root,
2648 struct btrfs_path *path,
2649 struct btrfs_block_group_cache *cache)
2652 struct btrfs_root *extent_root = root->fs_info->extent_root;
2654 struct extent_buffer *leaf;
2656 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2661 leaf = path->nodes[0];
2662 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2663 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2664 btrfs_mark_buffer_dirty(leaf);
2665 btrfs_release_path(extent_root, path);
2673 static struct btrfs_block_group_cache *
2674 next_block_group(struct btrfs_root *root,
2675 struct btrfs_block_group_cache *cache)
2677 struct rb_node *node;
2678 spin_lock(&root->fs_info->block_group_cache_lock);
2679 node = rb_next(&cache->cache_node);
2680 btrfs_put_block_group(cache);
2682 cache = rb_entry(node, struct btrfs_block_group_cache,
2684 btrfs_get_block_group(cache);
2687 spin_unlock(&root->fs_info->block_group_cache_lock);
2691 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2692 struct btrfs_trans_handle *trans,
2693 struct btrfs_path *path)
2695 struct btrfs_root *root = block_group->fs_info->tree_root;
2696 struct inode *inode = NULL;
2703 * If this block group is smaller than 100 megs don't bother caching the
2706 if (block_group->key.offset < (100 * 1024 * 1024)) {
2707 spin_lock(&block_group->lock);
2708 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2709 spin_unlock(&block_group->lock);
2714 inode = lookup_free_space_inode(root, block_group, path);
2715 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2716 ret = PTR_ERR(inode);
2717 btrfs_release_path(root, path);
2721 if (IS_ERR(inode)) {
2725 if (block_group->ro)
2728 ret = create_free_space_inode(root, trans, block_group, path);
2735 * We want to set the generation to 0, that way if anything goes wrong
2736 * from here on out we know not to trust this cache when we load up next
2739 BTRFS_I(inode)->generation = 0;
2740 ret = btrfs_update_inode(trans, root, inode);
2743 if (i_size_read(inode) > 0) {
2744 ret = btrfs_truncate_free_space_cache(root, trans, path,
2750 spin_lock(&block_group->lock);
2751 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2752 spin_unlock(&block_group->lock);
2755 spin_unlock(&block_group->lock);
2757 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2762 * Just to make absolutely sure we have enough space, we're going to
2763 * preallocate 12 pages worth of space for each block group. In
2764 * practice we ought to use at most 8, but we need extra space so we can
2765 * add our header and have a terminator between the extents and the
2769 num_pages *= PAGE_CACHE_SIZE;
2771 ret = btrfs_check_data_free_space(inode, num_pages);
2775 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2776 num_pages, num_pages,
2778 btrfs_free_reserved_data_space(inode, num_pages);
2782 btrfs_release_path(root, path);
2784 spin_lock(&block_group->lock);
2786 block_group->disk_cache_state = BTRFS_DC_ERROR;
2788 block_group->disk_cache_state = BTRFS_DC_SETUP;
2789 spin_unlock(&block_group->lock);
2794 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2795 struct btrfs_root *root)
2797 struct btrfs_block_group_cache *cache;
2799 struct btrfs_path *path;
2802 path = btrfs_alloc_path();
2808 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2810 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2812 cache = next_block_group(root, cache);
2820 err = cache_save_setup(cache, trans, path);
2821 last = cache->key.objectid + cache->key.offset;
2822 btrfs_put_block_group(cache);
2827 err = btrfs_run_delayed_refs(trans, root,
2832 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2834 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2835 btrfs_put_block_group(cache);
2841 cache = next_block_group(root, cache);
2851 last = cache->key.objectid + cache->key.offset;
2853 err = write_one_cache_group(trans, root, path, cache);
2855 btrfs_put_block_group(cache);
2858 btrfs_free_path(path);
2862 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2864 struct btrfs_block_group_cache *block_group;
2867 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2868 if (!block_group || block_group->ro)
2871 btrfs_put_block_group(block_group);
2875 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2876 u64 total_bytes, u64 bytes_used,
2877 struct btrfs_space_info **space_info)
2879 struct btrfs_space_info *found;
2883 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2884 BTRFS_BLOCK_GROUP_RAID10))
2889 found = __find_space_info(info, flags);
2891 spin_lock(&found->lock);
2892 found->total_bytes += total_bytes;
2893 found->bytes_used += bytes_used;
2894 found->disk_used += bytes_used * factor;
2896 spin_unlock(&found->lock);
2897 *space_info = found;
2900 found = kzalloc(sizeof(*found), GFP_NOFS);
2904 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2905 INIT_LIST_HEAD(&found->block_groups[i]);
2906 init_rwsem(&found->groups_sem);
2907 spin_lock_init(&found->lock);
2908 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2909 BTRFS_BLOCK_GROUP_SYSTEM |
2910 BTRFS_BLOCK_GROUP_METADATA);
2911 found->total_bytes = total_bytes;
2912 found->bytes_used = bytes_used;
2913 found->disk_used = bytes_used * factor;
2914 found->bytes_pinned = 0;
2915 found->bytes_reserved = 0;
2916 found->bytes_readonly = 0;
2917 found->bytes_may_use = 0;
2919 found->force_alloc = 0;
2920 *space_info = found;
2921 list_add_rcu(&found->list, &info->space_info);
2922 atomic_set(&found->caching_threads, 0);
2926 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2928 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2929 BTRFS_BLOCK_GROUP_RAID1 |
2930 BTRFS_BLOCK_GROUP_RAID10 |
2931 BTRFS_BLOCK_GROUP_DUP);
2933 if (flags & BTRFS_BLOCK_GROUP_DATA)
2934 fs_info->avail_data_alloc_bits |= extra_flags;
2935 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2936 fs_info->avail_metadata_alloc_bits |= extra_flags;
2937 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2938 fs_info->avail_system_alloc_bits |= extra_flags;
2942 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2944 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2946 if (num_devices == 1)
2947 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2948 if (num_devices < 4)
2949 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2951 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2952 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2953 BTRFS_BLOCK_GROUP_RAID10))) {
2954 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2957 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2958 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2959 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2962 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2963 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2964 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2965 (flags & BTRFS_BLOCK_GROUP_DUP)))
2966 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2970 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2972 if (flags & BTRFS_BLOCK_GROUP_DATA)
2973 flags |= root->fs_info->avail_data_alloc_bits &
2974 root->fs_info->data_alloc_profile;
2975 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2976 flags |= root->fs_info->avail_system_alloc_bits &
2977 root->fs_info->system_alloc_profile;
2978 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2979 flags |= root->fs_info->avail_metadata_alloc_bits &
2980 root->fs_info->metadata_alloc_profile;
2981 return btrfs_reduce_alloc_profile(root, flags);
2984 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
2989 flags = BTRFS_BLOCK_GROUP_DATA;
2990 else if (root == root->fs_info->chunk_root)
2991 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2993 flags = BTRFS_BLOCK_GROUP_METADATA;
2995 return get_alloc_profile(root, flags);
2998 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3000 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3001 BTRFS_BLOCK_GROUP_DATA);
3005 * This will check the space that the inode allocates from to make sure we have
3006 * enough space for bytes.
3008 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3010 struct btrfs_space_info *data_sinfo;
3011 struct btrfs_root *root = BTRFS_I(inode)->root;
3013 int ret = 0, committed = 0, alloc_chunk = 1;
3015 /* make sure bytes are sectorsize aligned */
3016 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3018 if (root == root->fs_info->tree_root) {
3023 data_sinfo = BTRFS_I(inode)->space_info;
3028 /* make sure we have enough space to handle the data first */
3029 spin_lock(&data_sinfo->lock);
3030 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3031 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3032 data_sinfo->bytes_may_use;
3034 if (used + bytes > data_sinfo->total_bytes) {
3035 struct btrfs_trans_handle *trans;
3038 * if we don't have enough free bytes in this space then we need
3039 * to alloc a new chunk.
3041 if (!data_sinfo->full && alloc_chunk) {
3044 data_sinfo->force_alloc = 1;
3045 spin_unlock(&data_sinfo->lock);
3047 alloc_target = btrfs_get_alloc_profile(root, 1);
3048 trans = btrfs_join_transaction(root, 1);
3050 return PTR_ERR(trans);
3052 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3053 bytes + 2 * 1024 * 1024,
3055 btrfs_end_transaction(trans, root);
3060 btrfs_set_inode_space_info(root, inode);
3061 data_sinfo = BTRFS_I(inode)->space_info;
3065 spin_unlock(&data_sinfo->lock);
3067 /* commit the current transaction and try again */
3068 if (!committed && !root->fs_info->open_ioctl_trans) {
3070 trans = btrfs_join_transaction(root, 1);
3072 return PTR_ERR(trans);
3073 ret = btrfs_commit_transaction(trans, root);
3079 #if 0 /* I hope we never need this code again, just in case */
3080 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3081 "%llu bytes_reserved, " "%llu bytes_pinned, "
3082 "%llu bytes_readonly, %llu may use %llu total\n",
3083 (unsigned long long)bytes,
3084 (unsigned long long)data_sinfo->bytes_used,
3085 (unsigned long long)data_sinfo->bytes_reserved,
3086 (unsigned long long)data_sinfo->bytes_pinned,
3087 (unsigned long long)data_sinfo->bytes_readonly,
3088 (unsigned long long)data_sinfo->bytes_may_use,
3089 (unsigned long long)data_sinfo->total_bytes);
3093 data_sinfo->bytes_may_use += bytes;
3094 BTRFS_I(inode)->reserved_bytes += bytes;
3095 spin_unlock(&data_sinfo->lock);
3101 * called when we are clearing an delalloc extent from the
3102 * inode's io_tree or there was an error for whatever reason
3103 * after calling btrfs_check_data_free_space
3105 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3107 struct btrfs_root *root = BTRFS_I(inode)->root;
3108 struct btrfs_space_info *data_sinfo;
3110 /* make sure bytes are sectorsize aligned */
3111 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3113 data_sinfo = BTRFS_I(inode)->space_info;
3114 spin_lock(&data_sinfo->lock);
3115 data_sinfo->bytes_may_use -= bytes;
3116 BTRFS_I(inode)->reserved_bytes -= bytes;
3117 spin_unlock(&data_sinfo->lock);
3120 static void force_metadata_allocation(struct btrfs_fs_info *info)
3122 struct list_head *head = &info->space_info;
3123 struct btrfs_space_info *found;
3126 list_for_each_entry_rcu(found, head, list) {
3127 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3128 found->force_alloc = 1;
3133 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3136 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3138 if (sinfo->bytes_used + sinfo->bytes_reserved +
3139 alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3142 if (sinfo->bytes_used + sinfo->bytes_reserved +
3143 alloc_bytes < div_factor(num_bytes, 8))
3149 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3150 struct btrfs_root *extent_root, u64 alloc_bytes,
3151 u64 flags, int force)
3153 struct btrfs_space_info *space_info;
3154 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3157 mutex_lock(&fs_info->chunk_mutex);
3159 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3161 space_info = __find_space_info(extent_root->fs_info, flags);
3163 ret = update_space_info(extent_root->fs_info, flags,
3167 BUG_ON(!space_info);
3169 spin_lock(&space_info->lock);
3170 if (space_info->force_alloc)
3172 if (space_info->full) {
3173 spin_unlock(&space_info->lock);
3177 if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3178 spin_unlock(&space_info->lock);
3181 spin_unlock(&space_info->lock);
3184 * if we're doing a data chunk, go ahead and make sure that
3185 * we keep a reasonable number of metadata chunks allocated in the
3188 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3189 fs_info->data_chunk_allocations++;
3190 if (!(fs_info->data_chunk_allocations %
3191 fs_info->metadata_ratio))
3192 force_metadata_allocation(fs_info);
3195 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3196 spin_lock(&space_info->lock);
3198 space_info->full = 1;
3201 space_info->force_alloc = 0;
3202 spin_unlock(&space_info->lock);
3204 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3208 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3209 struct btrfs_root *root,
3210 struct btrfs_space_info *sinfo, u64 num_bytes)
3218 spin_lock(&sinfo->lock);
3219 ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3220 spin_unlock(&sinfo->lock);
3225 trans = btrfs_join_transaction(root, 1);
3226 BUG_ON(IS_ERR(trans));
3230 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3231 num_bytes + 2 * 1024 * 1024,
3232 get_alloc_profile(root, sinfo->flags), 0);
3235 btrfs_end_transaction(trans, root);
3237 return ret == 1 ? 1 : 0;
3241 * shrink metadata reservation for delalloc
3243 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *root, u64 to_reclaim)
3246 struct btrfs_block_rsv *block_rsv;
3253 block_rsv = &root->fs_info->delalloc_block_rsv;
3254 spin_lock(&block_rsv->lock);
3255 reserved = block_rsv->reserved;
3256 spin_unlock(&block_rsv->lock);
3261 max_reclaim = min(reserved, to_reclaim);
3264 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3266 __set_current_state(TASK_INTERRUPTIBLE);
3267 schedule_timeout(pause);
3269 if (pause > HZ / 10)
3275 spin_lock(&block_rsv->lock);
3276 if (reserved > block_rsv->reserved)
3277 reclaimed = reserved - block_rsv->reserved;
3278 reserved = block_rsv->reserved;
3279 spin_unlock(&block_rsv->lock);
3281 if (reserved == 0 || reclaimed >= max_reclaim)
3284 if (trans && trans->transaction->blocked)
3287 return reclaimed >= to_reclaim;
3290 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3291 struct btrfs_root *root,
3292 struct btrfs_block_rsv *block_rsv,
3293 u64 num_bytes, int *retries)
3295 struct btrfs_space_info *space_info = block_rsv->space_info;
3301 ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3305 if (trans && trans->transaction->in_commit)
3308 ret = shrink_delalloc(trans, root, num_bytes);
3312 spin_lock(&space_info->lock);
3313 if (space_info->bytes_pinned < num_bytes)
3315 spin_unlock(&space_info->lock);
3324 trans = btrfs_join_transaction(root, 1);
3325 BUG_ON(IS_ERR(trans));
3326 ret = btrfs_commit_transaction(trans, root);
3332 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3335 struct btrfs_space_info *space_info = block_rsv->space_info;
3339 spin_lock(&space_info->lock);
3340 unused = space_info->bytes_used + space_info->bytes_reserved +
3341 space_info->bytes_pinned + space_info->bytes_readonly;
3343 if (unused < space_info->total_bytes)
3344 unused = space_info->total_bytes - unused;
3348 if (unused >= num_bytes) {
3349 if (block_rsv->priority >= 10) {
3350 space_info->bytes_reserved += num_bytes;
3353 if ((unused + block_rsv->reserved) *
3354 block_rsv->priority >=
3355 (num_bytes + block_rsv->reserved) * 10) {
3356 space_info->bytes_reserved += num_bytes;
3361 spin_unlock(&space_info->lock);
3366 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3367 struct btrfs_root *root)
3369 struct btrfs_block_rsv *block_rsv;
3371 block_rsv = trans->block_rsv;
3373 block_rsv = root->block_rsv;
3376 block_rsv = &root->fs_info->empty_block_rsv;
3381 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3385 spin_lock(&block_rsv->lock);
3386 if (block_rsv->reserved >= num_bytes) {
3387 block_rsv->reserved -= num_bytes;
3388 if (block_rsv->reserved < block_rsv->size)
3389 block_rsv->full = 0;
3392 spin_unlock(&block_rsv->lock);
3396 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3397 u64 num_bytes, int update_size)
3399 spin_lock(&block_rsv->lock);
3400 block_rsv->reserved += num_bytes;
3402 block_rsv->size += num_bytes;
3403 else if (block_rsv->reserved >= block_rsv->size)
3404 block_rsv->full = 1;
3405 spin_unlock(&block_rsv->lock);
3408 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3409 struct btrfs_block_rsv *dest, u64 num_bytes)
3411 struct btrfs_space_info *space_info = block_rsv->space_info;
3413 spin_lock(&block_rsv->lock);
3414 if (num_bytes == (u64)-1)
3415 num_bytes = block_rsv->size;
3416 block_rsv->size -= num_bytes;
3417 if (block_rsv->reserved >= block_rsv->size) {
3418 num_bytes = block_rsv->reserved - block_rsv->size;
3419 block_rsv->reserved = block_rsv->size;
3420 block_rsv->full = 1;
3424 spin_unlock(&block_rsv->lock);
3426 if (num_bytes > 0) {
3428 block_rsv_add_bytes(dest, num_bytes, 0);
3430 spin_lock(&space_info->lock);
3431 space_info->bytes_reserved -= num_bytes;
3432 spin_unlock(&space_info->lock);
3437 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3438 struct btrfs_block_rsv *dst, u64 num_bytes)
3442 ret = block_rsv_use_bytes(src, num_bytes);
3446 block_rsv_add_bytes(dst, num_bytes, 1);
3450 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3452 memset(rsv, 0, sizeof(*rsv));
3453 spin_lock_init(&rsv->lock);
3454 atomic_set(&rsv->usage, 1);
3456 INIT_LIST_HEAD(&rsv->list);
3459 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3461 struct btrfs_block_rsv *block_rsv;
3462 struct btrfs_fs_info *fs_info = root->fs_info;
3465 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3469 btrfs_init_block_rsv(block_rsv);
3471 alloc_target = btrfs_get_alloc_profile(root, 0);
3472 block_rsv->space_info = __find_space_info(fs_info,
3473 BTRFS_BLOCK_GROUP_METADATA);
3478 void btrfs_free_block_rsv(struct btrfs_root *root,
3479 struct btrfs_block_rsv *rsv)
3481 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3482 btrfs_block_rsv_release(root, rsv, (u64)-1);
3489 * make the block_rsv struct be able to capture freed space.
3490 * the captured space will re-add to the the block_rsv struct
3491 * after transaction commit
3493 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3494 struct btrfs_block_rsv *block_rsv)
3496 block_rsv->durable = 1;
3497 mutex_lock(&fs_info->durable_block_rsv_mutex);
3498 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3499 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3502 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3503 struct btrfs_root *root,
3504 struct btrfs_block_rsv *block_rsv,
3505 u64 num_bytes, int *retries)
3512 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3514 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3518 ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3525 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3526 struct btrfs_root *root,
3527 struct btrfs_block_rsv *block_rsv,
3528 u64 min_reserved, int min_factor)
3531 int commit_trans = 0;
3537 spin_lock(&block_rsv->lock);
3539 num_bytes = div_factor(block_rsv->size, min_factor);
3540 if (min_reserved > num_bytes)
3541 num_bytes = min_reserved;
3543 if (block_rsv->reserved >= num_bytes) {
3546 num_bytes -= block_rsv->reserved;
3547 if (block_rsv->durable &&
3548 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3551 spin_unlock(&block_rsv->lock);
3555 if (block_rsv->refill_used) {
3556 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3558 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3567 trans = btrfs_join_transaction(root, 1);
3568 BUG_ON(IS_ERR(trans));
3569 ret = btrfs_commit_transaction(trans, root);
3574 printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3575 block_rsv->size, block_rsv->reserved,
3576 block_rsv->freed[0], block_rsv->freed[1]);
3581 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3582 struct btrfs_block_rsv *dst_rsv,
3585 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3588 void btrfs_block_rsv_release(struct btrfs_root *root,
3589 struct btrfs_block_rsv *block_rsv,
3592 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3593 if (global_rsv->full || global_rsv == block_rsv ||
3594 block_rsv->space_info != global_rsv->space_info)
3596 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3600 * helper to calculate size of global block reservation.
3601 * the desired value is sum of space used by extent tree,
3602 * checksum tree and root tree
3604 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3606 struct btrfs_space_info *sinfo;
3610 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3613 * per tree used space accounting can be inaccuracy, so we
3616 spin_lock(&fs_info->extent_root->accounting_lock);
3617 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3618 spin_unlock(&fs_info->extent_root->accounting_lock);
3620 spin_lock(&fs_info->csum_root->accounting_lock);
3621 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3622 spin_unlock(&fs_info->csum_root->accounting_lock);
3624 spin_lock(&fs_info->tree_root->accounting_lock);
3625 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3626 spin_unlock(&fs_info->tree_root->accounting_lock);
3628 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3629 spin_lock(&sinfo->lock);
3630 data_used = sinfo->bytes_used;
3631 spin_unlock(&sinfo->lock);
3633 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3634 spin_lock(&sinfo->lock);
3635 meta_used = sinfo->bytes_used;
3636 spin_unlock(&sinfo->lock);
3638 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3640 num_bytes += div64_u64(data_used + meta_used, 50);
3642 if (num_bytes * 3 > meta_used)
3643 num_bytes = div64_u64(meta_used, 3);
3645 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3648 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3650 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3651 struct btrfs_space_info *sinfo = block_rsv->space_info;
3654 num_bytes = calc_global_metadata_size(fs_info);
3656 spin_lock(&block_rsv->lock);
3657 spin_lock(&sinfo->lock);
3659 block_rsv->size = num_bytes;
3661 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3662 sinfo->bytes_reserved + sinfo->bytes_readonly;
3664 if (sinfo->total_bytes > num_bytes) {
3665 num_bytes = sinfo->total_bytes - num_bytes;
3666 block_rsv->reserved += num_bytes;
3667 sinfo->bytes_reserved += num_bytes;
3670 if (block_rsv->reserved >= block_rsv->size) {
3671 num_bytes = block_rsv->reserved - block_rsv->size;
3672 sinfo->bytes_reserved -= num_bytes;
3673 block_rsv->reserved = block_rsv->size;
3674 block_rsv->full = 1;
3677 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3678 block_rsv->size, block_rsv->reserved);
3680 spin_unlock(&sinfo->lock);
3681 spin_unlock(&block_rsv->lock);
3684 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3686 struct btrfs_space_info *space_info;
3688 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3689 fs_info->chunk_block_rsv.space_info = space_info;
3690 fs_info->chunk_block_rsv.priority = 10;
3692 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3693 fs_info->global_block_rsv.space_info = space_info;
3694 fs_info->global_block_rsv.priority = 10;
3695 fs_info->global_block_rsv.refill_used = 1;
3696 fs_info->delalloc_block_rsv.space_info = space_info;
3697 fs_info->trans_block_rsv.space_info = space_info;
3698 fs_info->empty_block_rsv.space_info = space_info;
3699 fs_info->empty_block_rsv.priority = 10;
3701 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3702 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3703 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3704 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3705 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3707 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3709 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3711 update_global_block_rsv(fs_info);
3714 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3716 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3717 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3718 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3719 WARN_ON(fs_info->trans_block_rsv.size > 0);
3720 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3721 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3722 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3725 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3727 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3731 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3732 struct btrfs_root *root,
3733 int num_items, int *retries)
3738 if (num_items == 0 || root->fs_info->chunk_root == root)
3741 num_bytes = calc_trans_metadata_size(root, num_items);
3742 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3743 num_bytes, retries);
3745 trans->bytes_reserved += num_bytes;
3746 trans->block_rsv = &root->fs_info->trans_block_rsv;
3751 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3752 struct btrfs_root *root)
3754 if (!trans->bytes_reserved)
3757 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3758 btrfs_block_rsv_release(root, trans->block_rsv,
3759 trans->bytes_reserved);
3760 trans->bytes_reserved = 0;
3763 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3764 struct inode *inode)
3766 struct btrfs_root *root = BTRFS_I(inode)->root;
3767 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3768 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3771 * one for deleting orphan item, one for updating inode and
3772 * two for calling btrfs_truncate_inode_items.
3774 * btrfs_truncate_inode_items is a delete operation, it frees
3775 * more space than it uses in most cases. So two units of
3776 * metadata space should be enough for calling it many times.
3777 * If all of the metadata space is used, we can commit
3778 * transaction and use space it freed.
3780 u64 num_bytes = calc_trans_metadata_size(root, 4);
3781 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3784 void btrfs_orphan_release_metadata(struct inode *inode)
3786 struct btrfs_root *root = BTRFS_I(inode)->root;
3787 u64 num_bytes = calc_trans_metadata_size(root, 4);
3788 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3791 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3792 struct btrfs_pending_snapshot *pending)
3794 struct btrfs_root *root = pending->root;
3795 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3796 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3798 * two for root back/forward refs, two for directory entries
3799 * and one for root of the snapshot.
3801 u64 num_bytes = calc_trans_metadata_size(root, 5);
3802 dst_rsv->space_info = src_rsv->space_info;
3803 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3806 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3808 return num_bytes >>= 3;
3811 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3813 struct btrfs_root *root = BTRFS_I(inode)->root;
3814 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3820 if (btrfs_transaction_in_commit(root->fs_info))
3821 schedule_timeout(1);
3823 num_bytes = ALIGN(num_bytes, root->sectorsize);
3825 spin_lock(&BTRFS_I(inode)->accounting_lock);
3826 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3827 if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3828 nr_extents -= BTRFS_I(inode)->reserved_extents;
3829 to_reserve = calc_trans_metadata_size(root, nr_extents);
3835 to_reserve += calc_csum_metadata_size(inode, num_bytes);
3836 ret = reserve_metadata_bytes(block_rsv, to_reserve);
3838 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3839 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3846 BTRFS_I(inode)->reserved_extents += nr_extents;
3847 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3848 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3850 block_rsv_add_bytes(block_rsv, to_reserve, 1);
3852 if (block_rsv->size > 512 * 1024 * 1024)
3853 shrink_delalloc(NULL, root, to_reserve);
3858 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3860 struct btrfs_root *root = BTRFS_I(inode)->root;
3864 num_bytes = ALIGN(num_bytes, root->sectorsize);
3865 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3867 spin_lock(&BTRFS_I(inode)->accounting_lock);
3868 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3869 if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3870 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3871 BTRFS_I(inode)->reserved_extents -= nr_extents;
3875 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3877 to_free = calc_csum_metadata_size(inode, num_bytes);
3879 to_free += calc_trans_metadata_size(root, nr_extents);
3881 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3885 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3889 ret = btrfs_check_data_free_space(inode, num_bytes);
3893 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3895 btrfs_free_reserved_data_space(inode, num_bytes);
3902 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3904 btrfs_delalloc_release_metadata(inode, num_bytes);
3905 btrfs_free_reserved_data_space(inode, num_bytes);
3908 static int update_block_group(struct btrfs_trans_handle *trans,
3909 struct btrfs_root *root,
3910 u64 bytenr, u64 num_bytes, int alloc)
3912 struct btrfs_block_group_cache *cache = NULL;
3913 struct btrfs_fs_info *info = root->fs_info;
3914 u64 total = num_bytes;
3919 /* block accounting for super block */
3920 spin_lock(&info->delalloc_lock);
3921 old_val = btrfs_super_bytes_used(&info->super_copy);
3923 old_val += num_bytes;
3925 old_val -= num_bytes;
3926 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3927 spin_unlock(&info->delalloc_lock);
3930 cache = btrfs_lookup_block_group(info, bytenr);
3933 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3934 BTRFS_BLOCK_GROUP_RAID1 |
3935 BTRFS_BLOCK_GROUP_RAID10))
3940 byte_in_group = bytenr - cache->key.objectid;
3941 WARN_ON(byte_in_group > cache->key.offset);
3943 spin_lock(&cache->space_info->lock);
3944 spin_lock(&cache->lock);
3946 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
3947 cache->disk_cache_state < BTRFS_DC_CLEAR)
3948 cache->disk_cache_state = BTRFS_DC_CLEAR;
3951 old_val = btrfs_block_group_used(&cache->item);
3952 num_bytes = min(total, cache->key.offset - byte_in_group);
3954 old_val += num_bytes;
3955 btrfs_set_block_group_used(&cache->item, old_val);
3956 cache->reserved -= num_bytes;
3957 cache->space_info->bytes_reserved -= num_bytes;
3958 cache->space_info->bytes_used += num_bytes;
3959 cache->space_info->disk_used += num_bytes * factor;
3960 spin_unlock(&cache->lock);
3961 spin_unlock(&cache->space_info->lock);
3963 old_val -= num_bytes;
3964 btrfs_set_block_group_used(&cache->item, old_val);
3965 cache->pinned += num_bytes;
3966 cache->space_info->bytes_pinned += num_bytes;
3967 cache->space_info->bytes_used -= num_bytes;
3968 cache->space_info->disk_used -= num_bytes * factor;
3969 spin_unlock(&cache->lock);
3970 spin_unlock(&cache->space_info->lock);
3972 set_extent_dirty(info->pinned_extents,
3973 bytenr, bytenr + num_bytes - 1,
3974 GFP_NOFS | __GFP_NOFAIL);
3976 btrfs_put_block_group(cache);
3978 bytenr += num_bytes;
3983 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3985 struct btrfs_block_group_cache *cache;
3988 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3992 bytenr = cache->key.objectid;
3993 btrfs_put_block_group(cache);
3998 static int pin_down_extent(struct btrfs_root *root,
3999 struct btrfs_block_group_cache *cache,
4000 u64 bytenr, u64 num_bytes, int reserved)
4002 spin_lock(&cache->space_info->lock);
4003 spin_lock(&cache->lock);
4004 cache->pinned += num_bytes;
4005 cache->space_info->bytes_pinned += num_bytes;
4007 cache->reserved -= num_bytes;
4008 cache->space_info->bytes_reserved -= num_bytes;
4010 spin_unlock(&cache->lock);
4011 spin_unlock(&cache->space_info->lock);
4013 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4014 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4019 * this function must be called within transaction
4021 int btrfs_pin_extent(struct btrfs_root *root,
4022 u64 bytenr, u64 num_bytes, int reserved)
4024 struct btrfs_block_group_cache *cache;
4026 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4029 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4031 btrfs_put_block_group(cache);
4036 * update size of reserved extents. this function may return -EAGAIN
4037 * if 'reserve' is true or 'sinfo' is false.
4039 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4040 u64 num_bytes, int reserve, int sinfo)
4044 struct btrfs_space_info *space_info = cache->space_info;
4045 spin_lock(&space_info->lock);
4046 spin_lock(&cache->lock);
4051 cache->reserved += num_bytes;
4052 space_info->bytes_reserved += num_bytes;
4056 space_info->bytes_readonly += num_bytes;
4057 cache->reserved -= num_bytes;
4058 space_info->bytes_reserved -= num_bytes;
4060 spin_unlock(&cache->lock);
4061 spin_unlock(&space_info->lock);
4063 spin_lock(&cache->lock);
4068 cache->reserved += num_bytes;
4070 cache->reserved -= num_bytes;
4072 spin_unlock(&cache->lock);
4077 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4078 struct btrfs_root *root)
4080 struct btrfs_fs_info *fs_info = root->fs_info;
4081 struct btrfs_caching_control *next;
4082 struct btrfs_caching_control *caching_ctl;
4083 struct btrfs_block_group_cache *cache;
4085 down_write(&fs_info->extent_commit_sem);
4087 list_for_each_entry_safe(caching_ctl, next,
4088 &fs_info->caching_block_groups, list) {
4089 cache = caching_ctl->block_group;
4090 if (block_group_cache_done(cache)) {
4091 cache->last_byte_to_unpin = (u64)-1;
4092 list_del_init(&caching_ctl->list);
4093 put_caching_control(caching_ctl);
4095 cache->last_byte_to_unpin = caching_ctl->progress;
4099 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4100 fs_info->pinned_extents = &fs_info->freed_extents[1];
4102 fs_info->pinned_extents = &fs_info->freed_extents[0];
4104 up_write(&fs_info->extent_commit_sem);
4106 update_global_block_rsv(fs_info);
4110 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4112 struct btrfs_fs_info *fs_info = root->fs_info;
4113 struct btrfs_block_group_cache *cache = NULL;
4116 while (start <= end) {
4118 start >= cache->key.objectid + cache->key.offset) {
4120 btrfs_put_block_group(cache);
4121 cache = btrfs_lookup_block_group(fs_info, start);
4125 len = cache->key.objectid + cache->key.offset - start;
4126 len = min(len, end + 1 - start);
4128 if (start < cache->last_byte_to_unpin) {
4129 len = min(len, cache->last_byte_to_unpin - start);
4130 btrfs_add_free_space(cache, start, len);
4135 spin_lock(&cache->space_info->lock);
4136 spin_lock(&cache->lock);
4137 cache->pinned -= len;
4138 cache->space_info->bytes_pinned -= len;
4140 cache->space_info->bytes_readonly += len;
4141 } else if (cache->reserved_pinned > 0) {
4142 len = min(len, cache->reserved_pinned);
4143 cache->reserved_pinned -= len;
4144 cache->space_info->bytes_reserved += len;
4146 spin_unlock(&cache->lock);
4147 spin_unlock(&cache->space_info->lock);
4151 btrfs_put_block_group(cache);
4155 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4156 struct btrfs_root *root)
4158 struct btrfs_fs_info *fs_info = root->fs_info;
4159 struct extent_io_tree *unpin;
4160 struct btrfs_block_rsv *block_rsv;
4161 struct btrfs_block_rsv *next_rsv;
4167 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4168 unpin = &fs_info->freed_extents[1];
4170 unpin = &fs_info->freed_extents[0];
4173 ret = find_first_extent_bit(unpin, 0, &start, &end,
4178 ret = btrfs_discard_extent(root, start, end + 1 - start);
4180 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4181 unpin_extent_range(root, start, end);
4185 mutex_lock(&fs_info->durable_block_rsv_mutex);
4186 list_for_each_entry_safe(block_rsv, next_rsv,
4187 &fs_info->durable_block_rsv_list, list) {
4189 idx = trans->transid & 0x1;
4190 if (block_rsv->freed[idx] > 0) {
4191 block_rsv_add_bytes(block_rsv,
4192 block_rsv->freed[idx], 0);
4193 block_rsv->freed[idx] = 0;
4195 if (atomic_read(&block_rsv->usage) == 0) {
4196 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4198 if (block_rsv->freed[0] == 0 &&
4199 block_rsv->freed[1] == 0) {
4200 list_del_init(&block_rsv->list);
4204 btrfs_block_rsv_release(root, block_rsv, 0);
4207 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4212 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4213 struct btrfs_root *root,
4214 u64 bytenr, u64 num_bytes, u64 parent,
4215 u64 root_objectid, u64 owner_objectid,
4216 u64 owner_offset, int refs_to_drop,
4217 struct btrfs_delayed_extent_op *extent_op)
4219 struct btrfs_key key;
4220 struct btrfs_path *path;
4221 struct btrfs_fs_info *info = root->fs_info;
4222 struct btrfs_root *extent_root = info->extent_root;
4223 struct extent_buffer *leaf;
4224 struct btrfs_extent_item *ei;
4225 struct btrfs_extent_inline_ref *iref;
4228 int extent_slot = 0;
4229 int found_extent = 0;
4234 path = btrfs_alloc_path();
4239 path->leave_spinning = 1;
4241 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4242 BUG_ON(!is_data && refs_to_drop != 1);
4244 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4245 bytenr, num_bytes, parent,
4246 root_objectid, owner_objectid,
4249 extent_slot = path->slots[0];
4250 while (extent_slot >= 0) {
4251 btrfs_item_key_to_cpu(path->nodes[0], &key,
4253 if (key.objectid != bytenr)
4255 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4256 key.offset == num_bytes) {
4260 if (path->slots[0] - extent_slot > 5)
4264 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4265 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4266 if (found_extent && item_size < sizeof(*ei))
4269 if (!found_extent) {
4271 ret = remove_extent_backref(trans, extent_root, path,
4275 btrfs_release_path(extent_root, path);
4276 path->leave_spinning = 1;
4278 key.objectid = bytenr;
4279 key.type = BTRFS_EXTENT_ITEM_KEY;
4280 key.offset = num_bytes;
4282 ret = btrfs_search_slot(trans, extent_root,
4285 printk(KERN_ERR "umm, got %d back from search"
4286 ", was looking for %llu\n", ret,
4287 (unsigned long long)bytenr);
4288 btrfs_print_leaf(extent_root, path->nodes[0]);
4291 extent_slot = path->slots[0];
4294 btrfs_print_leaf(extent_root, path->nodes[0]);
4296 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4297 "parent %llu root %llu owner %llu offset %llu\n",
4298 (unsigned long long)bytenr,
4299 (unsigned long long)parent,
4300 (unsigned long long)root_objectid,
4301 (unsigned long long)owner_objectid,
4302 (unsigned long long)owner_offset);
4305 leaf = path->nodes[0];
4306 item_size = btrfs_item_size_nr(leaf, extent_slot);
4307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4308 if (item_size < sizeof(*ei)) {
4309 BUG_ON(found_extent || extent_slot != path->slots[0]);
4310 ret = convert_extent_item_v0(trans, extent_root, path,
4314 btrfs_release_path(extent_root, path);
4315 path->leave_spinning = 1;
4317 key.objectid = bytenr;
4318 key.type = BTRFS_EXTENT_ITEM_KEY;
4319 key.offset = num_bytes;
4321 ret = btrfs_search_slot(trans, extent_root, &key, path,
4324 printk(KERN_ERR "umm, got %d back from search"
4325 ", was looking for %llu\n", ret,
4326 (unsigned long long)bytenr);
4327 btrfs_print_leaf(extent_root, path->nodes[0]);
4330 extent_slot = path->slots[0];
4331 leaf = path->nodes[0];
4332 item_size = btrfs_item_size_nr(leaf, extent_slot);
4335 BUG_ON(item_size < sizeof(*ei));
4336 ei = btrfs_item_ptr(leaf, extent_slot,
4337 struct btrfs_extent_item);
4338 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4339 struct btrfs_tree_block_info *bi;
4340 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4341 bi = (struct btrfs_tree_block_info *)(ei + 1);
4342 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4345 refs = btrfs_extent_refs(leaf, ei);
4346 BUG_ON(refs < refs_to_drop);
4347 refs -= refs_to_drop;
4351 __run_delayed_extent_op(extent_op, leaf, ei);
4353 * In the case of inline back ref, reference count will
4354 * be updated by remove_extent_backref
4357 BUG_ON(!found_extent);
4359 btrfs_set_extent_refs(leaf, ei, refs);
4360 btrfs_mark_buffer_dirty(leaf);
4363 ret = remove_extent_backref(trans, extent_root, path,
4370 BUG_ON(is_data && refs_to_drop !=
4371 extent_data_ref_count(root, path, iref));
4373 BUG_ON(path->slots[0] != extent_slot);
4375 BUG_ON(path->slots[0] != extent_slot + 1);
4376 path->slots[0] = extent_slot;
4381 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4384 btrfs_release_path(extent_root, path);
4387 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4390 invalidate_mapping_pages(info->btree_inode->i_mapping,
4391 bytenr >> PAGE_CACHE_SHIFT,
4392 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4395 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4398 btrfs_free_path(path);
4403 * when we free an block, it is possible (and likely) that we free the last
4404 * delayed ref for that extent as well. This searches the delayed ref tree for
4405 * a given extent, and if there are no other delayed refs to be processed, it
4406 * removes it from the tree.
4408 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4409 struct btrfs_root *root, u64 bytenr)
4411 struct btrfs_delayed_ref_head *head;
4412 struct btrfs_delayed_ref_root *delayed_refs;
4413 struct btrfs_delayed_ref_node *ref;
4414 struct rb_node *node;
4417 delayed_refs = &trans->transaction->delayed_refs;
4418 spin_lock(&delayed_refs->lock);
4419 head = btrfs_find_delayed_ref_head(trans, bytenr);
4423 node = rb_prev(&head->node.rb_node);
4427 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4429 /* there are still entries for this ref, we can't drop it */
4430 if (ref->bytenr == bytenr)
4433 if (head->extent_op) {
4434 if (!head->must_insert_reserved)
4436 kfree(head->extent_op);
4437 head->extent_op = NULL;
4441 * waiting for the lock here would deadlock. If someone else has it
4442 * locked they are already in the process of dropping it anyway
4444 if (!mutex_trylock(&head->mutex))
4448 * at this point we have a head with no other entries. Go
4449 * ahead and process it.
4451 head->node.in_tree = 0;
4452 rb_erase(&head->node.rb_node, &delayed_refs->root);
4454 delayed_refs->num_entries--;
4457 * we don't take a ref on the node because we're removing it from the
4458 * tree, so we just steal the ref the tree was holding.
4460 delayed_refs->num_heads--;
4461 if (list_empty(&head->cluster))
4462 delayed_refs->num_heads_ready--;
4464 list_del_init(&head->cluster);
4465 spin_unlock(&delayed_refs->lock);
4467 BUG_ON(head->extent_op);
4468 if (head->must_insert_reserved)
4471 mutex_unlock(&head->mutex);
4472 btrfs_put_delayed_ref(&head->node);
4475 spin_unlock(&delayed_refs->lock);
4479 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4480 struct btrfs_root *root,
4481 struct extent_buffer *buf,
4482 u64 parent, int last_ref)
4484 struct btrfs_block_rsv *block_rsv;
4485 struct btrfs_block_group_cache *cache = NULL;
4488 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4489 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4490 parent, root->root_key.objectid,
4491 btrfs_header_level(buf),
4492 BTRFS_DROP_DELAYED_REF, NULL);
4499 block_rsv = get_block_rsv(trans, root);
4500 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4501 if (block_rsv->space_info != cache->space_info)
4504 if (btrfs_header_generation(buf) == trans->transid) {
4505 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4506 ret = check_ref_cleanup(trans, root, buf->start);
4511 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4512 pin_down_extent(root, cache, buf->start, buf->len, 1);
4516 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4518 btrfs_add_free_space(cache, buf->start, buf->len);
4519 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4520 if (ret == -EAGAIN) {
4521 /* block group became read-only */
4522 update_reserved_bytes(cache, buf->len, 0, 1);
4527 spin_lock(&block_rsv->lock);
4528 if (block_rsv->reserved < block_rsv->size) {
4529 block_rsv->reserved += buf->len;
4532 spin_unlock(&block_rsv->lock);
4535 spin_lock(&cache->space_info->lock);
4536 cache->space_info->bytes_reserved -= buf->len;
4537 spin_unlock(&cache->space_info->lock);
4542 if (block_rsv->durable && !cache->ro) {
4544 spin_lock(&cache->lock);
4546 cache->reserved_pinned += buf->len;
4549 spin_unlock(&cache->lock);
4552 spin_lock(&block_rsv->lock);
4553 block_rsv->freed[trans->transid & 0x1] += buf->len;
4554 spin_unlock(&block_rsv->lock);
4558 btrfs_put_block_group(cache);
4561 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4562 struct btrfs_root *root,
4563 u64 bytenr, u64 num_bytes, u64 parent,
4564 u64 root_objectid, u64 owner, u64 offset)
4569 * tree log blocks never actually go into the extent allocation
4570 * tree, just update pinning info and exit early.
4572 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4573 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4574 /* unlocks the pinned mutex */
4575 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4577 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4578 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4579 parent, root_objectid, (int)owner,
4580 BTRFS_DROP_DELAYED_REF, NULL);
4583 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4584 parent, root_objectid, owner,
4585 offset, BTRFS_DROP_DELAYED_REF, NULL);
4591 static u64 stripe_align(struct btrfs_root *root, u64 val)
4593 u64 mask = ((u64)root->stripesize - 1);
4594 u64 ret = (val + mask) & ~mask;
4599 * when we wait for progress in the block group caching, its because
4600 * our allocation attempt failed at least once. So, we must sleep
4601 * and let some progress happen before we try again.
4603 * This function will sleep at least once waiting for new free space to
4604 * show up, and then it will check the block group free space numbers
4605 * for our min num_bytes. Another option is to have it go ahead
4606 * and look in the rbtree for a free extent of a given size, but this
4610 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4613 struct btrfs_caching_control *caching_ctl;
4616 caching_ctl = get_caching_control(cache);
4620 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4621 (cache->free_space >= num_bytes));
4623 put_caching_control(caching_ctl);
4628 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4630 struct btrfs_caching_control *caching_ctl;
4633 caching_ctl = get_caching_control(cache);
4637 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4639 put_caching_control(caching_ctl);
4643 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4646 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4648 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4650 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4652 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4659 enum btrfs_loop_type {
4660 LOOP_FIND_IDEAL = 0,
4661 LOOP_CACHING_NOWAIT = 1,
4662 LOOP_CACHING_WAIT = 2,
4663 LOOP_ALLOC_CHUNK = 3,
4664 LOOP_NO_EMPTY_SIZE = 4,
4668 * walks the btree of allocated extents and find a hole of a given size.
4669 * The key ins is changed to record the hole:
4670 * ins->objectid == block start
4671 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4672 * ins->offset == number of blocks
4673 * Any available blocks before search_start are skipped.
4675 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4676 struct btrfs_root *orig_root,
4677 u64 num_bytes, u64 empty_size,
4678 u64 search_start, u64 search_end,
4679 u64 hint_byte, struct btrfs_key *ins,
4683 struct btrfs_root *root = orig_root->fs_info->extent_root;
4684 struct btrfs_free_cluster *last_ptr = NULL;
4685 struct btrfs_block_group_cache *block_group = NULL;
4686 int empty_cluster = 2 * 1024 * 1024;
4687 int allowed_chunk_alloc = 0;
4688 int done_chunk_alloc = 0;
4689 struct btrfs_space_info *space_info;
4690 int last_ptr_loop = 0;
4693 bool found_uncached_bg = false;
4694 bool failed_cluster_refill = false;
4695 bool failed_alloc = false;
4696 u64 ideal_cache_percent = 0;
4697 u64 ideal_cache_offset = 0;
4699 WARN_ON(num_bytes < root->sectorsize);
4700 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4704 space_info = __find_space_info(root->fs_info, data);
4706 printk(KERN_ERR "No space info for %d\n", data);
4710 if (orig_root->ref_cows || empty_size)
4711 allowed_chunk_alloc = 1;
4713 if (data & BTRFS_BLOCK_GROUP_METADATA) {
4714 last_ptr = &root->fs_info->meta_alloc_cluster;
4715 if (!btrfs_test_opt(root, SSD))
4716 empty_cluster = 64 * 1024;
4719 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4720 last_ptr = &root->fs_info->data_alloc_cluster;
4724 spin_lock(&last_ptr->lock);
4725 if (last_ptr->block_group)
4726 hint_byte = last_ptr->window_start;
4727 spin_unlock(&last_ptr->lock);
4730 search_start = max(search_start, first_logical_byte(root, 0));
4731 search_start = max(search_start, hint_byte);
4736 if (search_start == hint_byte) {
4738 block_group = btrfs_lookup_block_group(root->fs_info,
4741 * we don't want to use the block group if it doesn't match our
4742 * allocation bits, or if its not cached.
4744 * However if we are re-searching with an ideal block group
4745 * picked out then we don't care that the block group is cached.
4747 if (block_group && block_group_bits(block_group, data) &&
4748 (block_group->cached != BTRFS_CACHE_NO ||
4749 search_start == ideal_cache_offset)) {
4750 down_read(&space_info->groups_sem);
4751 if (list_empty(&block_group->list) ||
4754 * someone is removing this block group,
4755 * we can't jump into the have_block_group
4756 * target because our list pointers are not
4759 btrfs_put_block_group(block_group);
4760 up_read(&space_info->groups_sem);
4762 index = get_block_group_index(block_group);
4763 goto have_block_group;
4765 } else if (block_group) {
4766 btrfs_put_block_group(block_group);
4770 down_read(&space_info->groups_sem);
4771 list_for_each_entry(block_group, &space_info->block_groups[index],
4776 btrfs_get_block_group(block_group);
4777 search_start = block_group->key.objectid;
4780 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4783 free_percent = btrfs_block_group_used(&block_group->item);
4784 free_percent *= 100;
4785 free_percent = div64_u64(free_percent,
4786 block_group->key.offset);
4787 free_percent = 100 - free_percent;
4788 if (free_percent > ideal_cache_percent &&
4789 likely(!block_group->ro)) {
4790 ideal_cache_offset = block_group->key.objectid;
4791 ideal_cache_percent = free_percent;
4795 * We only want to start kthread caching if we are at
4796 * the point where we will wait for caching to make
4797 * progress, or if our ideal search is over and we've
4798 * found somebody to start caching.
4800 if (loop > LOOP_CACHING_NOWAIT ||
4801 (loop > LOOP_FIND_IDEAL &&
4802 atomic_read(&space_info->caching_threads) < 2)) {
4803 ret = cache_block_group(block_group);
4806 found_uncached_bg = true;
4809 * If loop is set for cached only, try the next block
4812 if (loop == LOOP_FIND_IDEAL)
4816 cached = block_group_cache_done(block_group);
4817 if (unlikely(!cached))
4818 found_uncached_bg = true;
4820 if (unlikely(block_group->ro))
4824 * Ok we want to try and use the cluster allocator, so lets look
4825 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4826 * have tried the cluster allocator plenty of times at this
4827 * point and not have found anything, so we are likely way too
4828 * fragmented for the clustering stuff to find anything, so lets
4829 * just skip it and let the allocator find whatever block it can
4832 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4834 * the refill lock keeps out other
4835 * people trying to start a new cluster
4837 spin_lock(&last_ptr->refill_lock);
4838 if (last_ptr->block_group &&
4839 (last_ptr->block_group->ro ||
4840 !block_group_bits(last_ptr->block_group, data))) {
4842 goto refill_cluster;
4845 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4846 num_bytes, search_start);
4848 /* we have a block, we're done */
4849 spin_unlock(&last_ptr->refill_lock);
4853 spin_lock(&last_ptr->lock);
4855 * whoops, this cluster doesn't actually point to
4856 * this block group. Get a ref on the block
4857 * group is does point to and try again
4859 if (!last_ptr_loop && last_ptr->block_group &&
4860 last_ptr->block_group != block_group) {
4862 btrfs_put_block_group(block_group);
4863 block_group = last_ptr->block_group;
4864 btrfs_get_block_group(block_group);
4865 spin_unlock(&last_ptr->lock);
4866 spin_unlock(&last_ptr->refill_lock);
4869 search_start = block_group->key.objectid;
4871 * we know this block group is properly
4872 * in the list because
4873 * btrfs_remove_block_group, drops the
4874 * cluster before it removes the block
4875 * group from the list
4877 goto have_block_group;
4879 spin_unlock(&last_ptr->lock);
4882 * this cluster didn't work out, free it and
4885 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4889 /* allocate a cluster in this block group */
4890 ret = btrfs_find_space_cluster(trans, root,
4891 block_group, last_ptr,
4893 empty_cluster + empty_size);
4896 * now pull our allocation out of this
4899 offset = btrfs_alloc_from_cluster(block_group,
4900 last_ptr, num_bytes,
4903 /* we found one, proceed */
4904 spin_unlock(&last_ptr->refill_lock);
4907 } else if (!cached && loop > LOOP_CACHING_NOWAIT
4908 && !failed_cluster_refill) {
4909 spin_unlock(&last_ptr->refill_lock);
4911 failed_cluster_refill = true;
4912 wait_block_group_cache_progress(block_group,
4913 num_bytes + empty_cluster + empty_size);
4914 goto have_block_group;
4918 * at this point we either didn't find a cluster
4919 * or we weren't able to allocate a block from our
4920 * cluster. Free the cluster we've been trying
4921 * to use, and go to the next block group
4923 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4924 spin_unlock(&last_ptr->refill_lock);
4928 offset = btrfs_find_space_for_alloc(block_group, search_start,
4929 num_bytes, empty_size);
4931 * If we didn't find a chunk, and we haven't failed on this
4932 * block group before, and this block group is in the middle of
4933 * caching and we are ok with waiting, then go ahead and wait
4934 * for progress to be made, and set failed_alloc to true.
4936 * If failed_alloc is true then we've already waited on this
4937 * block group once and should move on to the next block group.
4939 if (!offset && !failed_alloc && !cached &&
4940 loop > LOOP_CACHING_NOWAIT) {
4941 wait_block_group_cache_progress(block_group,
4942 num_bytes + empty_size);
4943 failed_alloc = true;
4944 goto have_block_group;
4945 } else if (!offset) {
4949 search_start = stripe_align(root, offset);
4950 /* move on to the next group */
4951 if (search_start + num_bytes >= search_end) {
4952 btrfs_add_free_space(block_group, offset, num_bytes);
4956 /* move on to the next group */
4957 if (search_start + num_bytes >
4958 block_group->key.objectid + block_group->key.offset) {
4959 btrfs_add_free_space(block_group, offset, num_bytes);
4963 ins->objectid = search_start;
4964 ins->offset = num_bytes;
4966 if (offset < search_start)
4967 btrfs_add_free_space(block_group, offset,
4968 search_start - offset);
4969 BUG_ON(offset > search_start);
4971 ret = update_reserved_bytes(block_group, num_bytes, 1,
4972 (data & BTRFS_BLOCK_GROUP_DATA));
4973 if (ret == -EAGAIN) {
4974 btrfs_add_free_space(block_group, offset, num_bytes);
4978 /* we are all good, lets return */
4979 ins->objectid = search_start;
4980 ins->offset = num_bytes;
4982 if (offset < search_start)
4983 btrfs_add_free_space(block_group, offset,
4984 search_start - offset);
4985 BUG_ON(offset > search_start);
4988 failed_cluster_refill = false;
4989 failed_alloc = false;
4990 BUG_ON(index != get_block_group_index(block_group));
4991 btrfs_put_block_group(block_group);
4993 up_read(&space_info->groups_sem);
4995 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4998 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4999 * for them to make caching progress. Also
5000 * determine the best possible bg to cache
5001 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5002 * caching kthreads as we move along
5003 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5004 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5005 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5008 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5009 (found_uncached_bg || empty_size || empty_cluster ||
5010 allowed_chunk_alloc)) {
5012 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5013 found_uncached_bg = false;
5015 if (!ideal_cache_percent &&
5016 atomic_read(&space_info->caching_threads))
5020 * 1 of the following 2 things have happened so far
5022 * 1) We found an ideal block group for caching that
5023 * is mostly full and will cache quickly, so we might
5024 * as well wait for it.
5026 * 2) We searched for cached only and we didn't find
5027 * anything, and we didn't start any caching kthreads
5028 * either, so chances are we will loop through and
5029 * start a couple caching kthreads, and then come back
5030 * around and just wait for them. This will be slower
5031 * because we will have 2 caching kthreads reading at
5032 * the same time when we could have just started one
5033 * and waited for it to get far enough to give us an
5034 * allocation, so go ahead and go to the wait caching
5037 loop = LOOP_CACHING_WAIT;
5038 search_start = ideal_cache_offset;
5039 ideal_cache_percent = 0;
5041 } else if (loop == LOOP_FIND_IDEAL) {
5043 * Didn't find a uncached bg, wait on anything we find
5046 loop = LOOP_CACHING_WAIT;
5050 if (loop < LOOP_CACHING_WAIT) {
5055 if (loop == LOOP_ALLOC_CHUNK) {
5060 if (allowed_chunk_alloc) {
5061 ret = do_chunk_alloc(trans, root, num_bytes +
5062 2 * 1024 * 1024, data, 1);
5063 allowed_chunk_alloc = 0;
5064 done_chunk_alloc = 1;
5065 } else if (!done_chunk_alloc) {
5066 space_info->force_alloc = 1;
5069 if (loop < LOOP_NO_EMPTY_SIZE) {
5074 } else if (!ins->objectid) {
5078 /* we found what we needed */
5079 if (ins->objectid) {
5080 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5081 trans->block_group = block_group->key.objectid;
5083 btrfs_put_block_group(block_group);
5090 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5091 int dump_block_groups)
5093 struct btrfs_block_group_cache *cache;
5096 spin_lock(&info->lock);
5097 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5098 (unsigned long long)(info->total_bytes - info->bytes_used -
5099 info->bytes_pinned - info->bytes_reserved -
5100 info->bytes_readonly),
5101 (info->full) ? "" : "not ");
5102 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5103 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5104 (unsigned long long)info->total_bytes,
5105 (unsigned long long)info->bytes_used,
5106 (unsigned long long)info->bytes_pinned,
5107 (unsigned long long)info->bytes_reserved,
5108 (unsigned long long)info->bytes_may_use,
5109 (unsigned long long)info->bytes_readonly);
5110 spin_unlock(&info->lock);
5112 if (!dump_block_groups)
5115 down_read(&info->groups_sem);
5117 list_for_each_entry(cache, &info->block_groups[index], list) {
5118 spin_lock(&cache->lock);
5119 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5120 "%llu pinned %llu reserved\n",
5121 (unsigned long long)cache->key.objectid,
5122 (unsigned long long)cache->key.offset,
5123 (unsigned long long)btrfs_block_group_used(&cache->item),
5124 (unsigned long long)cache->pinned,
5125 (unsigned long long)cache->reserved);
5126 btrfs_dump_free_space(cache, bytes);
5127 spin_unlock(&cache->lock);
5129 if (++index < BTRFS_NR_RAID_TYPES)
5131 up_read(&info->groups_sem);
5134 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5135 struct btrfs_root *root,
5136 u64 num_bytes, u64 min_alloc_size,
5137 u64 empty_size, u64 hint_byte,
5138 u64 search_end, struct btrfs_key *ins,
5142 u64 search_start = 0;
5144 data = btrfs_get_alloc_profile(root, data);
5147 * the only place that sets empty_size is btrfs_realloc_node, which
5148 * is not called recursively on allocations
5150 if (empty_size || root->ref_cows)
5151 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5152 num_bytes + 2 * 1024 * 1024, data, 0);
5154 WARN_ON(num_bytes < root->sectorsize);
5155 ret = find_free_extent(trans, root, num_bytes, empty_size,
5156 search_start, search_end, hint_byte,
5159 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5160 num_bytes = num_bytes >> 1;
5161 num_bytes = num_bytes & ~(root->sectorsize - 1);
5162 num_bytes = max(num_bytes, min_alloc_size);
5163 do_chunk_alloc(trans, root->fs_info->extent_root,
5164 num_bytes, data, 1);
5167 if (ret == -ENOSPC) {
5168 struct btrfs_space_info *sinfo;
5170 sinfo = __find_space_info(root->fs_info, data);
5171 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5172 "wanted %llu\n", (unsigned long long)data,
5173 (unsigned long long)num_bytes);
5174 dump_space_info(sinfo, num_bytes, 1);
5180 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5182 struct btrfs_block_group_cache *cache;
5185 cache = btrfs_lookup_block_group(root->fs_info, start);
5187 printk(KERN_ERR "Unable to find block group for %llu\n",
5188 (unsigned long long)start);
5192 ret = btrfs_discard_extent(root, start, len);
5194 btrfs_add_free_space(cache, start, len);
5195 update_reserved_bytes(cache, len, 0, 1);
5196 btrfs_put_block_group(cache);
5201 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5202 struct btrfs_root *root,
5203 u64 parent, u64 root_objectid,
5204 u64 flags, u64 owner, u64 offset,
5205 struct btrfs_key *ins, int ref_mod)
5208 struct btrfs_fs_info *fs_info = root->fs_info;
5209 struct btrfs_extent_item *extent_item;
5210 struct btrfs_extent_inline_ref *iref;
5211 struct btrfs_path *path;
5212 struct extent_buffer *leaf;
5217 type = BTRFS_SHARED_DATA_REF_KEY;
5219 type = BTRFS_EXTENT_DATA_REF_KEY;
5221 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5223 path = btrfs_alloc_path();
5226 path->leave_spinning = 1;
5227 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5231 leaf = path->nodes[0];
5232 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5233 struct btrfs_extent_item);
5234 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5235 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5236 btrfs_set_extent_flags(leaf, extent_item,
5237 flags | BTRFS_EXTENT_FLAG_DATA);
5239 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5240 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5242 struct btrfs_shared_data_ref *ref;
5243 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5244 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5245 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5247 struct btrfs_extent_data_ref *ref;
5248 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5249 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5250 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5251 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5252 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5255 btrfs_mark_buffer_dirty(path->nodes[0]);
5256 btrfs_free_path(path);
5258 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5260 printk(KERN_ERR "btrfs update block group failed for %llu "
5261 "%llu\n", (unsigned long long)ins->objectid,
5262 (unsigned long long)ins->offset);
5268 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5269 struct btrfs_root *root,
5270 u64 parent, u64 root_objectid,
5271 u64 flags, struct btrfs_disk_key *key,
5272 int level, struct btrfs_key *ins)
5275 struct btrfs_fs_info *fs_info = root->fs_info;
5276 struct btrfs_extent_item *extent_item;
5277 struct btrfs_tree_block_info *block_info;
5278 struct btrfs_extent_inline_ref *iref;
5279 struct btrfs_path *path;
5280 struct extent_buffer *leaf;
5281 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5283 path = btrfs_alloc_path();
5286 path->leave_spinning = 1;
5287 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5291 leaf = path->nodes[0];
5292 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5293 struct btrfs_extent_item);
5294 btrfs_set_extent_refs(leaf, extent_item, 1);
5295 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5296 btrfs_set_extent_flags(leaf, extent_item,
5297 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5298 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5300 btrfs_set_tree_block_key(leaf, block_info, key);
5301 btrfs_set_tree_block_level(leaf, block_info, level);
5303 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5305 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5306 btrfs_set_extent_inline_ref_type(leaf, iref,
5307 BTRFS_SHARED_BLOCK_REF_KEY);
5308 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5310 btrfs_set_extent_inline_ref_type(leaf, iref,
5311 BTRFS_TREE_BLOCK_REF_KEY);
5312 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5315 btrfs_mark_buffer_dirty(leaf);
5316 btrfs_free_path(path);
5318 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5320 printk(KERN_ERR "btrfs update block group failed for %llu "
5321 "%llu\n", (unsigned long long)ins->objectid,
5322 (unsigned long long)ins->offset);
5328 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5329 struct btrfs_root *root,
5330 u64 root_objectid, u64 owner,
5331 u64 offset, struct btrfs_key *ins)
5335 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5337 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5338 0, root_objectid, owner, offset,
5339 BTRFS_ADD_DELAYED_EXTENT, NULL);
5344 * this is used by the tree logging recovery code. It records that
5345 * an extent has been allocated and makes sure to clear the free
5346 * space cache bits as well
5348 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5349 struct btrfs_root *root,
5350 u64 root_objectid, u64 owner, u64 offset,
5351 struct btrfs_key *ins)
5354 struct btrfs_block_group_cache *block_group;
5355 struct btrfs_caching_control *caching_ctl;
5356 u64 start = ins->objectid;
5357 u64 num_bytes = ins->offset;
5359 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5360 cache_block_group(block_group);
5361 caching_ctl = get_caching_control(block_group);
5364 BUG_ON(!block_group_cache_done(block_group));
5365 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5368 mutex_lock(&caching_ctl->mutex);
5370 if (start >= caching_ctl->progress) {
5371 ret = add_excluded_extent(root, start, num_bytes);
5373 } else if (start + num_bytes <= caching_ctl->progress) {
5374 ret = btrfs_remove_free_space(block_group,
5378 num_bytes = caching_ctl->progress - start;
5379 ret = btrfs_remove_free_space(block_group,
5383 start = caching_ctl->progress;
5384 num_bytes = ins->objectid + ins->offset -
5385 caching_ctl->progress;
5386 ret = add_excluded_extent(root, start, num_bytes);
5390 mutex_unlock(&caching_ctl->mutex);
5391 put_caching_control(caching_ctl);
5394 ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5396 btrfs_put_block_group(block_group);
5397 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5398 0, owner, offset, ins, 1);
5402 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5403 struct btrfs_root *root,
5404 u64 bytenr, u32 blocksize,
5407 struct extent_buffer *buf;
5409 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5411 return ERR_PTR(-ENOMEM);
5412 btrfs_set_header_generation(buf, trans->transid);
5413 btrfs_set_buffer_lockdep_class(buf, level);
5414 btrfs_tree_lock(buf);
5415 clean_tree_block(trans, root, buf);
5417 btrfs_set_lock_blocking(buf);
5418 btrfs_set_buffer_uptodate(buf);
5420 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5422 * we allow two log transactions at a time, use different
5423 * EXENT bit to differentiate dirty pages.
5425 if (root->log_transid % 2 == 0)
5426 set_extent_dirty(&root->dirty_log_pages, buf->start,
5427 buf->start + buf->len - 1, GFP_NOFS);
5429 set_extent_new(&root->dirty_log_pages, buf->start,
5430 buf->start + buf->len - 1, GFP_NOFS);
5432 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5433 buf->start + buf->len - 1, GFP_NOFS);
5435 trans->blocks_used++;
5436 /* this returns a buffer locked for blocking */
5440 static struct btrfs_block_rsv *
5441 use_block_rsv(struct btrfs_trans_handle *trans,
5442 struct btrfs_root *root, u32 blocksize)
5444 struct btrfs_block_rsv *block_rsv;
5447 block_rsv = get_block_rsv(trans, root);
5449 if (block_rsv->size == 0) {
5450 ret = reserve_metadata_bytes(block_rsv, blocksize);
5452 return ERR_PTR(ret);
5456 ret = block_rsv_use_bytes(block_rsv, blocksize);
5461 printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5462 block_rsv->size, block_rsv->reserved,
5463 block_rsv->freed[0], block_rsv->freed[1]);
5465 return ERR_PTR(-ENOSPC);
5468 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5470 block_rsv_add_bytes(block_rsv, blocksize, 0);
5471 block_rsv_release_bytes(block_rsv, NULL, 0);
5475 * finds a free extent and does all the dirty work required for allocation
5476 * returns the key for the extent through ins, and a tree buffer for
5477 * the first block of the extent through buf.
5479 * returns the tree buffer or NULL.
5481 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5482 struct btrfs_root *root, u32 blocksize,
5483 u64 parent, u64 root_objectid,
5484 struct btrfs_disk_key *key, int level,
5485 u64 hint, u64 empty_size)
5487 struct btrfs_key ins;
5488 struct btrfs_block_rsv *block_rsv;
5489 struct extent_buffer *buf;
5494 block_rsv = use_block_rsv(trans, root, blocksize);
5495 if (IS_ERR(block_rsv))
5496 return ERR_CAST(block_rsv);
5498 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5499 empty_size, hint, (u64)-1, &ins, 0);
5501 unuse_block_rsv(block_rsv, blocksize);
5502 return ERR_PTR(ret);
5505 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5507 BUG_ON(IS_ERR(buf));
5509 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5511 parent = ins.objectid;
5512 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5516 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5517 struct btrfs_delayed_extent_op *extent_op;
5518 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5521 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5523 memset(&extent_op->key, 0, sizeof(extent_op->key));
5524 extent_op->flags_to_set = flags;
5525 extent_op->update_key = 1;
5526 extent_op->update_flags = 1;
5527 extent_op->is_data = 0;
5529 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5530 ins.offset, parent, root_objectid,
5531 level, BTRFS_ADD_DELAYED_EXTENT,
5538 struct walk_control {
5539 u64 refs[BTRFS_MAX_LEVEL];
5540 u64 flags[BTRFS_MAX_LEVEL];
5541 struct btrfs_key update_progress;
5551 #define DROP_REFERENCE 1
5552 #define UPDATE_BACKREF 2
5554 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5555 struct btrfs_root *root,
5556 struct walk_control *wc,
5557 struct btrfs_path *path)
5566 struct btrfs_key key;
5567 struct extent_buffer *eb;
5572 if (path->slots[wc->level] < wc->reada_slot) {
5573 wc->reada_count = wc->reada_count * 2 / 3;
5574 wc->reada_count = max(wc->reada_count, 2);
5576 wc->reada_count = wc->reada_count * 3 / 2;
5577 wc->reada_count = min_t(int, wc->reada_count,
5578 BTRFS_NODEPTRS_PER_BLOCK(root));
5581 eb = path->nodes[wc->level];
5582 nritems = btrfs_header_nritems(eb);
5583 blocksize = btrfs_level_size(root, wc->level - 1);
5585 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5586 if (nread >= wc->reada_count)
5590 bytenr = btrfs_node_blockptr(eb, slot);
5591 generation = btrfs_node_ptr_generation(eb, slot);
5593 if (slot == path->slots[wc->level])
5596 if (wc->stage == UPDATE_BACKREF &&
5597 generation <= root->root_key.offset)
5600 /* We don't lock the tree block, it's OK to be racy here */
5601 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5606 if (wc->stage == DROP_REFERENCE) {
5610 if (wc->level == 1 &&
5611 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5613 if (!wc->update_ref ||
5614 generation <= root->root_key.offset)
5616 btrfs_node_key_to_cpu(eb, &key, slot);
5617 ret = btrfs_comp_cpu_keys(&key,
5618 &wc->update_progress);
5622 if (wc->level == 1 &&
5623 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5627 ret = readahead_tree_block(root, bytenr, blocksize,
5631 last = bytenr + blocksize;
5634 wc->reada_slot = slot;
5638 * hepler to process tree block while walking down the tree.
5640 * when wc->stage == UPDATE_BACKREF, this function updates
5641 * back refs for pointers in the block.
5643 * NOTE: return value 1 means we should stop walking down.
5645 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5646 struct btrfs_root *root,
5647 struct btrfs_path *path,
5648 struct walk_control *wc, int lookup_info)
5650 int level = wc->level;
5651 struct extent_buffer *eb = path->nodes[level];
5652 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5655 if (wc->stage == UPDATE_BACKREF &&
5656 btrfs_header_owner(eb) != root->root_key.objectid)
5660 * when reference count of tree block is 1, it won't increase
5661 * again. once full backref flag is set, we never clear it.
5664 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5665 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5666 BUG_ON(!path->locks[level]);
5667 ret = btrfs_lookup_extent_info(trans, root,
5672 BUG_ON(wc->refs[level] == 0);
5675 if (wc->stage == DROP_REFERENCE) {
5676 if (wc->refs[level] > 1)
5679 if (path->locks[level] && !wc->keep_locks) {
5680 btrfs_tree_unlock(eb);
5681 path->locks[level] = 0;
5686 /* wc->stage == UPDATE_BACKREF */
5687 if (!(wc->flags[level] & flag)) {
5688 BUG_ON(!path->locks[level]);
5689 ret = btrfs_inc_ref(trans, root, eb, 1);
5691 ret = btrfs_dec_ref(trans, root, eb, 0);
5693 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5696 wc->flags[level] |= flag;
5700 * the block is shared by multiple trees, so it's not good to
5701 * keep the tree lock
5703 if (path->locks[level] && level > 0) {
5704 btrfs_tree_unlock(eb);
5705 path->locks[level] = 0;
5711 * hepler to process tree block pointer.
5713 * when wc->stage == DROP_REFERENCE, this function checks
5714 * reference count of the block pointed to. if the block
5715 * is shared and we need update back refs for the subtree
5716 * rooted at the block, this function changes wc->stage to
5717 * UPDATE_BACKREF. if the block is shared and there is no
5718 * need to update back, this function drops the reference
5721 * NOTE: return value 1 means we should stop walking down.
5723 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5724 struct btrfs_root *root,
5725 struct btrfs_path *path,
5726 struct walk_control *wc, int *lookup_info)
5732 struct btrfs_key key;
5733 struct extent_buffer *next;
5734 int level = wc->level;
5738 generation = btrfs_node_ptr_generation(path->nodes[level],
5739 path->slots[level]);
5741 * if the lower level block was created before the snapshot
5742 * was created, we know there is no need to update back refs
5745 if (wc->stage == UPDATE_BACKREF &&
5746 generation <= root->root_key.offset) {
5751 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5752 blocksize = btrfs_level_size(root, level - 1);
5754 next = btrfs_find_tree_block(root, bytenr, blocksize);
5756 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5761 btrfs_tree_lock(next);
5762 btrfs_set_lock_blocking(next);
5764 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5765 &wc->refs[level - 1],
5766 &wc->flags[level - 1]);
5768 BUG_ON(wc->refs[level - 1] == 0);
5771 if (wc->stage == DROP_REFERENCE) {
5772 if (wc->refs[level - 1] > 1) {
5774 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5777 if (!wc->update_ref ||
5778 generation <= root->root_key.offset)
5781 btrfs_node_key_to_cpu(path->nodes[level], &key,
5782 path->slots[level]);
5783 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5787 wc->stage = UPDATE_BACKREF;
5788 wc->shared_level = level - 1;
5792 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5796 if (!btrfs_buffer_uptodate(next, generation)) {
5797 btrfs_tree_unlock(next);
5798 free_extent_buffer(next);
5804 if (reada && level == 1)
5805 reada_walk_down(trans, root, wc, path);
5806 next = read_tree_block(root, bytenr, blocksize, generation);
5807 btrfs_tree_lock(next);
5808 btrfs_set_lock_blocking(next);
5812 BUG_ON(level != btrfs_header_level(next));
5813 path->nodes[level] = next;
5814 path->slots[level] = 0;
5815 path->locks[level] = 1;
5821 wc->refs[level - 1] = 0;
5822 wc->flags[level - 1] = 0;
5823 if (wc->stage == DROP_REFERENCE) {
5824 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5825 parent = path->nodes[level]->start;
5827 BUG_ON(root->root_key.objectid !=
5828 btrfs_header_owner(path->nodes[level]));
5832 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5833 root->root_key.objectid, level - 1, 0);
5836 btrfs_tree_unlock(next);
5837 free_extent_buffer(next);
5843 * hepler to process tree block while walking up the tree.
5845 * when wc->stage == DROP_REFERENCE, this function drops
5846 * reference count on the block.
5848 * when wc->stage == UPDATE_BACKREF, this function changes
5849 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5850 * to UPDATE_BACKREF previously while processing the block.
5852 * NOTE: return value 1 means we should stop walking up.
5854 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5855 struct btrfs_root *root,
5856 struct btrfs_path *path,
5857 struct walk_control *wc)
5860 int level = wc->level;
5861 struct extent_buffer *eb = path->nodes[level];
5864 if (wc->stage == UPDATE_BACKREF) {
5865 BUG_ON(wc->shared_level < level);
5866 if (level < wc->shared_level)
5869 ret = find_next_key(path, level + 1, &wc->update_progress);
5873 wc->stage = DROP_REFERENCE;
5874 wc->shared_level = -1;
5875 path->slots[level] = 0;
5878 * check reference count again if the block isn't locked.
5879 * we should start walking down the tree again if reference
5882 if (!path->locks[level]) {
5884 btrfs_tree_lock(eb);
5885 btrfs_set_lock_blocking(eb);
5886 path->locks[level] = 1;
5888 ret = btrfs_lookup_extent_info(trans, root,
5893 BUG_ON(wc->refs[level] == 0);
5894 if (wc->refs[level] == 1) {
5895 btrfs_tree_unlock(eb);
5896 path->locks[level] = 0;
5902 /* wc->stage == DROP_REFERENCE */
5903 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5905 if (wc->refs[level] == 1) {
5907 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5908 ret = btrfs_dec_ref(trans, root, eb, 1);
5910 ret = btrfs_dec_ref(trans, root, eb, 0);
5913 /* make block locked assertion in clean_tree_block happy */
5914 if (!path->locks[level] &&
5915 btrfs_header_generation(eb) == trans->transid) {
5916 btrfs_tree_lock(eb);
5917 btrfs_set_lock_blocking(eb);
5918 path->locks[level] = 1;
5920 clean_tree_block(trans, root, eb);
5923 if (eb == root->node) {
5924 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5927 BUG_ON(root->root_key.objectid !=
5928 btrfs_header_owner(eb));
5930 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5931 parent = path->nodes[level + 1]->start;
5933 BUG_ON(root->root_key.objectid !=
5934 btrfs_header_owner(path->nodes[level + 1]));
5937 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5939 wc->refs[level] = 0;
5940 wc->flags[level] = 0;
5944 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5945 struct btrfs_root *root,
5946 struct btrfs_path *path,
5947 struct walk_control *wc)
5949 int level = wc->level;
5950 int lookup_info = 1;
5953 while (level >= 0) {
5954 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5961 if (path->slots[level] >=
5962 btrfs_header_nritems(path->nodes[level]))
5965 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5967 path->slots[level]++;
5976 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5977 struct btrfs_root *root,
5978 struct btrfs_path *path,
5979 struct walk_control *wc, int max_level)
5981 int level = wc->level;
5984 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5985 while (level < max_level && path->nodes[level]) {
5987 if (path->slots[level] + 1 <
5988 btrfs_header_nritems(path->nodes[level])) {
5989 path->slots[level]++;
5992 ret = walk_up_proc(trans, root, path, wc);
5996 if (path->locks[level]) {
5997 btrfs_tree_unlock(path->nodes[level]);
5998 path->locks[level] = 0;
6000 free_extent_buffer(path->nodes[level]);
6001 path->nodes[level] = NULL;
6009 * drop a subvolume tree.
6011 * this function traverses the tree freeing any blocks that only
6012 * referenced by the tree.
6014 * when a shared tree block is found. this function decreases its
6015 * reference count by one. if update_ref is true, this function
6016 * also make sure backrefs for the shared block and all lower level
6017 * blocks are properly updated.
6019 int btrfs_drop_snapshot(struct btrfs_root *root,
6020 struct btrfs_block_rsv *block_rsv, int update_ref)
6022 struct btrfs_path *path;
6023 struct btrfs_trans_handle *trans;
6024 struct btrfs_root *tree_root = root->fs_info->tree_root;
6025 struct btrfs_root_item *root_item = &root->root_item;
6026 struct walk_control *wc;
6027 struct btrfs_key key;
6032 path = btrfs_alloc_path();
6035 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6038 trans = btrfs_start_transaction(tree_root, 0);
6040 trans->block_rsv = block_rsv;
6042 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6043 level = btrfs_header_level(root->node);
6044 path->nodes[level] = btrfs_lock_root_node(root);
6045 btrfs_set_lock_blocking(path->nodes[level]);
6046 path->slots[level] = 0;
6047 path->locks[level] = 1;
6048 memset(&wc->update_progress, 0,
6049 sizeof(wc->update_progress));
6051 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6052 memcpy(&wc->update_progress, &key,
6053 sizeof(wc->update_progress));
6055 level = root_item->drop_level;
6057 path->lowest_level = level;
6058 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6059 path->lowest_level = 0;
6067 * unlock our path, this is safe because only this
6068 * function is allowed to delete this snapshot
6070 btrfs_unlock_up_safe(path, 0);
6072 level = btrfs_header_level(root->node);
6074 btrfs_tree_lock(path->nodes[level]);
6075 btrfs_set_lock_blocking(path->nodes[level]);
6077 ret = btrfs_lookup_extent_info(trans, root,
6078 path->nodes[level]->start,
6079 path->nodes[level]->len,
6083 BUG_ON(wc->refs[level] == 0);
6085 if (level == root_item->drop_level)
6088 btrfs_tree_unlock(path->nodes[level]);
6089 WARN_ON(wc->refs[level] != 1);
6095 wc->shared_level = -1;
6096 wc->stage = DROP_REFERENCE;
6097 wc->update_ref = update_ref;
6099 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6102 ret = walk_down_tree(trans, root, path, wc);
6108 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6115 BUG_ON(wc->stage != DROP_REFERENCE);
6119 if (wc->stage == DROP_REFERENCE) {
6121 btrfs_node_key(path->nodes[level],
6122 &root_item->drop_progress,
6123 path->slots[level]);
6124 root_item->drop_level = level;
6127 BUG_ON(wc->level == 0);
6128 if (btrfs_should_end_transaction(trans, tree_root)) {
6129 ret = btrfs_update_root(trans, tree_root,
6134 btrfs_end_transaction_throttle(trans, tree_root);
6135 trans = btrfs_start_transaction(tree_root, 0);
6137 trans->block_rsv = block_rsv;
6140 btrfs_release_path(root, path);
6143 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6146 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6147 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6151 ret = btrfs_del_orphan_item(trans, tree_root,
6152 root->root_key.objectid);
6157 if (root->in_radix) {
6158 btrfs_free_fs_root(tree_root->fs_info, root);
6160 free_extent_buffer(root->node);
6161 free_extent_buffer(root->commit_root);
6165 btrfs_end_transaction_throttle(trans, tree_root);
6167 btrfs_free_path(path);
6172 * drop subtree rooted at tree block 'node'.
6174 * NOTE: this function will unlock and release tree block 'node'
6176 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6177 struct btrfs_root *root,
6178 struct extent_buffer *node,
6179 struct extent_buffer *parent)
6181 struct btrfs_path *path;
6182 struct walk_control *wc;
6188 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6190 path = btrfs_alloc_path();
6193 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6196 btrfs_assert_tree_locked(parent);
6197 parent_level = btrfs_header_level(parent);
6198 extent_buffer_get(parent);
6199 path->nodes[parent_level] = parent;
6200 path->slots[parent_level] = btrfs_header_nritems(parent);
6202 btrfs_assert_tree_locked(node);
6203 level = btrfs_header_level(node);
6204 path->nodes[level] = node;
6205 path->slots[level] = 0;
6206 path->locks[level] = 1;
6208 wc->refs[parent_level] = 1;
6209 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6211 wc->shared_level = -1;
6212 wc->stage = DROP_REFERENCE;
6215 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6218 wret = walk_down_tree(trans, root, path, wc);
6224 wret = walk_up_tree(trans, root, path, wc, parent_level);
6232 btrfs_free_path(path);
6237 static unsigned long calc_ra(unsigned long start, unsigned long last,
6240 return min(last, start + nr - 1);
6243 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6248 unsigned long first_index;
6249 unsigned long last_index;
6252 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6253 struct file_ra_state *ra;
6254 struct btrfs_ordered_extent *ordered;
6255 unsigned int total_read = 0;
6256 unsigned int total_dirty = 0;
6259 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6261 mutex_lock(&inode->i_mutex);
6262 first_index = start >> PAGE_CACHE_SHIFT;
6263 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6265 /* make sure the dirty trick played by the caller work */
6266 ret = invalidate_inode_pages2_range(inode->i_mapping,
6267 first_index, last_index);
6271 file_ra_state_init(ra, inode->i_mapping);
6273 for (i = first_index ; i <= last_index; i++) {
6274 if (total_read % ra->ra_pages == 0) {
6275 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6276 calc_ra(i, last_index, ra->ra_pages));
6280 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6282 page = grab_cache_page(inode->i_mapping, i);
6287 if (!PageUptodate(page)) {
6288 btrfs_readpage(NULL, page);
6290 if (!PageUptodate(page)) {
6292 page_cache_release(page);
6297 wait_on_page_writeback(page);
6299 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6300 page_end = page_start + PAGE_CACHE_SIZE - 1;
6301 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6303 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6305 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6307 page_cache_release(page);
6308 btrfs_start_ordered_extent(inode, ordered, 1);
6309 btrfs_put_ordered_extent(ordered);
6312 set_page_extent_mapped(page);
6314 if (i == first_index)
6315 set_extent_bits(io_tree, page_start, page_end,
6316 EXTENT_BOUNDARY, GFP_NOFS);
6317 btrfs_set_extent_delalloc(inode, page_start, page_end);
6319 set_page_dirty(page);
6322 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6324 page_cache_release(page);
6329 mutex_unlock(&inode->i_mutex);
6330 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6334 static noinline int relocate_data_extent(struct inode *reloc_inode,
6335 struct btrfs_key *extent_key,
6338 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6339 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6340 struct extent_map *em;
6341 u64 start = extent_key->objectid - offset;
6342 u64 end = start + extent_key->offset - 1;
6344 em = alloc_extent_map(GFP_NOFS);
6345 BUG_ON(!em || IS_ERR(em));
6348 em->len = extent_key->offset;
6349 em->block_len = extent_key->offset;
6350 em->block_start = extent_key->objectid;
6351 em->bdev = root->fs_info->fs_devices->latest_bdev;
6352 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6354 /* setup extent map to cheat btrfs_readpage */
6355 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6358 write_lock(&em_tree->lock);
6359 ret = add_extent_mapping(em_tree, em);
6360 write_unlock(&em_tree->lock);
6361 if (ret != -EEXIST) {
6362 free_extent_map(em);
6365 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6367 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6369 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6372 struct btrfs_ref_path {
6374 u64 nodes[BTRFS_MAX_LEVEL];
6376 u64 root_generation;
6383 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6384 u64 new_nodes[BTRFS_MAX_LEVEL];
6387 struct disk_extent {
6398 static int is_cowonly_root(u64 root_objectid)
6400 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6401 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6402 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6403 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6404 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6405 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6410 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6411 struct btrfs_root *extent_root,
6412 struct btrfs_ref_path *ref_path,
6415 struct extent_buffer *leaf;
6416 struct btrfs_path *path;
6417 struct btrfs_extent_ref *ref;
6418 struct btrfs_key key;
6419 struct btrfs_key found_key;
6425 path = btrfs_alloc_path();
6430 ref_path->lowest_level = -1;
6431 ref_path->current_level = -1;
6432 ref_path->shared_level = -1;
6436 level = ref_path->current_level - 1;
6437 while (level >= -1) {
6439 if (level < ref_path->lowest_level)
6443 bytenr = ref_path->nodes[level];
6445 bytenr = ref_path->extent_start;
6446 BUG_ON(bytenr == 0);
6448 parent = ref_path->nodes[level + 1];
6449 ref_path->nodes[level + 1] = 0;
6450 ref_path->current_level = level;
6451 BUG_ON(parent == 0);
6453 key.objectid = bytenr;
6454 key.offset = parent + 1;
6455 key.type = BTRFS_EXTENT_REF_KEY;
6457 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6462 leaf = path->nodes[0];
6463 nritems = btrfs_header_nritems(leaf);
6464 if (path->slots[0] >= nritems) {
6465 ret = btrfs_next_leaf(extent_root, path);
6470 leaf = path->nodes[0];
6473 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6474 if (found_key.objectid == bytenr &&
6475 found_key.type == BTRFS_EXTENT_REF_KEY) {
6476 if (level < ref_path->shared_level)
6477 ref_path->shared_level = level;
6482 btrfs_release_path(extent_root, path);
6485 /* reached lowest level */
6489 level = ref_path->current_level;
6490 while (level < BTRFS_MAX_LEVEL - 1) {
6494 bytenr = ref_path->nodes[level];
6496 bytenr = ref_path->extent_start;
6498 BUG_ON(bytenr == 0);
6500 key.objectid = bytenr;
6502 key.type = BTRFS_EXTENT_REF_KEY;
6504 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6508 leaf = path->nodes[0];
6509 nritems = btrfs_header_nritems(leaf);
6510 if (path->slots[0] >= nritems) {
6511 ret = btrfs_next_leaf(extent_root, path);
6515 /* the extent was freed by someone */
6516 if (ref_path->lowest_level == level)
6518 btrfs_release_path(extent_root, path);
6521 leaf = path->nodes[0];
6524 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6525 if (found_key.objectid != bytenr ||
6526 found_key.type != BTRFS_EXTENT_REF_KEY) {
6527 /* the extent was freed by someone */
6528 if (ref_path->lowest_level == level) {
6532 btrfs_release_path(extent_root, path);
6536 ref = btrfs_item_ptr(leaf, path->slots[0],
6537 struct btrfs_extent_ref);
6538 ref_objectid = btrfs_ref_objectid(leaf, ref);
6539 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6541 level = (int)ref_objectid;
6542 BUG_ON(level >= BTRFS_MAX_LEVEL);
6543 ref_path->lowest_level = level;
6544 ref_path->current_level = level;
6545 ref_path->nodes[level] = bytenr;
6547 WARN_ON(ref_objectid != level);
6550 WARN_ON(level != -1);
6554 if (ref_path->lowest_level == level) {
6555 ref_path->owner_objectid = ref_objectid;
6556 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6560 * the block is tree root or the block isn't in reference
6563 if (found_key.objectid == found_key.offset ||
6564 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6565 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6566 ref_path->root_generation =
6567 btrfs_ref_generation(leaf, ref);
6569 /* special reference from the tree log */
6570 ref_path->nodes[0] = found_key.offset;
6571 ref_path->current_level = 0;
6578 BUG_ON(ref_path->nodes[level] != 0);
6579 ref_path->nodes[level] = found_key.offset;
6580 ref_path->current_level = level;
6583 * the reference was created in the running transaction,
6584 * no need to continue walking up.
6586 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6587 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6588 ref_path->root_generation =
6589 btrfs_ref_generation(leaf, ref);
6594 btrfs_release_path(extent_root, path);
6597 /* reached max tree level, but no tree root found. */
6600 btrfs_free_path(path);
6604 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6605 struct btrfs_root *extent_root,
6606 struct btrfs_ref_path *ref_path,
6609 memset(ref_path, 0, sizeof(*ref_path));
6610 ref_path->extent_start = extent_start;
6612 return __next_ref_path(trans, extent_root, ref_path, 1);
6615 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6616 struct btrfs_root *extent_root,
6617 struct btrfs_ref_path *ref_path)
6619 return __next_ref_path(trans, extent_root, ref_path, 0);
6622 static noinline int get_new_locations(struct inode *reloc_inode,
6623 struct btrfs_key *extent_key,
6624 u64 offset, int no_fragment,
6625 struct disk_extent **extents,
6628 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6629 struct btrfs_path *path;
6630 struct btrfs_file_extent_item *fi;
6631 struct extent_buffer *leaf;
6632 struct disk_extent *exts = *extents;
6633 struct btrfs_key found_key;
6638 int max = *nr_extents;
6641 WARN_ON(!no_fragment && *extents);
6644 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6649 path = btrfs_alloc_path();
6652 cur_pos = extent_key->objectid - offset;
6653 last_byte = extent_key->objectid + extent_key->offset;
6654 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6664 leaf = path->nodes[0];
6665 nritems = btrfs_header_nritems(leaf);
6666 if (path->slots[0] >= nritems) {
6667 ret = btrfs_next_leaf(root, path);
6672 leaf = path->nodes[0];
6675 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6676 if (found_key.offset != cur_pos ||
6677 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6678 found_key.objectid != reloc_inode->i_ino)
6681 fi = btrfs_item_ptr(leaf, path->slots[0],
6682 struct btrfs_file_extent_item);
6683 if (btrfs_file_extent_type(leaf, fi) !=
6684 BTRFS_FILE_EXTENT_REG ||
6685 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6689 struct disk_extent *old = exts;
6691 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6692 memcpy(exts, old, sizeof(*exts) * nr);
6693 if (old != *extents)
6697 exts[nr].disk_bytenr =
6698 btrfs_file_extent_disk_bytenr(leaf, fi);
6699 exts[nr].disk_num_bytes =
6700 btrfs_file_extent_disk_num_bytes(leaf, fi);
6701 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6702 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6703 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6704 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6705 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6706 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6708 BUG_ON(exts[nr].offset > 0);
6709 BUG_ON(exts[nr].compression || exts[nr].encryption);
6710 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6712 cur_pos += exts[nr].num_bytes;
6715 if (cur_pos + offset >= last_byte)
6725 BUG_ON(cur_pos + offset > last_byte);
6726 if (cur_pos + offset < last_byte) {
6732 btrfs_free_path(path);
6734 if (exts != *extents)
6743 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6744 struct btrfs_root *root,
6745 struct btrfs_path *path,
6746 struct btrfs_key *extent_key,
6747 struct btrfs_key *leaf_key,
6748 struct btrfs_ref_path *ref_path,
6749 struct disk_extent *new_extents,
6752 struct extent_buffer *leaf;
6753 struct btrfs_file_extent_item *fi;
6754 struct inode *inode = NULL;
6755 struct btrfs_key key;
6760 u64 search_end = (u64)-1;
6763 int extent_locked = 0;
6767 memcpy(&key, leaf_key, sizeof(key));
6768 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6769 if (key.objectid < ref_path->owner_objectid ||
6770 (key.objectid == ref_path->owner_objectid &&
6771 key.type < BTRFS_EXTENT_DATA_KEY)) {
6772 key.objectid = ref_path->owner_objectid;
6773 key.type = BTRFS_EXTENT_DATA_KEY;
6779 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6783 leaf = path->nodes[0];
6784 nritems = btrfs_header_nritems(leaf);
6786 if (extent_locked && ret > 0) {
6788 * the file extent item was modified by someone
6789 * before the extent got locked.
6791 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6792 lock_end, GFP_NOFS);
6796 if (path->slots[0] >= nritems) {
6797 if (++nr_scaned > 2)
6800 BUG_ON(extent_locked);
6801 ret = btrfs_next_leaf(root, path);
6806 leaf = path->nodes[0];
6807 nritems = btrfs_header_nritems(leaf);
6810 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6812 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6813 if ((key.objectid > ref_path->owner_objectid) ||
6814 (key.objectid == ref_path->owner_objectid &&
6815 key.type > BTRFS_EXTENT_DATA_KEY) ||
6816 key.offset >= search_end)
6820 if (inode && key.objectid != inode->i_ino) {
6821 BUG_ON(extent_locked);
6822 btrfs_release_path(root, path);
6823 mutex_unlock(&inode->i_mutex);
6829 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6834 fi = btrfs_item_ptr(leaf, path->slots[0],
6835 struct btrfs_file_extent_item);
6836 extent_type = btrfs_file_extent_type(leaf, fi);
6837 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6838 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6839 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6840 extent_key->objectid)) {
6846 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6847 ext_offset = btrfs_file_extent_offset(leaf, fi);
6849 if (search_end == (u64)-1) {
6850 search_end = key.offset - ext_offset +
6851 btrfs_file_extent_ram_bytes(leaf, fi);
6854 if (!extent_locked) {
6855 lock_start = key.offset;
6856 lock_end = lock_start + num_bytes - 1;
6858 if (lock_start > key.offset ||
6859 lock_end + 1 < key.offset + num_bytes) {
6860 unlock_extent(&BTRFS_I(inode)->io_tree,
6861 lock_start, lock_end, GFP_NOFS);
6867 btrfs_release_path(root, path);
6869 inode = btrfs_iget_locked(root->fs_info->sb,
6870 key.objectid, root);
6871 if (inode->i_state & I_NEW) {
6872 BTRFS_I(inode)->root = root;
6873 BTRFS_I(inode)->location.objectid =
6875 BTRFS_I(inode)->location.type =
6876 BTRFS_INODE_ITEM_KEY;
6877 BTRFS_I(inode)->location.offset = 0;
6878 btrfs_read_locked_inode(inode);
6879 unlock_new_inode(inode);
6882 * some code call btrfs_commit_transaction while
6883 * holding the i_mutex, so we can't use mutex_lock
6886 if (is_bad_inode(inode) ||
6887 !mutex_trylock(&inode->i_mutex)) {
6890 key.offset = (u64)-1;
6895 if (!extent_locked) {
6896 struct btrfs_ordered_extent *ordered;
6898 btrfs_release_path(root, path);
6900 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6901 lock_end, GFP_NOFS);
6902 ordered = btrfs_lookup_first_ordered_extent(inode,
6905 ordered->file_offset <= lock_end &&
6906 ordered->file_offset + ordered->len > lock_start) {
6907 unlock_extent(&BTRFS_I(inode)->io_tree,
6908 lock_start, lock_end, GFP_NOFS);
6909 btrfs_start_ordered_extent(inode, ordered, 1);
6910 btrfs_put_ordered_extent(ordered);
6911 key.offset += num_bytes;
6915 btrfs_put_ordered_extent(ordered);
6921 if (nr_extents == 1) {
6922 /* update extent pointer in place */
6923 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6924 new_extents[0].disk_bytenr);
6925 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6926 new_extents[0].disk_num_bytes);
6927 btrfs_mark_buffer_dirty(leaf);
6929 btrfs_drop_extent_cache(inode, key.offset,
6930 key.offset + num_bytes - 1, 0);
6932 ret = btrfs_inc_extent_ref(trans, root,
6933 new_extents[0].disk_bytenr,
6934 new_extents[0].disk_num_bytes,
6936 root->root_key.objectid,
6941 ret = btrfs_free_extent(trans, root,
6942 extent_key->objectid,
6945 btrfs_header_owner(leaf),
6946 btrfs_header_generation(leaf),
6950 btrfs_release_path(root, path);
6951 key.offset += num_bytes;
6959 * drop old extent pointer at first, then insert the
6960 * new pointers one bye one
6962 btrfs_release_path(root, path);
6963 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6964 key.offset + num_bytes,
6965 key.offset, &alloc_hint);
6968 for (i = 0; i < nr_extents; i++) {
6969 if (ext_offset >= new_extents[i].num_bytes) {
6970 ext_offset -= new_extents[i].num_bytes;
6973 extent_len = min(new_extents[i].num_bytes -
6974 ext_offset, num_bytes);
6976 ret = btrfs_insert_empty_item(trans, root,
6981 leaf = path->nodes[0];
6982 fi = btrfs_item_ptr(leaf, path->slots[0],
6983 struct btrfs_file_extent_item);
6984 btrfs_set_file_extent_generation(leaf, fi,
6986 btrfs_set_file_extent_type(leaf, fi,
6987 BTRFS_FILE_EXTENT_REG);
6988 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6989 new_extents[i].disk_bytenr);
6990 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6991 new_extents[i].disk_num_bytes);
6992 btrfs_set_file_extent_ram_bytes(leaf, fi,
6993 new_extents[i].ram_bytes);
6995 btrfs_set_file_extent_compression(leaf, fi,
6996 new_extents[i].compression);
6997 btrfs_set_file_extent_encryption(leaf, fi,
6998 new_extents[i].encryption);
6999 btrfs_set_file_extent_other_encoding(leaf, fi,
7000 new_extents[i].other_encoding);
7002 btrfs_set_file_extent_num_bytes(leaf, fi,
7004 ext_offset += new_extents[i].offset;
7005 btrfs_set_file_extent_offset(leaf, fi,
7007 btrfs_mark_buffer_dirty(leaf);
7009 btrfs_drop_extent_cache(inode, key.offset,
7010 key.offset + extent_len - 1, 0);
7012 ret = btrfs_inc_extent_ref(trans, root,
7013 new_extents[i].disk_bytenr,
7014 new_extents[i].disk_num_bytes,
7016 root->root_key.objectid,
7017 trans->transid, key.objectid);
7019 btrfs_release_path(root, path);
7021 inode_add_bytes(inode, extent_len);
7024 num_bytes -= extent_len;
7025 key.offset += extent_len;
7030 BUG_ON(i >= nr_extents);
7034 if (extent_locked) {
7035 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7036 lock_end, GFP_NOFS);
7040 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7041 key.offset >= search_end)
7048 btrfs_release_path(root, path);
7050 mutex_unlock(&inode->i_mutex);
7051 if (extent_locked) {
7052 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7053 lock_end, GFP_NOFS);
7060 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7061 struct btrfs_root *root,
7062 struct extent_buffer *buf, u64 orig_start)
7067 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7068 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7070 level = btrfs_header_level(buf);
7072 struct btrfs_leaf_ref *ref;
7073 struct btrfs_leaf_ref *orig_ref;
7075 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7079 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7081 btrfs_free_leaf_ref(root, orig_ref);
7085 ref->nritems = orig_ref->nritems;
7086 memcpy(ref->extents, orig_ref->extents,
7087 sizeof(ref->extents[0]) * ref->nritems);
7089 btrfs_free_leaf_ref(root, orig_ref);
7091 ref->root_gen = trans->transid;
7092 ref->bytenr = buf->start;
7093 ref->owner = btrfs_header_owner(buf);
7094 ref->generation = btrfs_header_generation(buf);
7096 ret = btrfs_add_leaf_ref(root, ref, 0);
7098 btrfs_free_leaf_ref(root, ref);
7103 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7104 struct extent_buffer *leaf,
7105 struct btrfs_block_group_cache *group,
7106 struct btrfs_root *target_root)
7108 struct btrfs_key key;
7109 struct inode *inode = NULL;
7110 struct btrfs_file_extent_item *fi;
7111 struct extent_state *cached_state = NULL;
7113 u64 skip_objectid = 0;
7117 nritems = btrfs_header_nritems(leaf);
7118 for (i = 0; i < nritems; i++) {
7119 btrfs_item_key_to_cpu(leaf, &key, i);
7120 if (key.objectid == skip_objectid ||
7121 key.type != BTRFS_EXTENT_DATA_KEY)
7123 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7124 if (btrfs_file_extent_type(leaf, fi) ==
7125 BTRFS_FILE_EXTENT_INLINE)
7127 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7129 if (!inode || inode->i_ino != key.objectid) {
7131 inode = btrfs_ilookup(target_root->fs_info->sb,
7132 key.objectid, target_root, 1);
7135 skip_objectid = key.objectid;
7138 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7140 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7141 key.offset + num_bytes - 1, 0, &cached_state,
7143 btrfs_drop_extent_cache(inode, key.offset,
7144 key.offset + num_bytes - 1, 1);
7145 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7146 key.offset + num_bytes - 1, &cached_state,
7154 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7155 struct btrfs_root *root,
7156 struct extent_buffer *leaf,
7157 struct btrfs_block_group_cache *group,
7158 struct inode *reloc_inode)
7160 struct btrfs_key key;
7161 struct btrfs_key extent_key;
7162 struct btrfs_file_extent_item *fi;
7163 struct btrfs_leaf_ref *ref;
7164 struct disk_extent *new_extent;
7173 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7174 BUG_ON(!new_extent);
7176 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7180 nritems = btrfs_header_nritems(leaf);
7181 for (i = 0; i < nritems; i++) {
7182 btrfs_item_key_to_cpu(leaf, &key, i);
7183 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7185 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7186 if (btrfs_file_extent_type(leaf, fi) ==
7187 BTRFS_FILE_EXTENT_INLINE)
7189 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7190 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7195 if (bytenr >= group->key.objectid + group->key.offset ||
7196 bytenr + num_bytes <= group->key.objectid)
7199 extent_key.objectid = bytenr;
7200 extent_key.offset = num_bytes;
7201 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7203 ret = get_new_locations(reloc_inode, &extent_key,
7204 group->key.objectid, 1,
7205 &new_extent, &nr_extent);
7210 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7211 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7212 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7213 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7215 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7216 new_extent->disk_bytenr);
7217 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7218 new_extent->disk_num_bytes);
7219 btrfs_mark_buffer_dirty(leaf);
7221 ret = btrfs_inc_extent_ref(trans, root,
7222 new_extent->disk_bytenr,
7223 new_extent->disk_num_bytes,
7225 root->root_key.objectid,
7226 trans->transid, key.objectid);
7229 ret = btrfs_free_extent(trans, root,
7230 bytenr, num_bytes, leaf->start,
7231 btrfs_header_owner(leaf),
7232 btrfs_header_generation(leaf),
7238 BUG_ON(ext_index + 1 != ref->nritems);
7239 btrfs_free_leaf_ref(root, ref);
7243 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7244 struct btrfs_root *root)
7246 struct btrfs_root *reloc_root;
7249 if (root->reloc_root) {
7250 reloc_root = root->reloc_root;
7251 root->reloc_root = NULL;
7252 list_add(&reloc_root->dead_list,
7253 &root->fs_info->dead_reloc_roots);
7255 btrfs_set_root_bytenr(&reloc_root->root_item,
7256 reloc_root->node->start);
7257 btrfs_set_root_level(&root->root_item,
7258 btrfs_header_level(reloc_root->node));
7259 memset(&reloc_root->root_item.drop_progress, 0,
7260 sizeof(struct btrfs_disk_key));
7261 reloc_root->root_item.drop_level = 0;
7263 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7264 &reloc_root->root_key,
7265 &reloc_root->root_item);
7271 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7273 struct btrfs_trans_handle *trans;
7274 struct btrfs_root *reloc_root;
7275 struct btrfs_root *prev_root = NULL;
7276 struct list_head dead_roots;
7280 INIT_LIST_HEAD(&dead_roots);
7281 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7283 while (!list_empty(&dead_roots)) {
7284 reloc_root = list_entry(dead_roots.prev,
7285 struct btrfs_root, dead_list);
7286 list_del_init(&reloc_root->dead_list);
7288 BUG_ON(reloc_root->commit_root != NULL);
7290 trans = btrfs_join_transaction(root, 1);
7293 mutex_lock(&root->fs_info->drop_mutex);
7294 ret = btrfs_drop_snapshot(trans, reloc_root);
7297 mutex_unlock(&root->fs_info->drop_mutex);
7299 nr = trans->blocks_used;
7300 ret = btrfs_end_transaction(trans, root);
7302 btrfs_btree_balance_dirty(root, nr);
7305 free_extent_buffer(reloc_root->node);
7307 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7308 &reloc_root->root_key);
7310 mutex_unlock(&root->fs_info->drop_mutex);
7312 nr = trans->blocks_used;
7313 ret = btrfs_end_transaction(trans, root);
7315 btrfs_btree_balance_dirty(root, nr);
7318 prev_root = reloc_root;
7321 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7327 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7329 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7333 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7335 struct btrfs_root *reloc_root;
7336 struct btrfs_trans_handle *trans;
7337 struct btrfs_key location;
7341 mutex_lock(&root->fs_info->tree_reloc_mutex);
7342 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7344 found = !list_empty(&root->fs_info->dead_reloc_roots);
7345 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7348 trans = btrfs_start_transaction(root, 1);
7350 ret = btrfs_commit_transaction(trans, root);
7354 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7355 location.offset = (u64)-1;
7356 location.type = BTRFS_ROOT_ITEM_KEY;
7358 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7359 BUG_ON(!reloc_root);
7360 btrfs_orphan_cleanup(reloc_root);
7364 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7365 struct btrfs_root *root)
7367 struct btrfs_root *reloc_root;
7368 struct extent_buffer *eb;
7369 struct btrfs_root_item *root_item;
7370 struct btrfs_key root_key;
7373 BUG_ON(!root->ref_cows);
7374 if (root->reloc_root)
7377 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7380 ret = btrfs_copy_root(trans, root, root->commit_root,
7381 &eb, BTRFS_TREE_RELOC_OBJECTID);
7384 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7385 root_key.offset = root->root_key.objectid;
7386 root_key.type = BTRFS_ROOT_ITEM_KEY;
7388 memcpy(root_item, &root->root_item, sizeof(root_item));
7389 btrfs_set_root_refs(root_item, 0);
7390 btrfs_set_root_bytenr(root_item, eb->start);
7391 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7392 btrfs_set_root_generation(root_item, trans->transid);
7394 btrfs_tree_unlock(eb);
7395 free_extent_buffer(eb);
7397 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7398 &root_key, root_item);
7402 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7404 BUG_ON(!reloc_root);
7405 reloc_root->last_trans = trans->transid;
7406 reloc_root->commit_root = NULL;
7407 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7409 root->reloc_root = reloc_root;
7414 * Core function of space balance.
7416 * The idea is using reloc trees to relocate tree blocks in reference
7417 * counted roots. There is one reloc tree for each subvol, and all
7418 * reloc trees share same root key objectid. Reloc trees are snapshots
7419 * of the latest committed roots of subvols (root->commit_root).
7421 * To relocate a tree block referenced by a subvol, there are two steps.
7422 * COW the block through subvol's reloc tree, then update block pointer
7423 * in the subvol to point to the new block. Since all reloc trees share
7424 * same root key objectid, doing special handing for tree blocks owned
7425 * by them is easy. Once a tree block has been COWed in one reloc tree,
7426 * we can use the resulting new block directly when the same block is
7427 * required to COW again through other reloc trees. By this way, relocated
7428 * tree blocks are shared between reloc trees, so they are also shared
7431 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7432 struct btrfs_root *root,
7433 struct btrfs_path *path,
7434 struct btrfs_key *first_key,
7435 struct btrfs_ref_path *ref_path,
7436 struct btrfs_block_group_cache *group,
7437 struct inode *reloc_inode)
7439 struct btrfs_root *reloc_root;
7440 struct extent_buffer *eb = NULL;
7441 struct btrfs_key *keys;
7445 int lowest_level = 0;
7448 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7449 lowest_level = ref_path->owner_objectid;
7451 if (!root->ref_cows) {
7452 path->lowest_level = lowest_level;
7453 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7455 path->lowest_level = 0;
7456 btrfs_release_path(root, path);
7460 mutex_lock(&root->fs_info->tree_reloc_mutex);
7461 ret = init_reloc_tree(trans, root);
7463 reloc_root = root->reloc_root;
7465 shared_level = ref_path->shared_level;
7466 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7468 keys = ref_path->node_keys;
7469 nodes = ref_path->new_nodes;
7470 memset(&keys[shared_level + 1], 0,
7471 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7472 memset(&nodes[shared_level + 1], 0,
7473 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7475 if (nodes[lowest_level] == 0) {
7476 path->lowest_level = lowest_level;
7477 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7480 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7481 eb = path->nodes[level];
7482 if (!eb || eb == reloc_root->node)
7484 nodes[level] = eb->start;
7486 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7488 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7491 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7492 eb = path->nodes[0];
7493 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7494 group, reloc_inode);
7497 btrfs_release_path(reloc_root, path);
7499 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7505 * replace tree blocks in the fs tree with tree blocks in
7508 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7511 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7512 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7515 extent_buffer_get(path->nodes[0]);
7516 eb = path->nodes[0];
7517 btrfs_release_path(reloc_root, path);
7518 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7520 free_extent_buffer(eb);
7523 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7524 path->lowest_level = 0;
7528 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7529 struct btrfs_root *root,
7530 struct btrfs_path *path,
7531 struct btrfs_key *first_key,
7532 struct btrfs_ref_path *ref_path)
7536 ret = relocate_one_path(trans, root, path, first_key,
7537 ref_path, NULL, NULL);
7543 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7544 struct btrfs_root *extent_root,
7545 struct btrfs_path *path,
7546 struct btrfs_key *extent_key)
7550 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7553 ret = btrfs_del_item(trans, extent_root, path);
7555 btrfs_release_path(extent_root, path);
7559 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7560 struct btrfs_ref_path *ref_path)
7562 struct btrfs_key root_key;
7564 root_key.objectid = ref_path->root_objectid;
7565 root_key.type = BTRFS_ROOT_ITEM_KEY;
7566 if (is_cowonly_root(ref_path->root_objectid))
7567 root_key.offset = 0;
7569 root_key.offset = (u64)-1;
7571 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7574 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7575 struct btrfs_path *path,
7576 struct btrfs_key *extent_key,
7577 struct btrfs_block_group_cache *group,
7578 struct inode *reloc_inode, int pass)
7580 struct btrfs_trans_handle *trans;
7581 struct btrfs_root *found_root;
7582 struct btrfs_ref_path *ref_path = NULL;
7583 struct disk_extent *new_extents = NULL;
7588 struct btrfs_key first_key;
7592 trans = btrfs_start_transaction(extent_root, 1);
7595 if (extent_key->objectid == 0) {
7596 ret = del_extent_zero(trans, extent_root, path, extent_key);
7600 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7606 for (loops = 0; ; loops++) {
7608 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7609 extent_key->objectid);
7611 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7618 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7619 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7622 found_root = read_ref_root(extent_root->fs_info, ref_path);
7623 BUG_ON(!found_root);
7625 * for reference counted tree, only process reference paths
7626 * rooted at the latest committed root.
7628 if (found_root->ref_cows &&
7629 ref_path->root_generation != found_root->root_key.offset)
7632 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7635 * copy data extents to new locations
7637 u64 group_start = group->key.objectid;
7638 ret = relocate_data_extent(reloc_inode,
7647 level = ref_path->owner_objectid;
7650 if (prev_block != ref_path->nodes[level]) {
7651 struct extent_buffer *eb;
7652 u64 block_start = ref_path->nodes[level];
7653 u64 block_size = btrfs_level_size(found_root, level);
7655 eb = read_tree_block(found_root, block_start,
7657 btrfs_tree_lock(eb);
7658 BUG_ON(level != btrfs_header_level(eb));
7661 btrfs_item_key_to_cpu(eb, &first_key, 0);
7663 btrfs_node_key_to_cpu(eb, &first_key, 0);
7665 btrfs_tree_unlock(eb);
7666 free_extent_buffer(eb);
7667 prev_block = block_start;
7670 mutex_lock(&extent_root->fs_info->trans_mutex);
7671 btrfs_record_root_in_trans(found_root);
7672 mutex_unlock(&extent_root->fs_info->trans_mutex);
7673 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7675 * try to update data extent references while
7676 * keeping metadata shared between snapshots.
7679 ret = relocate_one_path(trans, found_root,
7680 path, &first_key, ref_path,
7681 group, reloc_inode);
7687 * use fallback method to process the remaining
7691 u64 group_start = group->key.objectid;
7692 new_extents = kmalloc(sizeof(*new_extents),
7695 ret = get_new_locations(reloc_inode,
7703 ret = replace_one_extent(trans, found_root,
7705 &first_key, ref_path,
7706 new_extents, nr_extents);
7708 ret = relocate_tree_block(trans, found_root, path,
7709 &first_key, ref_path);
7716 btrfs_end_transaction(trans, extent_root);
7723 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7726 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7727 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7729 num_devices = root->fs_info->fs_devices->rw_devices;
7730 if (num_devices == 1) {
7731 stripped |= BTRFS_BLOCK_GROUP_DUP;
7732 stripped = flags & ~stripped;
7734 /* turn raid0 into single device chunks */
7735 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7738 /* turn mirroring into duplication */
7739 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7740 BTRFS_BLOCK_GROUP_RAID10))
7741 return stripped | BTRFS_BLOCK_GROUP_DUP;
7744 /* they already had raid on here, just return */
7745 if (flags & stripped)
7748 stripped |= BTRFS_BLOCK_GROUP_DUP;
7749 stripped = flags & ~stripped;
7751 /* switch duplicated blocks with raid1 */
7752 if (flags & BTRFS_BLOCK_GROUP_DUP)
7753 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7755 /* turn single device chunks into raid0 */
7756 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7761 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7763 struct btrfs_space_info *sinfo = cache->space_info;
7770 spin_lock(&sinfo->lock);
7771 spin_lock(&cache->lock);
7772 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7773 cache->bytes_super - btrfs_block_group_used(&cache->item);
7775 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7776 sinfo->bytes_may_use + sinfo->bytes_readonly +
7777 cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7778 sinfo->bytes_readonly += num_bytes;
7779 sinfo->bytes_reserved += cache->reserved_pinned;
7780 cache->reserved_pinned = 0;
7784 spin_unlock(&cache->lock);
7785 spin_unlock(&sinfo->lock);
7789 int btrfs_set_block_group_ro(struct btrfs_root *root,
7790 struct btrfs_block_group_cache *cache)
7793 struct btrfs_trans_handle *trans;
7799 trans = btrfs_join_transaction(root, 1);
7800 BUG_ON(IS_ERR(trans));
7802 alloc_flags = update_block_group_flags(root, cache->flags);
7803 if (alloc_flags != cache->flags)
7804 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7806 ret = set_block_group_ro(cache);
7809 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7810 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7813 ret = set_block_group_ro(cache);
7815 btrfs_end_transaction(trans, root);
7819 int btrfs_set_block_group_rw(struct btrfs_root *root,
7820 struct btrfs_block_group_cache *cache)
7822 struct btrfs_space_info *sinfo = cache->space_info;
7827 spin_lock(&sinfo->lock);
7828 spin_lock(&cache->lock);
7829 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7830 cache->bytes_super - btrfs_block_group_used(&cache->item);
7831 sinfo->bytes_readonly -= num_bytes;
7833 spin_unlock(&cache->lock);
7834 spin_unlock(&sinfo->lock);
7839 * checks to see if its even possible to relocate this block group.
7841 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7842 * ok to go ahead and try.
7844 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7846 struct btrfs_block_group_cache *block_group;
7847 struct btrfs_space_info *space_info;
7848 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7849 struct btrfs_device *device;
7853 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7855 /* odd, couldn't find the block group, leave it alone */
7859 /* no bytes used, we're good */
7860 if (!btrfs_block_group_used(&block_group->item))
7863 space_info = block_group->space_info;
7864 spin_lock(&space_info->lock);
7866 full = space_info->full;
7869 * if this is the last block group we have in this space, we can't
7870 * relocate it unless we're able to allocate a new chunk below.
7872 * Otherwise, we need to make sure we have room in the space to handle
7873 * all of the extents from this block group. If we can, we're good
7875 if ((space_info->total_bytes != block_group->key.offset) &&
7876 (space_info->bytes_used + space_info->bytes_reserved +
7877 space_info->bytes_pinned + space_info->bytes_readonly +
7878 btrfs_block_group_used(&block_group->item) <
7879 space_info->total_bytes)) {
7880 spin_unlock(&space_info->lock);
7883 spin_unlock(&space_info->lock);
7886 * ok we don't have enough space, but maybe we have free space on our
7887 * devices to allocate new chunks for relocation, so loop through our
7888 * alloc devices and guess if we have enough space. However, if we
7889 * were marked as full, then we know there aren't enough chunks, and we
7896 mutex_lock(&root->fs_info->chunk_mutex);
7897 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7898 u64 min_free = btrfs_block_group_used(&block_group->item);
7899 u64 dev_offset, max_avail;
7902 * check to make sure we can actually find a chunk with enough
7903 * space to fit our block group in.
7905 if (device->total_bytes > device->bytes_used + min_free) {
7906 ret = find_free_dev_extent(NULL, device, min_free,
7907 &dev_offset, &max_avail);
7913 mutex_unlock(&root->fs_info->chunk_mutex);
7915 btrfs_put_block_group(block_group);
7919 static int find_first_block_group(struct btrfs_root *root,
7920 struct btrfs_path *path, struct btrfs_key *key)
7923 struct btrfs_key found_key;
7924 struct extent_buffer *leaf;
7927 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7932 slot = path->slots[0];
7933 leaf = path->nodes[0];
7934 if (slot >= btrfs_header_nritems(leaf)) {
7935 ret = btrfs_next_leaf(root, path);
7942 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7944 if (found_key.objectid >= key->objectid &&
7945 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7955 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7957 struct btrfs_block_group_cache *block_group;
7961 struct inode *inode;
7963 block_group = btrfs_lookup_first_block_group(info, last);
7964 while (block_group) {
7965 spin_lock(&block_group->lock);
7966 if (block_group->iref)
7968 spin_unlock(&block_group->lock);
7969 block_group = next_block_group(info->tree_root,
7979 inode = block_group->inode;
7980 block_group->iref = 0;
7981 block_group->inode = NULL;
7982 spin_unlock(&block_group->lock);
7984 last = block_group->key.objectid + block_group->key.offset;
7985 btrfs_put_block_group(block_group);
7989 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7991 struct btrfs_block_group_cache *block_group;
7992 struct btrfs_space_info *space_info;
7993 struct btrfs_caching_control *caching_ctl;
7996 down_write(&info->extent_commit_sem);
7997 while (!list_empty(&info->caching_block_groups)) {
7998 caching_ctl = list_entry(info->caching_block_groups.next,
7999 struct btrfs_caching_control, list);
8000 list_del(&caching_ctl->list);
8001 put_caching_control(caching_ctl);
8003 up_write(&info->extent_commit_sem);
8005 spin_lock(&info->block_group_cache_lock);
8006 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8007 block_group = rb_entry(n, struct btrfs_block_group_cache,
8009 rb_erase(&block_group->cache_node,
8010 &info->block_group_cache_tree);
8011 spin_unlock(&info->block_group_cache_lock);
8013 down_write(&block_group->space_info->groups_sem);
8014 list_del(&block_group->list);
8015 up_write(&block_group->space_info->groups_sem);
8017 if (block_group->cached == BTRFS_CACHE_STARTED)
8018 wait_block_group_cache_done(block_group);
8020 btrfs_remove_free_space_cache(block_group);
8021 btrfs_put_block_group(block_group);
8023 spin_lock(&info->block_group_cache_lock);
8025 spin_unlock(&info->block_group_cache_lock);
8027 /* now that all the block groups are freed, go through and
8028 * free all the space_info structs. This is only called during
8029 * the final stages of unmount, and so we know nobody is
8030 * using them. We call synchronize_rcu() once before we start,
8031 * just to be on the safe side.
8035 release_global_block_rsv(info);
8037 while(!list_empty(&info->space_info)) {
8038 space_info = list_entry(info->space_info.next,
8039 struct btrfs_space_info,
8041 if (space_info->bytes_pinned > 0 ||
8042 space_info->bytes_reserved > 0) {
8044 dump_space_info(space_info, 0, 0);
8046 list_del(&space_info->list);
8052 static void __link_block_group(struct btrfs_space_info *space_info,
8053 struct btrfs_block_group_cache *cache)
8055 int index = get_block_group_index(cache);
8057 down_write(&space_info->groups_sem);
8058 list_add_tail(&cache->list, &space_info->block_groups[index]);
8059 up_write(&space_info->groups_sem);
8062 int btrfs_read_block_groups(struct btrfs_root *root)
8064 struct btrfs_path *path;
8066 struct btrfs_block_group_cache *cache;
8067 struct btrfs_fs_info *info = root->fs_info;
8068 struct btrfs_space_info *space_info;
8069 struct btrfs_key key;
8070 struct btrfs_key found_key;
8071 struct extent_buffer *leaf;
8075 root = info->extent_root;
8078 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8079 path = btrfs_alloc_path();
8083 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8084 if (cache_gen != 0 &&
8085 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8089 ret = find_first_block_group(root, path, &key);
8095 leaf = path->nodes[0];
8096 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8097 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8103 atomic_set(&cache->count, 1);
8104 spin_lock_init(&cache->lock);
8105 spin_lock_init(&cache->tree_lock);
8106 cache->fs_info = info;
8107 INIT_LIST_HEAD(&cache->list);
8108 INIT_LIST_HEAD(&cache->cluster_list);
8111 cache->disk_cache_state = BTRFS_DC_CLEAR;
8114 * we only want to have 32k of ram per block group for keeping
8115 * track of free space, and if we pass 1/2 of that we want to
8116 * start converting things over to using bitmaps
8118 cache->extents_thresh = ((1024 * 32) / 2) /
8119 sizeof(struct btrfs_free_space);
8121 read_extent_buffer(leaf, &cache->item,
8122 btrfs_item_ptr_offset(leaf, path->slots[0]),
8123 sizeof(cache->item));
8124 memcpy(&cache->key, &found_key, sizeof(found_key));
8126 key.objectid = found_key.objectid + found_key.offset;
8127 btrfs_release_path(root, path);
8128 cache->flags = btrfs_block_group_flags(&cache->item);
8129 cache->sectorsize = root->sectorsize;
8132 * check for two cases, either we are full, and therefore
8133 * don't need to bother with the caching work since we won't
8134 * find any space, or we are empty, and we can just add all
8135 * the space in and be done with it. This saves us _alot_ of
8136 * time, particularly in the full case.
8138 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8139 exclude_super_stripes(root, cache);
8140 cache->last_byte_to_unpin = (u64)-1;
8141 cache->cached = BTRFS_CACHE_FINISHED;
8142 free_excluded_extents(root, cache);
8143 } else if (btrfs_block_group_used(&cache->item) == 0) {
8144 exclude_super_stripes(root, cache);
8145 cache->last_byte_to_unpin = (u64)-1;
8146 cache->cached = BTRFS_CACHE_FINISHED;
8147 add_new_free_space(cache, root->fs_info,
8149 found_key.objectid +
8151 free_excluded_extents(root, cache);
8154 ret = update_space_info(info, cache->flags, found_key.offset,
8155 btrfs_block_group_used(&cache->item),
8158 cache->space_info = space_info;
8159 spin_lock(&cache->space_info->lock);
8160 cache->space_info->bytes_readonly += cache->bytes_super;
8161 spin_unlock(&cache->space_info->lock);
8163 __link_block_group(space_info, cache);
8165 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8168 set_avail_alloc_bits(root->fs_info, cache->flags);
8169 if (btrfs_chunk_readonly(root, cache->key.objectid))
8170 set_block_group_ro(cache);
8173 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8174 if (!(get_alloc_profile(root, space_info->flags) &
8175 (BTRFS_BLOCK_GROUP_RAID10 |
8176 BTRFS_BLOCK_GROUP_RAID1 |
8177 BTRFS_BLOCK_GROUP_DUP)))
8180 * avoid allocating from un-mirrored block group if there are
8181 * mirrored block groups.
8183 list_for_each_entry(cache, &space_info->block_groups[3], list)
8184 set_block_group_ro(cache);
8185 list_for_each_entry(cache, &space_info->block_groups[4], list)
8186 set_block_group_ro(cache);
8189 init_global_block_rsv(info);
8192 btrfs_free_path(path);
8196 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8197 struct btrfs_root *root, u64 bytes_used,
8198 u64 type, u64 chunk_objectid, u64 chunk_offset,
8202 struct btrfs_root *extent_root;
8203 struct btrfs_block_group_cache *cache;
8205 extent_root = root->fs_info->extent_root;
8207 root->fs_info->last_trans_log_full_commit = trans->transid;
8209 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8213 cache->key.objectid = chunk_offset;
8214 cache->key.offset = size;
8215 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8216 cache->sectorsize = root->sectorsize;
8217 cache->fs_info = root->fs_info;
8220 * we only want to have 32k of ram per block group for keeping track
8221 * of free space, and if we pass 1/2 of that we want to start
8222 * converting things over to using bitmaps
8224 cache->extents_thresh = ((1024 * 32) / 2) /
8225 sizeof(struct btrfs_free_space);
8226 atomic_set(&cache->count, 1);
8227 spin_lock_init(&cache->lock);
8228 spin_lock_init(&cache->tree_lock);
8229 INIT_LIST_HEAD(&cache->list);
8230 INIT_LIST_HEAD(&cache->cluster_list);
8232 btrfs_set_block_group_used(&cache->item, bytes_used);
8233 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8234 cache->flags = type;
8235 btrfs_set_block_group_flags(&cache->item, type);
8237 cache->last_byte_to_unpin = (u64)-1;
8238 cache->cached = BTRFS_CACHE_FINISHED;
8239 exclude_super_stripes(root, cache);
8241 add_new_free_space(cache, root->fs_info, chunk_offset,
8242 chunk_offset + size);
8244 free_excluded_extents(root, cache);
8246 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8247 &cache->space_info);
8250 spin_lock(&cache->space_info->lock);
8251 cache->space_info->bytes_readonly += cache->bytes_super;
8252 spin_unlock(&cache->space_info->lock);
8254 __link_block_group(cache->space_info, cache);
8256 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8259 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8260 sizeof(cache->item));
8263 set_avail_alloc_bits(extent_root->fs_info, type);
8268 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8269 struct btrfs_root *root, u64 group_start)
8271 struct btrfs_path *path;
8272 struct btrfs_block_group_cache *block_group;
8273 struct btrfs_free_cluster *cluster;
8274 struct btrfs_root *tree_root = root->fs_info->tree_root;
8275 struct btrfs_key key;
8276 struct inode *inode;
8279 root = root->fs_info->extent_root;
8281 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8282 BUG_ON(!block_group);
8283 BUG_ON(!block_group->ro);
8285 /* make sure this block group isn't part of an allocation cluster */
8286 cluster = &root->fs_info->data_alloc_cluster;
8287 spin_lock(&cluster->refill_lock);
8288 btrfs_return_cluster_to_free_space(block_group, cluster);
8289 spin_unlock(&cluster->refill_lock);
8292 * make sure this block group isn't part of a metadata
8293 * allocation cluster
8295 cluster = &root->fs_info->meta_alloc_cluster;
8296 spin_lock(&cluster->refill_lock);
8297 btrfs_return_cluster_to_free_space(block_group, cluster);
8298 spin_unlock(&cluster->refill_lock);
8300 path = btrfs_alloc_path();
8303 inode = lookup_free_space_inode(root, block_group, path);
8304 if (!IS_ERR(inode)) {
8305 btrfs_orphan_add(trans, inode);
8307 /* One for the block groups ref */
8308 spin_lock(&block_group->lock);
8309 if (block_group->iref) {
8310 block_group->iref = 0;
8311 block_group->inode = NULL;
8312 spin_unlock(&block_group->lock);
8315 spin_unlock(&block_group->lock);
8317 /* One for our lookup ref */
8321 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8322 key.offset = block_group->key.objectid;
8325 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8329 btrfs_release_path(tree_root, path);
8331 ret = btrfs_del_item(trans, tree_root, path);
8334 btrfs_release_path(tree_root, path);
8337 spin_lock(&root->fs_info->block_group_cache_lock);
8338 rb_erase(&block_group->cache_node,
8339 &root->fs_info->block_group_cache_tree);
8340 spin_unlock(&root->fs_info->block_group_cache_lock);
8342 down_write(&block_group->space_info->groups_sem);
8344 * we must use list_del_init so people can check to see if they
8345 * are still on the list after taking the semaphore
8347 list_del_init(&block_group->list);
8348 up_write(&block_group->space_info->groups_sem);
8350 if (block_group->cached == BTRFS_CACHE_STARTED)
8351 wait_block_group_cache_done(block_group);
8353 btrfs_remove_free_space_cache(block_group);
8355 spin_lock(&block_group->space_info->lock);
8356 block_group->space_info->total_bytes -= block_group->key.offset;
8357 block_group->space_info->bytes_readonly -= block_group->key.offset;
8358 spin_unlock(&block_group->space_info->lock);
8360 memcpy(&key, &block_group->key, sizeof(key));
8362 btrfs_clear_space_info_full(root->fs_info);
8364 btrfs_put_block_group(block_group);
8365 btrfs_put_block_group(block_group);
8367 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8373 ret = btrfs_del_item(trans, root, path);
8375 btrfs_free_path(path);