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>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "ref-cache.h"
35 #define PENDING_EXTENT_INSERT 0
36 #define PENDING_EXTENT_DELETE 1
37 #define PENDING_BACKREF_UPDATE 2
39 struct pending_extent_op {
48 struct list_head list;
52 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
53 struct btrfs_root *root, u64 parent,
54 u64 root_objectid, u64 ref_generation,
55 u64 owner, struct btrfs_key *ins,
57 static int update_reserved_extents(struct btrfs_root *root,
58 u64 bytenr, u64 num, int reserve);
59 static int update_block_group(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root,
61 u64 bytenr, u64 num_bytes, int alloc,
63 static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 u64 bytenr, u64 num_bytes, u64 parent,
66 u64 root_objectid, u64 ref_generation,
67 u64 owner_objectid, int pin,
70 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
71 struct btrfs_root *extent_root, u64 alloc_bytes,
72 u64 flags, int force);
74 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
76 return (cache->flags & bits) == bits;
80 * this adds the block group to the fs_info rb tree for the block group
83 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
84 struct btrfs_block_group_cache *block_group)
87 struct rb_node *parent = NULL;
88 struct btrfs_block_group_cache *cache;
90 spin_lock(&info->block_group_cache_lock);
91 p = &info->block_group_cache_tree.rb_node;
95 cache = rb_entry(parent, struct btrfs_block_group_cache,
97 if (block_group->key.objectid < cache->key.objectid) {
99 } else if (block_group->key.objectid > cache->key.objectid) {
102 spin_unlock(&info->block_group_cache_lock);
107 rb_link_node(&block_group->cache_node, parent, p);
108 rb_insert_color(&block_group->cache_node,
109 &info->block_group_cache_tree);
110 spin_unlock(&info->block_group_cache_lock);
116 * This will return the block group at or after bytenr if contains is 0, else
117 * it will return the block group that contains the bytenr
119 static struct btrfs_block_group_cache *
120 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
123 struct btrfs_block_group_cache *cache, *ret = NULL;
127 spin_lock(&info->block_group_cache_lock);
128 n = info->block_group_cache_tree.rb_node;
131 cache = rb_entry(n, struct btrfs_block_group_cache,
133 end = cache->key.objectid + cache->key.offset - 1;
134 start = cache->key.objectid;
136 if (bytenr < start) {
137 if (!contains && (!ret || start < ret->key.objectid))
140 } else if (bytenr > start) {
141 if (contains && bytenr <= end) {
152 atomic_inc(&ret->count);
153 spin_unlock(&info->block_group_cache_lock);
159 * this is only called by cache_block_group, since we could have freed extents
160 * we need to check the pinned_extents for any extents that can't be used yet
161 * since their free space will be released as soon as the transaction commits.
163 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
164 struct btrfs_fs_info *info, u64 start, u64 end)
166 u64 extent_start, extent_end, size;
169 mutex_lock(&info->pinned_mutex);
170 while (start < end) {
171 ret = find_first_extent_bit(&info->pinned_extents, start,
172 &extent_start, &extent_end,
177 if (extent_start == start) {
178 start = extent_end + 1;
179 } else if (extent_start > start && extent_start < end) {
180 size = extent_start - start;
181 ret = btrfs_add_free_space(block_group, start,
184 start = extent_end + 1;
192 ret = btrfs_add_free_space(block_group, start, size);
195 mutex_unlock(&info->pinned_mutex);
200 static int remove_sb_from_cache(struct btrfs_root *root,
201 struct btrfs_block_group_cache *cache)
208 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
209 bytenr = btrfs_sb_offset(i);
210 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
211 cache->key.objectid, bytenr, 0,
212 &logical, &nr, &stripe_len);
215 btrfs_remove_free_space(cache, logical[nr],
223 static int cache_block_group(struct btrfs_root *root,
224 struct btrfs_block_group_cache *block_group)
226 struct btrfs_path *path;
228 struct btrfs_key key;
229 struct extent_buffer *leaf;
236 root = root->fs_info->extent_root;
238 if (block_group->cached)
241 path = btrfs_alloc_path();
247 * we get into deadlocks with paths held by callers of this function.
248 * since the alloc_mutex is protecting things right now, just
249 * skip the locking here
251 path->skip_locking = 1;
252 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
255 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
256 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
261 leaf = path->nodes[0];
262 slot = path->slots[0];
263 if (slot >= btrfs_header_nritems(leaf)) {
264 ret = btrfs_next_leaf(root, path);
272 btrfs_item_key_to_cpu(leaf, &key, slot);
273 if (key.objectid < block_group->key.objectid)
276 if (key.objectid >= block_group->key.objectid +
277 block_group->key.offset)
280 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
281 add_new_free_space(block_group, root->fs_info, last,
284 last = key.objectid + key.offset;
290 add_new_free_space(block_group, root->fs_info, last,
291 block_group->key.objectid +
292 block_group->key.offset);
294 remove_sb_from_cache(root, block_group);
295 block_group->cached = 1;
298 btrfs_free_path(path);
303 * return the block group that starts at or after bytenr
305 static struct btrfs_block_group_cache *
306 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
308 struct btrfs_block_group_cache *cache;
310 cache = block_group_cache_tree_search(info, bytenr, 0);
316 * return the block group that contains teh given bytenr
318 struct btrfs_block_group_cache *btrfs_lookup_block_group(
319 struct btrfs_fs_info *info,
322 struct btrfs_block_group_cache *cache;
324 cache = block_group_cache_tree_search(info, bytenr, 1);
329 static inline void put_block_group(struct btrfs_block_group_cache *cache)
331 if (atomic_dec_and_test(&cache->count))
335 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
338 struct list_head *head = &info->space_info;
339 struct btrfs_space_info *found;
342 list_for_each_entry_rcu(found, head, list) {
343 if (found->flags == flags) {
353 * after adding space to the filesystem, we need to clear the full flags
354 * on all the space infos.
356 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
358 struct list_head *head = &info->space_info;
359 struct btrfs_space_info *found;
362 list_for_each_entry_rcu(found, head, list)
367 static u64 div_factor(u64 num, int factor)
376 u64 btrfs_find_block_group(struct btrfs_root *root,
377 u64 search_start, u64 search_hint, int owner)
379 struct btrfs_block_group_cache *cache;
381 u64 last = max(search_hint, search_start);
388 cache = btrfs_lookup_first_block_group(root->fs_info, last);
392 spin_lock(&cache->lock);
393 last = cache->key.objectid + cache->key.offset;
394 used = btrfs_block_group_used(&cache->item);
396 if ((full_search || !cache->ro) &&
397 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
398 if (used + cache->pinned + cache->reserved <
399 div_factor(cache->key.offset, factor)) {
400 group_start = cache->key.objectid;
401 spin_unlock(&cache->lock);
402 put_block_group(cache);
406 spin_unlock(&cache->lock);
407 put_block_group(cache);
415 if (!full_search && factor < 10) {
425 /* simple helper to search for an existing extent at a given offset */
426 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
429 struct btrfs_key key;
430 struct btrfs_path *path;
432 path = btrfs_alloc_path();
434 key.objectid = start;
436 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
437 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
439 btrfs_free_path(path);
444 * Back reference rules. Back refs have three main goals:
446 * 1) differentiate between all holders of references to an extent so that
447 * when a reference is dropped we can make sure it was a valid reference
448 * before freeing the extent.
450 * 2) Provide enough information to quickly find the holders of an extent
451 * if we notice a given block is corrupted or bad.
453 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
454 * maintenance. This is actually the same as #2, but with a slightly
455 * different use case.
457 * File extents can be referenced by:
459 * - multiple snapshots, subvolumes, or different generations in one subvol
460 * - different files inside a single subvolume
461 * - different offsets inside a file (bookend extents in file.c)
463 * The extent ref structure has fields for:
465 * - Objectid of the subvolume root
466 * - Generation number of the tree holding the reference
467 * - objectid of the file holding the reference
468 * - number of references holding by parent node (alway 1 for tree blocks)
470 * Btree leaf may hold multiple references to a file extent. In most cases,
471 * these references are from same file and the corresponding offsets inside
472 * the file are close together.
474 * When a file extent is allocated the fields are filled in:
475 * (root_key.objectid, trans->transid, inode objectid, 1)
477 * When a leaf is cow'd new references are added for every file extent found
478 * in the leaf. It looks similar to the create case, but trans->transid will
479 * be different when the block is cow'd.
481 * (root_key.objectid, trans->transid, inode objectid,
482 * number of references in the leaf)
484 * When a file extent is removed either during snapshot deletion or
485 * file truncation, we find the corresponding back reference and check
486 * the following fields:
488 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
491 * Btree extents can be referenced by:
493 * - Different subvolumes
494 * - Different generations of the same subvolume
496 * When a tree block is created, back references are inserted:
498 * (root->root_key.objectid, trans->transid, level, 1)
500 * When a tree block is cow'd, new back references are added for all the
501 * blocks it points to. If the tree block isn't in reference counted root,
502 * the old back references are removed. These new back references are of
503 * the form (trans->transid will have increased since creation):
505 * (root->root_key.objectid, trans->transid, level, 1)
507 * When a backref is in deleting, the following fields are checked:
509 * if backref was for a tree root:
510 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
512 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
514 * Back Reference Key composing:
516 * The key objectid corresponds to the first byte in the extent, the key
517 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
518 * byte of parent extent. If a extent is tree root, the key offset is set
519 * to the key objectid.
522 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
523 struct btrfs_root *root,
524 struct btrfs_path *path,
525 u64 bytenr, u64 parent,
526 u64 ref_root, u64 ref_generation,
527 u64 owner_objectid, int del)
529 struct btrfs_key key;
530 struct btrfs_extent_ref *ref;
531 struct extent_buffer *leaf;
535 key.objectid = bytenr;
536 key.type = BTRFS_EXTENT_REF_KEY;
539 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
547 leaf = path->nodes[0];
548 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
549 ref_objectid = btrfs_ref_objectid(leaf, ref);
550 if (btrfs_ref_root(leaf, ref) != ref_root ||
551 btrfs_ref_generation(leaf, ref) != ref_generation ||
552 (ref_objectid != owner_objectid &&
553 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
563 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
564 struct btrfs_root *root,
565 struct btrfs_path *path,
566 u64 bytenr, u64 parent,
567 u64 ref_root, u64 ref_generation,
571 struct btrfs_key key;
572 struct extent_buffer *leaf;
573 struct btrfs_extent_ref *ref;
577 key.objectid = bytenr;
578 key.type = BTRFS_EXTENT_REF_KEY;
581 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
583 leaf = path->nodes[0];
584 ref = btrfs_item_ptr(leaf, path->slots[0],
585 struct btrfs_extent_ref);
586 btrfs_set_ref_root(leaf, ref, ref_root);
587 btrfs_set_ref_generation(leaf, ref, ref_generation);
588 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
589 btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
590 } else if (ret == -EEXIST) {
593 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
594 leaf = path->nodes[0];
595 ref = btrfs_item_ptr(leaf, path->slots[0],
596 struct btrfs_extent_ref);
597 if (btrfs_ref_root(leaf, ref) != ref_root ||
598 btrfs_ref_generation(leaf, ref) != ref_generation) {
604 num_refs = btrfs_ref_num_refs(leaf, ref);
605 BUG_ON(num_refs == 0);
606 btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
608 existing_owner = btrfs_ref_objectid(leaf, ref);
609 if (existing_owner != owner_objectid &&
610 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
611 btrfs_set_ref_objectid(leaf, ref,
612 BTRFS_MULTIPLE_OBJECTIDS);
618 btrfs_unlock_up_safe(path, 1);
619 btrfs_mark_buffer_dirty(path->nodes[0]);
621 btrfs_release_path(root, path);
625 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
626 struct btrfs_root *root,
627 struct btrfs_path *path,
630 struct extent_buffer *leaf;
631 struct btrfs_extent_ref *ref;
635 leaf = path->nodes[0];
636 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
637 num_refs = btrfs_ref_num_refs(leaf, ref);
638 BUG_ON(num_refs < refs_to_drop);
639 num_refs -= refs_to_drop;
641 ret = btrfs_del_item(trans, root, path);
643 btrfs_set_ref_num_refs(leaf, ref, num_refs);
644 btrfs_mark_buffer_dirty(leaf);
646 btrfs_release_path(root, path);
650 #ifdef BIO_RW_DISCARD
651 static void btrfs_issue_discard(struct block_device *bdev,
654 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
658 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
661 #ifdef BIO_RW_DISCARD
663 u64 map_length = num_bytes;
664 struct btrfs_multi_bio *multi = NULL;
666 /* Tell the block device(s) that the sectors can be discarded */
667 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
668 bytenr, &map_length, &multi, 0);
670 struct btrfs_bio_stripe *stripe = multi->stripes;
673 if (map_length > num_bytes)
674 map_length = num_bytes;
676 for (i = 0; i < multi->num_stripes; i++, stripe++) {
677 btrfs_issue_discard(stripe->dev->bdev,
690 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
691 struct btrfs_root *root, u64 bytenr,
693 u64 orig_parent, u64 parent,
694 u64 orig_root, u64 ref_root,
695 u64 orig_generation, u64 ref_generation,
699 int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
701 ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
702 orig_parent, parent, orig_root,
703 ref_root, orig_generation,
704 ref_generation, owner_objectid, pin);
709 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
710 struct btrfs_root *root, u64 bytenr,
711 u64 num_bytes, u64 orig_parent, u64 parent,
712 u64 ref_root, u64 ref_generation,
716 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
717 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
720 ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
721 orig_parent, parent, ref_root,
722 ref_root, ref_generation,
723 ref_generation, owner_objectid);
726 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
727 struct btrfs_root *root, u64 bytenr,
729 u64 orig_parent, u64 parent,
730 u64 orig_root, u64 ref_root,
731 u64 orig_generation, u64 ref_generation,
736 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
737 ref_generation, owner_objectid,
738 BTRFS_ADD_DELAYED_REF, 0);
743 static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 num_bytes, u64 parent, u64 ref_root,
746 u64 ref_generation, u64 owner_objectid,
749 struct btrfs_path *path;
751 struct btrfs_key key;
752 struct extent_buffer *l;
753 struct btrfs_extent_item *item;
756 path = btrfs_alloc_path();
761 path->leave_spinning = 1;
762 key.objectid = bytenr;
763 key.type = BTRFS_EXTENT_ITEM_KEY;
764 key.offset = num_bytes;
766 /* first find the extent item and update its reference count */
767 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
770 btrfs_set_path_blocking(path);
776 btrfs_free_path(path);
781 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
782 if (key.objectid != bytenr) {
783 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
784 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
785 (unsigned long long)bytenr,
786 (unsigned long long)key.objectid);
789 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
791 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
793 refs = btrfs_extent_refs(l, item);
794 btrfs_set_extent_refs(l, item, refs + refs_to_add);
795 btrfs_unlock_up_safe(path, 1);
797 btrfs_mark_buffer_dirty(path->nodes[0]);
799 btrfs_release_path(root->fs_info->extent_root, path);
802 path->leave_spinning = 1;
804 /* now insert the actual backref */
805 ret = insert_extent_backref(trans, root->fs_info->extent_root,
806 path, bytenr, parent,
807 ref_root, ref_generation,
808 owner_objectid, refs_to_add);
810 btrfs_free_path(path);
814 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root,
816 u64 bytenr, u64 num_bytes, u64 parent,
817 u64 ref_root, u64 ref_generation,
821 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
822 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
825 ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
826 0, ref_root, 0, ref_generation,
831 static int drop_delayed_ref(struct btrfs_trans_handle *trans,
832 struct btrfs_root *root,
833 struct btrfs_delayed_ref_node *node)
836 struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
838 BUG_ON(node->ref_mod == 0);
839 ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
840 node->parent, ref->root, ref->generation,
841 ref->owner_objectid, ref->pin, node->ref_mod);
846 /* helper function to actually process a single delayed ref entry */
847 static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
848 struct btrfs_root *root,
849 struct btrfs_delayed_ref_node *node,
853 struct btrfs_delayed_ref *ref;
855 if (node->parent == (u64)-1) {
856 struct btrfs_delayed_ref_head *head;
858 * we've hit the end of the chain and we were supposed
859 * to insert this extent into the tree. But, it got
860 * deleted before we ever needed to insert it, so all
861 * we have to do is clean up the accounting
863 if (insert_reserved) {
864 update_reserved_extents(root, node->bytenr,
867 head = btrfs_delayed_node_to_head(node);
868 mutex_unlock(&head->mutex);
872 ref = btrfs_delayed_node_to_ref(node);
873 if (ref->action == BTRFS_ADD_DELAYED_REF) {
874 if (insert_reserved) {
875 struct btrfs_key ins;
877 ins.objectid = node->bytenr;
878 ins.offset = node->num_bytes;
879 ins.type = BTRFS_EXTENT_ITEM_KEY;
881 /* record the full extent allocation */
882 ret = __btrfs_alloc_reserved_extent(trans, root,
883 node->parent, ref->root,
884 ref->generation, ref->owner_objectid,
885 &ins, node->ref_mod);
886 update_reserved_extents(root, node->bytenr,
889 /* just add one backref */
890 ret = add_extent_ref(trans, root, node->bytenr,
892 node->parent, ref->root, ref->generation,
893 ref->owner_objectid, node->ref_mod);
896 } else if (ref->action == BTRFS_DROP_DELAYED_REF) {
897 WARN_ON(insert_reserved);
898 ret = drop_delayed_ref(trans, root, node);
903 static noinline struct btrfs_delayed_ref_node *
904 select_delayed_ref(struct btrfs_delayed_ref_head *head)
906 struct rb_node *node;
907 struct btrfs_delayed_ref_node *ref;
908 int action = BTRFS_ADD_DELAYED_REF;
911 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
912 * this prevents ref count from going down to zero when
913 * there still are pending delayed ref.
915 node = rb_prev(&head->node.rb_node);
919 ref = rb_entry(node, struct btrfs_delayed_ref_node,
921 if (ref->bytenr != head->node.bytenr)
923 if (btrfs_delayed_node_to_ref(ref)->action == action)
925 node = rb_prev(node);
927 if (action == BTRFS_ADD_DELAYED_REF) {
928 action = BTRFS_DROP_DELAYED_REF;
934 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
935 struct btrfs_root *root,
936 struct list_head *cluster)
938 struct btrfs_delayed_ref_root *delayed_refs;
939 struct btrfs_delayed_ref_node *ref;
940 struct btrfs_delayed_ref_head *locked_ref = NULL;
943 int must_insert_reserved = 0;
945 delayed_refs = &trans->transaction->delayed_refs;
948 /* pick a new head ref from the cluster list */
949 if (list_empty(cluster))
952 locked_ref = list_entry(cluster->next,
953 struct btrfs_delayed_ref_head, cluster);
955 /* grab the lock that says we are going to process
956 * all the refs for this head */
957 ret = btrfs_delayed_ref_lock(trans, locked_ref);
960 * we may have dropped the spin lock to get the head
961 * mutex lock, and that might have given someone else
962 * time to free the head. If that's true, it has been
963 * removed from our list and we can move on.
965 if (ret == -EAGAIN) {
973 * record the must insert reserved flag before we
974 * drop the spin lock.
976 must_insert_reserved = locked_ref->must_insert_reserved;
977 locked_ref->must_insert_reserved = 0;
980 * locked_ref is the head node, so we have to go one
981 * node back for any delayed ref updates
983 ref = select_delayed_ref(locked_ref);
985 /* All delayed refs have been processed, Go ahead
986 * and send the head node to run_one_delayed_ref,
987 * so that any accounting fixes can happen
989 ref = &locked_ref->node;
990 list_del_init(&locked_ref->cluster);
995 rb_erase(&ref->rb_node, &delayed_refs->root);
996 delayed_refs->num_entries--;
997 spin_unlock(&delayed_refs->lock);
999 ret = run_one_delayed_ref(trans, root, ref,
1000 must_insert_reserved);
1002 btrfs_put_delayed_ref(ref);
1006 spin_lock(&delayed_refs->lock);
1012 * this starts processing the delayed reference count updates and
1013 * extent insertions we have queued up so far. count can be
1014 * 0, which means to process everything in the tree at the start
1015 * of the run (but not newly added entries), or it can be some target
1016 * number you'd like to process.
1018 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root, unsigned long count)
1021 struct rb_node *node;
1022 struct btrfs_delayed_ref_root *delayed_refs;
1023 struct btrfs_delayed_ref_node *ref;
1024 struct list_head cluster;
1026 int run_all = count == (unsigned long)-1;
1029 if (root == root->fs_info->extent_root)
1030 root = root->fs_info->tree_root;
1032 delayed_refs = &trans->transaction->delayed_refs;
1033 INIT_LIST_HEAD(&cluster);
1035 spin_lock(&delayed_refs->lock);
1037 count = delayed_refs->num_entries * 2;
1041 if (!(run_all || run_most) &&
1042 delayed_refs->num_heads_ready < 64)
1046 * go find something we can process in the rbtree. We start at
1047 * the beginning of the tree, and then build a cluster
1048 * of refs to process starting at the first one we are able to
1051 ret = btrfs_find_ref_cluster(trans, &cluster,
1052 delayed_refs->run_delayed_start);
1056 ret = run_clustered_refs(trans, root, &cluster);
1059 count -= min_t(unsigned long, ret, count);
1066 node = rb_first(&delayed_refs->root);
1069 count = (unsigned long)-1;
1072 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1074 if (btrfs_delayed_ref_is_head(ref)) {
1075 struct btrfs_delayed_ref_head *head;
1077 head = btrfs_delayed_node_to_head(ref);
1078 atomic_inc(&ref->refs);
1080 spin_unlock(&delayed_refs->lock);
1081 mutex_lock(&head->mutex);
1082 mutex_unlock(&head->mutex);
1084 btrfs_put_delayed_ref(ref);
1088 node = rb_next(node);
1090 spin_unlock(&delayed_refs->lock);
1091 schedule_timeout(1);
1095 spin_unlock(&delayed_refs->lock);
1099 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1100 struct btrfs_root *root, u64 objectid, u64 bytenr)
1102 struct btrfs_root *extent_root = root->fs_info->extent_root;
1103 struct btrfs_path *path;
1104 struct extent_buffer *leaf;
1105 struct btrfs_extent_ref *ref_item;
1106 struct btrfs_key key;
1107 struct btrfs_key found_key;
1113 key.objectid = bytenr;
1114 key.offset = (u64)-1;
1115 key.type = BTRFS_EXTENT_ITEM_KEY;
1117 path = btrfs_alloc_path();
1118 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1124 if (path->slots[0] == 0)
1128 leaf = path->nodes[0];
1129 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1131 if (found_key.objectid != bytenr ||
1132 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1135 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1137 leaf = path->nodes[0];
1138 nritems = btrfs_header_nritems(leaf);
1139 if (path->slots[0] >= nritems) {
1140 ret = btrfs_next_leaf(extent_root, path);
1147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1148 if (found_key.objectid != bytenr)
1151 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1156 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1157 struct btrfs_extent_ref);
1158 ref_root = btrfs_ref_root(leaf, ref_item);
1159 if ((ref_root != root->root_key.objectid &&
1160 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1161 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1165 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1174 btrfs_free_path(path);
1178 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1179 struct extent_buffer *buf, u32 nr_extents)
1181 struct btrfs_key key;
1182 struct btrfs_file_extent_item *fi;
1190 if (!root->ref_cows)
1193 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1195 root_gen = root->root_key.offset;
1198 root_gen = trans->transid - 1;
1201 level = btrfs_header_level(buf);
1202 nritems = btrfs_header_nritems(buf);
1205 struct btrfs_leaf_ref *ref;
1206 struct btrfs_extent_info *info;
1208 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1214 ref->root_gen = root_gen;
1215 ref->bytenr = buf->start;
1216 ref->owner = btrfs_header_owner(buf);
1217 ref->generation = btrfs_header_generation(buf);
1218 ref->nritems = nr_extents;
1219 info = ref->extents;
1221 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1223 btrfs_item_key_to_cpu(buf, &key, i);
1224 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1226 fi = btrfs_item_ptr(buf, i,
1227 struct btrfs_file_extent_item);
1228 if (btrfs_file_extent_type(buf, fi) ==
1229 BTRFS_FILE_EXTENT_INLINE)
1231 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1232 if (disk_bytenr == 0)
1235 info->bytenr = disk_bytenr;
1237 btrfs_file_extent_disk_num_bytes(buf, fi);
1238 info->objectid = key.objectid;
1239 info->offset = key.offset;
1243 ret = btrfs_add_leaf_ref(root, ref, shared);
1244 if (ret == -EEXIST && shared) {
1245 struct btrfs_leaf_ref *old;
1246 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1248 btrfs_remove_leaf_ref(root, old);
1249 btrfs_free_leaf_ref(root, old);
1250 ret = btrfs_add_leaf_ref(root, ref, shared);
1253 btrfs_free_leaf_ref(root, ref);
1259 /* when a block goes through cow, we update the reference counts of
1260 * everything that block points to. The internal pointers of the block
1261 * can be in just about any order, and it is likely to have clusters of
1262 * things that are close together and clusters of things that are not.
1264 * To help reduce the seeks that come with updating all of these reference
1265 * counts, sort them by byte number before actual updates are done.
1267 * struct refsort is used to match byte number to slot in the btree block.
1268 * we sort based on the byte number and then use the slot to actually
1271 * struct refsort is smaller than strcut btrfs_item and smaller than
1272 * struct btrfs_key_ptr. Since we're currently limited to the page size
1273 * for a btree block, there's no way for a kmalloc of refsorts for a
1274 * single node to be bigger than a page.
1282 * for passing into sort()
1284 static int refsort_cmp(const void *a_void, const void *b_void)
1286 const struct refsort *a = a_void;
1287 const struct refsort *b = b_void;
1289 if (a->bytenr < b->bytenr)
1291 if (a->bytenr > b->bytenr)
1297 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1298 struct btrfs_root *root,
1299 struct extent_buffer *orig_buf,
1300 struct extent_buffer *buf, u32 *nr_extents)
1306 u64 orig_generation;
1307 struct refsort *sorted;
1309 u32 nr_file_extents = 0;
1310 struct btrfs_key key;
1311 struct btrfs_file_extent_item *fi;
1318 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1319 u64, u64, u64, u64, u64, u64, u64, u64, u64);
1321 ref_root = btrfs_header_owner(buf);
1322 ref_generation = btrfs_header_generation(buf);
1323 orig_root = btrfs_header_owner(orig_buf);
1324 orig_generation = btrfs_header_generation(orig_buf);
1326 nritems = btrfs_header_nritems(buf);
1327 level = btrfs_header_level(buf);
1329 sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1332 if (root->ref_cows) {
1333 process_func = __btrfs_inc_extent_ref;
1336 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1339 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1341 process_func = __btrfs_update_extent_ref;
1345 * we make two passes through the items. In the first pass we
1346 * only record the byte number and slot. Then we sort based on
1347 * byte number and do the actual work based on the sorted results
1349 for (i = 0; i < nritems; i++) {
1352 btrfs_item_key_to_cpu(buf, &key, i);
1353 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1355 fi = btrfs_item_ptr(buf, i,
1356 struct btrfs_file_extent_item);
1357 if (btrfs_file_extent_type(buf, fi) ==
1358 BTRFS_FILE_EXTENT_INLINE)
1360 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1365 sorted[refi].bytenr = bytenr;
1366 sorted[refi].slot = i;
1369 bytenr = btrfs_node_blockptr(buf, i);
1370 sorted[refi].bytenr = bytenr;
1371 sorted[refi].slot = i;
1376 * if refi == 0, we didn't actually put anything into the sorted
1377 * array and we're done
1382 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1384 for (i = 0; i < refi; i++) {
1386 slot = sorted[i].slot;
1387 bytenr = sorted[i].bytenr;
1390 btrfs_item_key_to_cpu(buf, &key, slot);
1391 fi = btrfs_item_ptr(buf, slot,
1392 struct btrfs_file_extent_item);
1394 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1398 ret = process_func(trans, root, bytenr,
1399 btrfs_file_extent_disk_num_bytes(buf, fi),
1400 orig_buf->start, buf->start,
1401 orig_root, ref_root,
1402 orig_generation, ref_generation,
1411 ret = process_func(trans, root, bytenr, buf->len,
1412 orig_buf->start, buf->start,
1413 orig_root, ref_root,
1414 orig_generation, ref_generation,
1427 *nr_extents = nr_file_extents;
1429 *nr_extents = nritems;
1438 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1439 struct btrfs_root *root, struct extent_buffer *orig_buf,
1440 struct extent_buffer *buf, int start_slot, int nr)
1447 u64 orig_generation;
1448 struct btrfs_key key;
1449 struct btrfs_file_extent_item *fi;
1455 BUG_ON(start_slot < 0);
1456 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1458 ref_root = btrfs_header_owner(buf);
1459 ref_generation = btrfs_header_generation(buf);
1460 orig_root = btrfs_header_owner(orig_buf);
1461 orig_generation = btrfs_header_generation(orig_buf);
1462 level = btrfs_header_level(buf);
1464 if (!root->ref_cows) {
1466 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1469 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1473 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1476 btrfs_item_key_to_cpu(buf, &key, slot);
1477 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1479 fi = btrfs_item_ptr(buf, slot,
1480 struct btrfs_file_extent_item);
1481 if (btrfs_file_extent_type(buf, fi) ==
1482 BTRFS_FILE_EXTENT_INLINE)
1484 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1487 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1488 btrfs_file_extent_disk_num_bytes(buf, fi),
1489 orig_buf->start, buf->start,
1490 orig_root, ref_root, orig_generation,
1491 ref_generation, key.objectid);
1495 bytenr = btrfs_node_blockptr(buf, slot);
1496 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1497 buf->len, orig_buf->start,
1498 buf->start, orig_root, ref_root,
1499 orig_generation, ref_generation,
1511 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1512 struct btrfs_root *root,
1513 struct btrfs_path *path,
1514 struct btrfs_block_group_cache *cache)
1517 struct btrfs_root *extent_root = root->fs_info->extent_root;
1519 struct extent_buffer *leaf;
1521 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1526 leaf = path->nodes[0];
1527 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1528 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1529 btrfs_mark_buffer_dirty(leaf);
1530 btrfs_release_path(extent_root, path);
1538 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1539 struct btrfs_root *root)
1541 struct btrfs_block_group_cache *cache, *entry;
1545 struct btrfs_path *path;
1548 path = btrfs_alloc_path();
1554 spin_lock(&root->fs_info->block_group_cache_lock);
1555 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1556 n; n = rb_next(n)) {
1557 entry = rb_entry(n, struct btrfs_block_group_cache,
1564 spin_unlock(&root->fs_info->block_group_cache_lock);
1570 last += cache->key.offset;
1572 err = write_one_cache_group(trans, root,
1575 * if we fail to write the cache group, we want
1576 * to keep it marked dirty in hopes that a later
1584 btrfs_free_path(path);
1588 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1590 struct btrfs_block_group_cache *block_group;
1593 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1594 if (!block_group || block_group->ro)
1597 put_block_group(block_group);
1601 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1602 u64 total_bytes, u64 bytes_used,
1603 struct btrfs_space_info **space_info)
1605 struct btrfs_space_info *found;
1607 found = __find_space_info(info, flags);
1609 spin_lock(&found->lock);
1610 found->total_bytes += total_bytes;
1611 found->bytes_used += bytes_used;
1613 spin_unlock(&found->lock);
1614 *space_info = found;
1617 found = kzalloc(sizeof(*found), GFP_NOFS);
1621 INIT_LIST_HEAD(&found->block_groups);
1622 init_rwsem(&found->groups_sem);
1623 spin_lock_init(&found->lock);
1624 found->flags = flags;
1625 found->total_bytes = total_bytes;
1626 found->bytes_used = bytes_used;
1627 found->bytes_pinned = 0;
1628 found->bytes_reserved = 0;
1629 found->bytes_readonly = 0;
1630 found->bytes_delalloc = 0;
1632 found->force_alloc = 0;
1633 *space_info = found;
1634 list_add_rcu(&found->list, &info->space_info);
1638 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1640 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1641 BTRFS_BLOCK_GROUP_RAID1 |
1642 BTRFS_BLOCK_GROUP_RAID10 |
1643 BTRFS_BLOCK_GROUP_DUP);
1645 if (flags & BTRFS_BLOCK_GROUP_DATA)
1646 fs_info->avail_data_alloc_bits |= extra_flags;
1647 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1648 fs_info->avail_metadata_alloc_bits |= extra_flags;
1649 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1650 fs_info->avail_system_alloc_bits |= extra_flags;
1654 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1656 spin_lock(&cache->space_info->lock);
1657 spin_lock(&cache->lock);
1659 cache->space_info->bytes_readonly += cache->key.offset -
1660 btrfs_block_group_used(&cache->item);
1663 spin_unlock(&cache->lock);
1664 spin_unlock(&cache->space_info->lock);
1667 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1669 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1671 if (num_devices == 1)
1672 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1673 if (num_devices < 4)
1674 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1676 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1677 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1678 BTRFS_BLOCK_GROUP_RAID10))) {
1679 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1682 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1683 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1684 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1687 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1688 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1689 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1690 (flags & BTRFS_BLOCK_GROUP_DUP)))
1691 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1695 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
1697 struct btrfs_fs_info *info = root->fs_info;
1701 alloc_profile = info->avail_data_alloc_bits &
1702 info->data_alloc_profile;
1703 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1704 } else if (root == root->fs_info->chunk_root) {
1705 alloc_profile = info->avail_system_alloc_bits &
1706 info->system_alloc_profile;
1707 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1709 alloc_profile = info->avail_metadata_alloc_bits &
1710 info->metadata_alloc_profile;
1711 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1714 return btrfs_reduce_alloc_profile(root, data);
1717 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
1721 alloc_target = btrfs_get_alloc_profile(root, 1);
1722 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
1727 * for now this just makes sure we have at least 5% of our metadata space free
1730 int btrfs_check_metadata_free_space(struct btrfs_root *root)
1732 struct btrfs_fs_info *info = root->fs_info;
1733 struct btrfs_space_info *meta_sinfo;
1734 u64 alloc_target, thresh;
1735 int committed = 0, ret;
1737 /* get the space info for where the metadata will live */
1738 alloc_target = btrfs_get_alloc_profile(root, 0);
1739 meta_sinfo = __find_space_info(info, alloc_target);
1742 spin_lock(&meta_sinfo->lock);
1743 if (!meta_sinfo->full)
1744 thresh = meta_sinfo->total_bytes * 80;
1746 thresh = meta_sinfo->total_bytes * 95;
1748 do_div(thresh, 100);
1750 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
1751 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
1752 struct btrfs_trans_handle *trans;
1753 if (!meta_sinfo->full) {
1754 meta_sinfo->force_alloc = 1;
1755 spin_unlock(&meta_sinfo->lock);
1757 trans = btrfs_start_transaction(root, 1);
1761 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1762 2 * 1024 * 1024, alloc_target, 0);
1763 btrfs_end_transaction(trans, root);
1766 spin_unlock(&meta_sinfo->lock);
1770 trans = btrfs_join_transaction(root, 1);
1773 ret = btrfs_commit_transaction(trans, root);
1780 spin_unlock(&meta_sinfo->lock);
1786 * This will check the space that the inode allocates from to make sure we have
1787 * enough space for bytes.
1789 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
1792 struct btrfs_space_info *data_sinfo;
1793 int ret = 0, committed = 0;
1795 /* make sure bytes are sectorsize aligned */
1796 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1798 data_sinfo = BTRFS_I(inode)->space_info;
1800 /* make sure we have enough space to handle the data first */
1801 spin_lock(&data_sinfo->lock);
1802 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
1803 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
1804 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
1805 data_sinfo->bytes_may_use < bytes) {
1806 struct btrfs_trans_handle *trans;
1809 * if we don't have enough free bytes in this space then we need
1810 * to alloc a new chunk.
1812 if (!data_sinfo->full) {
1815 data_sinfo->force_alloc = 1;
1816 spin_unlock(&data_sinfo->lock);
1818 alloc_target = btrfs_get_alloc_profile(root, 1);
1819 trans = btrfs_start_transaction(root, 1);
1823 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1824 bytes + 2 * 1024 * 1024,
1826 btrfs_end_transaction(trans, root);
1831 spin_unlock(&data_sinfo->lock);
1833 /* commit the current transaction and try again */
1836 trans = btrfs_join_transaction(root, 1);
1839 ret = btrfs_commit_transaction(trans, root);
1845 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
1846 ", %llu bytes_used, %llu bytes_reserved, "
1847 "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
1848 "%llu total\n", bytes, data_sinfo->bytes_delalloc,
1849 data_sinfo->bytes_used, data_sinfo->bytes_reserved,
1850 data_sinfo->bytes_pinned, data_sinfo->bytes_readonly,
1851 data_sinfo->bytes_may_use, data_sinfo->total_bytes);
1854 data_sinfo->bytes_may_use += bytes;
1855 BTRFS_I(inode)->reserved_bytes += bytes;
1856 spin_unlock(&data_sinfo->lock);
1858 return btrfs_check_metadata_free_space(root);
1862 * if there was an error for whatever reason after calling
1863 * btrfs_check_data_free_space, call this so we can cleanup the counters.
1865 void btrfs_free_reserved_data_space(struct btrfs_root *root,
1866 struct inode *inode, u64 bytes)
1868 struct btrfs_space_info *data_sinfo;
1870 /* make sure bytes are sectorsize aligned */
1871 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1873 data_sinfo = BTRFS_I(inode)->space_info;
1874 spin_lock(&data_sinfo->lock);
1875 data_sinfo->bytes_may_use -= bytes;
1876 BTRFS_I(inode)->reserved_bytes -= bytes;
1877 spin_unlock(&data_sinfo->lock);
1880 /* called when we are adding a delalloc extent to the inode's io_tree */
1881 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
1884 struct btrfs_space_info *data_sinfo;
1886 /* get the space info for where this inode will be storing its data */
1887 data_sinfo = BTRFS_I(inode)->space_info;
1889 /* make sure we have enough space to handle the data first */
1890 spin_lock(&data_sinfo->lock);
1891 data_sinfo->bytes_delalloc += bytes;
1894 * we are adding a delalloc extent without calling
1895 * btrfs_check_data_free_space first. This happens on a weird
1896 * writepage condition, but shouldn't hurt our accounting
1898 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
1899 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
1900 BTRFS_I(inode)->reserved_bytes = 0;
1902 data_sinfo->bytes_may_use -= bytes;
1903 BTRFS_I(inode)->reserved_bytes -= bytes;
1906 spin_unlock(&data_sinfo->lock);
1909 /* called when we are clearing an delalloc extent from the inode's io_tree */
1910 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
1913 struct btrfs_space_info *info;
1915 info = BTRFS_I(inode)->space_info;
1917 spin_lock(&info->lock);
1918 info->bytes_delalloc -= bytes;
1919 spin_unlock(&info->lock);
1922 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1923 struct btrfs_root *extent_root, u64 alloc_bytes,
1924 u64 flags, int force)
1926 struct btrfs_space_info *space_info;
1930 mutex_lock(&extent_root->fs_info->chunk_mutex);
1932 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1934 space_info = __find_space_info(extent_root->fs_info, flags);
1936 ret = update_space_info(extent_root->fs_info, flags,
1940 BUG_ON(!space_info);
1942 spin_lock(&space_info->lock);
1943 if (space_info->force_alloc) {
1945 space_info->force_alloc = 0;
1947 if (space_info->full) {
1948 spin_unlock(&space_info->lock);
1952 thresh = space_info->total_bytes - space_info->bytes_readonly;
1953 thresh = div_factor(thresh, 6);
1955 (space_info->bytes_used + space_info->bytes_pinned +
1956 space_info->bytes_reserved + alloc_bytes) < thresh) {
1957 spin_unlock(&space_info->lock);
1960 spin_unlock(&space_info->lock);
1962 ret = btrfs_alloc_chunk(trans, extent_root, flags);
1964 space_info->full = 1;
1966 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1970 static int update_block_group(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes, int alloc,
1975 struct btrfs_block_group_cache *cache;
1976 struct btrfs_fs_info *info = root->fs_info;
1977 u64 total = num_bytes;
1982 cache = btrfs_lookup_block_group(info, bytenr);
1985 byte_in_group = bytenr - cache->key.objectid;
1986 WARN_ON(byte_in_group > cache->key.offset);
1988 spin_lock(&cache->space_info->lock);
1989 spin_lock(&cache->lock);
1991 old_val = btrfs_block_group_used(&cache->item);
1992 num_bytes = min(total, cache->key.offset - byte_in_group);
1994 old_val += num_bytes;
1995 cache->space_info->bytes_used += num_bytes;
1997 cache->space_info->bytes_readonly -= num_bytes;
1998 btrfs_set_block_group_used(&cache->item, old_val);
1999 spin_unlock(&cache->lock);
2000 spin_unlock(&cache->space_info->lock);
2002 old_val -= num_bytes;
2003 cache->space_info->bytes_used -= num_bytes;
2005 cache->space_info->bytes_readonly += num_bytes;
2006 btrfs_set_block_group_used(&cache->item, old_val);
2007 spin_unlock(&cache->lock);
2008 spin_unlock(&cache->space_info->lock);
2012 ret = btrfs_discard_extent(root, bytenr,
2016 ret = btrfs_add_free_space(cache, bytenr,
2021 put_block_group(cache);
2023 bytenr += num_bytes;
2028 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2030 struct btrfs_block_group_cache *cache;
2033 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2037 bytenr = cache->key.objectid;
2038 put_block_group(cache);
2043 int btrfs_update_pinned_extents(struct btrfs_root *root,
2044 u64 bytenr, u64 num, int pin)
2047 struct btrfs_block_group_cache *cache;
2048 struct btrfs_fs_info *fs_info = root->fs_info;
2050 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2052 set_extent_dirty(&fs_info->pinned_extents,
2053 bytenr, bytenr + num - 1, GFP_NOFS);
2055 clear_extent_dirty(&fs_info->pinned_extents,
2056 bytenr, bytenr + num - 1, GFP_NOFS);
2058 mutex_unlock(&root->fs_info->pinned_mutex);
2061 cache = btrfs_lookup_block_group(fs_info, bytenr);
2063 len = min(num, cache->key.offset -
2064 (bytenr - cache->key.objectid));
2066 spin_lock(&cache->space_info->lock);
2067 spin_lock(&cache->lock);
2068 cache->pinned += len;
2069 cache->space_info->bytes_pinned += len;
2070 spin_unlock(&cache->lock);
2071 spin_unlock(&cache->space_info->lock);
2072 fs_info->total_pinned += len;
2074 spin_lock(&cache->space_info->lock);
2075 spin_lock(&cache->lock);
2076 cache->pinned -= len;
2077 cache->space_info->bytes_pinned -= len;
2078 spin_unlock(&cache->lock);
2079 spin_unlock(&cache->space_info->lock);
2080 fs_info->total_pinned -= len;
2082 btrfs_add_free_space(cache, bytenr, len);
2084 put_block_group(cache);
2091 static int update_reserved_extents(struct btrfs_root *root,
2092 u64 bytenr, u64 num, int reserve)
2095 struct btrfs_block_group_cache *cache;
2096 struct btrfs_fs_info *fs_info = root->fs_info;
2099 cache = btrfs_lookup_block_group(fs_info, bytenr);
2101 len = min(num, cache->key.offset -
2102 (bytenr - cache->key.objectid));
2104 spin_lock(&cache->space_info->lock);
2105 spin_lock(&cache->lock);
2107 cache->reserved += len;
2108 cache->space_info->bytes_reserved += len;
2110 cache->reserved -= len;
2111 cache->space_info->bytes_reserved -= len;
2113 spin_unlock(&cache->lock);
2114 spin_unlock(&cache->space_info->lock);
2115 put_block_group(cache);
2122 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2127 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2130 mutex_lock(&root->fs_info->pinned_mutex);
2132 ret = find_first_extent_bit(pinned_extents, last,
2133 &start, &end, EXTENT_DIRTY);
2136 set_extent_dirty(copy, start, end, GFP_NOFS);
2139 mutex_unlock(&root->fs_info->pinned_mutex);
2143 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2144 struct btrfs_root *root,
2145 struct extent_io_tree *unpin)
2152 mutex_lock(&root->fs_info->pinned_mutex);
2153 ret = find_first_extent_bit(unpin, 0, &start, &end,
2158 ret = btrfs_discard_extent(root, start, end + 1 - start);
2160 /* unlocks the pinned mutex */
2161 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2162 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2166 mutex_unlock(&root->fs_info->pinned_mutex);
2170 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *root,
2172 struct btrfs_path *path,
2173 u64 bytenr, u64 num_bytes, int is_data,
2174 struct extent_buffer **must_clean)
2177 struct extent_buffer *buf;
2182 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2186 /* we can reuse a block if it hasn't been written
2187 * and it is from this transaction. We can't
2188 * reuse anything from the tree log root because
2189 * it has tiny sub-transactions.
2191 if (btrfs_buffer_uptodate(buf, 0) &&
2192 btrfs_try_tree_lock(buf)) {
2193 u64 header_owner = btrfs_header_owner(buf);
2194 u64 header_transid = btrfs_header_generation(buf);
2195 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2196 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2197 header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2198 header_transid == trans->transid &&
2199 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2203 btrfs_tree_unlock(buf);
2205 free_extent_buffer(buf);
2207 btrfs_set_path_blocking(path);
2208 mutex_lock(&root->fs_info->pinned_mutex);
2209 /* unlocks the pinned mutex */
2210 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2217 * remove an extent from the root, returns 0 on success
2219 static int __free_extent(struct btrfs_trans_handle *trans,
2220 struct btrfs_root *root,
2221 u64 bytenr, u64 num_bytes, u64 parent,
2222 u64 root_objectid, u64 ref_generation,
2223 u64 owner_objectid, int pin, int mark_free,
2226 struct btrfs_path *path;
2227 struct btrfs_key key;
2228 struct btrfs_fs_info *info = root->fs_info;
2229 struct btrfs_root *extent_root = info->extent_root;
2230 struct extent_buffer *leaf;
2232 int extent_slot = 0;
2233 int found_extent = 0;
2235 struct btrfs_extent_item *ei;
2238 key.objectid = bytenr;
2239 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2240 key.offset = num_bytes;
2241 path = btrfs_alloc_path();
2246 path->leave_spinning = 1;
2247 ret = lookup_extent_backref(trans, extent_root, path,
2248 bytenr, parent, root_objectid,
2249 ref_generation, owner_objectid, 1);
2251 struct btrfs_key found_key;
2252 extent_slot = path->slots[0];
2253 while (extent_slot > 0) {
2255 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2257 if (found_key.objectid != bytenr)
2259 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2260 found_key.offset == num_bytes) {
2264 if (path->slots[0] - extent_slot > 5)
2267 if (!found_extent) {
2268 ret = remove_extent_backref(trans, extent_root, path,
2271 btrfs_release_path(extent_root, path);
2272 path->leave_spinning = 1;
2273 ret = btrfs_search_slot(trans, extent_root,
2276 printk(KERN_ERR "umm, got %d back from search"
2277 ", was looking for %llu\n", ret,
2278 (unsigned long long)bytenr);
2279 btrfs_print_leaf(extent_root, path->nodes[0]);
2282 extent_slot = path->slots[0];
2285 btrfs_print_leaf(extent_root, path->nodes[0]);
2287 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2288 "parent %llu root %llu gen %llu owner %llu\n",
2289 (unsigned long long)bytenr,
2290 (unsigned long long)parent,
2291 (unsigned long long)root_objectid,
2292 (unsigned long long)ref_generation,
2293 (unsigned long long)owner_objectid);
2296 leaf = path->nodes[0];
2297 ei = btrfs_item_ptr(leaf, extent_slot,
2298 struct btrfs_extent_item);
2299 refs = btrfs_extent_refs(leaf, ei);
2302 * we're not allowed to delete the extent item if there
2303 * are other delayed ref updates pending
2306 BUG_ON(refs < refs_to_drop);
2307 refs -= refs_to_drop;
2308 btrfs_set_extent_refs(leaf, ei, refs);
2309 btrfs_mark_buffer_dirty(leaf);
2311 if (refs == 0 && found_extent &&
2312 path->slots[0] == extent_slot + 1) {
2313 struct btrfs_extent_ref *ref;
2314 ref = btrfs_item_ptr(leaf, path->slots[0],
2315 struct btrfs_extent_ref);
2316 BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
2317 /* if the back ref and the extent are next to each other
2318 * they get deleted below in one shot
2320 path->slots[0] = extent_slot;
2322 } else if (found_extent) {
2323 /* otherwise delete the extent back ref */
2324 ret = remove_extent_backref(trans, extent_root, path,
2327 /* if refs are 0, we need to setup the path for deletion */
2329 btrfs_release_path(extent_root, path);
2330 path->leave_spinning = 1;
2331 ret = btrfs_search_slot(trans, extent_root, &key, path,
2340 struct extent_buffer *must_clean = NULL;
2343 ret = pin_down_bytes(trans, root, path,
2345 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID,
2352 /* block accounting for super block */
2353 spin_lock(&info->delalloc_lock);
2354 super_used = btrfs_super_bytes_used(&info->super_copy);
2355 btrfs_set_super_bytes_used(&info->super_copy,
2356 super_used - num_bytes);
2358 /* block accounting for root item */
2359 root_used = btrfs_root_used(&root->root_item);
2360 btrfs_set_root_used(&root->root_item,
2361 root_used - num_bytes);
2362 spin_unlock(&info->delalloc_lock);
2365 * it is going to be very rare for someone to be waiting
2366 * on the block we're freeing. del_items might need to
2367 * schedule, so rather than get fancy, just force it
2371 btrfs_set_lock_blocking(must_clean);
2373 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2376 btrfs_release_path(extent_root, path);
2379 clean_tree_block(NULL, root, must_clean);
2380 btrfs_tree_unlock(must_clean);
2381 free_extent_buffer(must_clean);
2384 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2385 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2389 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2393 btrfs_free_path(path);
2398 * remove an extent from the root, returns 0 on success
2400 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2401 struct btrfs_root *root,
2402 u64 bytenr, u64 num_bytes, u64 parent,
2403 u64 root_objectid, u64 ref_generation,
2404 u64 owner_objectid, int pin,
2407 WARN_ON(num_bytes < root->sectorsize);
2410 * if metadata always pin
2411 * if data pin when any transaction has committed this
2413 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2414 ref_generation != trans->transid)
2417 if (ref_generation != trans->transid)
2420 return __free_extent(trans, root, bytenr, num_bytes, parent,
2421 root_objectid, ref_generation,
2422 owner_objectid, pin, pin == 0, refs_to_drop);
2426 * when we free an extent, it is possible (and likely) that we free the last
2427 * delayed ref for that extent as well. This searches the delayed ref tree for
2428 * a given extent, and if there are no other delayed refs to be processed, it
2429 * removes it from the tree.
2431 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2432 struct btrfs_root *root, u64 bytenr)
2434 struct btrfs_delayed_ref_head *head;
2435 struct btrfs_delayed_ref_root *delayed_refs;
2436 struct btrfs_delayed_ref_node *ref;
2437 struct rb_node *node;
2440 delayed_refs = &trans->transaction->delayed_refs;
2441 spin_lock(&delayed_refs->lock);
2442 head = btrfs_find_delayed_ref_head(trans, bytenr);
2446 node = rb_prev(&head->node.rb_node);
2450 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2452 /* there are still entries for this ref, we can't drop it */
2453 if (ref->bytenr == bytenr)
2457 * waiting for the lock here would deadlock. If someone else has it
2458 * locked they are already in the process of dropping it anyway
2460 if (!mutex_trylock(&head->mutex))
2464 * at this point we have a head with no other entries. Go
2465 * ahead and process it.
2467 head->node.in_tree = 0;
2468 rb_erase(&head->node.rb_node, &delayed_refs->root);
2470 delayed_refs->num_entries--;
2473 * we don't take a ref on the node because we're removing it from the
2474 * tree, so we just steal the ref the tree was holding.
2476 delayed_refs->num_heads--;
2477 if (list_empty(&head->cluster))
2478 delayed_refs->num_heads_ready--;
2480 list_del_init(&head->cluster);
2481 spin_unlock(&delayed_refs->lock);
2483 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2484 &head->node, head->must_insert_reserved);
2486 btrfs_put_delayed_ref(&head->node);
2489 spin_unlock(&delayed_refs->lock);
2493 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2494 struct btrfs_root *root,
2495 u64 bytenr, u64 num_bytes, u64 parent,
2496 u64 root_objectid, u64 ref_generation,
2497 u64 owner_objectid, int pin)
2502 * tree log blocks never actually go into the extent allocation
2503 * tree, just update pinning info and exit early.
2505 * data extents referenced by the tree log do need to have
2506 * their reference counts bumped.
2508 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2509 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2510 mutex_lock(&root->fs_info->pinned_mutex);
2512 /* unlocks the pinned mutex */
2513 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2514 update_reserved_extents(root, bytenr, num_bytes, 0);
2517 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2518 root_objectid, ref_generation,
2520 BTRFS_DROP_DELAYED_REF, 1);
2522 ret = check_ref_cleanup(trans, root, bytenr);
2528 static u64 stripe_align(struct btrfs_root *root, u64 val)
2530 u64 mask = ((u64)root->stripesize - 1);
2531 u64 ret = (val + mask) & ~mask;
2536 * walks the btree of allocated extents and find a hole of a given size.
2537 * The key ins is changed to record the hole:
2538 * ins->objectid == block start
2539 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2540 * ins->offset == number of blocks
2541 * Any available blocks before search_start are skipped.
2543 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2544 struct btrfs_root *orig_root,
2545 u64 num_bytes, u64 empty_size,
2546 u64 search_start, u64 search_end,
2547 u64 hint_byte, struct btrfs_key *ins,
2548 u64 exclude_start, u64 exclude_nr,
2552 struct btrfs_root *root = orig_root->fs_info->extent_root;
2553 u64 total_needed = num_bytes;
2554 u64 *last_ptr = NULL;
2555 u64 last_wanted = 0;
2556 struct btrfs_block_group_cache *block_group = NULL;
2557 int chunk_alloc_done = 0;
2558 int empty_cluster = 2 * 1024 * 1024;
2559 int allowed_chunk_alloc = 0;
2560 struct list_head *head = NULL, *cur = NULL;
2563 struct btrfs_space_info *space_info;
2565 WARN_ON(num_bytes < root->sectorsize);
2566 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2570 if (orig_root->ref_cows || empty_size)
2571 allowed_chunk_alloc = 1;
2573 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2574 last_ptr = &root->fs_info->last_alloc;
2575 if (!btrfs_test_opt(root, SSD))
2576 empty_cluster = 64 * 1024;
2579 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2580 last_ptr = &root->fs_info->last_data_alloc;
2584 hint_byte = *last_ptr;
2585 last_wanted = *last_ptr;
2587 empty_size += empty_cluster;
2591 search_start = max(search_start, first_logical_byte(root, 0));
2592 search_start = max(search_start, hint_byte);
2594 if (last_wanted && search_start != last_wanted) {
2596 empty_size += empty_cluster;
2599 total_needed += empty_size;
2600 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2602 block_group = btrfs_lookup_first_block_group(root->fs_info,
2604 space_info = __find_space_info(root->fs_info, data);
2606 down_read(&space_info->groups_sem);
2608 struct btrfs_free_space *free_space;
2610 * the only way this happens if our hint points to a block
2611 * group thats not of the proper type, while looping this
2612 * should never happen
2618 goto new_group_no_lock;
2620 if (unlikely(!block_group->cached)) {
2621 mutex_lock(&block_group->cache_mutex);
2622 ret = cache_block_group(root, block_group);
2623 mutex_unlock(&block_group->cache_mutex);
2628 mutex_lock(&block_group->alloc_mutex);
2629 if (unlikely(!block_group_bits(block_group, data)))
2632 if (unlikely(block_group->ro))
2635 free_space = btrfs_find_free_space(block_group, search_start,
2638 u64 start = block_group->key.objectid;
2639 u64 end = block_group->key.objectid +
2640 block_group->key.offset;
2642 search_start = stripe_align(root, free_space->offset);
2644 /* move on to the next group */
2645 if (search_start + num_bytes >= search_end)
2648 /* move on to the next group */
2649 if (search_start + num_bytes > end)
2652 if (last_wanted && search_start != last_wanted) {
2653 total_needed += empty_cluster;
2654 empty_size += empty_cluster;
2657 * if search_start is still in this block group
2658 * then we just re-search this block group
2660 if (search_start >= start &&
2661 search_start < end) {
2662 mutex_unlock(&block_group->alloc_mutex);
2666 /* else we go to the next block group */
2670 if (exclude_nr > 0 &&
2671 (search_start + num_bytes > exclude_start &&
2672 search_start < exclude_start + exclude_nr)) {
2673 search_start = exclude_start + exclude_nr;
2675 * if search_start is still in this block group
2676 * then we just re-search this block group
2678 if (search_start >= start &&
2679 search_start < end) {
2680 mutex_unlock(&block_group->alloc_mutex);
2685 /* else we go to the next block group */
2689 ins->objectid = search_start;
2690 ins->offset = num_bytes;
2692 btrfs_remove_free_space_lock(block_group, search_start,
2694 /* we are all good, lets return */
2695 mutex_unlock(&block_group->alloc_mutex);
2699 mutex_unlock(&block_group->alloc_mutex);
2700 put_block_group(block_group);
2703 /* don't try to compare new allocations against the
2704 * last allocation any more
2709 * Here's how this works.
2710 * loop == 0: we were searching a block group via a hint
2711 * and didn't find anything, so we start at
2712 * the head of the block groups and keep searching
2713 * loop == 1: we're searching through all of the block groups
2714 * if we hit the head again we have searched
2715 * all of the block groups for this space and we
2716 * need to try and allocate, if we cant error out.
2717 * loop == 2: we allocated more space and are looping through
2718 * all of the block groups again.
2721 head = &space_info->block_groups;
2724 } else if (loop == 1 && cur == head) {
2727 /* at this point we give up on the empty_size
2728 * allocations and just try to allocate the min
2731 * The extra_loop field was set if an empty_size
2732 * allocation was attempted above, and if this
2733 * is try we need to try the loop again without
2734 * the additional empty_size.
2736 total_needed -= empty_size;
2738 keep_going = extra_loop;
2741 if (allowed_chunk_alloc && !chunk_alloc_done) {
2742 up_read(&space_info->groups_sem);
2743 ret = do_chunk_alloc(trans, root, num_bytes +
2744 2 * 1024 * 1024, data, 1);
2745 down_read(&space_info->groups_sem);
2748 head = &space_info->block_groups;
2750 * we've allocated a new chunk, keep
2754 chunk_alloc_done = 1;
2755 } else if (!allowed_chunk_alloc) {
2756 space_info->force_alloc = 1;
2765 } else if (cur == head) {
2769 block_group = list_entry(cur, struct btrfs_block_group_cache,
2771 atomic_inc(&block_group->count);
2773 search_start = block_group->key.objectid;
2777 /* we found what we needed */
2778 if (ins->objectid) {
2779 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2780 trans->block_group = block_group->key.objectid;
2783 *last_ptr = ins->objectid + ins->offset;
2786 printk(KERN_ERR "btrfs searching for %llu bytes, "
2787 "num_bytes %llu, loop %d, allowed_alloc %d\n",
2788 (unsigned long long)total_needed,
2789 (unsigned long long)num_bytes,
2790 loop, allowed_chunk_alloc);
2794 put_block_group(block_group);
2796 up_read(&space_info->groups_sem);
2800 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2802 struct btrfs_block_group_cache *cache;
2804 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
2805 (unsigned long long)(info->total_bytes - info->bytes_used -
2806 info->bytes_pinned - info->bytes_reserved),
2807 (info->full) ? "" : "not ");
2808 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
2809 " may_use=%llu, used=%llu\n", info->total_bytes,
2810 info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
2813 down_read(&info->groups_sem);
2814 list_for_each_entry(cache, &info->block_groups, list) {
2815 spin_lock(&cache->lock);
2816 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
2817 "%llu pinned %llu reserved\n",
2818 (unsigned long long)cache->key.objectid,
2819 (unsigned long long)cache->key.offset,
2820 (unsigned long long)btrfs_block_group_used(&cache->item),
2821 (unsigned long long)cache->pinned,
2822 (unsigned long long)cache->reserved);
2823 btrfs_dump_free_space(cache, bytes);
2824 spin_unlock(&cache->lock);
2826 up_read(&info->groups_sem);
2829 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2830 struct btrfs_root *root,
2831 u64 num_bytes, u64 min_alloc_size,
2832 u64 empty_size, u64 hint_byte,
2833 u64 search_end, struct btrfs_key *ins,
2837 u64 search_start = 0;
2838 struct btrfs_fs_info *info = root->fs_info;
2840 data = btrfs_get_alloc_profile(root, data);
2843 * the only place that sets empty_size is btrfs_realloc_node, which
2844 * is not called recursively on allocations
2846 if (empty_size || root->ref_cows) {
2847 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2848 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2850 BTRFS_BLOCK_GROUP_METADATA |
2851 (info->metadata_alloc_profile &
2852 info->avail_metadata_alloc_bits), 0);
2854 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2855 num_bytes + 2 * 1024 * 1024, data, 0);
2858 WARN_ON(num_bytes < root->sectorsize);
2859 ret = find_free_extent(trans, root, num_bytes, empty_size,
2860 search_start, search_end, hint_byte, ins,
2861 trans->alloc_exclude_start,
2862 trans->alloc_exclude_nr, data);
2864 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2865 num_bytes = num_bytes >> 1;
2866 num_bytes = num_bytes & ~(root->sectorsize - 1);
2867 num_bytes = max(num_bytes, min_alloc_size);
2868 do_chunk_alloc(trans, root->fs_info->extent_root,
2869 num_bytes, data, 1);
2873 struct btrfs_space_info *sinfo;
2875 sinfo = __find_space_info(root->fs_info, data);
2876 printk(KERN_ERR "btrfs allocation failed flags %llu, "
2877 "wanted %llu\n", (unsigned long long)data,
2878 (unsigned long long)num_bytes);
2879 dump_space_info(sinfo, num_bytes);
2886 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2888 struct btrfs_block_group_cache *cache;
2891 cache = btrfs_lookup_block_group(root->fs_info, start);
2893 printk(KERN_ERR "Unable to find block group for %llu\n",
2894 (unsigned long long)start);
2898 ret = btrfs_discard_extent(root, start, len);
2900 btrfs_add_free_space(cache, start, len);
2901 put_block_group(cache);
2902 update_reserved_extents(root, start, len, 0);
2907 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2908 struct btrfs_root *root,
2909 u64 num_bytes, u64 min_alloc_size,
2910 u64 empty_size, u64 hint_byte,
2911 u64 search_end, struct btrfs_key *ins,
2915 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2916 empty_size, hint_byte, search_end, ins,
2918 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2922 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2923 struct btrfs_root *root, u64 parent,
2924 u64 root_objectid, u64 ref_generation,
2925 u64 owner, struct btrfs_key *ins,
2931 u64 num_bytes = ins->offset;
2933 struct btrfs_fs_info *info = root->fs_info;
2934 struct btrfs_root *extent_root = info->extent_root;
2935 struct btrfs_extent_item *extent_item;
2936 struct btrfs_extent_ref *ref;
2937 struct btrfs_path *path;
2938 struct btrfs_key keys[2];
2941 parent = ins->objectid;
2943 /* block accounting for super block */
2944 spin_lock(&info->delalloc_lock);
2945 super_used = btrfs_super_bytes_used(&info->super_copy);
2946 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2948 /* block accounting for root item */
2949 root_used = btrfs_root_used(&root->root_item);
2950 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2951 spin_unlock(&info->delalloc_lock);
2953 memcpy(&keys[0], ins, sizeof(*ins));
2954 keys[1].objectid = ins->objectid;
2955 keys[1].type = BTRFS_EXTENT_REF_KEY;
2956 keys[1].offset = parent;
2957 sizes[0] = sizeof(*extent_item);
2958 sizes[1] = sizeof(*ref);
2960 path = btrfs_alloc_path();
2963 path->leave_spinning = 1;
2964 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2968 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2969 struct btrfs_extent_item);
2970 btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
2971 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2972 struct btrfs_extent_ref);
2974 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2975 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2976 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2977 btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
2979 btrfs_mark_buffer_dirty(path->nodes[0]);
2981 trans->alloc_exclude_start = 0;
2982 trans->alloc_exclude_nr = 0;
2983 btrfs_free_path(path);
2988 ret = update_block_group(trans, root, ins->objectid,
2991 printk(KERN_ERR "btrfs update block group failed for %llu "
2992 "%llu\n", (unsigned long long)ins->objectid,
2993 (unsigned long long)ins->offset);
3000 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3001 struct btrfs_root *root, u64 parent,
3002 u64 root_objectid, u64 ref_generation,
3003 u64 owner, struct btrfs_key *ins)
3007 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3010 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3011 ins->offset, parent, root_objectid,
3012 ref_generation, owner,
3013 BTRFS_ADD_DELAYED_EXTENT, 0);
3019 * this is used by the tree logging recovery code. It records that
3020 * an extent has been allocated and makes sure to clear the free
3021 * space cache bits as well
3023 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3024 struct btrfs_root *root, u64 parent,
3025 u64 root_objectid, u64 ref_generation,
3026 u64 owner, struct btrfs_key *ins)
3029 struct btrfs_block_group_cache *block_group;
3031 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3032 mutex_lock(&block_group->cache_mutex);
3033 cache_block_group(root, block_group);
3034 mutex_unlock(&block_group->cache_mutex);
3036 ret = btrfs_remove_free_space(block_group, ins->objectid,
3039 put_block_group(block_group);
3040 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3041 ref_generation, owner, ins, 1);
3046 * finds a free extent and does all the dirty work required for allocation
3047 * returns the key for the extent through ins, and a tree buffer for
3048 * the first block of the extent through buf.
3050 * returns 0 if everything worked, non-zero otherwise.
3052 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3053 struct btrfs_root *root,
3054 u64 num_bytes, u64 parent, u64 min_alloc_size,
3055 u64 root_objectid, u64 ref_generation,
3056 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3057 u64 search_end, struct btrfs_key *ins, u64 data)
3060 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3061 min_alloc_size, empty_size, hint_byte,
3062 search_end, ins, data);
3064 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3065 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3066 ins->offset, parent, root_objectid,
3067 ref_generation, owner_objectid,
3068 BTRFS_ADD_DELAYED_EXTENT, 0);
3071 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3075 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3076 struct btrfs_root *root,
3077 u64 bytenr, u32 blocksize,
3080 struct extent_buffer *buf;
3082 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3084 return ERR_PTR(-ENOMEM);
3085 btrfs_set_header_generation(buf, trans->transid);
3086 btrfs_set_buffer_lockdep_class(buf, level);
3087 btrfs_tree_lock(buf);
3088 clean_tree_block(trans, root, buf);
3090 btrfs_set_lock_blocking(buf);
3091 btrfs_set_buffer_uptodate(buf);
3093 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3094 set_extent_dirty(&root->dirty_log_pages, buf->start,
3095 buf->start + buf->len - 1, GFP_NOFS);
3097 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3098 buf->start + buf->len - 1, GFP_NOFS);
3100 trans->blocks_used++;
3101 /* this returns a buffer locked for blocking */
3106 * helper function to allocate a block for a given tree
3107 * returns the tree buffer or NULL.
3109 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3110 struct btrfs_root *root,
3111 u32 blocksize, u64 parent,
3118 struct btrfs_key ins;
3120 struct extent_buffer *buf;
3122 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3123 root_objectid, ref_generation, level,
3124 empty_size, hint, (u64)-1, &ins, 0);
3127 return ERR_PTR(ret);
3130 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3135 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3136 struct btrfs_root *root, struct extent_buffer *leaf)
3139 u64 leaf_generation;
3140 struct refsort *sorted;
3141 struct btrfs_key key;
3142 struct btrfs_file_extent_item *fi;
3149 BUG_ON(!btrfs_is_leaf(leaf));
3150 nritems = btrfs_header_nritems(leaf);
3151 leaf_owner = btrfs_header_owner(leaf);
3152 leaf_generation = btrfs_header_generation(leaf);
3154 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3155 /* we do this loop twice. The first time we build a list
3156 * of the extents we have a reference on, then we sort the list
3157 * by bytenr. The second time around we actually do the
3160 for (i = 0; i < nritems; i++) {
3164 btrfs_item_key_to_cpu(leaf, &key, i);
3166 /* only extents have references, skip everything else */
3167 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3170 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3172 /* inline extents live in the btree, they don't have refs */
3173 if (btrfs_file_extent_type(leaf, fi) ==
3174 BTRFS_FILE_EXTENT_INLINE)
3177 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3179 /* holes don't have refs */
3180 if (disk_bytenr == 0)
3183 sorted[refi].bytenr = disk_bytenr;
3184 sorted[refi].slot = i;
3191 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3193 for (i = 0; i < refi; i++) {
3196 disk_bytenr = sorted[i].bytenr;
3197 slot = sorted[i].slot;
3201 btrfs_item_key_to_cpu(leaf, &key, slot);
3202 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3205 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3207 ret = btrfs_free_extent(trans, root, disk_bytenr,
3208 btrfs_file_extent_disk_num_bytes(leaf, fi),
3209 leaf->start, leaf_owner, leaf_generation,
3213 atomic_inc(&root->fs_info->throttle_gen);
3214 wake_up(&root->fs_info->transaction_throttle);
3222 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3223 struct btrfs_root *root,
3224 struct btrfs_leaf_ref *ref)
3228 struct btrfs_extent_info *info;
3229 struct refsort *sorted;
3231 if (ref->nritems == 0)
3234 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3235 for (i = 0; i < ref->nritems; i++) {
3236 sorted[i].bytenr = ref->extents[i].bytenr;
3239 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3242 * the items in the ref were sorted when the ref was inserted
3243 * into the ref cache, so this is already in order
3245 for (i = 0; i < ref->nritems; i++) {
3246 info = ref->extents + sorted[i].slot;
3247 ret = btrfs_free_extent(trans, root, info->bytenr,
3248 info->num_bytes, ref->bytenr,
3249 ref->owner, ref->generation,
3252 atomic_inc(&root->fs_info->throttle_gen);
3253 wake_up(&root->fs_info->transaction_throttle);
3264 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3265 struct btrfs_root *root, u64 start,
3270 ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3273 #if 0 /* some debugging code in case we see problems here */
3274 /* if the refs count is one, it won't get increased again. But
3275 * if the ref count is > 1, someone may be decreasing it at
3276 * the same time we are.
3279 struct extent_buffer *eb = NULL;
3280 eb = btrfs_find_create_tree_block(root, start, len);
3282 btrfs_tree_lock(eb);
3284 mutex_lock(&root->fs_info->alloc_mutex);
3285 ret = lookup_extent_ref(NULL, root, start, len, refs);
3287 mutex_unlock(&root->fs_info->alloc_mutex);
3290 btrfs_tree_unlock(eb);
3291 free_extent_buffer(eb);
3294 printk(KERN_ERR "btrfs block %llu went down to one "
3295 "during drop_snap\n", (unsigned long long)start);
3306 * this is used while deleting old snapshots, and it drops the refs
3307 * on a whole subtree starting from a level 1 node.
3309 * The idea is to sort all the leaf pointers, and then drop the
3310 * ref on all the leaves in order. Most of the time the leaves
3311 * will have ref cache entries, so no leaf IOs will be required to
3312 * find the extents they have references on.
3314 * For each leaf, any references it has are also dropped in order
3316 * This ends up dropping the references in something close to optimal
3317 * order for reading and modifying the extent allocation tree.
3319 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3320 struct btrfs_root *root,
3321 struct btrfs_path *path)
3326 struct extent_buffer *eb = path->nodes[1];
3327 struct extent_buffer *leaf;
3328 struct btrfs_leaf_ref *ref;
3329 struct refsort *sorted = NULL;
3330 int nritems = btrfs_header_nritems(eb);
3334 int slot = path->slots[1];
3335 u32 blocksize = btrfs_level_size(root, 0);
3341 root_owner = btrfs_header_owner(eb);
3342 root_gen = btrfs_header_generation(eb);
3343 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3346 * step one, sort all the leaf pointers so we don't scribble
3347 * randomly into the extent allocation tree
3349 for (i = slot; i < nritems; i++) {
3350 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3351 sorted[refi].slot = i;
3356 * nritems won't be zero, but if we're picking up drop_snapshot
3357 * after a crash, slot might be > 0, so double check things
3363 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3366 * the first loop frees everything the leaves point to
3368 for (i = 0; i < refi; i++) {
3371 bytenr = sorted[i].bytenr;
3374 * check the reference count on this leaf. If it is > 1
3375 * we just decrement it below and don't update any
3376 * of the refs the leaf points to.
3378 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3384 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3387 * the leaf only had one reference, which means the
3388 * only thing pointing to this leaf is the snapshot
3389 * we're deleting. It isn't possible for the reference
3390 * count to increase again later
3392 * The reference cache is checked for the leaf,
3393 * and if found we'll be able to drop any refs held by
3394 * the leaf without needing to read it in.
3396 ref = btrfs_lookup_leaf_ref(root, bytenr);
3397 if (ref && ref->generation != ptr_gen) {
3398 btrfs_free_leaf_ref(root, ref);
3402 ret = cache_drop_leaf_ref(trans, root, ref);
3404 btrfs_remove_leaf_ref(root, ref);
3405 btrfs_free_leaf_ref(root, ref);
3408 * the leaf wasn't in the reference cache, so
3409 * we have to read it.
3411 leaf = read_tree_block(root, bytenr, blocksize,
3413 ret = btrfs_drop_leaf_ref(trans, root, leaf);
3415 free_extent_buffer(leaf);
3417 atomic_inc(&root->fs_info->throttle_gen);
3418 wake_up(&root->fs_info->transaction_throttle);
3423 * run through the loop again to free the refs on the leaves.
3424 * This is faster than doing it in the loop above because
3425 * the leaves are likely to be clustered together. We end up
3426 * working in nice chunks on the extent allocation tree.
3428 for (i = 0; i < refi; i++) {
3429 bytenr = sorted[i].bytenr;
3430 ret = btrfs_free_extent(trans, root, bytenr,
3431 blocksize, eb->start,
3432 root_owner, root_gen, 0, 1);
3435 atomic_inc(&root->fs_info->throttle_gen);
3436 wake_up(&root->fs_info->transaction_throttle);
3443 * update the path to show we've processed the entire level 1
3444 * node. This will get saved into the root's drop_snapshot_progress
3445 * field so these drops are not repeated again if this transaction
3448 path->slots[1] = nritems;
3453 * helper function for drop_snapshot, this walks down the tree dropping ref
3454 * counts as it goes.
3456 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3457 struct btrfs_root *root,
3458 struct btrfs_path *path, int *level)
3464 struct extent_buffer *next;
3465 struct extent_buffer *cur;
3466 struct extent_buffer *parent;
3471 WARN_ON(*level < 0);
3472 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3473 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
3474 path->nodes[*level]->len, &refs);
3480 * walk down to the last node level and free all the leaves
3482 while (*level >= 0) {
3483 WARN_ON(*level < 0);
3484 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3485 cur = path->nodes[*level];
3487 if (btrfs_header_level(cur) != *level)
3490 if (path->slots[*level] >=
3491 btrfs_header_nritems(cur))
3494 /* the new code goes down to level 1 and does all the
3495 * leaves pointed to that node in bulk. So, this check
3496 * for level 0 will always be false.
3498 * But, the disk format allows the drop_snapshot_progress
3499 * field in the root to leave things in a state where
3500 * a leaf will need cleaning up here. If someone crashes
3501 * with the old code and then boots with the new code,
3502 * we might find a leaf here.
3505 ret = btrfs_drop_leaf_ref(trans, root, cur);
3511 * once we get to level one, process the whole node
3512 * at once, including everything below it.
3515 ret = drop_level_one_refs(trans, root, path);
3520 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3521 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3522 blocksize = btrfs_level_size(root, *level - 1);
3524 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3529 * if there is more than one reference, we don't need
3530 * to read that node to drop any references it has. We
3531 * just drop the ref we hold on that node and move on to the
3532 * next slot in this level.
3535 parent = path->nodes[*level];
3536 root_owner = btrfs_header_owner(parent);
3537 root_gen = btrfs_header_generation(parent);
3538 path->slots[*level]++;
3540 ret = btrfs_free_extent(trans, root, bytenr,
3541 blocksize, parent->start,
3542 root_owner, root_gen,
3546 atomic_inc(&root->fs_info->throttle_gen);
3547 wake_up(&root->fs_info->transaction_throttle);
3554 * we need to keep freeing things in the next level down.
3555 * read the block and loop around to process it
3557 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3558 WARN_ON(*level <= 0);
3559 if (path->nodes[*level-1])
3560 free_extent_buffer(path->nodes[*level-1]);
3561 path->nodes[*level-1] = next;
3562 *level = btrfs_header_level(next);
3563 path->slots[*level] = 0;
3567 WARN_ON(*level < 0);
3568 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3570 if (path->nodes[*level] == root->node) {
3571 parent = path->nodes[*level];
3572 bytenr = path->nodes[*level]->start;
3574 parent = path->nodes[*level + 1];
3575 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3578 blocksize = btrfs_level_size(root, *level);
3579 root_owner = btrfs_header_owner(parent);
3580 root_gen = btrfs_header_generation(parent);
3583 * cleanup and free the reference on the last node
3586 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3587 parent->start, root_owner, root_gen,
3589 free_extent_buffer(path->nodes[*level]);
3590 path->nodes[*level] = NULL;
3600 * helper function for drop_subtree, this function is similar to
3601 * walk_down_tree. The main difference is that it checks reference
3602 * counts while tree blocks are locked.
3604 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3605 struct btrfs_root *root,
3606 struct btrfs_path *path, int *level)
3608 struct extent_buffer *next;
3609 struct extent_buffer *cur;
3610 struct extent_buffer *parent;
3617 cur = path->nodes[*level];
3618 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3624 while (*level >= 0) {
3625 cur = path->nodes[*level];
3627 ret = btrfs_drop_leaf_ref(trans, root, cur);
3629 clean_tree_block(trans, root, cur);
3632 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3633 clean_tree_block(trans, root, cur);
3637 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3638 blocksize = btrfs_level_size(root, *level - 1);
3639 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3641 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3642 btrfs_tree_lock(next);
3643 btrfs_set_lock_blocking(next);
3645 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3649 parent = path->nodes[*level];
3650 ret = btrfs_free_extent(trans, root, bytenr,
3651 blocksize, parent->start,
3652 btrfs_header_owner(parent),
3653 btrfs_header_generation(parent),
3656 path->slots[*level]++;
3657 btrfs_tree_unlock(next);
3658 free_extent_buffer(next);
3662 *level = btrfs_header_level(next);
3663 path->nodes[*level] = next;
3664 path->slots[*level] = 0;
3665 path->locks[*level] = 1;
3669 parent = path->nodes[*level + 1];
3670 bytenr = path->nodes[*level]->start;
3671 blocksize = path->nodes[*level]->len;
3673 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3674 parent->start, btrfs_header_owner(parent),
3675 btrfs_header_generation(parent), *level, 1);
3678 if (path->locks[*level]) {
3679 btrfs_tree_unlock(path->nodes[*level]);
3680 path->locks[*level] = 0;
3682 free_extent_buffer(path->nodes[*level]);
3683 path->nodes[*level] = NULL;
3690 * helper for dropping snapshots. This walks back up the tree in the path
3691 * to find the first node higher up where we haven't yet gone through
3694 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3695 struct btrfs_root *root,
3696 struct btrfs_path *path,
3697 int *level, int max_level)
3701 struct btrfs_root_item *root_item = &root->root_item;
3706 for (i = *level; i < max_level && path->nodes[i]; i++) {
3707 slot = path->slots[i];
3708 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3709 struct extent_buffer *node;
3710 struct btrfs_disk_key disk_key;
3713 * there is more work to do in this level.
3714 * Update the drop_progress marker to reflect
3715 * the work we've done so far, and then bump
3718 node = path->nodes[i];
3721 WARN_ON(*level == 0);
3722 btrfs_node_key(node, &disk_key, path->slots[i]);
3723 memcpy(&root_item->drop_progress,
3724 &disk_key, sizeof(disk_key));
3725 root_item->drop_level = i;
3728 struct extent_buffer *parent;
3731 * this whole node is done, free our reference
3732 * on it and go up one level
3734 if (path->nodes[*level] == root->node)
3735 parent = path->nodes[*level];
3737 parent = path->nodes[*level + 1];
3739 root_owner = btrfs_header_owner(parent);
3740 root_gen = btrfs_header_generation(parent);
3742 clean_tree_block(trans, root, path->nodes[*level]);
3743 ret = btrfs_free_extent(trans, root,
3744 path->nodes[*level]->start,
3745 path->nodes[*level]->len,
3746 parent->start, root_owner,
3747 root_gen, *level, 1);
3749 if (path->locks[*level]) {
3750 btrfs_tree_unlock(path->nodes[*level]);
3751 path->locks[*level] = 0;
3753 free_extent_buffer(path->nodes[*level]);
3754 path->nodes[*level] = NULL;
3762 * drop the reference count on the tree rooted at 'snap'. This traverses
3763 * the tree freeing any blocks that have a ref count of zero after being
3766 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3772 struct btrfs_path *path;
3776 struct btrfs_root_item *root_item = &root->root_item;
3778 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3779 path = btrfs_alloc_path();
3782 level = btrfs_header_level(root->node);
3784 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3785 path->nodes[level] = root->node;
3786 extent_buffer_get(root->node);
3787 path->slots[level] = 0;
3789 struct btrfs_key key;
3790 struct btrfs_disk_key found_key;
3791 struct extent_buffer *node;
3793 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3794 level = root_item->drop_level;
3795 path->lowest_level = level;
3796 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3801 node = path->nodes[level];
3802 btrfs_node_key(node, &found_key, path->slots[level]);
3803 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3804 sizeof(found_key)));
3806 * unlock our path, this is safe because only this
3807 * function is allowed to delete this snapshot
3809 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3810 if (path->nodes[i] && path->locks[i]) {
3812 btrfs_tree_unlock(path->nodes[i]);
3817 unsigned long update;
3818 wret = walk_down_tree(trans, root, path, &level);
3824 wret = walk_up_tree(trans, root, path, &level,
3830 if (trans->transaction->in_commit ||
3831 trans->transaction->delayed_refs.flushing) {
3835 atomic_inc(&root->fs_info->throttle_gen);
3836 wake_up(&root->fs_info->transaction_throttle);
3837 for (update_count = 0; update_count < 16; update_count++) {
3838 update = trans->delayed_ref_updates;
3839 trans->delayed_ref_updates = 0;
3841 btrfs_run_delayed_refs(trans, root, update);
3846 for (i = 0; i <= orig_level; i++) {
3847 if (path->nodes[i]) {
3848 free_extent_buffer(path->nodes[i]);
3849 path->nodes[i] = NULL;
3853 btrfs_free_path(path);
3857 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3858 struct btrfs_root *root,
3859 struct extent_buffer *node,
3860 struct extent_buffer *parent)
3862 struct btrfs_path *path;
3868 path = btrfs_alloc_path();
3871 btrfs_assert_tree_locked(parent);
3872 parent_level = btrfs_header_level(parent);
3873 extent_buffer_get(parent);
3874 path->nodes[parent_level] = parent;
3875 path->slots[parent_level] = btrfs_header_nritems(parent);
3877 btrfs_assert_tree_locked(node);
3878 level = btrfs_header_level(node);
3879 extent_buffer_get(node);
3880 path->nodes[level] = node;
3881 path->slots[level] = 0;
3884 wret = walk_down_subtree(trans, root, path, &level);
3890 wret = walk_up_tree(trans, root, path, &level, parent_level);
3897 btrfs_free_path(path);
3901 static unsigned long calc_ra(unsigned long start, unsigned long last,
3904 return min(last, start + nr - 1);
3907 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3912 unsigned long first_index;
3913 unsigned long last_index;
3916 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3917 struct file_ra_state *ra;
3918 struct btrfs_ordered_extent *ordered;
3919 unsigned int total_read = 0;
3920 unsigned int total_dirty = 0;
3923 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3925 mutex_lock(&inode->i_mutex);
3926 first_index = start >> PAGE_CACHE_SHIFT;
3927 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3929 /* make sure the dirty trick played by the caller work */
3930 ret = invalidate_inode_pages2_range(inode->i_mapping,
3931 first_index, last_index);
3935 file_ra_state_init(ra, inode->i_mapping);
3937 for (i = first_index ; i <= last_index; i++) {
3938 if (total_read % ra->ra_pages == 0) {
3939 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3940 calc_ra(i, last_index, ra->ra_pages));
3944 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3946 page = grab_cache_page(inode->i_mapping, i);
3951 if (!PageUptodate(page)) {
3952 btrfs_readpage(NULL, page);
3954 if (!PageUptodate(page)) {
3956 page_cache_release(page);
3961 wait_on_page_writeback(page);
3963 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3964 page_end = page_start + PAGE_CACHE_SIZE - 1;
3965 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3967 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3969 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3971 page_cache_release(page);
3972 btrfs_start_ordered_extent(inode, ordered, 1);
3973 btrfs_put_ordered_extent(ordered);
3976 set_page_extent_mapped(page);
3978 if (i == first_index)
3979 set_extent_bits(io_tree, page_start, page_end,
3980 EXTENT_BOUNDARY, GFP_NOFS);
3981 btrfs_set_extent_delalloc(inode, page_start, page_end);
3983 set_page_dirty(page);
3986 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3988 page_cache_release(page);
3993 mutex_unlock(&inode->i_mutex);
3994 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3998 static noinline int relocate_data_extent(struct inode *reloc_inode,
3999 struct btrfs_key *extent_key,
4002 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4003 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4004 struct extent_map *em;
4005 u64 start = extent_key->objectid - offset;
4006 u64 end = start + extent_key->offset - 1;
4008 em = alloc_extent_map(GFP_NOFS);
4009 BUG_ON(!em || IS_ERR(em));
4012 em->len = extent_key->offset;
4013 em->block_len = extent_key->offset;
4014 em->block_start = extent_key->objectid;
4015 em->bdev = root->fs_info->fs_devices->latest_bdev;
4016 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4018 /* setup extent map to cheat btrfs_readpage */
4019 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4022 spin_lock(&em_tree->lock);
4023 ret = add_extent_mapping(em_tree, em);
4024 spin_unlock(&em_tree->lock);
4025 if (ret != -EEXIST) {
4026 free_extent_map(em);
4029 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4031 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4033 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4036 struct btrfs_ref_path {
4038 u64 nodes[BTRFS_MAX_LEVEL];
4040 u64 root_generation;
4047 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4048 u64 new_nodes[BTRFS_MAX_LEVEL];
4051 struct disk_extent {
4062 static int is_cowonly_root(u64 root_objectid)
4064 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4065 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4066 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4067 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4068 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4069 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4074 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4075 struct btrfs_root *extent_root,
4076 struct btrfs_ref_path *ref_path,
4079 struct extent_buffer *leaf;
4080 struct btrfs_path *path;
4081 struct btrfs_extent_ref *ref;
4082 struct btrfs_key key;
4083 struct btrfs_key found_key;
4089 path = btrfs_alloc_path();
4094 ref_path->lowest_level = -1;
4095 ref_path->current_level = -1;
4096 ref_path->shared_level = -1;
4100 level = ref_path->current_level - 1;
4101 while (level >= -1) {
4103 if (level < ref_path->lowest_level)
4107 bytenr = ref_path->nodes[level];
4109 bytenr = ref_path->extent_start;
4110 BUG_ON(bytenr == 0);
4112 parent = ref_path->nodes[level + 1];
4113 ref_path->nodes[level + 1] = 0;
4114 ref_path->current_level = level;
4115 BUG_ON(parent == 0);
4117 key.objectid = bytenr;
4118 key.offset = parent + 1;
4119 key.type = BTRFS_EXTENT_REF_KEY;
4121 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4126 leaf = path->nodes[0];
4127 nritems = btrfs_header_nritems(leaf);
4128 if (path->slots[0] >= nritems) {
4129 ret = btrfs_next_leaf(extent_root, path);
4134 leaf = path->nodes[0];
4137 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4138 if (found_key.objectid == bytenr &&
4139 found_key.type == BTRFS_EXTENT_REF_KEY) {
4140 if (level < ref_path->shared_level)
4141 ref_path->shared_level = level;
4146 btrfs_release_path(extent_root, path);
4149 /* reached lowest level */
4153 level = ref_path->current_level;
4154 while (level < BTRFS_MAX_LEVEL - 1) {
4158 bytenr = ref_path->nodes[level];
4160 bytenr = ref_path->extent_start;
4162 BUG_ON(bytenr == 0);
4164 key.objectid = bytenr;
4166 key.type = BTRFS_EXTENT_REF_KEY;
4168 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4172 leaf = path->nodes[0];
4173 nritems = btrfs_header_nritems(leaf);
4174 if (path->slots[0] >= nritems) {
4175 ret = btrfs_next_leaf(extent_root, path);
4179 /* the extent was freed by someone */
4180 if (ref_path->lowest_level == level)
4182 btrfs_release_path(extent_root, path);
4185 leaf = path->nodes[0];
4188 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4189 if (found_key.objectid != bytenr ||
4190 found_key.type != BTRFS_EXTENT_REF_KEY) {
4191 /* the extent was freed by someone */
4192 if (ref_path->lowest_level == level) {
4196 btrfs_release_path(extent_root, path);
4200 ref = btrfs_item_ptr(leaf, path->slots[0],
4201 struct btrfs_extent_ref);
4202 ref_objectid = btrfs_ref_objectid(leaf, ref);
4203 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4205 level = (int)ref_objectid;
4206 BUG_ON(level >= BTRFS_MAX_LEVEL);
4207 ref_path->lowest_level = level;
4208 ref_path->current_level = level;
4209 ref_path->nodes[level] = bytenr;
4211 WARN_ON(ref_objectid != level);
4214 WARN_ON(level != -1);
4218 if (ref_path->lowest_level == level) {
4219 ref_path->owner_objectid = ref_objectid;
4220 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4224 * the block is tree root or the block isn't in reference
4227 if (found_key.objectid == found_key.offset ||
4228 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4229 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4230 ref_path->root_generation =
4231 btrfs_ref_generation(leaf, ref);
4233 /* special reference from the tree log */
4234 ref_path->nodes[0] = found_key.offset;
4235 ref_path->current_level = 0;
4242 BUG_ON(ref_path->nodes[level] != 0);
4243 ref_path->nodes[level] = found_key.offset;
4244 ref_path->current_level = level;
4247 * the reference was created in the running transaction,
4248 * no need to continue walking up.
4250 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4251 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4252 ref_path->root_generation =
4253 btrfs_ref_generation(leaf, ref);
4258 btrfs_release_path(extent_root, path);
4261 /* reached max tree level, but no tree root found. */
4264 btrfs_free_path(path);
4268 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4269 struct btrfs_root *extent_root,
4270 struct btrfs_ref_path *ref_path,
4273 memset(ref_path, 0, sizeof(*ref_path));
4274 ref_path->extent_start = extent_start;
4276 return __next_ref_path(trans, extent_root, ref_path, 1);
4279 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4280 struct btrfs_root *extent_root,
4281 struct btrfs_ref_path *ref_path)
4283 return __next_ref_path(trans, extent_root, ref_path, 0);
4286 static noinline int get_new_locations(struct inode *reloc_inode,
4287 struct btrfs_key *extent_key,
4288 u64 offset, int no_fragment,
4289 struct disk_extent **extents,
4292 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4293 struct btrfs_path *path;
4294 struct btrfs_file_extent_item *fi;
4295 struct extent_buffer *leaf;
4296 struct disk_extent *exts = *extents;
4297 struct btrfs_key found_key;
4302 int max = *nr_extents;
4305 WARN_ON(!no_fragment && *extents);
4308 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4313 path = btrfs_alloc_path();
4316 cur_pos = extent_key->objectid - offset;
4317 last_byte = extent_key->objectid + extent_key->offset;
4318 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4328 leaf = path->nodes[0];
4329 nritems = btrfs_header_nritems(leaf);
4330 if (path->slots[0] >= nritems) {
4331 ret = btrfs_next_leaf(root, path);
4336 leaf = path->nodes[0];
4339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4340 if (found_key.offset != cur_pos ||
4341 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4342 found_key.objectid != reloc_inode->i_ino)
4345 fi = btrfs_item_ptr(leaf, path->slots[0],
4346 struct btrfs_file_extent_item);
4347 if (btrfs_file_extent_type(leaf, fi) !=
4348 BTRFS_FILE_EXTENT_REG ||
4349 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4353 struct disk_extent *old = exts;
4355 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4356 memcpy(exts, old, sizeof(*exts) * nr);
4357 if (old != *extents)
4361 exts[nr].disk_bytenr =
4362 btrfs_file_extent_disk_bytenr(leaf, fi);
4363 exts[nr].disk_num_bytes =
4364 btrfs_file_extent_disk_num_bytes(leaf, fi);
4365 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4366 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4367 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4368 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4369 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4370 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4372 BUG_ON(exts[nr].offset > 0);
4373 BUG_ON(exts[nr].compression || exts[nr].encryption);
4374 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4376 cur_pos += exts[nr].num_bytes;
4379 if (cur_pos + offset >= last_byte)
4389 BUG_ON(cur_pos + offset > last_byte);
4390 if (cur_pos + offset < last_byte) {
4396 btrfs_free_path(path);
4398 if (exts != *extents)
4407 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4408 struct btrfs_root *root,
4409 struct btrfs_path *path,
4410 struct btrfs_key *extent_key,
4411 struct btrfs_key *leaf_key,
4412 struct btrfs_ref_path *ref_path,
4413 struct disk_extent *new_extents,
4416 struct extent_buffer *leaf;
4417 struct btrfs_file_extent_item *fi;
4418 struct inode *inode = NULL;
4419 struct btrfs_key key;
4424 u64 search_end = (u64)-1;
4427 int extent_locked = 0;
4431 memcpy(&key, leaf_key, sizeof(key));
4432 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4433 if (key.objectid < ref_path->owner_objectid ||
4434 (key.objectid == ref_path->owner_objectid &&
4435 key.type < BTRFS_EXTENT_DATA_KEY)) {
4436 key.objectid = ref_path->owner_objectid;
4437 key.type = BTRFS_EXTENT_DATA_KEY;
4443 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4447 leaf = path->nodes[0];
4448 nritems = btrfs_header_nritems(leaf);
4450 if (extent_locked && ret > 0) {
4452 * the file extent item was modified by someone
4453 * before the extent got locked.
4455 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4456 lock_end, GFP_NOFS);
4460 if (path->slots[0] >= nritems) {
4461 if (++nr_scaned > 2)
4464 BUG_ON(extent_locked);
4465 ret = btrfs_next_leaf(root, path);
4470 leaf = path->nodes[0];
4471 nritems = btrfs_header_nritems(leaf);
4474 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4476 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4477 if ((key.objectid > ref_path->owner_objectid) ||
4478 (key.objectid == ref_path->owner_objectid &&
4479 key.type > BTRFS_EXTENT_DATA_KEY) ||
4480 key.offset >= search_end)
4484 if (inode && key.objectid != inode->i_ino) {
4485 BUG_ON(extent_locked);
4486 btrfs_release_path(root, path);
4487 mutex_unlock(&inode->i_mutex);
4493 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4498 fi = btrfs_item_ptr(leaf, path->slots[0],
4499 struct btrfs_file_extent_item);
4500 extent_type = btrfs_file_extent_type(leaf, fi);
4501 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4502 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4503 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4504 extent_key->objectid)) {
4510 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4511 ext_offset = btrfs_file_extent_offset(leaf, fi);
4513 if (search_end == (u64)-1) {
4514 search_end = key.offset - ext_offset +
4515 btrfs_file_extent_ram_bytes(leaf, fi);
4518 if (!extent_locked) {
4519 lock_start = key.offset;
4520 lock_end = lock_start + num_bytes - 1;
4522 if (lock_start > key.offset ||
4523 lock_end + 1 < key.offset + num_bytes) {
4524 unlock_extent(&BTRFS_I(inode)->io_tree,
4525 lock_start, lock_end, GFP_NOFS);
4531 btrfs_release_path(root, path);
4533 inode = btrfs_iget_locked(root->fs_info->sb,
4534 key.objectid, root);
4535 if (inode->i_state & I_NEW) {
4536 BTRFS_I(inode)->root = root;
4537 BTRFS_I(inode)->location.objectid =
4539 BTRFS_I(inode)->location.type =
4540 BTRFS_INODE_ITEM_KEY;
4541 BTRFS_I(inode)->location.offset = 0;
4542 btrfs_read_locked_inode(inode);
4543 unlock_new_inode(inode);
4546 * some code call btrfs_commit_transaction while
4547 * holding the i_mutex, so we can't use mutex_lock
4550 if (is_bad_inode(inode) ||
4551 !mutex_trylock(&inode->i_mutex)) {
4554 key.offset = (u64)-1;
4559 if (!extent_locked) {
4560 struct btrfs_ordered_extent *ordered;
4562 btrfs_release_path(root, path);
4564 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4565 lock_end, GFP_NOFS);
4566 ordered = btrfs_lookup_first_ordered_extent(inode,
4569 ordered->file_offset <= lock_end &&
4570 ordered->file_offset + ordered->len > lock_start) {
4571 unlock_extent(&BTRFS_I(inode)->io_tree,
4572 lock_start, lock_end, GFP_NOFS);
4573 btrfs_start_ordered_extent(inode, ordered, 1);
4574 btrfs_put_ordered_extent(ordered);
4575 key.offset += num_bytes;
4579 btrfs_put_ordered_extent(ordered);
4585 if (nr_extents == 1) {
4586 /* update extent pointer in place */
4587 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4588 new_extents[0].disk_bytenr);
4589 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4590 new_extents[0].disk_num_bytes);
4591 btrfs_mark_buffer_dirty(leaf);
4593 btrfs_drop_extent_cache(inode, key.offset,
4594 key.offset + num_bytes - 1, 0);
4596 ret = btrfs_inc_extent_ref(trans, root,
4597 new_extents[0].disk_bytenr,
4598 new_extents[0].disk_num_bytes,
4600 root->root_key.objectid,
4605 ret = btrfs_free_extent(trans, root,
4606 extent_key->objectid,
4609 btrfs_header_owner(leaf),
4610 btrfs_header_generation(leaf),
4614 btrfs_release_path(root, path);
4615 key.offset += num_bytes;
4623 * drop old extent pointer at first, then insert the
4624 * new pointers one bye one
4626 btrfs_release_path(root, path);
4627 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4628 key.offset + num_bytes,
4629 key.offset, &alloc_hint);
4632 for (i = 0; i < nr_extents; i++) {
4633 if (ext_offset >= new_extents[i].num_bytes) {
4634 ext_offset -= new_extents[i].num_bytes;
4637 extent_len = min(new_extents[i].num_bytes -
4638 ext_offset, num_bytes);
4640 ret = btrfs_insert_empty_item(trans, root,
4645 leaf = path->nodes[0];
4646 fi = btrfs_item_ptr(leaf, path->slots[0],
4647 struct btrfs_file_extent_item);
4648 btrfs_set_file_extent_generation(leaf, fi,
4650 btrfs_set_file_extent_type(leaf, fi,
4651 BTRFS_FILE_EXTENT_REG);
4652 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4653 new_extents[i].disk_bytenr);
4654 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4655 new_extents[i].disk_num_bytes);
4656 btrfs_set_file_extent_ram_bytes(leaf, fi,
4657 new_extents[i].ram_bytes);
4659 btrfs_set_file_extent_compression(leaf, fi,
4660 new_extents[i].compression);
4661 btrfs_set_file_extent_encryption(leaf, fi,
4662 new_extents[i].encryption);
4663 btrfs_set_file_extent_other_encoding(leaf, fi,
4664 new_extents[i].other_encoding);
4666 btrfs_set_file_extent_num_bytes(leaf, fi,
4668 ext_offset += new_extents[i].offset;
4669 btrfs_set_file_extent_offset(leaf, fi,
4671 btrfs_mark_buffer_dirty(leaf);
4673 btrfs_drop_extent_cache(inode, key.offset,
4674 key.offset + extent_len - 1, 0);
4676 ret = btrfs_inc_extent_ref(trans, root,
4677 new_extents[i].disk_bytenr,
4678 new_extents[i].disk_num_bytes,
4680 root->root_key.objectid,
4681 trans->transid, key.objectid);
4683 btrfs_release_path(root, path);
4685 inode_add_bytes(inode, extent_len);
4688 num_bytes -= extent_len;
4689 key.offset += extent_len;
4694 BUG_ON(i >= nr_extents);
4698 if (extent_locked) {
4699 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4700 lock_end, GFP_NOFS);
4704 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4705 key.offset >= search_end)
4712 btrfs_release_path(root, path);
4714 mutex_unlock(&inode->i_mutex);
4715 if (extent_locked) {
4716 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4717 lock_end, GFP_NOFS);
4724 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4725 struct btrfs_root *root,
4726 struct extent_buffer *buf, u64 orig_start)
4731 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4732 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4734 level = btrfs_header_level(buf);
4736 struct btrfs_leaf_ref *ref;
4737 struct btrfs_leaf_ref *orig_ref;
4739 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4743 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4745 btrfs_free_leaf_ref(root, orig_ref);
4749 ref->nritems = orig_ref->nritems;
4750 memcpy(ref->extents, orig_ref->extents,
4751 sizeof(ref->extents[0]) * ref->nritems);
4753 btrfs_free_leaf_ref(root, orig_ref);
4755 ref->root_gen = trans->transid;
4756 ref->bytenr = buf->start;
4757 ref->owner = btrfs_header_owner(buf);
4758 ref->generation = btrfs_header_generation(buf);
4760 ret = btrfs_add_leaf_ref(root, ref, 0);
4762 btrfs_free_leaf_ref(root, ref);
4767 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4768 struct extent_buffer *leaf,
4769 struct btrfs_block_group_cache *group,
4770 struct btrfs_root *target_root)
4772 struct btrfs_key key;
4773 struct inode *inode = NULL;
4774 struct btrfs_file_extent_item *fi;
4776 u64 skip_objectid = 0;
4780 nritems = btrfs_header_nritems(leaf);
4781 for (i = 0; i < nritems; i++) {
4782 btrfs_item_key_to_cpu(leaf, &key, i);
4783 if (key.objectid == skip_objectid ||
4784 key.type != BTRFS_EXTENT_DATA_KEY)
4786 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4787 if (btrfs_file_extent_type(leaf, fi) ==
4788 BTRFS_FILE_EXTENT_INLINE)
4790 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4792 if (!inode || inode->i_ino != key.objectid) {
4794 inode = btrfs_ilookup(target_root->fs_info->sb,
4795 key.objectid, target_root, 1);
4798 skip_objectid = key.objectid;
4801 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4803 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4804 key.offset + num_bytes - 1, GFP_NOFS);
4805 btrfs_drop_extent_cache(inode, key.offset,
4806 key.offset + num_bytes - 1, 1);
4807 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4808 key.offset + num_bytes - 1, GFP_NOFS);
4815 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4816 struct btrfs_root *root,
4817 struct extent_buffer *leaf,
4818 struct btrfs_block_group_cache *group,
4819 struct inode *reloc_inode)
4821 struct btrfs_key key;
4822 struct btrfs_key extent_key;
4823 struct btrfs_file_extent_item *fi;
4824 struct btrfs_leaf_ref *ref;
4825 struct disk_extent *new_extent;
4834 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4835 BUG_ON(!new_extent);
4837 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4841 nritems = btrfs_header_nritems(leaf);
4842 for (i = 0; i < nritems; i++) {
4843 btrfs_item_key_to_cpu(leaf, &key, i);
4844 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4846 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4847 if (btrfs_file_extent_type(leaf, fi) ==
4848 BTRFS_FILE_EXTENT_INLINE)
4850 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4851 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4856 if (bytenr >= group->key.objectid + group->key.offset ||
4857 bytenr + num_bytes <= group->key.objectid)
4860 extent_key.objectid = bytenr;
4861 extent_key.offset = num_bytes;
4862 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4864 ret = get_new_locations(reloc_inode, &extent_key,
4865 group->key.objectid, 1,
4866 &new_extent, &nr_extent);
4871 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4872 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4873 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4874 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4876 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4877 new_extent->disk_bytenr);
4878 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4879 new_extent->disk_num_bytes);
4880 btrfs_mark_buffer_dirty(leaf);
4882 ret = btrfs_inc_extent_ref(trans, root,
4883 new_extent->disk_bytenr,
4884 new_extent->disk_num_bytes,
4886 root->root_key.objectid,
4887 trans->transid, key.objectid);
4890 ret = btrfs_free_extent(trans, root,
4891 bytenr, num_bytes, leaf->start,
4892 btrfs_header_owner(leaf),
4893 btrfs_header_generation(leaf),
4899 BUG_ON(ext_index + 1 != ref->nritems);
4900 btrfs_free_leaf_ref(root, ref);
4904 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4905 struct btrfs_root *root)
4907 struct btrfs_root *reloc_root;
4910 if (root->reloc_root) {
4911 reloc_root = root->reloc_root;
4912 root->reloc_root = NULL;
4913 list_add(&reloc_root->dead_list,
4914 &root->fs_info->dead_reloc_roots);
4916 btrfs_set_root_bytenr(&reloc_root->root_item,
4917 reloc_root->node->start);
4918 btrfs_set_root_level(&root->root_item,
4919 btrfs_header_level(reloc_root->node));
4920 memset(&reloc_root->root_item.drop_progress, 0,
4921 sizeof(struct btrfs_disk_key));
4922 reloc_root->root_item.drop_level = 0;
4924 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4925 &reloc_root->root_key,
4926 &reloc_root->root_item);
4932 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4934 struct btrfs_trans_handle *trans;
4935 struct btrfs_root *reloc_root;
4936 struct btrfs_root *prev_root = NULL;
4937 struct list_head dead_roots;
4941 INIT_LIST_HEAD(&dead_roots);
4942 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4944 while (!list_empty(&dead_roots)) {
4945 reloc_root = list_entry(dead_roots.prev,
4946 struct btrfs_root, dead_list);
4947 list_del_init(&reloc_root->dead_list);
4949 BUG_ON(reloc_root->commit_root != NULL);
4951 trans = btrfs_join_transaction(root, 1);
4954 mutex_lock(&root->fs_info->drop_mutex);
4955 ret = btrfs_drop_snapshot(trans, reloc_root);
4958 mutex_unlock(&root->fs_info->drop_mutex);
4960 nr = trans->blocks_used;
4961 ret = btrfs_end_transaction(trans, root);
4963 btrfs_btree_balance_dirty(root, nr);
4966 free_extent_buffer(reloc_root->node);
4968 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4969 &reloc_root->root_key);
4971 mutex_unlock(&root->fs_info->drop_mutex);
4973 nr = trans->blocks_used;
4974 ret = btrfs_end_transaction(trans, root);
4976 btrfs_btree_balance_dirty(root, nr);
4979 prev_root = reloc_root;
4982 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4988 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4990 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4994 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4996 struct btrfs_root *reloc_root;
4997 struct btrfs_trans_handle *trans;
4998 struct btrfs_key location;
5002 mutex_lock(&root->fs_info->tree_reloc_mutex);
5003 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5005 found = !list_empty(&root->fs_info->dead_reloc_roots);
5006 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5009 trans = btrfs_start_transaction(root, 1);
5011 ret = btrfs_commit_transaction(trans, root);
5015 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5016 location.offset = (u64)-1;
5017 location.type = BTRFS_ROOT_ITEM_KEY;
5019 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5020 BUG_ON(!reloc_root);
5021 btrfs_orphan_cleanup(reloc_root);
5025 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5026 struct btrfs_root *root)
5028 struct btrfs_root *reloc_root;
5029 struct extent_buffer *eb;
5030 struct btrfs_root_item *root_item;
5031 struct btrfs_key root_key;
5034 BUG_ON(!root->ref_cows);
5035 if (root->reloc_root)
5038 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5041 ret = btrfs_copy_root(trans, root, root->commit_root,
5042 &eb, BTRFS_TREE_RELOC_OBJECTID);
5045 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5046 root_key.offset = root->root_key.objectid;
5047 root_key.type = BTRFS_ROOT_ITEM_KEY;
5049 memcpy(root_item, &root->root_item, sizeof(root_item));
5050 btrfs_set_root_refs(root_item, 0);
5051 btrfs_set_root_bytenr(root_item, eb->start);
5052 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5053 btrfs_set_root_generation(root_item, trans->transid);
5055 btrfs_tree_unlock(eb);
5056 free_extent_buffer(eb);
5058 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5059 &root_key, root_item);
5063 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5065 BUG_ON(!reloc_root);
5066 reloc_root->last_trans = trans->transid;
5067 reloc_root->commit_root = NULL;
5068 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5070 root->reloc_root = reloc_root;
5075 * Core function of space balance.
5077 * The idea is using reloc trees to relocate tree blocks in reference
5078 * counted roots. There is one reloc tree for each subvol, and all
5079 * reloc trees share same root key objectid. Reloc trees are snapshots
5080 * of the latest committed roots of subvols (root->commit_root).
5082 * To relocate a tree block referenced by a subvol, there are two steps.
5083 * COW the block through subvol's reloc tree, then update block pointer
5084 * in the subvol to point to the new block. Since all reloc trees share
5085 * same root key objectid, doing special handing for tree blocks owned
5086 * by them is easy. Once a tree block has been COWed in one reloc tree,
5087 * we can use the resulting new block directly when the same block is
5088 * required to COW again through other reloc trees. By this way, relocated
5089 * tree blocks are shared between reloc trees, so they are also shared
5092 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5093 struct btrfs_root *root,
5094 struct btrfs_path *path,
5095 struct btrfs_key *first_key,
5096 struct btrfs_ref_path *ref_path,
5097 struct btrfs_block_group_cache *group,
5098 struct inode *reloc_inode)
5100 struct btrfs_root *reloc_root;
5101 struct extent_buffer *eb = NULL;
5102 struct btrfs_key *keys;
5106 int lowest_level = 0;
5109 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5110 lowest_level = ref_path->owner_objectid;
5112 if (!root->ref_cows) {
5113 path->lowest_level = lowest_level;
5114 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5116 path->lowest_level = 0;
5117 btrfs_release_path(root, path);
5121 mutex_lock(&root->fs_info->tree_reloc_mutex);
5122 ret = init_reloc_tree(trans, root);
5124 reloc_root = root->reloc_root;
5126 shared_level = ref_path->shared_level;
5127 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5129 keys = ref_path->node_keys;
5130 nodes = ref_path->new_nodes;
5131 memset(&keys[shared_level + 1], 0,
5132 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5133 memset(&nodes[shared_level + 1], 0,
5134 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5136 if (nodes[lowest_level] == 0) {
5137 path->lowest_level = lowest_level;
5138 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5141 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5142 eb = path->nodes[level];
5143 if (!eb || eb == reloc_root->node)
5145 nodes[level] = eb->start;
5147 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5149 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5152 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5153 eb = path->nodes[0];
5154 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5155 group, reloc_inode);
5158 btrfs_release_path(reloc_root, path);
5160 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5166 * replace tree blocks in the fs tree with tree blocks in
5169 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5172 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5173 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5176 extent_buffer_get(path->nodes[0]);
5177 eb = path->nodes[0];
5178 btrfs_release_path(reloc_root, path);
5179 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5181 free_extent_buffer(eb);
5184 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5185 path->lowest_level = 0;
5189 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5190 struct btrfs_root *root,
5191 struct btrfs_path *path,
5192 struct btrfs_key *first_key,
5193 struct btrfs_ref_path *ref_path)
5197 ret = relocate_one_path(trans, root, path, first_key,
5198 ref_path, NULL, NULL);
5204 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5205 struct btrfs_root *extent_root,
5206 struct btrfs_path *path,
5207 struct btrfs_key *extent_key)
5211 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5214 ret = btrfs_del_item(trans, extent_root, path);
5216 btrfs_release_path(extent_root, path);
5220 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5221 struct btrfs_ref_path *ref_path)
5223 struct btrfs_key root_key;
5225 root_key.objectid = ref_path->root_objectid;
5226 root_key.type = BTRFS_ROOT_ITEM_KEY;
5227 if (is_cowonly_root(ref_path->root_objectid))
5228 root_key.offset = 0;
5230 root_key.offset = (u64)-1;
5232 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5235 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5236 struct btrfs_path *path,
5237 struct btrfs_key *extent_key,
5238 struct btrfs_block_group_cache *group,
5239 struct inode *reloc_inode, int pass)
5241 struct btrfs_trans_handle *trans;
5242 struct btrfs_root *found_root;
5243 struct btrfs_ref_path *ref_path = NULL;
5244 struct disk_extent *new_extents = NULL;
5249 struct btrfs_key first_key;
5253 trans = btrfs_start_transaction(extent_root, 1);
5256 if (extent_key->objectid == 0) {
5257 ret = del_extent_zero(trans, extent_root, path, extent_key);
5261 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5267 for (loops = 0; ; loops++) {
5269 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5270 extent_key->objectid);
5272 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5279 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5280 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5283 found_root = read_ref_root(extent_root->fs_info, ref_path);
5284 BUG_ON(!found_root);
5286 * for reference counted tree, only process reference paths
5287 * rooted at the latest committed root.
5289 if (found_root->ref_cows &&
5290 ref_path->root_generation != found_root->root_key.offset)
5293 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5296 * copy data extents to new locations
5298 u64 group_start = group->key.objectid;
5299 ret = relocate_data_extent(reloc_inode,
5308 level = ref_path->owner_objectid;
5311 if (prev_block != ref_path->nodes[level]) {
5312 struct extent_buffer *eb;
5313 u64 block_start = ref_path->nodes[level];
5314 u64 block_size = btrfs_level_size(found_root, level);
5316 eb = read_tree_block(found_root, block_start,
5318 btrfs_tree_lock(eb);
5319 BUG_ON(level != btrfs_header_level(eb));
5322 btrfs_item_key_to_cpu(eb, &first_key, 0);
5324 btrfs_node_key_to_cpu(eb, &first_key, 0);
5326 btrfs_tree_unlock(eb);
5327 free_extent_buffer(eb);
5328 prev_block = block_start;
5331 mutex_lock(&extent_root->fs_info->trans_mutex);
5332 btrfs_record_root_in_trans(found_root);
5333 mutex_unlock(&extent_root->fs_info->trans_mutex);
5334 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5336 * try to update data extent references while
5337 * keeping metadata shared between snapshots.
5340 ret = relocate_one_path(trans, found_root,
5341 path, &first_key, ref_path,
5342 group, reloc_inode);
5348 * use fallback method to process the remaining
5352 u64 group_start = group->key.objectid;
5353 new_extents = kmalloc(sizeof(*new_extents),
5356 ret = get_new_locations(reloc_inode,
5364 ret = replace_one_extent(trans, found_root,
5366 &first_key, ref_path,
5367 new_extents, nr_extents);
5369 ret = relocate_tree_block(trans, found_root, path,
5370 &first_key, ref_path);
5377 btrfs_end_transaction(trans, extent_root);
5383 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5386 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5387 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5389 num_devices = root->fs_info->fs_devices->rw_devices;
5390 if (num_devices == 1) {
5391 stripped |= BTRFS_BLOCK_GROUP_DUP;
5392 stripped = flags & ~stripped;
5394 /* turn raid0 into single device chunks */
5395 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5398 /* turn mirroring into duplication */
5399 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5400 BTRFS_BLOCK_GROUP_RAID10))
5401 return stripped | BTRFS_BLOCK_GROUP_DUP;
5404 /* they already had raid on here, just return */
5405 if (flags & stripped)
5408 stripped |= BTRFS_BLOCK_GROUP_DUP;
5409 stripped = flags & ~stripped;
5411 /* switch duplicated blocks with raid1 */
5412 if (flags & BTRFS_BLOCK_GROUP_DUP)
5413 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5415 /* turn single device chunks into raid0 */
5416 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5421 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5422 struct btrfs_block_group_cache *shrink_block_group,
5425 struct btrfs_trans_handle *trans;
5426 u64 new_alloc_flags;
5429 spin_lock(&shrink_block_group->lock);
5430 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5431 spin_unlock(&shrink_block_group->lock);
5433 trans = btrfs_start_transaction(root, 1);
5434 spin_lock(&shrink_block_group->lock);
5436 new_alloc_flags = update_block_group_flags(root,
5437 shrink_block_group->flags);
5438 if (new_alloc_flags != shrink_block_group->flags) {
5440 btrfs_block_group_used(&shrink_block_group->item);
5442 calc = shrink_block_group->key.offset;
5444 spin_unlock(&shrink_block_group->lock);
5446 do_chunk_alloc(trans, root->fs_info->extent_root,
5447 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5449 btrfs_end_transaction(trans, root);
5451 spin_unlock(&shrink_block_group->lock);
5455 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5456 struct btrfs_root *root,
5457 u64 objectid, u64 size)
5459 struct btrfs_path *path;
5460 struct btrfs_inode_item *item;
5461 struct extent_buffer *leaf;
5464 path = btrfs_alloc_path();
5468 path->leave_spinning = 1;
5469 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5473 leaf = path->nodes[0];
5474 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5475 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5476 btrfs_set_inode_generation(leaf, item, 1);
5477 btrfs_set_inode_size(leaf, item, size);
5478 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5479 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5480 btrfs_mark_buffer_dirty(leaf);
5481 btrfs_release_path(root, path);
5483 btrfs_free_path(path);
5487 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5488 struct btrfs_block_group_cache *group)
5490 struct inode *inode = NULL;
5491 struct btrfs_trans_handle *trans;
5492 struct btrfs_root *root;
5493 struct btrfs_key root_key;
5494 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5497 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5498 root_key.type = BTRFS_ROOT_ITEM_KEY;
5499 root_key.offset = (u64)-1;
5500 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5502 return ERR_CAST(root);
5504 trans = btrfs_start_transaction(root, 1);
5507 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5511 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5514 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5515 group->key.offset, 0, group->key.offset,
5519 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5520 if (inode->i_state & I_NEW) {
5521 BTRFS_I(inode)->root = root;
5522 BTRFS_I(inode)->location.objectid = objectid;
5523 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5524 BTRFS_I(inode)->location.offset = 0;
5525 btrfs_read_locked_inode(inode);
5526 unlock_new_inode(inode);
5527 BUG_ON(is_bad_inode(inode));
5531 BTRFS_I(inode)->index_cnt = group->key.objectid;
5533 err = btrfs_orphan_add(trans, inode);
5535 btrfs_end_transaction(trans, root);
5539 inode = ERR_PTR(err);
5544 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5547 struct btrfs_ordered_sum *sums;
5548 struct btrfs_sector_sum *sector_sum;
5549 struct btrfs_ordered_extent *ordered;
5550 struct btrfs_root *root = BTRFS_I(inode)->root;
5551 struct list_head list;
5556 INIT_LIST_HEAD(&list);
5558 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5559 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5561 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5562 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5563 disk_bytenr + len - 1, &list);
5565 while (!list_empty(&list)) {
5566 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5567 list_del_init(&sums->list);
5569 sector_sum = sums->sums;
5570 sums->bytenr = ordered->start;
5573 while (offset < sums->len) {
5574 sector_sum->bytenr += ordered->start - disk_bytenr;
5576 offset += root->sectorsize;
5579 btrfs_add_ordered_sum(inode, ordered, sums);
5581 btrfs_put_ordered_extent(ordered);
5585 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5587 struct btrfs_trans_handle *trans;
5588 struct btrfs_path *path;
5589 struct btrfs_fs_info *info = root->fs_info;
5590 struct extent_buffer *leaf;
5591 struct inode *reloc_inode;
5592 struct btrfs_block_group_cache *block_group;
5593 struct btrfs_key key;
5602 root = root->fs_info->extent_root;
5604 block_group = btrfs_lookup_block_group(info, group_start);
5605 BUG_ON(!block_group);
5607 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5608 (unsigned long long)block_group->key.objectid,
5609 (unsigned long long)block_group->flags);
5611 path = btrfs_alloc_path();
5614 reloc_inode = create_reloc_inode(info, block_group);
5615 BUG_ON(IS_ERR(reloc_inode));
5617 __alloc_chunk_for_shrink(root, block_group, 1);
5618 set_block_group_readonly(block_group);
5620 btrfs_start_delalloc_inodes(info->tree_root);
5621 btrfs_wait_ordered_extents(info->tree_root, 0);
5626 key.objectid = block_group->key.objectid;
5629 cur_byte = key.objectid;
5631 trans = btrfs_start_transaction(info->tree_root, 1);
5632 btrfs_commit_transaction(trans, info->tree_root);
5634 mutex_lock(&root->fs_info->cleaner_mutex);
5635 btrfs_clean_old_snapshots(info->tree_root);
5636 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5637 mutex_unlock(&root->fs_info->cleaner_mutex);
5639 trans = btrfs_start_transaction(info->tree_root, 1);
5640 btrfs_commit_transaction(trans, info->tree_root);
5643 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5647 leaf = path->nodes[0];
5648 nritems = btrfs_header_nritems(leaf);
5649 if (path->slots[0] >= nritems) {
5650 ret = btrfs_next_leaf(root, path);
5657 leaf = path->nodes[0];
5658 nritems = btrfs_header_nritems(leaf);
5661 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5663 if (key.objectid >= block_group->key.objectid +
5664 block_group->key.offset)
5667 if (progress && need_resched()) {
5668 btrfs_release_path(root, path);
5675 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5676 key.objectid + key.offset <= cur_byte) {
5682 cur_byte = key.objectid + key.offset;
5683 btrfs_release_path(root, path);
5685 __alloc_chunk_for_shrink(root, block_group, 0);
5686 ret = relocate_one_extent(root, path, &key, block_group,
5692 key.objectid = cur_byte;
5697 btrfs_release_path(root, path);
5700 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5701 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5704 if (total_found > 0) {
5705 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5706 (unsigned long long)total_found, pass);
5708 if (total_found == skipped && pass > 2) {
5710 reloc_inode = create_reloc_inode(info, block_group);
5716 /* delete reloc_inode */
5719 /* unpin extents in this range */
5720 trans = btrfs_start_transaction(info->tree_root, 1);
5721 btrfs_commit_transaction(trans, info->tree_root);
5723 spin_lock(&block_group->lock);
5724 WARN_ON(block_group->pinned > 0);
5725 WARN_ON(block_group->reserved > 0);
5726 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5727 spin_unlock(&block_group->lock);
5728 put_block_group(block_group);
5731 btrfs_free_path(path);
5735 static int find_first_block_group(struct btrfs_root *root,
5736 struct btrfs_path *path, struct btrfs_key *key)
5739 struct btrfs_key found_key;
5740 struct extent_buffer *leaf;
5743 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5748 slot = path->slots[0];
5749 leaf = path->nodes[0];
5750 if (slot >= btrfs_header_nritems(leaf)) {
5751 ret = btrfs_next_leaf(root, path);
5758 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5760 if (found_key.objectid >= key->objectid &&
5761 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5772 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5774 struct btrfs_block_group_cache *block_group;
5775 struct btrfs_space_info *space_info;
5778 spin_lock(&info->block_group_cache_lock);
5779 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5780 block_group = rb_entry(n, struct btrfs_block_group_cache,
5782 rb_erase(&block_group->cache_node,
5783 &info->block_group_cache_tree);
5784 spin_unlock(&info->block_group_cache_lock);
5786 btrfs_remove_free_space_cache(block_group);
5787 down_write(&block_group->space_info->groups_sem);
5788 list_del(&block_group->list);
5789 up_write(&block_group->space_info->groups_sem);
5791 WARN_ON(atomic_read(&block_group->count) != 1);
5794 spin_lock(&info->block_group_cache_lock);
5796 spin_unlock(&info->block_group_cache_lock);
5798 /* now that all the block groups are freed, go through and
5799 * free all the space_info structs. This is only called during
5800 * the final stages of unmount, and so we know nobody is
5801 * using them. We call synchronize_rcu() once before we start,
5802 * just to be on the safe side.
5806 while(!list_empty(&info->space_info)) {
5807 space_info = list_entry(info->space_info.next,
5808 struct btrfs_space_info,
5811 list_del(&space_info->list);
5817 int btrfs_read_block_groups(struct btrfs_root *root)
5819 struct btrfs_path *path;
5821 struct btrfs_block_group_cache *cache;
5822 struct btrfs_fs_info *info = root->fs_info;
5823 struct btrfs_space_info *space_info;
5824 struct btrfs_key key;
5825 struct btrfs_key found_key;
5826 struct extent_buffer *leaf;
5828 root = info->extent_root;
5831 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5832 path = btrfs_alloc_path();
5837 ret = find_first_block_group(root, path, &key);
5845 leaf = path->nodes[0];
5846 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5847 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5853 atomic_set(&cache->count, 1);
5854 spin_lock_init(&cache->lock);
5855 mutex_init(&cache->alloc_mutex);
5856 mutex_init(&cache->cache_mutex);
5857 INIT_LIST_HEAD(&cache->list);
5858 read_extent_buffer(leaf, &cache->item,
5859 btrfs_item_ptr_offset(leaf, path->slots[0]),
5860 sizeof(cache->item));
5861 memcpy(&cache->key, &found_key, sizeof(found_key));
5863 key.objectid = found_key.objectid + found_key.offset;
5864 btrfs_release_path(root, path);
5865 cache->flags = btrfs_block_group_flags(&cache->item);
5867 ret = update_space_info(info, cache->flags, found_key.offset,
5868 btrfs_block_group_used(&cache->item),
5871 cache->space_info = space_info;
5872 down_write(&space_info->groups_sem);
5873 list_add_tail(&cache->list, &space_info->block_groups);
5874 up_write(&space_info->groups_sem);
5876 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5879 set_avail_alloc_bits(root->fs_info, cache->flags);
5880 if (btrfs_chunk_readonly(root, cache->key.objectid))
5881 set_block_group_readonly(cache);
5885 btrfs_free_path(path);
5889 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5890 struct btrfs_root *root, u64 bytes_used,
5891 u64 type, u64 chunk_objectid, u64 chunk_offset,
5895 struct btrfs_root *extent_root;
5896 struct btrfs_block_group_cache *cache;
5898 extent_root = root->fs_info->extent_root;
5900 root->fs_info->last_trans_log_full_commit = trans->transid;
5902 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5906 cache->key.objectid = chunk_offset;
5907 cache->key.offset = size;
5908 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5909 atomic_set(&cache->count, 1);
5910 spin_lock_init(&cache->lock);
5911 mutex_init(&cache->alloc_mutex);
5912 mutex_init(&cache->cache_mutex);
5913 INIT_LIST_HEAD(&cache->list);
5915 btrfs_set_block_group_used(&cache->item, bytes_used);
5916 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5917 cache->flags = type;
5918 btrfs_set_block_group_flags(&cache->item, type);
5920 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5921 &cache->space_info);
5923 down_write(&cache->space_info->groups_sem);
5924 list_add_tail(&cache->list, &cache->space_info->block_groups);
5925 up_write(&cache->space_info->groups_sem);
5927 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5930 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5931 sizeof(cache->item));
5934 set_avail_alloc_bits(extent_root->fs_info, type);
5939 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5940 struct btrfs_root *root, u64 group_start)
5942 struct btrfs_path *path;
5943 struct btrfs_block_group_cache *block_group;
5944 struct btrfs_key key;
5947 root = root->fs_info->extent_root;
5949 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5950 BUG_ON(!block_group);
5951 BUG_ON(!block_group->ro);
5953 memcpy(&key, &block_group->key, sizeof(key));
5955 path = btrfs_alloc_path();
5958 spin_lock(&root->fs_info->block_group_cache_lock);
5959 rb_erase(&block_group->cache_node,
5960 &root->fs_info->block_group_cache_tree);
5961 spin_unlock(&root->fs_info->block_group_cache_lock);
5962 btrfs_remove_free_space_cache(block_group);
5963 down_write(&block_group->space_info->groups_sem);
5964 list_del(&block_group->list);
5965 up_write(&block_group->space_info->groups_sem);
5967 spin_lock(&block_group->space_info->lock);
5968 block_group->space_info->total_bytes -= block_group->key.offset;
5969 block_group->space_info->bytes_readonly -= block_group->key.offset;
5970 spin_unlock(&block_group->space_info->lock);
5971 block_group->space_info->full = 0;
5973 put_block_group(block_group);
5974 put_block_group(block_group);
5976 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5982 ret = btrfs_del_item(trans, root, path);
5984 btrfs_free_path(path);