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>
27 #include "print-tree.h"
28 #include "transaction.h"
31 #include "ref-cache.h"
33 #define PENDING_EXTENT_INSERT 0
34 #define PENDING_EXTENT_DELETE 1
35 #define PENDING_BACKREF_UPDATE 2
37 struct pending_extent_op {
46 struct list_head list;
50 static int finish_current_insert(struct btrfs_trans_handle *trans,
51 struct btrfs_root *extent_root, int all);
52 static int del_pending_extents(struct btrfs_trans_handle *trans,
53 struct btrfs_root *extent_root, int all);
54 static int pin_down_bytes(struct btrfs_trans_handle *trans,
55 struct btrfs_root *root,
56 u64 bytenr, u64 num_bytes, int is_data);
57 static int update_block_group(struct btrfs_trans_handle *trans,
58 struct btrfs_root *root,
59 u64 bytenr, u64 num_bytes, int alloc,
62 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
64 return (cache->flags & bits) == bits;
68 * this adds the block group to the fs_info rb tree for the block group
71 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
72 struct btrfs_block_group_cache *block_group)
75 struct rb_node *parent = NULL;
76 struct btrfs_block_group_cache *cache;
78 spin_lock(&info->block_group_cache_lock);
79 p = &info->block_group_cache_tree.rb_node;
83 cache = rb_entry(parent, struct btrfs_block_group_cache,
85 if (block_group->key.objectid < cache->key.objectid) {
87 } else if (block_group->key.objectid > cache->key.objectid) {
90 spin_unlock(&info->block_group_cache_lock);
95 rb_link_node(&block_group->cache_node, parent, p);
96 rb_insert_color(&block_group->cache_node,
97 &info->block_group_cache_tree);
98 spin_unlock(&info->block_group_cache_lock);
104 * This will return the block group at or after bytenr if contains is 0, else
105 * it will return the block group that contains the bytenr
107 static struct btrfs_block_group_cache *
108 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
111 struct btrfs_block_group_cache *cache, *ret = NULL;
115 spin_lock(&info->block_group_cache_lock);
116 n = info->block_group_cache_tree.rb_node;
119 cache = rb_entry(n, struct btrfs_block_group_cache,
121 end = cache->key.objectid + cache->key.offset - 1;
122 start = cache->key.objectid;
124 if (bytenr < start) {
125 if (!contains && (!ret || start < ret->key.objectid))
128 } else if (bytenr > start) {
129 if (contains && bytenr <= end) {
140 atomic_inc(&ret->count);
141 spin_unlock(&info->block_group_cache_lock);
147 * this is only called by cache_block_group, since we could have freed extents
148 * we need to check the pinned_extents for any extents that can't be used yet
149 * since their free space will be released as soon as the transaction commits.
151 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
152 struct btrfs_fs_info *info, u64 start, u64 end)
154 u64 extent_start, extent_end, size;
157 mutex_lock(&info->pinned_mutex);
158 while (start < end) {
159 ret = find_first_extent_bit(&info->pinned_extents, start,
160 &extent_start, &extent_end,
165 if (extent_start == start) {
166 start = extent_end + 1;
167 } else if (extent_start > start && extent_start < end) {
168 size = extent_start - start;
169 ret = btrfs_add_free_space(block_group, start,
172 start = extent_end + 1;
180 ret = btrfs_add_free_space(block_group, start, size);
183 mutex_unlock(&info->pinned_mutex);
188 static int remove_sb_from_cache(struct btrfs_root *root,
189 struct btrfs_block_group_cache *cache)
196 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
197 bytenr = btrfs_sb_offset(i);
198 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
199 cache->key.objectid, bytenr, 0,
200 &logical, &nr, &stripe_len);
203 btrfs_remove_free_space(cache, logical[nr],
211 static int cache_block_group(struct btrfs_root *root,
212 struct btrfs_block_group_cache *block_group)
214 struct btrfs_path *path;
216 struct btrfs_key key;
217 struct extent_buffer *leaf;
224 root = root->fs_info->extent_root;
226 if (block_group->cached)
229 path = btrfs_alloc_path();
235 * we get into deadlocks with paths held by callers of this function.
236 * since the alloc_mutex is protecting things right now, just
237 * skip the locking here
239 path->skip_locking = 1;
240 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
243 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
244 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
249 leaf = path->nodes[0];
250 slot = path->slots[0];
251 if (slot >= btrfs_header_nritems(leaf)) {
252 ret = btrfs_next_leaf(root, path);
260 btrfs_item_key_to_cpu(leaf, &key, slot);
261 if (key.objectid < block_group->key.objectid)
264 if (key.objectid >= block_group->key.objectid +
265 block_group->key.offset)
268 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
269 add_new_free_space(block_group, root->fs_info, last,
272 last = key.objectid + key.offset;
278 add_new_free_space(block_group, root->fs_info, last,
279 block_group->key.objectid +
280 block_group->key.offset);
282 remove_sb_from_cache(root, block_group);
283 block_group->cached = 1;
286 btrfs_free_path(path);
291 * return the block group that starts at or after bytenr
293 static struct btrfs_block_group_cache *
294 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
296 struct btrfs_block_group_cache *cache;
298 cache = block_group_cache_tree_search(info, bytenr, 0);
304 * return the block group that contains teh given bytenr
306 struct btrfs_block_group_cache *btrfs_lookup_block_group(
307 struct btrfs_fs_info *info,
310 struct btrfs_block_group_cache *cache;
312 cache = block_group_cache_tree_search(info, bytenr, 1);
317 static inline void put_block_group(struct btrfs_block_group_cache *cache)
319 if (atomic_dec_and_test(&cache->count))
323 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
326 struct list_head *head = &info->space_info;
327 struct btrfs_space_info *found;
328 list_for_each_entry(found, head, list) {
329 if (found->flags == flags)
335 static u64 div_factor(u64 num, int factor)
344 u64 btrfs_find_block_group(struct btrfs_root *root,
345 u64 search_start, u64 search_hint, int owner)
347 struct btrfs_block_group_cache *cache;
349 u64 last = max(search_hint, search_start);
356 cache = btrfs_lookup_first_block_group(root->fs_info, last);
360 spin_lock(&cache->lock);
361 last = cache->key.objectid + cache->key.offset;
362 used = btrfs_block_group_used(&cache->item);
364 if ((full_search || !cache->ro) &&
365 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
366 if (used + cache->pinned + cache->reserved <
367 div_factor(cache->key.offset, factor)) {
368 group_start = cache->key.objectid;
369 spin_unlock(&cache->lock);
370 put_block_group(cache);
374 spin_unlock(&cache->lock);
375 put_block_group(cache);
383 if (!full_search && factor < 10) {
393 /* simple helper to search for an existing extent at a given offset */
394 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
397 struct btrfs_key key;
398 struct btrfs_path *path;
400 path = btrfs_alloc_path();
402 key.objectid = start;
404 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
405 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
407 btrfs_free_path(path);
412 * Back reference rules. Back refs have three main goals:
414 * 1) differentiate between all holders of references to an extent so that
415 * when a reference is dropped we can make sure it was a valid reference
416 * before freeing the extent.
418 * 2) Provide enough information to quickly find the holders of an extent
419 * if we notice a given block is corrupted or bad.
421 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
422 * maintenance. This is actually the same as #2, but with a slightly
423 * different use case.
425 * File extents can be referenced by:
427 * - multiple snapshots, subvolumes, or different generations in one subvol
428 * - different files inside a single subvolume
429 * - different offsets inside a file (bookend extents in file.c)
431 * The extent ref structure has fields for:
433 * - Objectid of the subvolume root
434 * - Generation number of the tree holding the reference
435 * - objectid of the file holding the reference
436 * - number of references holding by parent node (alway 1 for tree blocks)
438 * Btree leaf may hold multiple references to a file extent. In most cases,
439 * these references are from same file and the corresponding offsets inside
440 * the file are close together.
442 * When a file extent is allocated the fields are filled in:
443 * (root_key.objectid, trans->transid, inode objectid, 1)
445 * When a leaf is cow'd new references are added for every file extent found
446 * in the leaf. It looks similar to the create case, but trans->transid will
447 * be different when the block is cow'd.
449 * (root_key.objectid, trans->transid, inode objectid,
450 * number of references in the leaf)
452 * When a file extent is removed either during snapshot deletion or
453 * file truncation, we find the corresponding back reference and check
454 * the following fields:
456 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
459 * Btree extents can be referenced by:
461 * - Different subvolumes
462 * - Different generations of the same subvolume
464 * When a tree block is created, back references are inserted:
466 * (root->root_key.objectid, trans->transid, level, 1)
468 * When a tree block is cow'd, new back references are added for all the
469 * blocks it points to. If the tree block isn't in reference counted root,
470 * the old back references are removed. These new back references are of
471 * the form (trans->transid will have increased since creation):
473 * (root->root_key.objectid, trans->transid, level, 1)
475 * When a backref is in deleting, the following fields are checked:
477 * if backref was for a tree root:
478 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
480 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
482 * Back Reference Key composing:
484 * The key objectid corresponds to the first byte in the extent, the key
485 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
486 * byte of parent extent. If a extent is tree root, the key offset is set
487 * to the key objectid.
490 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
491 struct btrfs_root *root,
492 struct btrfs_path *path,
493 u64 bytenr, u64 parent,
494 u64 ref_root, u64 ref_generation,
495 u64 owner_objectid, int del)
497 struct btrfs_key key;
498 struct btrfs_extent_ref *ref;
499 struct extent_buffer *leaf;
503 key.objectid = bytenr;
504 key.type = BTRFS_EXTENT_REF_KEY;
507 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
515 leaf = path->nodes[0];
516 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
517 ref_objectid = btrfs_ref_objectid(leaf, ref);
518 if (btrfs_ref_root(leaf, ref) != ref_root ||
519 btrfs_ref_generation(leaf, ref) != ref_generation ||
520 (ref_objectid != owner_objectid &&
521 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
532 * updates all the backrefs that are pending on update_list for the
535 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
536 struct btrfs_root *extent_root,
537 struct btrfs_path *path,
538 struct list_head *update_list)
540 struct btrfs_key key;
541 struct btrfs_extent_ref *ref;
542 struct btrfs_fs_info *info = extent_root->fs_info;
543 struct pending_extent_op *op;
544 struct extent_buffer *leaf;
546 struct list_head *cur = update_list->next;
548 u64 ref_root = extent_root->root_key.objectid;
550 op = list_entry(cur, struct pending_extent_op, list);
553 key.objectid = op->bytenr;
554 key.type = BTRFS_EXTENT_REF_KEY;
555 key.offset = op->orig_parent;
557 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
560 leaf = path->nodes[0];
563 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
565 ref_objectid = btrfs_ref_objectid(leaf, ref);
567 if (btrfs_ref_root(leaf, ref) != ref_root ||
568 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
569 (ref_objectid != op->level &&
570 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
571 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
572 "root %llu, owner %u\n",
573 (unsigned long long)op->bytenr,
574 (unsigned long long)op->orig_parent,
575 (unsigned long long)ref_root, op->level);
576 btrfs_print_leaf(extent_root, leaf);
580 key.objectid = op->bytenr;
581 key.offset = op->parent;
582 key.type = BTRFS_EXTENT_REF_KEY;
583 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
585 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
586 btrfs_set_ref_generation(leaf, ref, op->generation);
590 list_del_init(&op->list);
591 unlock_extent(&info->extent_ins, op->bytenr,
592 op->bytenr + op->num_bytes - 1, GFP_NOFS);
595 if (cur == update_list) {
596 btrfs_mark_buffer_dirty(path->nodes[0]);
597 btrfs_release_path(extent_root, path);
601 op = list_entry(cur, struct pending_extent_op, list);
604 while (path->slots[0] < btrfs_header_nritems(leaf)) {
605 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
606 if (key.objectid == op->bytenr &&
607 key.type == BTRFS_EXTENT_REF_KEY)
612 btrfs_mark_buffer_dirty(path->nodes[0]);
613 btrfs_release_path(extent_root, path);
620 static noinline int insert_extents(struct btrfs_trans_handle *trans,
621 struct btrfs_root *extent_root,
622 struct btrfs_path *path,
623 struct list_head *insert_list, int nr)
625 struct btrfs_key *keys;
627 struct pending_extent_op *op;
628 struct extent_buffer *leaf;
629 struct list_head *cur = insert_list->next;
630 struct btrfs_fs_info *info = extent_root->fs_info;
631 u64 ref_root = extent_root->root_key.objectid;
632 int i = 0, last = 0, ret;
638 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
642 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
648 list_for_each_entry(op, insert_list, list) {
649 keys[i].objectid = op->bytenr;
650 keys[i].offset = op->num_bytes;
651 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
652 data_size[i] = sizeof(struct btrfs_extent_item);
655 keys[i].objectid = op->bytenr;
656 keys[i].offset = op->parent;
657 keys[i].type = BTRFS_EXTENT_REF_KEY;
658 data_size[i] = sizeof(struct btrfs_extent_ref);
662 op = list_entry(cur, struct pending_extent_op, list);
666 ret = btrfs_insert_some_items(trans, extent_root, path,
667 keys+i, data_size+i, total-i);
673 leaf = path->nodes[0];
674 for (c = 0; c < ret; c++) {
675 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
678 * if the first item we inserted was a backref, then
679 * the EXTENT_ITEM will be the odd c's, else it will
682 if ((ref_first && (c % 2)) ||
683 (!ref_first && !(c % 2))) {
684 struct btrfs_extent_item *itm;
686 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
687 struct btrfs_extent_item);
688 btrfs_set_extent_refs(path->nodes[0], itm, 1);
691 struct btrfs_extent_ref *ref;
693 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
694 struct btrfs_extent_ref);
695 btrfs_set_ref_root(leaf, ref, ref_root);
696 btrfs_set_ref_generation(leaf, ref,
698 btrfs_set_ref_objectid(leaf, ref, op->level);
699 btrfs_set_ref_num_refs(leaf, ref, 1);
704 * using del to see when its ok to free up the
705 * pending_extent_op. In the case where we insert the
706 * last item on the list in order to help do batching
707 * we need to not free the extent op until we actually
708 * insert the extent_item
711 unlock_extent(&info->extent_ins, op->bytenr,
712 op->bytenr + op->num_bytes - 1,
715 list_del_init(&op->list);
717 if (cur != insert_list)
719 struct pending_extent_op,
723 btrfs_mark_buffer_dirty(leaf);
724 btrfs_release_path(extent_root, path);
727 * Ok backref's and items usually go right next to eachother,
728 * but if we could only insert 1 item that means that we
729 * inserted on the end of a leaf, and we have no idea what may
730 * be on the next leaf so we just play it safe. In order to
731 * try and help this case we insert the last thing on our
732 * insert list so hopefully it will end up being the last
733 * thing on the leaf and everything else will be before it,
734 * which will let us insert a whole bunch of items at the same
737 if (ret == 1 && !last && (i + ret < total)) {
739 * last: where we will pick up the next time around
740 * i: our current key to insert, will be total - 1
741 * cur: the current op we are screwing with
746 cur = insert_list->prev;
747 op = list_entry(cur, struct pending_extent_op, list);
750 * ok we successfully inserted the last item on the
751 * list, lets reset everything
753 * i: our current key to insert, so where we left off
755 * last: done with this
756 * cur: the op we are messing with
758 * total: since we inserted the last key, we need to
759 * decrement total so we dont overflow
765 cur = insert_list->next;
766 op = list_entry(cur, struct pending_extent_op,
781 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root,
783 struct btrfs_path *path,
784 u64 bytenr, u64 parent,
785 u64 ref_root, u64 ref_generation,
788 struct btrfs_key key;
789 struct extent_buffer *leaf;
790 struct btrfs_extent_ref *ref;
794 key.objectid = bytenr;
795 key.type = BTRFS_EXTENT_REF_KEY;
798 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
800 leaf = path->nodes[0];
801 ref = btrfs_item_ptr(leaf, path->slots[0],
802 struct btrfs_extent_ref);
803 btrfs_set_ref_root(leaf, ref, ref_root);
804 btrfs_set_ref_generation(leaf, ref, ref_generation);
805 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
806 btrfs_set_ref_num_refs(leaf, ref, 1);
807 } else if (ret == -EEXIST) {
809 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
810 leaf = path->nodes[0];
811 ref = btrfs_item_ptr(leaf, path->slots[0],
812 struct btrfs_extent_ref);
813 if (btrfs_ref_root(leaf, ref) != ref_root ||
814 btrfs_ref_generation(leaf, ref) != ref_generation) {
820 num_refs = btrfs_ref_num_refs(leaf, ref);
821 BUG_ON(num_refs == 0);
822 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
824 existing_owner = btrfs_ref_objectid(leaf, ref);
825 if (existing_owner != owner_objectid &&
826 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
827 btrfs_set_ref_objectid(leaf, ref,
828 BTRFS_MULTIPLE_OBJECTIDS);
834 btrfs_mark_buffer_dirty(path->nodes[0]);
836 btrfs_release_path(root, path);
840 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
841 struct btrfs_root *root,
842 struct btrfs_path *path)
844 struct extent_buffer *leaf;
845 struct btrfs_extent_ref *ref;
849 leaf = path->nodes[0];
850 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
851 num_refs = btrfs_ref_num_refs(leaf, ref);
852 BUG_ON(num_refs == 0);
855 ret = btrfs_del_item(trans, root, path);
857 btrfs_set_ref_num_refs(leaf, ref, num_refs);
858 btrfs_mark_buffer_dirty(leaf);
860 btrfs_release_path(root, path);
864 #ifdef BIO_RW_DISCARD
865 static void btrfs_issue_discard(struct block_device *bdev,
868 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
872 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
875 #ifdef BIO_RW_DISCARD
877 u64 map_length = num_bytes;
878 struct btrfs_multi_bio *multi = NULL;
880 /* Tell the block device(s) that the sectors can be discarded */
881 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
882 bytenr, &map_length, &multi, 0);
884 struct btrfs_bio_stripe *stripe = multi->stripes;
887 if (map_length > num_bytes)
888 map_length = num_bytes;
890 for (i = 0; i < multi->num_stripes; i++, stripe++) {
891 btrfs_issue_discard(stripe->dev->bdev,
904 static noinline int free_extents(struct btrfs_trans_handle *trans,
905 struct btrfs_root *extent_root,
906 struct list_head *del_list)
908 struct btrfs_fs_info *info = extent_root->fs_info;
909 struct btrfs_path *path;
910 struct btrfs_key key, found_key;
911 struct extent_buffer *leaf;
912 struct list_head *cur;
913 struct pending_extent_op *op;
914 struct btrfs_extent_item *ei;
915 int ret, num_to_del, extent_slot = 0, found_extent = 0;
919 path = btrfs_alloc_path();
925 /* search for the backref for the current ref we want to delete */
926 cur = del_list->next;
927 op = list_entry(cur, struct pending_extent_op, list);
928 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
930 extent_root->root_key.objectid,
931 op->orig_generation, op->level, 1);
933 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
934 "root %llu gen %llu owner %u\n",
935 (unsigned long long)op->bytenr,
936 (unsigned long long)extent_root->root_key.objectid,
937 (unsigned long long)op->orig_generation, op->level);
938 btrfs_print_leaf(extent_root, path->nodes[0]);
943 extent_slot = path->slots[0];
948 * if we aren't the first item on the leaf we can move back one and see
949 * if our ref is right next to our extent item
951 if (likely(extent_slot)) {
953 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
955 if (found_key.objectid == op->bytenr &&
956 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
957 found_key.offset == op->num_bytes) {
964 * if we didn't find the extent we need to delete the backref and then
965 * search for the extent item key so we can update its ref count
968 key.objectid = op->bytenr;
969 key.type = BTRFS_EXTENT_ITEM_KEY;
970 key.offset = op->num_bytes;
972 ret = remove_extent_backref(trans, extent_root, path);
974 btrfs_release_path(extent_root, path);
975 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
977 extent_slot = path->slots[0];
980 /* this is where we update the ref count for the extent */
981 leaf = path->nodes[0];
982 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
983 refs = btrfs_extent_refs(leaf, ei);
986 btrfs_set_extent_refs(leaf, ei, refs);
988 btrfs_mark_buffer_dirty(leaf);
991 * This extent needs deleting. The reason cur_slot is extent_slot +
992 * num_to_del is because extent_slot points to the slot where the extent
993 * is, and if the backref was not right next to the extent we will be
994 * deleting at least 1 item, and will want to start searching at the
995 * slot directly next to extent_slot. However if we did find the
996 * backref next to the extent item them we will be deleting at least 2
997 * items and will want to start searching directly after the ref slot
1000 struct list_head *pos, *n, *end;
1001 int cur_slot = extent_slot+num_to_del;
1005 path->slots[0] = extent_slot;
1006 bytes_freed = op->num_bytes;
1008 mutex_lock(&info->pinned_mutex);
1009 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1010 op->num_bytes, op->level >=
1011 BTRFS_FIRST_FREE_OBJECTID);
1012 mutex_unlock(&info->pinned_mutex);
1017 * we need to see if we can delete multiple things at once, so
1018 * start looping through the list of extents we are wanting to
1019 * delete and see if their extent/backref's are right next to
1020 * eachother and the extents only have 1 ref
1022 for (pos = cur->next; pos != del_list; pos = pos->next) {
1023 struct pending_extent_op *tmp;
1025 tmp = list_entry(pos, struct pending_extent_op, list);
1027 /* we only want to delete extent+ref at this stage */
1028 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1031 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1032 if (found_key.objectid != tmp->bytenr ||
1033 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1034 found_key.offset != tmp->num_bytes)
1037 /* check to make sure this extent only has one ref */
1038 ei = btrfs_item_ptr(leaf, cur_slot,
1039 struct btrfs_extent_item);
1040 if (btrfs_extent_refs(leaf, ei) != 1)
1043 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1044 if (found_key.objectid != tmp->bytenr ||
1045 found_key.type != BTRFS_EXTENT_REF_KEY ||
1046 found_key.offset != tmp->orig_parent)
1050 * the ref is right next to the extent, we can set the
1051 * ref count to 0 since we will delete them both now
1053 btrfs_set_extent_refs(leaf, ei, 0);
1055 /* pin down the bytes for this extent */
1056 mutex_lock(&info->pinned_mutex);
1057 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1058 tmp->num_bytes, tmp->level >=
1059 BTRFS_FIRST_FREE_OBJECTID);
1060 mutex_unlock(&info->pinned_mutex);
1064 * use the del field to tell if we need to go ahead and
1065 * free up the extent when we delete the item or not.
1068 bytes_freed += tmp->num_bytes;
1075 /* update the free space counters */
1076 spin_lock(&info->delalloc_lock);
1077 super_used = btrfs_super_bytes_used(&info->super_copy);
1078 btrfs_set_super_bytes_used(&info->super_copy,
1079 super_used - bytes_freed);
1081 root_used = btrfs_root_used(&extent_root->root_item);
1082 btrfs_set_root_used(&extent_root->root_item,
1083 root_used - bytes_freed);
1084 spin_unlock(&info->delalloc_lock);
1086 /* delete the items */
1087 ret = btrfs_del_items(trans, extent_root, path,
1088 path->slots[0], num_to_del);
1092 * loop through the extents we deleted and do the cleanup work
1095 for (pos = cur, n = pos->next; pos != end;
1096 pos = n, n = pos->next) {
1097 struct pending_extent_op *tmp;
1098 tmp = list_entry(pos, struct pending_extent_op, list);
1101 * remember tmp->del tells us wether or not we pinned
1104 ret = update_block_group(trans, extent_root,
1105 tmp->bytenr, tmp->num_bytes, 0,
1109 list_del_init(&tmp->list);
1110 unlock_extent(&info->extent_ins, tmp->bytenr,
1111 tmp->bytenr + tmp->num_bytes - 1,
1115 } else if (refs && found_extent) {
1117 * the ref and extent were right next to eachother, but the
1118 * extent still has a ref, so just free the backref and keep
1121 ret = remove_extent_backref(trans, extent_root, path);
1124 list_del_init(&op->list);
1125 unlock_extent(&info->extent_ins, op->bytenr,
1126 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1130 * the extent has multiple refs and the backref we were looking
1131 * for was not right next to it, so just unlock and go next,
1134 list_del_init(&op->list);
1135 unlock_extent(&info->extent_ins, op->bytenr,
1136 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1140 btrfs_release_path(extent_root, path);
1141 if (!list_empty(del_list))
1145 btrfs_free_path(path);
1149 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1150 struct btrfs_root *root, u64 bytenr,
1151 u64 orig_parent, u64 parent,
1152 u64 orig_root, u64 ref_root,
1153 u64 orig_generation, u64 ref_generation,
1157 struct btrfs_root *extent_root = root->fs_info->extent_root;
1158 struct btrfs_path *path;
1160 if (root == root->fs_info->extent_root) {
1161 struct pending_extent_op *extent_op;
1164 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1165 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1166 mutex_lock(&root->fs_info->extent_ins_mutex);
1167 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1168 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1170 ret = get_state_private(&root->fs_info->extent_ins,
1173 extent_op = (struct pending_extent_op *)
1174 (unsigned long)priv;
1175 BUG_ON(extent_op->parent != orig_parent);
1176 BUG_ON(extent_op->generation != orig_generation);
1178 extent_op->parent = parent;
1179 extent_op->generation = ref_generation;
1181 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1184 extent_op->type = PENDING_BACKREF_UPDATE;
1185 extent_op->bytenr = bytenr;
1186 extent_op->num_bytes = num_bytes;
1187 extent_op->parent = parent;
1188 extent_op->orig_parent = orig_parent;
1189 extent_op->generation = ref_generation;
1190 extent_op->orig_generation = orig_generation;
1191 extent_op->level = (int)owner_objectid;
1192 INIT_LIST_HEAD(&extent_op->list);
1195 set_extent_bits(&root->fs_info->extent_ins,
1196 bytenr, bytenr + num_bytes - 1,
1197 EXTENT_WRITEBACK, GFP_NOFS);
1198 set_state_private(&root->fs_info->extent_ins,
1199 bytenr, (unsigned long)extent_op);
1201 mutex_unlock(&root->fs_info->extent_ins_mutex);
1205 path = btrfs_alloc_path();
1208 ret = lookup_extent_backref(trans, extent_root, path,
1209 bytenr, orig_parent, orig_root,
1210 orig_generation, owner_objectid, 1);
1213 ret = remove_extent_backref(trans, extent_root, path);
1216 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1217 parent, ref_root, ref_generation,
1220 finish_current_insert(trans, extent_root, 0);
1221 del_pending_extents(trans, extent_root, 0);
1223 btrfs_free_path(path);
1227 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1228 struct btrfs_root *root, u64 bytenr,
1229 u64 orig_parent, u64 parent,
1230 u64 ref_root, u64 ref_generation,
1234 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1235 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1237 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1238 parent, ref_root, ref_root,
1239 ref_generation, ref_generation,
1244 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1245 struct btrfs_root *root, u64 bytenr,
1246 u64 orig_parent, u64 parent,
1247 u64 orig_root, u64 ref_root,
1248 u64 orig_generation, u64 ref_generation,
1251 struct btrfs_path *path;
1253 struct btrfs_key key;
1254 struct extent_buffer *l;
1255 struct btrfs_extent_item *item;
1258 path = btrfs_alloc_path();
1263 key.objectid = bytenr;
1264 key.type = BTRFS_EXTENT_ITEM_KEY;
1265 key.offset = (u64)-1;
1267 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1271 BUG_ON(ret == 0 || path->slots[0] == 0);
1276 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1277 if (key.objectid != bytenr) {
1278 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1279 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1280 (unsigned long long)bytenr,
1281 (unsigned long long)key.objectid);
1284 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1286 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1287 refs = btrfs_extent_refs(l, item);
1288 btrfs_set_extent_refs(l, item, refs + 1);
1289 btrfs_mark_buffer_dirty(path->nodes[0]);
1291 btrfs_release_path(root->fs_info->extent_root, path);
1294 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1295 path, bytenr, parent,
1296 ref_root, ref_generation,
1299 finish_current_insert(trans, root->fs_info->extent_root, 0);
1300 del_pending_extents(trans, root->fs_info->extent_root, 0);
1302 btrfs_free_path(path);
1306 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1307 struct btrfs_root *root,
1308 u64 bytenr, u64 num_bytes, u64 parent,
1309 u64 ref_root, u64 ref_generation,
1313 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1314 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1316 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1317 0, ref_root, 0, ref_generation,
1322 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *root)
1325 finish_current_insert(trans, root->fs_info->extent_root, 1);
1326 del_pending_extents(trans, root->fs_info->extent_root, 1);
1330 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1331 struct btrfs_root *root, u64 bytenr,
1332 u64 num_bytes, u32 *refs)
1334 struct btrfs_path *path;
1336 struct btrfs_key key;
1337 struct extent_buffer *l;
1338 struct btrfs_extent_item *item;
1340 WARN_ON(num_bytes < root->sectorsize);
1341 path = btrfs_alloc_path();
1343 key.objectid = bytenr;
1344 key.offset = num_bytes;
1345 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1346 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1351 btrfs_print_leaf(root, path->nodes[0]);
1352 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1353 (unsigned long long)bytenr);
1357 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1358 *refs = btrfs_extent_refs(l, item);
1360 btrfs_free_path(path);
1364 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1365 struct btrfs_root *root, u64 objectid, u64 bytenr)
1367 struct btrfs_root *extent_root = root->fs_info->extent_root;
1368 struct btrfs_path *path;
1369 struct extent_buffer *leaf;
1370 struct btrfs_extent_ref *ref_item;
1371 struct btrfs_key key;
1372 struct btrfs_key found_key;
1378 key.objectid = bytenr;
1379 key.offset = (u64)-1;
1380 key.type = BTRFS_EXTENT_ITEM_KEY;
1382 path = btrfs_alloc_path();
1383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1389 if (path->slots[0] == 0)
1393 leaf = path->nodes[0];
1394 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1396 if (found_key.objectid != bytenr ||
1397 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1400 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1402 leaf = path->nodes[0];
1403 nritems = btrfs_header_nritems(leaf);
1404 if (path->slots[0] >= nritems) {
1405 ret = btrfs_next_leaf(extent_root, path);
1412 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1413 if (found_key.objectid != bytenr)
1416 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1421 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1422 struct btrfs_extent_ref);
1423 ref_root = btrfs_ref_root(leaf, ref_item);
1424 if ((ref_root != root->root_key.objectid &&
1425 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1426 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1430 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1439 btrfs_free_path(path);
1443 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1444 struct extent_buffer *buf, u32 nr_extents)
1446 struct btrfs_key key;
1447 struct btrfs_file_extent_item *fi;
1455 if (!root->ref_cows)
1458 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1460 root_gen = root->root_key.offset;
1463 root_gen = trans->transid - 1;
1466 level = btrfs_header_level(buf);
1467 nritems = btrfs_header_nritems(buf);
1470 struct btrfs_leaf_ref *ref;
1471 struct btrfs_extent_info *info;
1473 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1479 ref->root_gen = root_gen;
1480 ref->bytenr = buf->start;
1481 ref->owner = btrfs_header_owner(buf);
1482 ref->generation = btrfs_header_generation(buf);
1483 ref->nritems = nr_extents;
1484 info = ref->extents;
1486 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1488 btrfs_item_key_to_cpu(buf, &key, i);
1489 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1491 fi = btrfs_item_ptr(buf, i,
1492 struct btrfs_file_extent_item);
1493 if (btrfs_file_extent_type(buf, fi) ==
1494 BTRFS_FILE_EXTENT_INLINE)
1496 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1497 if (disk_bytenr == 0)
1500 info->bytenr = disk_bytenr;
1502 btrfs_file_extent_disk_num_bytes(buf, fi);
1503 info->objectid = key.objectid;
1504 info->offset = key.offset;
1508 ret = btrfs_add_leaf_ref(root, ref, shared);
1509 if (ret == -EEXIST && shared) {
1510 struct btrfs_leaf_ref *old;
1511 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1513 btrfs_remove_leaf_ref(root, old);
1514 btrfs_free_leaf_ref(root, old);
1515 ret = btrfs_add_leaf_ref(root, ref, shared);
1518 btrfs_free_leaf_ref(root, ref);
1524 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1525 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1532 u64 orig_generation;
1534 u32 nr_file_extents = 0;
1535 struct btrfs_key key;
1536 struct btrfs_file_extent_item *fi;
1541 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1542 u64, u64, u64, u64, u64, u64, u64, u64);
1544 ref_root = btrfs_header_owner(buf);
1545 ref_generation = btrfs_header_generation(buf);
1546 orig_root = btrfs_header_owner(orig_buf);
1547 orig_generation = btrfs_header_generation(orig_buf);
1549 nritems = btrfs_header_nritems(buf);
1550 level = btrfs_header_level(buf);
1552 if (root->ref_cows) {
1553 process_func = __btrfs_inc_extent_ref;
1556 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1559 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1561 process_func = __btrfs_update_extent_ref;
1564 for (i = 0; i < nritems; i++) {
1567 btrfs_item_key_to_cpu(buf, &key, i);
1568 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1570 fi = btrfs_item_ptr(buf, i,
1571 struct btrfs_file_extent_item);
1572 if (btrfs_file_extent_type(buf, fi) ==
1573 BTRFS_FILE_EXTENT_INLINE)
1575 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1581 ret = process_func(trans, root, bytenr,
1582 orig_buf->start, buf->start,
1583 orig_root, ref_root,
1584 orig_generation, ref_generation,
1593 bytenr = btrfs_node_blockptr(buf, i);
1594 ret = process_func(trans, root, bytenr,
1595 orig_buf->start, buf->start,
1596 orig_root, ref_root,
1597 orig_generation, ref_generation,
1609 *nr_extents = nr_file_extents;
1611 *nr_extents = nritems;
1619 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1620 struct btrfs_root *root, struct extent_buffer *orig_buf,
1621 struct extent_buffer *buf, int start_slot, int nr)
1628 u64 orig_generation;
1629 struct btrfs_key key;
1630 struct btrfs_file_extent_item *fi;
1636 BUG_ON(start_slot < 0);
1637 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1639 ref_root = btrfs_header_owner(buf);
1640 ref_generation = btrfs_header_generation(buf);
1641 orig_root = btrfs_header_owner(orig_buf);
1642 orig_generation = btrfs_header_generation(orig_buf);
1643 level = btrfs_header_level(buf);
1645 if (!root->ref_cows) {
1647 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1650 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1654 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1657 btrfs_item_key_to_cpu(buf, &key, slot);
1658 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1660 fi = btrfs_item_ptr(buf, slot,
1661 struct btrfs_file_extent_item);
1662 if (btrfs_file_extent_type(buf, fi) ==
1663 BTRFS_FILE_EXTENT_INLINE)
1665 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1668 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1669 orig_buf->start, buf->start,
1670 orig_root, ref_root,
1671 orig_generation, ref_generation,
1676 bytenr = btrfs_node_blockptr(buf, slot);
1677 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1678 orig_buf->start, buf->start,
1679 orig_root, ref_root,
1680 orig_generation, ref_generation,
1692 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_path *path,
1695 struct btrfs_block_group_cache *cache)
1699 struct btrfs_root *extent_root = root->fs_info->extent_root;
1701 struct extent_buffer *leaf;
1703 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1708 leaf = path->nodes[0];
1709 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1710 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1711 btrfs_mark_buffer_dirty(leaf);
1712 btrfs_release_path(extent_root, path);
1714 finish_current_insert(trans, extent_root, 0);
1715 pending_ret = del_pending_extents(trans, extent_root, 0);
1724 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1725 struct btrfs_root *root)
1727 struct btrfs_block_group_cache *cache, *entry;
1731 struct btrfs_path *path;
1734 path = btrfs_alloc_path();
1740 spin_lock(&root->fs_info->block_group_cache_lock);
1741 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1742 n; n = rb_next(n)) {
1743 entry = rb_entry(n, struct btrfs_block_group_cache,
1750 spin_unlock(&root->fs_info->block_group_cache_lock);
1756 last += cache->key.offset;
1758 err = write_one_cache_group(trans, root,
1761 * if we fail to write the cache group, we want
1762 * to keep it marked dirty in hopes that a later
1770 btrfs_free_path(path);
1774 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1776 struct btrfs_block_group_cache *block_group;
1779 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1780 if (!block_group || block_group->ro)
1783 put_block_group(block_group);
1787 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1788 u64 total_bytes, u64 bytes_used,
1789 struct btrfs_space_info **space_info)
1791 struct btrfs_space_info *found;
1793 found = __find_space_info(info, flags);
1795 spin_lock(&found->lock);
1796 found->total_bytes += total_bytes;
1797 found->bytes_used += bytes_used;
1799 spin_unlock(&found->lock);
1800 *space_info = found;
1803 found = kzalloc(sizeof(*found), GFP_NOFS);
1807 list_add(&found->list, &info->space_info);
1808 INIT_LIST_HEAD(&found->block_groups);
1809 init_rwsem(&found->groups_sem);
1810 spin_lock_init(&found->lock);
1811 found->flags = flags;
1812 found->total_bytes = total_bytes;
1813 found->bytes_used = bytes_used;
1814 found->bytes_pinned = 0;
1815 found->bytes_reserved = 0;
1816 found->bytes_readonly = 0;
1818 found->force_alloc = 0;
1819 *space_info = found;
1823 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1825 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1826 BTRFS_BLOCK_GROUP_RAID1 |
1827 BTRFS_BLOCK_GROUP_RAID10 |
1828 BTRFS_BLOCK_GROUP_DUP);
1830 if (flags & BTRFS_BLOCK_GROUP_DATA)
1831 fs_info->avail_data_alloc_bits |= extra_flags;
1832 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1833 fs_info->avail_metadata_alloc_bits |= extra_flags;
1834 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1835 fs_info->avail_system_alloc_bits |= extra_flags;
1839 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1841 spin_lock(&cache->space_info->lock);
1842 spin_lock(&cache->lock);
1844 cache->space_info->bytes_readonly += cache->key.offset -
1845 btrfs_block_group_used(&cache->item);
1848 spin_unlock(&cache->lock);
1849 spin_unlock(&cache->space_info->lock);
1852 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1854 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1856 if (num_devices == 1)
1857 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1858 if (num_devices < 4)
1859 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1861 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1862 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1863 BTRFS_BLOCK_GROUP_RAID10))) {
1864 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1867 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1868 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1869 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1872 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1873 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1874 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1875 (flags & BTRFS_BLOCK_GROUP_DUP)))
1876 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1880 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *extent_root, u64 alloc_bytes,
1882 u64 flags, int force)
1884 struct btrfs_space_info *space_info;
1888 mutex_lock(&extent_root->fs_info->chunk_mutex);
1890 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1892 space_info = __find_space_info(extent_root->fs_info, flags);
1894 ret = update_space_info(extent_root->fs_info, flags,
1898 BUG_ON(!space_info);
1900 spin_lock(&space_info->lock);
1901 if (space_info->force_alloc) {
1903 space_info->force_alloc = 0;
1905 if (space_info->full) {
1906 spin_unlock(&space_info->lock);
1910 thresh = space_info->total_bytes - space_info->bytes_readonly;
1911 thresh = div_factor(thresh, 6);
1913 (space_info->bytes_used + space_info->bytes_pinned +
1914 space_info->bytes_reserved + alloc_bytes) < thresh) {
1915 spin_unlock(&space_info->lock);
1918 spin_unlock(&space_info->lock);
1920 ret = btrfs_alloc_chunk(trans, extent_root, flags);
1922 space_info->full = 1;
1924 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1928 static int update_block_group(struct btrfs_trans_handle *trans,
1929 struct btrfs_root *root,
1930 u64 bytenr, u64 num_bytes, int alloc,
1933 struct btrfs_block_group_cache *cache;
1934 struct btrfs_fs_info *info = root->fs_info;
1935 u64 total = num_bytes;
1940 cache = btrfs_lookup_block_group(info, bytenr);
1943 byte_in_group = bytenr - cache->key.objectid;
1944 WARN_ON(byte_in_group > cache->key.offset);
1946 spin_lock(&cache->space_info->lock);
1947 spin_lock(&cache->lock);
1949 old_val = btrfs_block_group_used(&cache->item);
1950 num_bytes = min(total, cache->key.offset - byte_in_group);
1952 old_val += num_bytes;
1953 cache->space_info->bytes_used += num_bytes;
1955 cache->space_info->bytes_readonly -= num_bytes;
1956 btrfs_set_block_group_used(&cache->item, old_val);
1957 spin_unlock(&cache->lock);
1958 spin_unlock(&cache->space_info->lock);
1960 old_val -= num_bytes;
1961 cache->space_info->bytes_used -= num_bytes;
1963 cache->space_info->bytes_readonly += num_bytes;
1964 btrfs_set_block_group_used(&cache->item, old_val);
1965 spin_unlock(&cache->lock);
1966 spin_unlock(&cache->space_info->lock);
1970 ret = btrfs_discard_extent(root, bytenr,
1974 ret = btrfs_add_free_space(cache, bytenr,
1979 put_block_group(cache);
1981 bytenr += num_bytes;
1986 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1988 struct btrfs_block_group_cache *cache;
1991 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1995 bytenr = cache->key.objectid;
1996 put_block_group(cache);
2001 int btrfs_update_pinned_extents(struct btrfs_root *root,
2002 u64 bytenr, u64 num, int pin)
2005 struct btrfs_block_group_cache *cache;
2006 struct btrfs_fs_info *fs_info = root->fs_info;
2008 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2010 set_extent_dirty(&fs_info->pinned_extents,
2011 bytenr, bytenr + num - 1, GFP_NOFS);
2013 clear_extent_dirty(&fs_info->pinned_extents,
2014 bytenr, bytenr + num - 1, GFP_NOFS);
2017 cache = btrfs_lookup_block_group(fs_info, bytenr);
2019 len = min(num, cache->key.offset -
2020 (bytenr - cache->key.objectid));
2022 spin_lock(&cache->space_info->lock);
2023 spin_lock(&cache->lock);
2024 cache->pinned += len;
2025 cache->space_info->bytes_pinned += len;
2026 spin_unlock(&cache->lock);
2027 spin_unlock(&cache->space_info->lock);
2028 fs_info->total_pinned += len;
2030 spin_lock(&cache->space_info->lock);
2031 spin_lock(&cache->lock);
2032 cache->pinned -= len;
2033 cache->space_info->bytes_pinned -= len;
2034 spin_unlock(&cache->lock);
2035 spin_unlock(&cache->space_info->lock);
2036 fs_info->total_pinned -= len;
2038 btrfs_add_free_space(cache, bytenr, len);
2040 put_block_group(cache);
2047 static int update_reserved_extents(struct btrfs_root *root,
2048 u64 bytenr, u64 num, int reserve)
2051 struct btrfs_block_group_cache *cache;
2052 struct btrfs_fs_info *fs_info = root->fs_info;
2055 cache = btrfs_lookup_block_group(fs_info, bytenr);
2057 len = min(num, cache->key.offset -
2058 (bytenr - cache->key.objectid));
2060 spin_lock(&cache->space_info->lock);
2061 spin_lock(&cache->lock);
2063 cache->reserved += len;
2064 cache->space_info->bytes_reserved += len;
2066 cache->reserved -= len;
2067 cache->space_info->bytes_reserved -= len;
2069 spin_unlock(&cache->lock);
2070 spin_unlock(&cache->space_info->lock);
2071 put_block_group(cache);
2078 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2083 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2086 mutex_lock(&root->fs_info->pinned_mutex);
2088 ret = find_first_extent_bit(pinned_extents, last,
2089 &start, &end, EXTENT_DIRTY);
2092 set_extent_dirty(copy, start, end, GFP_NOFS);
2095 mutex_unlock(&root->fs_info->pinned_mutex);
2099 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct extent_io_tree *unpin)
2107 mutex_lock(&root->fs_info->pinned_mutex);
2109 ret = find_first_extent_bit(unpin, 0, &start, &end,
2114 ret = btrfs_discard_extent(root, start, end + 1 - start);
2116 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2117 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2119 if (need_resched()) {
2120 mutex_unlock(&root->fs_info->pinned_mutex);
2122 mutex_lock(&root->fs_info->pinned_mutex);
2125 mutex_unlock(&root->fs_info->pinned_mutex);
2129 static int finish_current_insert(struct btrfs_trans_handle *trans,
2130 struct btrfs_root *extent_root, int all)
2137 struct btrfs_fs_info *info = extent_root->fs_info;
2138 struct btrfs_path *path;
2139 struct pending_extent_op *extent_op, *tmp;
2140 struct list_head insert_list, update_list;
2142 int num_inserts = 0, max_inserts;
2144 path = btrfs_alloc_path();
2145 INIT_LIST_HEAD(&insert_list);
2146 INIT_LIST_HEAD(&update_list);
2148 max_inserts = extent_root->leafsize /
2149 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2150 sizeof(struct btrfs_extent_ref) +
2151 sizeof(struct btrfs_extent_item));
2153 mutex_lock(&info->extent_ins_mutex);
2155 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2156 &end, EXTENT_WRITEBACK);
2158 if (skipped && all && !num_inserts &&
2159 list_empty(&update_list)) {
2164 mutex_unlock(&info->extent_ins_mutex);
2168 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2172 if (need_resched()) {
2173 mutex_unlock(&info->extent_ins_mutex);
2175 mutex_lock(&info->extent_ins_mutex);
2180 ret = get_state_private(&info->extent_ins, start, &priv);
2182 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2184 if (extent_op->type == PENDING_EXTENT_INSERT) {
2186 list_add_tail(&extent_op->list, &insert_list);
2188 if (num_inserts == max_inserts) {
2189 mutex_unlock(&info->extent_ins_mutex);
2192 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2193 list_add_tail(&extent_op->list, &update_list);
2201 * process the update list, clear the writeback bit for it, and if
2202 * somebody marked this thing for deletion then just unlock it and be
2203 * done, the free_extents will handle it
2205 mutex_lock(&info->extent_ins_mutex);
2206 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2207 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2208 extent_op->bytenr + extent_op->num_bytes - 1,
2209 EXTENT_WRITEBACK, GFP_NOFS);
2210 if (extent_op->del) {
2211 list_del_init(&extent_op->list);
2212 unlock_extent(&info->extent_ins, extent_op->bytenr,
2213 extent_op->bytenr + extent_op->num_bytes
2218 mutex_unlock(&info->extent_ins_mutex);
2221 * still have things left on the update list, go ahead an update
2224 if (!list_empty(&update_list)) {
2225 ret = update_backrefs(trans, extent_root, path, &update_list);
2230 * if no inserts need to be done, but we skipped some extents and we
2231 * need to make sure everything is cleaned then reset everything and
2232 * go back to the beginning
2234 if (!num_inserts && all && skipped) {
2237 INIT_LIST_HEAD(&update_list);
2238 INIT_LIST_HEAD(&insert_list);
2240 } else if (!num_inserts) {
2245 * process the insert extents list. Again if we are deleting this
2246 * extent, then just unlock it, pin down the bytes if need be, and be
2247 * done with it. Saves us from having to actually insert the extent
2248 * into the tree and then subsequently come along and delete it
2250 mutex_lock(&info->extent_ins_mutex);
2251 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2252 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2253 extent_op->bytenr + extent_op->num_bytes - 1,
2254 EXTENT_WRITEBACK, GFP_NOFS);
2255 if (extent_op->del) {
2257 list_del_init(&extent_op->list);
2258 unlock_extent(&info->extent_ins, extent_op->bytenr,
2259 extent_op->bytenr + extent_op->num_bytes
2262 mutex_lock(&extent_root->fs_info->pinned_mutex);
2263 ret = pin_down_bytes(trans, extent_root,
2265 extent_op->num_bytes, 0);
2266 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2268 spin_lock(&info->delalloc_lock);
2269 used = btrfs_super_bytes_used(&info->super_copy);
2270 btrfs_set_super_bytes_used(&info->super_copy,
2271 used - extent_op->num_bytes);
2272 used = btrfs_root_used(&extent_root->root_item);
2273 btrfs_set_root_used(&extent_root->root_item,
2274 used - extent_op->num_bytes);
2275 spin_unlock(&info->delalloc_lock);
2277 ret = update_block_group(trans, extent_root,
2279 extent_op->num_bytes,
2286 mutex_unlock(&info->extent_ins_mutex);
2288 ret = insert_extents(trans, extent_root, path, &insert_list,
2293 * if we broke out of the loop in order to insert stuff because we hit
2294 * the maximum number of inserts at a time we can handle, then loop
2295 * back and pick up where we left off
2297 if (num_inserts == max_inserts) {
2298 INIT_LIST_HEAD(&insert_list);
2299 INIT_LIST_HEAD(&update_list);
2305 * again, if we need to make absolutely sure there are no more pending
2306 * extent operations left and we know that we skipped some, go back to
2307 * the beginning and do it all again
2309 if (all && skipped) {
2310 INIT_LIST_HEAD(&insert_list);
2311 INIT_LIST_HEAD(&update_list);
2318 btrfs_free_path(path);
2322 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root,
2324 u64 bytenr, u64 num_bytes, int is_data)
2327 struct extent_buffer *buf;
2332 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2336 /* we can reuse a block if it hasn't been written
2337 * and it is from this transaction. We can't
2338 * reuse anything from the tree log root because
2339 * it has tiny sub-transactions.
2341 if (btrfs_buffer_uptodate(buf, 0) &&
2342 btrfs_try_tree_lock(buf)) {
2343 u64 header_owner = btrfs_header_owner(buf);
2344 u64 header_transid = btrfs_header_generation(buf);
2345 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2346 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2347 header_transid == trans->transid &&
2348 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2349 clean_tree_block(NULL, root, buf);
2350 btrfs_tree_unlock(buf);
2351 free_extent_buffer(buf);
2354 btrfs_tree_unlock(buf);
2356 free_extent_buffer(buf);
2358 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2365 * remove an extent from the root, returns 0 on success
2367 static int __free_extent(struct btrfs_trans_handle *trans,
2368 struct btrfs_root *root,
2369 u64 bytenr, u64 num_bytes, u64 parent,
2370 u64 root_objectid, u64 ref_generation,
2371 u64 owner_objectid, int pin, int mark_free)
2373 struct btrfs_path *path;
2374 struct btrfs_key key;
2375 struct btrfs_fs_info *info = root->fs_info;
2376 struct btrfs_root *extent_root = info->extent_root;
2377 struct extent_buffer *leaf;
2379 int extent_slot = 0;
2380 int found_extent = 0;
2382 struct btrfs_extent_item *ei;
2385 key.objectid = bytenr;
2386 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2387 key.offset = num_bytes;
2388 path = btrfs_alloc_path();
2393 ret = lookup_extent_backref(trans, extent_root, path,
2394 bytenr, parent, root_objectid,
2395 ref_generation, owner_objectid, 1);
2397 struct btrfs_key found_key;
2398 extent_slot = path->slots[0];
2399 while (extent_slot > 0) {
2401 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2403 if (found_key.objectid != bytenr)
2405 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2406 found_key.offset == num_bytes) {
2410 if (path->slots[0] - extent_slot > 5)
2413 if (!found_extent) {
2414 ret = remove_extent_backref(trans, extent_root, path);
2416 btrfs_release_path(extent_root, path);
2417 ret = btrfs_search_slot(trans, extent_root,
2420 printk(KERN_ERR "umm, got %d back from search"
2421 ", was looking for %llu\n", ret,
2422 (unsigned long long)bytenr);
2423 btrfs_print_leaf(extent_root, path->nodes[0]);
2426 extent_slot = path->slots[0];
2429 btrfs_print_leaf(extent_root, path->nodes[0]);
2431 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2432 "root %llu gen %llu owner %llu\n",
2433 (unsigned long long)bytenr,
2434 (unsigned long long)root_objectid,
2435 (unsigned long long)ref_generation,
2436 (unsigned long long)owner_objectid);
2439 leaf = path->nodes[0];
2440 ei = btrfs_item_ptr(leaf, extent_slot,
2441 struct btrfs_extent_item);
2442 refs = btrfs_extent_refs(leaf, ei);
2445 btrfs_set_extent_refs(leaf, ei, refs);
2447 btrfs_mark_buffer_dirty(leaf);
2449 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2450 struct btrfs_extent_ref *ref;
2451 ref = btrfs_item_ptr(leaf, path->slots[0],
2452 struct btrfs_extent_ref);
2453 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2454 /* if the back ref and the extent are next to each other
2455 * they get deleted below in one shot
2457 path->slots[0] = extent_slot;
2459 } else if (found_extent) {
2460 /* otherwise delete the extent back ref */
2461 ret = remove_extent_backref(trans, extent_root, path);
2463 /* if refs are 0, we need to setup the path for deletion */
2465 btrfs_release_path(extent_root, path);
2466 ret = btrfs_search_slot(trans, extent_root, &key, path,
2477 mutex_lock(&root->fs_info->pinned_mutex);
2478 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2479 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2480 mutex_unlock(&root->fs_info->pinned_mutex);
2485 /* block accounting for super block */
2486 spin_lock(&info->delalloc_lock);
2487 super_used = btrfs_super_bytes_used(&info->super_copy);
2488 btrfs_set_super_bytes_used(&info->super_copy,
2489 super_used - num_bytes);
2491 /* block accounting for root item */
2492 root_used = btrfs_root_used(&root->root_item);
2493 btrfs_set_root_used(&root->root_item,
2494 root_used - num_bytes);
2495 spin_unlock(&info->delalloc_lock);
2496 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2499 btrfs_release_path(extent_root, path);
2501 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2502 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2506 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2510 btrfs_free_path(path);
2511 finish_current_insert(trans, extent_root, 0);
2516 * find all the blocks marked as pending in the radix tree and remove
2517 * them from the extent map
2519 static int del_pending_extents(struct btrfs_trans_handle *trans,
2520 struct btrfs_root *extent_root, int all)
2528 int nr = 0, skipped = 0;
2529 struct extent_io_tree *pending_del;
2530 struct extent_io_tree *extent_ins;
2531 struct pending_extent_op *extent_op;
2532 struct btrfs_fs_info *info = extent_root->fs_info;
2533 struct list_head delete_list;
2535 INIT_LIST_HEAD(&delete_list);
2536 extent_ins = &extent_root->fs_info->extent_ins;
2537 pending_del = &extent_root->fs_info->pending_del;
2540 mutex_lock(&info->extent_ins_mutex);
2542 ret = find_first_extent_bit(pending_del, search, &start, &end,
2545 if (all && skipped && !nr) {
2550 mutex_unlock(&info->extent_ins_mutex);
2554 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2559 if (need_resched()) {
2560 mutex_unlock(&info->extent_ins_mutex);
2562 mutex_lock(&info->extent_ins_mutex);
2569 ret = get_state_private(pending_del, start, &priv);
2571 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2573 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2575 if (!test_range_bit(extent_ins, start, end,
2576 EXTENT_WRITEBACK, 0)) {
2577 list_add_tail(&extent_op->list, &delete_list);
2582 ret = get_state_private(&info->extent_ins, start,
2585 extent_op = (struct pending_extent_op *)
2586 (unsigned long)priv;
2588 clear_extent_bits(&info->extent_ins, start, end,
2589 EXTENT_WRITEBACK, GFP_NOFS);
2591 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2592 list_add_tail(&extent_op->list, &delete_list);
2598 mutex_lock(&extent_root->fs_info->pinned_mutex);
2599 ret = pin_down_bytes(trans, extent_root, start,
2600 end + 1 - start, 0);
2601 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2603 ret = update_block_group(trans, extent_root, start,
2604 end + 1 - start, 0, ret > 0);
2606 unlock_extent(extent_ins, start, end, GFP_NOFS);
2615 if (need_resched()) {
2616 mutex_unlock(&info->extent_ins_mutex);
2618 mutex_lock(&info->extent_ins_mutex);
2623 ret = free_extents(trans, extent_root, &delete_list);
2627 if (all && skipped) {
2628 INIT_LIST_HEAD(&delete_list);
2638 * remove an extent from the root, returns 0 on success
2640 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root,
2642 u64 bytenr, u64 num_bytes, u64 parent,
2643 u64 root_objectid, u64 ref_generation,
2644 u64 owner_objectid, int pin)
2646 struct btrfs_root *extent_root = root->fs_info->extent_root;
2650 WARN_ON(num_bytes < root->sectorsize);
2651 if (root == extent_root) {
2652 struct pending_extent_op *extent_op = NULL;
2654 mutex_lock(&root->fs_info->extent_ins_mutex);
2655 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2656 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2658 ret = get_state_private(&root->fs_info->extent_ins,
2661 extent_op = (struct pending_extent_op *)
2662 (unsigned long)priv;
2665 if (extent_op->type == PENDING_EXTENT_INSERT) {
2666 mutex_unlock(&root->fs_info->extent_ins_mutex);
2672 ref_generation = extent_op->orig_generation;
2673 parent = extent_op->orig_parent;
2676 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2679 extent_op->type = PENDING_EXTENT_DELETE;
2680 extent_op->bytenr = bytenr;
2681 extent_op->num_bytes = num_bytes;
2682 extent_op->parent = parent;
2683 extent_op->orig_parent = parent;
2684 extent_op->generation = ref_generation;
2685 extent_op->orig_generation = ref_generation;
2686 extent_op->level = (int)owner_objectid;
2687 INIT_LIST_HEAD(&extent_op->list);
2690 set_extent_bits(&root->fs_info->pending_del,
2691 bytenr, bytenr + num_bytes - 1,
2692 EXTENT_WRITEBACK, GFP_NOFS);
2693 set_state_private(&root->fs_info->pending_del,
2694 bytenr, (unsigned long)extent_op);
2695 mutex_unlock(&root->fs_info->extent_ins_mutex);
2698 /* if metadata always pin */
2699 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2700 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2701 struct btrfs_block_group_cache *cache;
2703 /* btrfs_free_reserved_extent */
2704 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2706 btrfs_add_free_space(cache, bytenr, num_bytes);
2707 put_block_group(cache);
2708 update_reserved_extents(root, bytenr, num_bytes, 0);
2714 /* if data pin when any transaction has committed this */
2715 if (ref_generation != trans->transid)
2718 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2719 root_objectid, ref_generation,
2720 owner_objectid, pin, pin == 0);
2722 finish_current_insert(trans, root->fs_info->extent_root, 0);
2723 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2724 return ret ? ret : pending_ret;
2727 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2728 struct btrfs_root *root,
2729 u64 bytenr, u64 num_bytes, u64 parent,
2730 u64 root_objectid, u64 ref_generation,
2731 u64 owner_objectid, int pin)
2735 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2736 root_objectid, ref_generation,
2737 owner_objectid, pin);
2741 static u64 stripe_align(struct btrfs_root *root, u64 val)
2743 u64 mask = ((u64)root->stripesize - 1);
2744 u64 ret = (val + mask) & ~mask;
2749 * walks the btree of allocated extents and find a hole of a given size.
2750 * The key ins is changed to record the hole:
2751 * ins->objectid == block start
2752 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2753 * ins->offset == number of blocks
2754 * Any available blocks before search_start are skipped.
2756 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2757 struct btrfs_root *orig_root,
2758 u64 num_bytes, u64 empty_size,
2759 u64 search_start, u64 search_end,
2760 u64 hint_byte, struct btrfs_key *ins,
2761 u64 exclude_start, u64 exclude_nr,
2765 struct btrfs_root *root = orig_root->fs_info->extent_root;
2766 u64 total_needed = num_bytes;
2767 u64 *last_ptr = NULL;
2768 u64 last_wanted = 0;
2769 struct btrfs_block_group_cache *block_group = NULL;
2770 int chunk_alloc_done = 0;
2771 int empty_cluster = 2 * 1024 * 1024;
2772 int allowed_chunk_alloc = 0;
2773 struct list_head *head = NULL, *cur = NULL;
2776 struct btrfs_space_info *space_info;
2778 WARN_ON(num_bytes < root->sectorsize);
2779 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2783 if (orig_root->ref_cows || empty_size)
2784 allowed_chunk_alloc = 1;
2786 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2787 last_ptr = &root->fs_info->last_alloc;
2788 empty_cluster = 64 * 1024;
2791 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2792 last_ptr = &root->fs_info->last_data_alloc;
2796 hint_byte = *last_ptr;
2797 last_wanted = *last_ptr;
2799 empty_size += empty_cluster;
2803 search_start = max(search_start, first_logical_byte(root, 0));
2804 search_start = max(search_start, hint_byte);
2806 if (last_wanted && search_start != last_wanted) {
2808 empty_size += empty_cluster;
2811 total_needed += empty_size;
2812 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2814 block_group = btrfs_lookup_first_block_group(root->fs_info,
2816 space_info = __find_space_info(root->fs_info, data);
2818 down_read(&space_info->groups_sem);
2820 struct btrfs_free_space *free_space;
2822 * the only way this happens if our hint points to a block
2823 * group thats not of the proper type, while looping this
2824 * should never happen
2830 goto new_group_no_lock;
2832 if (unlikely(!block_group->cached)) {
2833 mutex_lock(&block_group->cache_mutex);
2834 ret = cache_block_group(root, block_group);
2835 mutex_unlock(&block_group->cache_mutex);
2840 mutex_lock(&block_group->alloc_mutex);
2841 if (unlikely(!block_group_bits(block_group, data)))
2844 if (unlikely(block_group->ro))
2847 free_space = btrfs_find_free_space(block_group, search_start,
2850 u64 start = block_group->key.objectid;
2851 u64 end = block_group->key.objectid +
2852 block_group->key.offset;
2854 search_start = stripe_align(root, free_space->offset);
2856 /* move on to the next group */
2857 if (search_start + num_bytes >= search_end)
2860 /* move on to the next group */
2861 if (search_start + num_bytes > end)
2864 if (last_wanted && search_start != last_wanted) {
2865 total_needed += empty_cluster;
2866 empty_size += empty_cluster;
2869 * if search_start is still in this block group
2870 * then we just re-search this block group
2872 if (search_start >= start &&
2873 search_start < end) {
2874 mutex_unlock(&block_group->alloc_mutex);
2878 /* else we go to the next block group */
2882 if (exclude_nr > 0 &&
2883 (search_start + num_bytes > exclude_start &&
2884 search_start < exclude_start + exclude_nr)) {
2885 search_start = exclude_start + exclude_nr;
2887 * if search_start is still in this block group
2888 * then we just re-search this block group
2890 if (search_start >= start &&
2891 search_start < end) {
2892 mutex_unlock(&block_group->alloc_mutex);
2897 /* else we go to the next block group */
2901 ins->objectid = search_start;
2902 ins->offset = num_bytes;
2904 btrfs_remove_free_space_lock(block_group, search_start,
2906 /* we are all good, lets return */
2907 mutex_unlock(&block_group->alloc_mutex);
2911 mutex_unlock(&block_group->alloc_mutex);
2912 put_block_group(block_group);
2915 /* don't try to compare new allocations against the
2916 * last allocation any more
2921 * Here's how this works.
2922 * loop == 0: we were searching a block group via a hint
2923 * and didn't find anything, so we start at
2924 * the head of the block groups and keep searching
2925 * loop == 1: we're searching through all of the block groups
2926 * if we hit the head again we have searched
2927 * all of the block groups for this space and we
2928 * need to try and allocate, if we cant error out.
2929 * loop == 2: we allocated more space and are looping through
2930 * all of the block groups again.
2933 head = &space_info->block_groups;
2936 } else if (loop == 1 && cur == head) {
2939 /* at this point we give up on the empty_size
2940 * allocations and just try to allocate the min
2943 * The extra_loop field was set if an empty_size
2944 * allocation was attempted above, and if this
2945 * is try we need to try the loop again without
2946 * the additional empty_size.
2948 total_needed -= empty_size;
2950 keep_going = extra_loop;
2953 if (allowed_chunk_alloc && !chunk_alloc_done) {
2954 up_read(&space_info->groups_sem);
2955 ret = do_chunk_alloc(trans, root, num_bytes +
2956 2 * 1024 * 1024, data, 1);
2957 down_read(&space_info->groups_sem);
2960 head = &space_info->block_groups;
2962 * we've allocated a new chunk, keep
2966 chunk_alloc_done = 1;
2967 } else if (!allowed_chunk_alloc) {
2968 space_info->force_alloc = 1;
2977 } else if (cur == head) {
2981 block_group = list_entry(cur, struct btrfs_block_group_cache,
2983 atomic_inc(&block_group->count);
2985 search_start = block_group->key.objectid;
2989 /* we found what we needed */
2990 if (ins->objectid) {
2991 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2992 trans->block_group = block_group->key.objectid;
2995 *last_ptr = ins->objectid + ins->offset;
2998 printk(KERN_ERR "btrfs searching for %llu bytes, "
2999 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3000 (unsigned long long)total_needed,
3001 (unsigned long long)num_bytes,
3002 loop, allowed_chunk_alloc);
3006 put_block_group(block_group);
3008 up_read(&space_info->groups_sem);
3012 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3014 struct btrfs_block_group_cache *cache;
3016 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3017 (unsigned long long)(info->total_bytes - info->bytes_used -
3018 info->bytes_pinned - info->bytes_reserved),
3019 (info->full) ? "" : "not ");
3021 down_read(&info->groups_sem);
3022 list_for_each_entry(cache, &info->block_groups, list) {
3023 spin_lock(&cache->lock);
3024 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3025 "%llu pinned %llu reserved\n",
3026 (unsigned long long)cache->key.objectid,
3027 (unsigned long long)cache->key.offset,
3028 (unsigned long long)btrfs_block_group_used(&cache->item),
3029 (unsigned long long)cache->pinned,
3030 (unsigned long long)cache->reserved);
3031 btrfs_dump_free_space(cache, bytes);
3032 spin_unlock(&cache->lock);
3034 up_read(&info->groups_sem);
3037 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 num_bytes, u64 min_alloc_size,
3040 u64 empty_size, u64 hint_byte,
3041 u64 search_end, struct btrfs_key *ins,
3045 u64 search_start = 0;
3047 struct btrfs_fs_info *info = root->fs_info;
3050 alloc_profile = info->avail_data_alloc_bits &
3051 info->data_alloc_profile;
3052 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3053 } else if (root == root->fs_info->chunk_root) {
3054 alloc_profile = info->avail_system_alloc_bits &
3055 info->system_alloc_profile;
3056 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3058 alloc_profile = info->avail_metadata_alloc_bits &
3059 info->metadata_alloc_profile;
3060 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3063 data = btrfs_reduce_alloc_profile(root, data);
3065 * the only place that sets empty_size is btrfs_realloc_node, which
3066 * is not called recursively on allocations
3068 if (empty_size || root->ref_cows) {
3069 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3070 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3072 BTRFS_BLOCK_GROUP_METADATA |
3073 (info->metadata_alloc_profile &
3074 info->avail_metadata_alloc_bits), 0);
3076 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3077 num_bytes + 2 * 1024 * 1024, data, 0);
3080 WARN_ON(num_bytes < root->sectorsize);
3081 ret = find_free_extent(trans, root, num_bytes, empty_size,
3082 search_start, search_end, hint_byte, ins,
3083 trans->alloc_exclude_start,
3084 trans->alloc_exclude_nr, data);
3086 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3087 num_bytes = num_bytes >> 1;
3088 num_bytes = num_bytes & ~(root->sectorsize - 1);
3089 num_bytes = max(num_bytes, min_alloc_size);
3090 do_chunk_alloc(trans, root->fs_info->extent_root,
3091 num_bytes, data, 1);
3095 struct btrfs_space_info *sinfo;
3097 sinfo = __find_space_info(root->fs_info, data);
3098 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3099 "wanted %llu\n", (unsigned long long)data,
3100 (unsigned long long)num_bytes);
3101 dump_space_info(sinfo, num_bytes);
3108 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3110 struct btrfs_block_group_cache *cache;
3113 cache = btrfs_lookup_block_group(root->fs_info, start);
3115 printk(KERN_ERR "Unable to find block group for %llu\n",
3116 (unsigned long long)start);
3120 ret = btrfs_discard_extent(root, start, len);
3122 btrfs_add_free_space(cache, start, len);
3123 put_block_group(cache);
3124 update_reserved_extents(root, start, len, 0);
3129 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3130 struct btrfs_root *root,
3131 u64 num_bytes, u64 min_alloc_size,
3132 u64 empty_size, u64 hint_byte,
3133 u64 search_end, struct btrfs_key *ins,
3137 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3138 empty_size, hint_byte, search_end, ins,
3140 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3144 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3145 struct btrfs_root *root, u64 parent,
3146 u64 root_objectid, u64 ref_generation,
3147 u64 owner, struct btrfs_key *ins)
3153 u64 num_bytes = ins->offset;
3155 struct btrfs_fs_info *info = root->fs_info;
3156 struct btrfs_root *extent_root = info->extent_root;
3157 struct btrfs_extent_item *extent_item;
3158 struct btrfs_extent_ref *ref;
3159 struct btrfs_path *path;
3160 struct btrfs_key keys[2];
3163 parent = ins->objectid;
3165 /* block accounting for super block */
3166 spin_lock(&info->delalloc_lock);
3167 super_used = btrfs_super_bytes_used(&info->super_copy);
3168 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3170 /* block accounting for root item */
3171 root_used = btrfs_root_used(&root->root_item);
3172 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3173 spin_unlock(&info->delalloc_lock);
3175 if (root == extent_root) {
3176 struct pending_extent_op *extent_op;
3178 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3181 extent_op->type = PENDING_EXTENT_INSERT;
3182 extent_op->bytenr = ins->objectid;
3183 extent_op->num_bytes = ins->offset;
3184 extent_op->parent = parent;
3185 extent_op->orig_parent = 0;
3186 extent_op->generation = ref_generation;
3187 extent_op->orig_generation = 0;
3188 extent_op->level = (int)owner;
3189 INIT_LIST_HEAD(&extent_op->list);
3192 mutex_lock(&root->fs_info->extent_ins_mutex);
3193 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3194 ins->objectid + ins->offset - 1,
3195 EXTENT_WRITEBACK, GFP_NOFS);
3196 set_state_private(&root->fs_info->extent_ins,
3197 ins->objectid, (unsigned long)extent_op);
3198 mutex_unlock(&root->fs_info->extent_ins_mutex);
3202 memcpy(&keys[0], ins, sizeof(*ins));
3203 keys[1].objectid = ins->objectid;
3204 keys[1].type = BTRFS_EXTENT_REF_KEY;
3205 keys[1].offset = parent;
3206 sizes[0] = sizeof(*extent_item);
3207 sizes[1] = sizeof(*ref);
3209 path = btrfs_alloc_path();
3212 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3216 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3217 struct btrfs_extent_item);
3218 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3219 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3220 struct btrfs_extent_ref);
3222 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3223 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3224 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3225 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3227 btrfs_mark_buffer_dirty(path->nodes[0]);
3229 trans->alloc_exclude_start = 0;
3230 trans->alloc_exclude_nr = 0;
3231 btrfs_free_path(path);
3232 finish_current_insert(trans, extent_root, 0);
3233 pending_ret = del_pending_extents(trans, extent_root, 0);
3243 ret = update_block_group(trans, root, ins->objectid,
3246 printk(KERN_ERR "btrfs update block group failed for %llu "
3247 "%llu\n", (unsigned long long)ins->objectid,
3248 (unsigned long long)ins->offset);
3255 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3256 struct btrfs_root *root, u64 parent,
3257 u64 root_objectid, u64 ref_generation,
3258 u64 owner, struct btrfs_key *ins)
3262 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3264 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3265 ref_generation, owner, ins);
3266 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3271 * this is used by the tree logging recovery code. It records that
3272 * an extent has been allocated and makes sure to clear the free
3273 * space cache bits as well
3275 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3276 struct btrfs_root *root, u64 parent,
3277 u64 root_objectid, u64 ref_generation,
3278 u64 owner, struct btrfs_key *ins)
3281 struct btrfs_block_group_cache *block_group;
3283 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3284 mutex_lock(&block_group->cache_mutex);
3285 cache_block_group(root, block_group);
3286 mutex_unlock(&block_group->cache_mutex);
3288 ret = btrfs_remove_free_space(block_group, ins->objectid,
3291 put_block_group(block_group);
3292 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3293 ref_generation, owner, ins);
3298 * finds a free extent and does all the dirty work required for allocation
3299 * returns the key for the extent through ins, and a tree buffer for
3300 * the first block of the extent through buf.
3302 * returns 0 if everything worked, non-zero otherwise.
3304 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3305 struct btrfs_root *root,
3306 u64 num_bytes, u64 parent, u64 min_alloc_size,
3307 u64 root_objectid, u64 ref_generation,
3308 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3309 u64 search_end, struct btrfs_key *ins, u64 data)
3313 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3314 min_alloc_size, empty_size, hint_byte,
3315 search_end, ins, data);
3317 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3318 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3319 root_objectid, ref_generation,
3320 owner_objectid, ins);
3324 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3329 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root,
3331 u64 bytenr, u32 blocksize)
3333 struct extent_buffer *buf;
3335 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3337 return ERR_PTR(-ENOMEM);
3338 btrfs_set_header_generation(buf, trans->transid);
3339 btrfs_tree_lock(buf);
3340 clean_tree_block(trans, root, buf);
3341 btrfs_set_buffer_uptodate(buf);
3342 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3343 set_extent_dirty(&root->dirty_log_pages, buf->start,
3344 buf->start + buf->len - 1, GFP_NOFS);
3346 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3347 buf->start + buf->len - 1, GFP_NOFS);
3349 trans->blocks_used++;
3354 * helper function to allocate a block for a given tree
3355 * returns the tree buffer or NULL.
3357 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3358 struct btrfs_root *root,
3359 u32 blocksize, u64 parent,
3366 struct btrfs_key ins;
3368 struct extent_buffer *buf;
3370 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3371 root_objectid, ref_generation, level,
3372 empty_size, hint, (u64)-1, &ins, 0);
3375 return ERR_PTR(ret);
3378 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3382 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3383 struct btrfs_root *root, struct extent_buffer *leaf)
3386 u64 leaf_generation;
3387 struct btrfs_key key;
3388 struct btrfs_file_extent_item *fi;
3393 BUG_ON(!btrfs_is_leaf(leaf));
3394 nritems = btrfs_header_nritems(leaf);
3395 leaf_owner = btrfs_header_owner(leaf);
3396 leaf_generation = btrfs_header_generation(leaf);
3398 for (i = 0; i < nritems; i++) {
3402 btrfs_item_key_to_cpu(leaf, &key, i);
3403 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3405 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3406 if (btrfs_file_extent_type(leaf, fi) ==
3407 BTRFS_FILE_EXTENT_INLINE)
3410 * FIXME make sure to insert a trans record that
3411 * repeats the snapshot del on crash
3413 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3414 if (disk_bytenr == 0)
3417 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3418 btrfs_file_extent_disk_num_bytes(leaf, fi),
3419 leaf->start, leaf_owner, leaf_generation,
3423 atomic_inc(&root->fs_info->throttle_gen);
3424 wake_up(&root->fs_info->transaction_throttle);
3430 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3431 struct btrfs_root *root,
3432 struct btrfs_leaf_ref *ref)
3436 struct btrfs_extent_info *info = ref->extents;
3438 for (i = 0; i < ref->nritems; i++) {
3439 ret = __btrfs_free_extent(trans, root, info->bytenr,
3440 info->num_bytes, ref->bytenr,
3441 ref->owner, ref->generation,
3444 atomic_inc(&root->fs_info->throttle_gen);
3445 wake_up(&root->fs_info->transaction_throttle);
3455 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3460 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3463 #if 0 /* some debugging code in case we see problems here */
3464 /* if the refs count is one, it won't get increased again. But
3465 * if the ref count is > 1, someone may be decreasing it at
3466 * the same time we are.
3469 struct extent_buffer *eb = NULL;
3470 eb = btrfs_find_create_tree_block(root, start, len);
3472 btrfs_tree_lock(eb);
3474 mutex_lock(&root->fs_info->alloc_mutex);
3475 ret = lookup_extent_ref(NULL, root, start, len, refs);
3477 mutex_unlock(&root->fs_info->alloc_mutex);
3480 btrfs_tree_unlock(eb);
3481 free_extent_buffer(eb);
3484 printk(KERN_ERR "btrfs block %llu went down to one "
3485 "during drop_snap\n", (unsigned long long)start);
3496 * helper function for drop_snapshot, this walks down the tree dropping ref
3497 * counts as it goes.
3499 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3500 struct btrfs_root *root,
3501 struct btrfs_path *path, int *level)
3507 struct extent_buffer *next;
3508 struct extent_buffer *cur;
3509 struct extent_buffer *parent;
3510 struct btrfs_leaf_ref *ref;
3515 WARN_ON(*level < 0);
3516 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3517 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3518 path->nodes[*level]->len, &refs);
3524 * walk down to the last node level and free all the leaves
3526 while (*level >= 0) {
3527 WARN_ON(*level < 0);
3528 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3529 cur = path->nodes[*level];
3531 if (btrfs_header_level(cur) != *level)
3534 if (path->slots[*level] >=
3535 btrfs_header_nritems(cur))
3538 ret = btrfs_drop_leaf_ref(trans, root, cur);
3542 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3543 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3544 blocksize = btrfs_level_size(root, *level - 1);
3546 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3549 parent = path->nodes[*level];
3550 root_owner = btrfs_header_owner(parent);
3551 root_gen = btrfs_header_generation(parent);
3552 path->slots[*level]++;
3554 ret = __btrfs_free_extent(trans, root, bytenr,
3555 blocksize, parent->start,
3556 root_owner, root_gen,
3560 atomic_inc(&root->fs_info->throttle_gen);
3561 wake_up(&root->fs_info->transaction_throttle);
3567 * at this point, we have a single ref, and since the
3568 * only place referencing this extent is a dead root
3569 * the reference count should never go higher.
3570 * So, we don't need to check it again
3573 ref = btrfs_lookup_leaf_ref(root, bytenr);
3574 if (ref && ref->generation != ptr_gen) {
3575 btrfs_free_leaf_ref(root, ref);
3579 ret = cache_drop_leaf_ref(trans, root, ref);
3581 btrfs_remove_leaf_ref(root, ref);
3582 btrfs_free_leaf_ref(root, ref);
3587 next = btrfs_find_tree_block(root, bytenr, blocksize);
3588 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3589 free_extent_buffer(next);
3591 next = read_tree_block(root, bytenr, blocksize,
3596 * this is a debugging check and can go away
3597 * the ref should never go all the way down to 1
3600 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3606 WARN_ON(*level <= 0);
3607 if (path->nodes[*level-1])
3608 free_extent_buffer(path->nodes[*level-1]);
3609 path->nodes[*level-1] = next;
3610 *level = btrfs_header_level(next);
3611 path->slots[*level] = 0;
3615 WARN_ON(*level < 0);
3616 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3618 if (path->nodes[*level] == root->node) {
3619 parent = path->nodes[*level];
3620 bytenr = path->nodes[*level]->start;
3622 parent = path->nodes[*level + 1];
3623 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3626 blocksize = btrfs_level_size(root, *level);
3627 root_owner = btrfs_header_owner(parent);
3628 root_gen = btrfs_header_generation(parent);
3630 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3631 parent->start, root_owner, root_gen,
3633 free_extent_buffer(path->nodes[*level]);
3634 path->nodes[*level] = NULL;
3643 * helper function for drop_subtree, this function is similar to
3644 * walk_down_tree. The main difference is that it checks reference
3645 * counts while tree blocks are locked.
3647 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3648 struct btrfs_root *root,
3649 struct btrfs_path *path, int *level)
3651 struct extent_buffer *next;
3652 struct extent_buffer *cur;
3653 struct extent_buffer *parent;
3660 cur = path->nodes[*level];
3661 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3667 while (*level >= 0) {
3668 cur = path->nodes[*level];
3670 ret = btrfs_drop_leaf_ref(trans, root, cur);
3672 clean_tree_block(trans, root, cur);
3675 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3676 clean_tree_block(trans, root, cur);
3680 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3681 blocksize = btrfs_level_size(root, *level - 1);
3682 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3684 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3685 btrfs_tree_lock(next);
3687 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3691 parent = path->nodes[*level];
3692 ret = btrfs_free_extent(trans, root, bytenr,
3693 blocksize, parent->start,
3694 btrfs_header_owner(parent),
3695 btrfs_header_generation(parent),
3698 path->slots[*level]++;
3699 btrfs_tree_unlock(next);
3700 free_extent_buffer(next);
3704 *level = btrfs_header_level(next);
3705 path->nodes[*level] = next;
3706 path->slots[*level] = 0;
3707 path->locks[*level] = 1;
3711 parent = path->nodes[*level + 1];
3712 bytenr = path->nodes[*level]->start;
3713 blocksize = path->nodes[*level]->len;
3715 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3716 parent->start, btrfs_header_owner(parent),
3717 btrfs_header_generation(parent), *level, 1);
3720 if (path->locks[*level]) {
3721 btrfs_tree_unlock(path->nodes[*level]);
3722 path->locks[*level] = 0;
3724 free_extent_buffer(path->nodes[*level]);
3725 path->nodes[*level] = NULL;
3732 * helper for dropping snapshots. This walks back up the tree in the path
3733 * to find the first node higher up where we haven't yet gone through
3736 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3737 struct btrfs_root *root,
3738 struct btrfs_path *path,
3739 int *level, int max_level)
3743 struct btrfs_root_item *root_item = &root->root_item;
3748 for (i = *level; i < max_level && path->nodes[i]; i++) {
3749 slot = path->slots[i];
3750 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3751 struct extent_buffer *node;
3752 struct btrfs_disk_key disk_key;
3753 node = path->nodes[i];
3756 WARN_ON(*level == 0);
3757 btrfs_node_key(node, &disk_key, path->slots[i]);
3758 memcpy(&root_item->drop_progress,
3759 &disk_key, sizeof(disk_key));
3760 root_item->drop_level = i;
3763 struct extent_buffer *parent;
3764 if (path->nodes[*level] == root->node)
3765 parent = path->nodes[*level];
3767 parent = path->nodes[*level + 1];
3769 root_owner = btrfs_header_owner(parent);
3770 root_gen = btrfs_header_generation(parent);
3772 clean_tree_block(trans, root, path->nodes[*level]);
3773 ret = btrfs_free_extent(trans, root,
3774 path->nodes[*level]->start,
3775 path->nodes[*level]->len,
3776 parent->start, root_owner,
3777 root_gen, *level, 1);
3779 if (path->locks[*level]) {
3780 btrfs_tree_unlock(path->nodes[*level]);
3781 path->locks[*level] = 0;
3783 free_extent_buffer(path->nodes[*level]);
3784 path->nodes[*level] = NULL;
3792 * drop the reference count on the tree rooted at 'snap'. This traverses
3793 * the tree freeing any blocks that have a ref count of zero after being
3796 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3802 struct btrfs_path *path;
3805 struct btrfs_root_item *root_item = &root->root_item;
3807 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3808 path = btrfs_alloc_path();
3811 level = btrfs_header_level(root->node);
3813 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3814 path->nodes[level] = root->node;
3815 extent_buffer_get(root->node);
3816 path->slots[level] = 0;
3818 struct btrfs_key key;
3819 struct btrfs_disk_key found_key;
3820 struct extent_buffer *node;
3822 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3823 level = root_item->drop_level;
3824 path->lowest_level = level;
3825 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3830 node = path->nodes[level];
3831 btrfs_node_key(node, &found_key, path->slots[level]);
3832 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3833 sizeof(found_key)));
3835 * unlock our path, this is safe because only this
3836 * function is allowed to delete this snapshot
3838 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3839 if (path->nodes[i] && path->locks[i]) {
3841 btrfs_tree_unlock(path->nodes[i]);
3846 wret = walk_down_tree(trans, root, path, &level);
3852 wret = walk_up_tree(trans, root, path, &level,
3858 if (trans->transaction->in_commit) {
3862 atomic_inc(&root->fs_info->throttle_gen);
3863 wake_up(&root->fs_info->transaction_throttle);
3865 for (i = 0; i <= orig_level; i++) {
3866 if (path->nodes[i]) {
3867 free_extent_buffer(path->nodes[i]);
3868 path->nodes[i] = NULL;
3872 btrfs_free_path(path);
3876 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3877 struct btrfs_root *root,
3878 struct extent_buffer *node,
3879 struct extent_buffer *parent)
3881 struct btrfs_path *path;
3887 path = btrfs_alloc_path();
3890 BUG_ON(!btrfs_tree_locked(parent));
3891 parent_level = btrfs_header_level(parent);
3892 extent_buffer_get(parent);
3893 path->nodes[parent_level] = parent;
3894 path->slots[parent_level] = btrfs_header_nritems(parent);
3896 BUG_ON(!btrfs_tree_locked(node));
3897 level = btrfs_header_level(node);
3898 extent_buffer_get(node);
3899 path->nodes[level] = node;
3900 path->slots[level] = 0;
3903 wret = walk_down_subtree(trans, root, path, &level);
3909 wret = walk_up_tree(trans, root, path, &level, parent_level);
3916 btrfs_free_path(path);
3920 static unsigned long calc_ra(unsigned long start, unsigned long last,
3923 return min(last, start + nr - 1);
3926 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3931 unsigned long first_index;
3932 unsigned long last_index;
3935 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3936 struct file_ra_state *ra;
3937 struct btrfs_ordered_extent *ordered;
3938 unsigned int total_read = 0;
3939 unsigned int total_dirty = 0;
3942 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3944 mutex_lock(&inode->i_mutex);
3945 first_index = start >> PAGE_CACHE_SHIFT;
3946 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3948 /* make sure the dirty trick played by the caller work */
3949 ret = invalidate_inode_pages2_range(inode->i_mapping,
3950 first_index, last_index);
3954 file_ra_state_init(ra, inode->i_mapping);
3956 for (i = first_index ; i <= last_index; i++) {
3957 if (total_read % ra->ra_pages == 0) {
3958 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3959 calc_ra(i, last_index, ra->ra_pages));
3963 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3965 page = grab_cache_page(inode->i_mapping, i);
3970 if (!PageUptodate(page)) {
3971 btrfs_readpage(NULL, page);
3973 if (!PageUptodate(page)) {
3975 page_cache_release(page);
3980 wait_on_page_writeback(page);
3982 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3983 page_end = page_start + PAGE_CACHE_SIZE - 1;
3984 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3986 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3988 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3990 page_cache_release(page);
3991 btrfs_start_ordered_extent(inode, ordered, 1);
3992 btrfs_put_ordered_extent(ordered);
3995 set_page_extent_mapped(page);
3997 if (i == first_index)
3998 set_extent_bits(io_tree, page_start, page_end,
3999 EXTENT_BOUNDARY, GFP_NOFS);
4000 btrfs_set_extent_delalloc(inode, page_start, page_end);
4002 set_page_dirty(page);
4005 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4007 page_cache_release(page);
4012 mutex_unlock(&inode->i_mutex);
4013 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4017 static noinline int relocate_data_extent(struct inode *reloc_inode,
4018 struct btrfs_key *extent_key,
4021 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4022 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4023 struct extent_map *em;
4024 u64 start = extent_key->objectid - offset;
4025 u64 end = start + extent_key->offset - 1;
4027 em = alloc_extent_map(GFP_NOFS);
4028 BUG_ON(!em || IS_ERR(em));
4031 em->len = extent_key->offset;
4032 em->block_len = extent_key->offset;
4033 em->block_start = extent_key->objectid;
4034 em->bdev = root->fs_info->fs_devices->latest_bdev;
4035 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4037 /* setup extent map to cheat btrfs_readpage */
4038 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4041 spin_lock(&em_tree->lock);
4042 ret = add_extent_mapping(em_tree, em);
4043 spin_unlock(&em_tree->lock);
4044 if (ret != -EEXIST) {
4045 free_extent_map(em);
4048 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4050 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4052 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4055 struct btrfs_ref_path {
4057 u64 nodes[BTRFS_MAX_LEVEL];
4059 u64 root_generation;
4066 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4067 u64 new_nodes[BTRFS_MAX_LEVEL];
4070 struct disk_extent {
4081 static int is_cowonly_root(u64 root_objectid)
4083 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4084 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4085 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4086 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4087 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4088 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4093 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4094 struct btrfs_root *extent_root,
4095 struct btrfs_ref_path *ref_path,
4098 struct extent_buffer *leaf;
4099 struct btrfs_path *path;
4100 struct btrfs_extent_ref *ref;
4101 struct btrfs_key key;
4102 struct btrfs_key found_key;
4108 path = btrfs_alloc_path();
4113 ref_path->lowest_level = -1;
4114 ref_path->current_level = -1;
4115 ref_path->shared_level = -1;
4119 level = ref_path->current_level - 1;
4120 while (level >= -1) {
4122 if (level < ref_path->lowest_level)
4126 bytenr = ref_path->nodes[level];
4128 bytenr = ref_path->extent_start;
4129 BUG_ON(bytenr == 0);
4131 parent = ref_path->nodes[level + 1];
4132 ref_path->nodes[level + 1] = 0;
4133 ref_path->current_level = level;
4134 BUG_ON(parent == 0);
4136 key.objectid = bytenr;
4137 key.offset = parent + 1;
4138 key.type = BTRFS_EXTENT_REF_KEY;
4140 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4145 leaf = path->nodes[0];
4146 nritems = btrfs_header_nritems(leaf);
4147 if (path->slots[0] >= nritems) {
4148 ret = btrfs_next_leaf(extent_root, path);
4153 leaf = path->nodes[0];
4156 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4157 if (found_key.objectid == bytenr &&
4158 found_key.type == BTRFS_EXTENT_REF_KEY) {
4159 if (level < ref_path->shared_level)
4160 ref_path->shared_level = level;
4165 btrfs_release_path(extent_root, path);
4168 /* reached lowest level */
4172 level = ref_path->current_level;
4173 while (level < BTRFS_MAX_LEVEL - 1) {
4177 bytenr = ref_path->nodes[level];
4179 bytenr = ref_path->extent_start;
4181 BUG_ON(bytenr == 0);
4183 key.objectid = bytenr;
4185 key.type = BTRFS_EXTENT_REF_KEY;
4187 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4191 leaf = path->nodes[0];
4192 nritems = btrfs_header_nritems(leaf);
4193 if (path->slots[0] >= nritems) {
4194 ret = btrfs_next_leaf(extent_root, path);
4198 /* the extent was freed by someone */
4199 if (ref_path->lowest_level == level)
4201 btrfs_release_path(extent_root, path);
4204 leaf = path->nodes[0];
4207 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4208 if (found_key.objectid != bytenr ||
4209 found_key.type != BTRFS_EXTENT_REF_KEY) {
4210 /* the extent was freed by someone */
4211 if (ref_path->lowest_level == level) {
4215 btrfs_release_path(extent_root, path);
4219 ref = btrfs_item_ptr(leaf, path->slots[0],
4220 struct btrfs_extent_ref);
4221 ref_objectid = btrfs_ref_objectid(leaf, ref);
4222 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4224 level = (int)ref_objectid;
4225 BUG_ON(level >= BTRFS_MAX_LEVEL);
4226 ref_path->lowest_level = level;
4227 ref_path->current_level = level;
4228 ref_path->nodes[level] = bytenr;
4230 WARN_ON(ref_objectid != level);
4233 WARN_ON(level != -1);
4237 if (ref_path->lowest_level == level) {
4238 ref_path->owner_objectid = ref_objectid;
4239 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4243 * the block is tree root or the block isn't in reference
4246 if (found_key.objectid == found_key.offset ||
4247 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4248 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4249 ref_path->root_generation =
4250 btrfs_ref_generation(leaf, ref);
4252 /* special reference from the tree log */
4253 ref_path->nodes[0] = found_key.offset;
4254 ref_path->current_level = 0;
4261 BUG_ON(ref_path->nodes[level] != 0);
4262 ref_path->nodes[level] = found_key.offset;
4263 ref_path->current_level = level;
4266 * the reference was created in the running transaction,
4267 * no need to continue walking up.
4269 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4270 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4271 ref_path->root_generation =
4272 btrfs_ref_generation(leaf, ref);
4277 btrfs_release_path(extent_root, path);
4280 /* reached max tree level, but no tree root found. */
4283 btrfs_free_path(path);
4287 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4288 struct btrfs_root *extent_root,
4289 struct btrfs_ref_path *ref_path,
4292 memset(ref_path, 0, sizeof(*ref_path));
4293 ref_path->extent_start = extent_start;
4295 return __next_ref_path(trans, extent_root, ref_path, 1);
4298 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4299 struct btrfs_root *extent_root,
4300 struct btrfs_ref_path *ref_path)
4302 return __next_ref_path(trans, extent_root, ref_path, 0);
4305 static noinline int get_new_locations(struct inode *reloc_inode,
4306 struct btrfs_key *extent_key,
4307 u64 offset, int no_fragment,
4308 struct disk_extent **extents,
4311 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4312 struct btrfs_path *path;
4313 struct btrfs_file_extent_item *fi;
4314 struct extent_buffer *leaf;
4315 struct disk_extent *exts = *extents;
4316 struct btrfs_key found_key;
4321 int max = *nr_extents;
4324 WARN_ON(!no_fragment && *extents);
4327 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4332 path = btrfs_alloc_path();
4335 cur_pos = extent_key->objectid - offset;
4336 last_byte = extent_key->objectid + extent_key->offset;
4337 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4347 leaf = path->nodes[0];
4348 nritems = btrfs_header_nritems(leaf);
4349 if (path->slots[0] >= nritems) {
4350 ret = btrfs_next_leaf(root, path);
4355 leaf = path->nodes[0];
4358 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4359 if (found_key.offset != cur_pos ||
4360 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4361 found_key.objectid != reloc_inode->i_ino)
4364 fi = btrfs_item_ptr(leaf, path->slots[0],
4365 struct btrfs_file_extent_item);
4366 if (btrfs_file_extent_type(leaf, fi) !=
4367 BTRFS_FILE_EXTENT_REG ||
4368 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4372 struct disk_extent *old = exts;
4374 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4375 memcpy(exts, old, sizeof(*exts) * nr);
4376 if (old != *extents)
4380 exts[nr].disk_bytenr =
4381 btrfs_file_extent_disk_bytenr(leaf, fi);
4382 exts[nr].disk_num_bytes =
4383 btrfs_file_extent_disk_num_bytes(leaf, fi);
4384 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4385 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4386 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4387 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4388 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4389 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4391 BUG_ON(exts[nr].offset > 0);
4392 BUG_ON(exts[nr].compression || exts[nr].encryption);
4393 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4395 cur_pos += exts[nr].num_bytes;
4398 if (cur_pos + offset >= last_byte)
4408 BUG_ON(cur_pos + offset > last_byte);
4409 if (cur_pos + offset < last_byte) {
4415 btrfs_free_path(path);
4417 if (exts != *extents)
4426 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4427 struct btrfs_root *root,
4428 struct btrfs_path *path,
4429 struct btrfs_key *extent_key,
4430 struct btrfs_key *leaf_key,
4431 struct btrfs_ref_path *ref_path,
4432 struct disk_extent *new_extents,
4435 struct extent_buffer *leaf;
4436 struct btrfs_file_extent_item *fi;
4437 struct inode *inode = NULL;
4438 struct btrfs_key key;
4443 u64 search_end = (u64)-1;
4446 int extent_locked = 0;
4450 memcpy(&key, leaf_key, sizeof(key));
4451 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4452 if (key.objectid < ref_path->owner_objectid ||
4453 (key.objectid == ref_path->owner_objectid &&
4454 key.type < BTRFS_EXTENT_DATA_KEY)) {
4455 key.objectid = ref_path->owner_objectid;
4456 key.type = BTRFS_EXTENT_DATA_KEY;
4462 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4466 leaf = path->nodes[0];
4467 nritems = btrfs_header_nritems(leaf);
4469 if (extent_locked && ret > 0) {
4471 * the file extent item was modified by someone
4472 * before the extent got locked.
4474 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4475 lock_end, GFP_NOFS);
4479 if (path->slots[0] >= nritems) {
4480 if (++nr_scaned > 2)
4483 BUG_ON(extent_locked);
4484 ret = btrfs_next_leaf(root, path);
4489 leaf = path->nodes[0];
4490 nritems = btrfs_header_nritems(leaf);
4493 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4495 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4496 if ((key.objectid > ref_path->owner_objectid) ||
4497 (key.objectid == ref_path->owner_objectid &&
4498 key.type > BTRFS_EXTENT_DATA_KEY) ||
4499 key.offset >= search_end)
4503 if (inode && key.objectid != inode->i_ino) {
4504 BUG_ON(extent_locked);
4505 btrfs_release_path(root, path);
4506 mutex_unlock(&inode->i_mutex);
4512 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4517 fi = btrfs_item_ptr(leaf, path->slots[0],
4518 struct btrfs_file_extent_item);
4519 extent_type = btrfs_file_extent_type(leaf, fi);
4520 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4521 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4522 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4523 extent_key->objectid)) {
4529 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4530 ext_offset = btrfs_file_extent_offset(leaf, fi);
4532 if (search_end == (u64)-1) {
4533 search_end = key.offset - ext_offset +
4534 btrfs_file_extent_ram_bytes(leaf, fi);
4537 if (!extent_locked) {
4538 lock_start = key.offset;
4539 lock_end = lock_start + num_bytes - 1;
4541 if (lock_start > key.offset ||
4542 lock_end + 1 < key.offset + num_bytes) {
4543 unlock_extent(&BTRFS_I(inode)->io_tree,
4544 lock_start, lock_end, GFP_NOFS);
4550 btrfs_release_path(root, path);
4552 inode = btrfs_iget_locked(root->fs_info->sb,
4553 key.objectid, root);
4554 if (inode->i_state & I_NEW) {
4555 BTRFS_I(inode)->root = root;
4556 BTRFS_I(inode)->location.objectid =
4558 BTRFS_I(inode)->location.type =
4559 BTRFS_INODE_ITEM_KEY;
4560 BTRFS_I(inode)->location.offset = 0;
4561 btrfs_read_locked_inode(inode);
4562 unlock_new_inode(inode);
4565 * some code call btrfs_commit_transaction while
4566 * holding the i_mutex, so we can't use mutex_lock
4569 if (is_bad_inode(inode) ||
4570 !mutex_trylock(&inode->i_mutex)) {
4573 key.offset = (u64)-1;
4578 if (!extent_locked) {
4579 struct btrfs_ordered_extent *ordered;
4581 btrfs_release_path(root, path);
4583 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4584 lock_end, GFP_NOFS);
4585 ordered = btrfs_lookup_first_ordered_extent(inode,
4588 ordered->file_offset <= lock_end &&
4589 ordered->file_offset + ordered->len > lock_start) {
4590 unlock_extent(&BTRFS_I(inode)->io_tree,
4591 lock_start, lock_end, GFP_NOFS);
4592 btrfs_start_ordered_extent(inode, ordered, 1);
4593 btrfs_put_ordered_extent(ordered);
4594 key.offset += num_bytes;
4598 btrfs_put_ordered_extent(ordered);
4604 if (nr_extents == 1) {
4605 /* update extent pointer in place */
4606 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4607 new_extents[0].disk_bytenr);
4608 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4609 new_extents[0].disk_num_bytes);
4610 btrfs_mark_buffer_dirty(leaf);
4612 btrfs_drop_extent_cache(inode, key.offset,
4613 key.offset + num_bytes - 1, 0);
4615 ret = btrfs_inc_extent_ref(trans, root,
4616 new_extents[0].disk_bytenr,
4617 new_extents[0].disk_num_bytes,
4619 root->root_key.objectid,
4624 ret = btrfs_free_extent(trans, root,
4625 extent_key->objectid,
4628 btrfs_header_owner(leaf),
4629 btrfs_header_generation(leaf),
4633 btrfs_release_path(root, path);
4634 key.offset += num_bytes;
4642 * drop old extent pointer at first, then insert the
4643 * new pointers one bye one
4645 btrfs_release_path(root, path);
4646 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4647 key.offset + num_bytes,
4648 key.offset, &alloc_hint);
4651 for (i = 0; i < nr_extents; i++) {
4652 if (ext_offset >= new_extents[i].num_bytes) {
4653 ext_offset -= new_extents[i].num_bytes;
4656 extent_len = min(new_extents[i].num_bytes -
4657 ext_offset, num_bytes);
4659 ret = btrfs_insert_empty_item(trans, root,
4664 leaf = path->nodes[0];
4665 fi = btrfs_item_ptr(leaf, path->slots[0],
4666 struct btrfs_file_extent_item);
4667 btrfs_set_file_extent_generation(leaf, fi,
4669 btrfs_set_file_extent_type(leaf, fi,
4670 BTRFS_FILE_EXTENT_REG);
4671 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4672 new_extents[i].disk_bytenr);
4673 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4674 new_extents[i].disk_num_bytes);
4675 btrfs_set_file_extent_ram_bytes(leaf, fi,
4676 new_extents[i].ram_bytes);
4678 btrfs_set_file_extent_compression(leaf, fi,
4679 new_extents[i].compression);
4680 btrfs_set_file_extent_encryption(leaf, fi,
4681 new_extents[i].encryption);
4682 btrfs_set_file_extent_other_encoding(leaf, fi,
4683 new_extents[i].other_encoding);
4685 btrfs_set_file_extent_num_bytes(leaf, fi,
4687 ext_offset += new_extents[i].offset;
4688 btrfs_set_file_extent_offset(leaf, fi,
4690 btrfs_mark_buffer_dirty(leaf);
4692 btrfs_drop_extent_cache(inode, key.offset,
4693 key.offset + extent_len - 1, 0);
4695 ret = btrfs_inc_extent_ref(trans, root,
4696 new_extents[i].disk_bytenr,
4697 new_extents[i].disk_num_bytes,
4699 root->root_key.objectid,
4700 trans->transid, key.objectid);
4702 btrfs_release_path(root, path);
4704 inode_add_bytes(inode, extent_len);
4707 num_bytes -= extent_len;
4708 key.offset += extent_len;
4713 BUG_ON(i >= nr_extents);
4717 if (extent_locked) {
4718 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4719 lock_end, GFP_NOFS);
4723 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4724 key.offset >= search_end)
4731 btrfs_release_path(root, path);
4733 mutex_unlock(&inode->i_mutex);
4734 if (extent_locked) {
4735 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4736 lock_end, GFP_NOFS);
4743 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4744 struct btrfs_root *root,
4745 struct extent_buffer *buf, u64 orig_start)
4750 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4751 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4753 level = btrfs_header_level(buf);
4755 struct btrfs_leaf_ref *ref;
4756 struct btrfs_leaf_ref *orig_ref;
4758 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4762 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4764 btrfs_free_leaf_ref(root, orig_ref);
4768 ref->nritems = orig_ref->nritems;
4769 memcpy(ref->extents, orig_ref->extents,
4770 sizeof(ref->extents[0]) * ref->nritems);
4772 btrfs_free_leaf_ref(root, orig_ref);
4774 ref->root_gen = trans->transid;
4775 ref->bytenr = buf->start;
4776 ref->owner = btrfs_header_owner(buf);
4777 ref->generation = btrfs_header_generation(buf);
4778 ret = btrfs_add_leaf_ref(root, ref, 0);
4780 btrfs_free_leaf_ref(root, ref);
4785 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4786 struct extent_buffer *leaf,
4787 struct btrfs_block_group_cache *group,
4788 struct btrfs_root *target_root)
4790 struct btrfs_key key;
4791 struct inode *inode = NULL;
4792 struct btrfs_file_extent_item *fi;
4794 u64 skip_objectid = 0;
4798 nritems = btrfs_header_nritems(leaf);
4799 for (i = 0; i < nritems; i++) {
4800 btrfs_item_key_to_cpu(leaf, &key, i);
4801 if (key.objectid == skip_objectid ||
4802 key.type != BTRFS_EXTENT_DATA_KEY)
4804 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4805 if (btrfs_file_extent_type(leaf, fi) ==
4806 BTRFS_FILE_EXTENT_INLINE)
4808 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4810 if (!inode || inode->i_ino != key.objectid) {
4812 inode = btrfs_ilookup(target_root->fs_info->sb,
4813 key.objectid, target_root, 1);
4816 skip_objectid = key.objectid;
4819 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4821 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4822 key.offset + num_bytes - 1, GFP_NOFS);
4823 btrfs_drop_extent_cache(inode, key.offset,
4824 key.offset + num_bytes - 1, 1);
4825 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4826 key.offset + num_bytes - 1, GFP_NOFS);
4833 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4834 struct btrfs_root *root,
4835 struct extent_buffer *leaf,
4836 struct btrfs_block_group_cache *group,
4837 struct inode *reloc_inode)
4839 struct btrfs_key key;
4840 struct btrfs_key extent_key;
4841 struct btrfs_file_extent_item *fi;
4842 struct btrfs_leaf_ref *ref;
4843 struct disk_extent *new_extent;
4852 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4853 BUG_ON(!new_extent);
4855 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4859 nritems = btrfs_header_nritems(leaf);
4860 for (i = 0; i < nritems; i++) {
4861 btrfs_item_key_to_cpu(leaf, &key, i);
4862 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4864 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4865 if (btrfs_file_extent_type(leaf, fi) ==
4866 BTRFS_FILE_EXTENT_INLINE)
4868 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4869 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4874 if (bytenr >= group->key.objectid + group->key.offset ||
4875 bytenr + num_bytes <= group->key.objectid)
4878 extent_key.objectid = bytenr;
4879 extent_key.offset = num_bytes;
4880 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4882 ret = get_new_locations(reloc_inode, &extent_key,
4883 group->key.objectid, 1,
4884 &new_extent, &nr_extent);
4889 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4890 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4891 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4892 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4894 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4895 new_extent->disk_bytenr);
4896 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4897 new_extent->disk_num_bytes);
4898 btrfs_mark_buffer_dirty(leaf);
4900 ret = btrfs_inc_extent_ref(trans, root,
4901 new_extent->disk_bytenr,
4902 new_extent->disk_num_bytes,
4904 root->root_key.objectid,
4905 trans->transid, key.objectid);
4907 ret = btrfs_free_extent(trans, root,
4908 bytenr, num_bytes, leaf->start,
4909 btrfs_header_owner(leaf),
4910 btrfs_header_generation(leaf),
4916 BUG_ON(ext_index + 1 != ref->nritems);
4917 btrfs_free_leaf_ref(root, ref);
4921 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4922 struct btrfs_root *root)
4924 struct btrfs_root *reloc_root;
4927 if (root->reloc_root) {
4928 reloc_root = root->reloc_root;
4929 root->reloc_root = NULL;
4930 list_add(&reloc_root->dead_list,
4931 &root->fs_info->dead_reloc_roots);
4933 btrfs_set_root_bytenr(&reloc_root->root_item,
4934 reloc_root->node->start);
4935 btrfs_set_root_level(&root->root_item,
4936 btrfs_header_level(reloc_root->node));
4937 memset(&reloc_root->root_item.drop_progress, 0,
4938 sizeof(struct btrfs_disk_key));
4939 reloc_root->root_item.drop_level = 0;
4941 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4942 &reloc_root->root_key,
4943 &reloc_root->root_item);
4949 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4951 struct btrfs_trans_handle *trans;
4952 struct btrfs_root *reloc_root;
4953 struct btrfs_root *prev_root = NULL;
4954 struct list_head dead_roots;
4958 INIT_LIST_HEAD(&dead_roots);
4959 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4961 while (!list_empty(&dead_roots)) {
4962 reloc_root = list_entry(dead_roots.prev,
4963 struct btrfs_root, dead_list);
4964 list_del_init(&reloc_root->dead_list);
4966 BUG_ON(reloc_root->commit_root != NULL);
4968 trans = btrfs_join_transaction(root, 1);
4971 mutex_lock(&root->fs_info->drop_mutex);
4972 ret = btrfs_drop_snapshot(trans, reloc_root);
4975 mutex_unlock(&root->fs_info->drop_mutex);
4977 nr = trans->blocks_used;
4978 ret = btrfs_end_transaction(trans, root);
4980 btrfs_btree_balance_dirty(root, nr);
4983 free_extent_buffer(reloc_root->node);
4985 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4986 &reloc_root->root_key);
4988 mutex_unlock(&root->fs_info->drop_mutex);
4990 nr = trans->blocks_used;
4991 ret = btrfs_end_transaction(trans, root);
4993 btrfs_btree_balance_dirty(root, nr);
4996 prev_root = reloc_root;
4999 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5005 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5007 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5011 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5013 struct btrfs_root *reloc_root;
5014 struct btrfs_trans_handle *trans;
5015 struct btrfs_key location;
5019 mutex_lock(&root->fs_info->tree_reloc_mutex);
5020 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5022 found = !list_empty(&root->fs_info->dead_reloc_roots);
5023 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5026 trans = btrfs_start_transaction(root, 1);
5028 ret = btrfs_commit_transaction(trans, root);
5032 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5033 location.offset = (u64)-1;
5034 location.type = BTRFS_ROOT_ITEM_KEY;
5036 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5037 BUG_ON(!reloc_root);
5038 btrfs_orphan_cleanup(reloc_root);
5042 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5043 struct btrfs_root *root)
5045 struct btrfs_root *reloc_root;
5046 struct extent_buffer *eb;
5047 struct btrfs_root_item *root_item;
5048 struct btrfs_key root_key;
5051 BUG_ON(!root->ref_cows);
5052 if (root->reloc_root)
5055 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5058 ret = btrfs_copy_root(trans, root, root->commit_root,
5059 &eb, BTRFS_TREE_RELOC_OBJECTID);
5062 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5063 root_key.offset = root->root_key.objectid;
5064 root_key.type = BTRFS_ROOT_ITEM_KEY;
5066 memcpy(root_item, &root->root_item, sizeof(root_item));
5067 btrfs_set_root_refs(root_item, 0);
5068 btrfs_set_root_bytenr(root_item, eb->start);
5069 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5070 btrfs_set_root_generation(root_item, trans->transid);
5072 btrfs_tree_unlock(eb);
5073 free_extent_buffer(eb);
5075 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5076 &root_key, root_item);
5080 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5082 BUG_ON(!reloc_root);
5083 reloc_root->last_trans = trans->transid;
5084 reloc_root->commit_root = NULL;
5085 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5087 root->reloc_root = reloc_root;
5092 * Core function of space balance.
5094 * The idea is using reloc trees to relocate tree blocks in reference
5095 * counted roots. There is one reloc tree for each subvol, and all
5096 * reloc trees share same root key objectid. Reloc trees are snapshots
5097 * of the latest committed roots of subvols (root->commit_root).
5099 * To relocate a tree block referenced by a subvol, there are two steps.
5100 * COW the block through subvol's reloc tree, then update block pointer
5101 * in the subvol to point to the new block. Since all reloc trees share
5102 * same root key objectid, doing special handing for tree blocks owned
5103 * by them is easy. Once a tree block has been COWed in one reloc tree,
5104 * we can use the resulting new block directly when the same block is
5105 * required to COW again through other reloc trees. By this way, relocated
5106 * tree blocks are shared between reloc trees, so they are also shared
5109 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5110 struct btrfs_root *root,
5111 struct btrfs_path *path,
5112 struct btrfs_key *first_key,
5113 struct btrfs_ref_path *ref_path,
5114 struct btrfs_block_group_cache *group,
5115 struct inode *reloc_inode)
5117 struct btrfs_root *reloc_root;
5118 struct extent_buffer *eb = NULL;
5119 struct btrfs_key *keys;
5123 int lowest_level = 0;
5126 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5127 lowest_level = ref_path->owner_objectid;
5129 if (!root->ref_cows) {
5130 path->lowest_level = lowest_level;
5131 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5133 path->lowest_level = 0;
5134 btrfs_release_path(root, path);
5138 mutex_lock(&root->fs_info->tree_reloc_mutex);
5139 ret = init_reloc_tree(trans, root);
5141 reloc_root = root->reloc_root;
5143 shared_level = ref_path->shared_level;
5144 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5146 keys = ref_path->node_keys;
5147 nodes = ref_path->new_nodes;
5148 memset(&keys[shared_level + 1], 0,
5149 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5150 memset(&nodes[shared_level + 1], 0,
5151 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5153 if (nodes[lowest_level] == 0) {
5154 path->lowest_level = lowest_level;
5155 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5158 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5159 eb = path->nodes[level];
5160 if (!eb || eb == reloc_root->node)
5162 nodes[level] = eb->start;
5164 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5166 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5169 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5170 eb = path->nodes[0];
5171 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5172 group, reloc_inode);
5175 btrfs_release_path(reloc_root, path);
5177 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5183 * replace tree blocks in the fs tree with tree blocks in
5186 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5189 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5190 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5193 extent_buffer_get(path->nodes[0]);
5194 eb = path->nodes[0];
5195 btrfs_release_path(reloc_root, path);
5196 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5198 free_extent_buffer(eb);
5201 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5202 path->lowest_level = 0;
5206 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5207 struct btrfs_root *root,
5208 struct btrfs_path *path,
5209 struct btrfs_key *first_key,
5210 struct btrfs_ref_path *ref_path)
5214 ret = relocate_one_path(trans, root, path, first_key,
5215 ref_path, NULL, NULL);
5218 if (root == root->fs_info->extent_root)
5219 btrfs_extent_post_op(trans, root);
5224 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5225 struct btrfs_root *extent_root,
5226 struct btrfs_path *path,
5227 struct btrfs_key *extent_key)
5231 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5234 ret = btrfs_del_item(trans, extent_root, path);
5236 btrfs_release_path(extent_root, path);
5240 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5241 struct btrfs_ref_path *ref_path)
5243 struct btrfs_key root_key;
5245 root_key.objectid = ref_path->root_objectid;
5246 root_key.type = BTRFS_ROOT_ITEM_KEY;
5247 if (is_cowonly_root(ref_path->root_objectid))
5248 root_key.offset = 0;
5250 root_key.offset = (u64)-1;
5252 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5255 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5256 struct btrfs_path *path,
5257 struct btrfs_key *extent_key,
5258 struct btrfs_block_group_cache *group,
5259 struct inode *reloc_inode, int pass)
5261 struct btrfs_trans_handle *trans;
5262 struct btrfs_root *found_root;
5263 struct btrfs_ref_path *ref_path = NULL;
5264 struct disk_extent *new_extents = NULL;
5269 struct btrfs_key first_key;
5273 trans = btrfs_start_transaction(extent_root, 1);
5276 if (extent_key->objectid == 0) {
5277 ret = del_extent_zero(trans, extent_root, path, extent_key);
5281 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5287 for (loops = 0; ; loops++) {
5289 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5290 extent_key->objectid);
5292 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5299 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5300 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5303 found_root = read_ref_root(extent_root->fs_info, ref_path);
5304 BUG_ON(!found_root);
5306 * for reference counted tree, only process reference paths
5307 * rooted at the latest committed root.
5309 if (found_root->ref_cows &&
5310 ref_path->root_generation != found_root->root_key.offset)
5313 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5316 * copy data extents to new locations
5318 u64 group_start = group->key.objectid;
5319 ret = relocate_data_extent(reloc_inode,
5328 level = ref_path->owner_objectid;
5331 if (prev_block != ref_path->nodes[level]) {
5332 struct extent_buffer *eb;
5333 u64 block_start = ref_path->nodes[level];
5334 u64 block_size = btrfs_level_size(found_root, level);
5336 eb = read_tree_block(found_root, block_start,
5338 btrfs_tree_lock(eb);
5339 BUG_ON(level != btrfs_header_level(eb));
5342 btrfs_item_key_to_cpu(eb, &first_key, 0);
5344 btrfs_node_key_to_cpu(eb, &first_key, 0);
5346 btrfs_tree_unlock(eb);
5347 free_extent_buffer(eb);
5348 prev_block = block_start;
5351 btrfs_record_root_in_trans(found_root);
5352 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5354 * try to update data extent references while
5355 * keeping metadata shared between snapshots.
5358 ret = relocate_one_path(trans, found_root,
5359 path, &first_key, ref_path,
5360 group, reloc_inode);
5366 * use fallback method to process the remaining
5370 u64 group_start = group->key.objectid;
5371 new_extents = kmalloc(sizeof(*new_extents),
5374 ret = get_new_locations(reloc_inode,
5382 ret = replace_one_extent(trans, found_root,
5384 &first_key, ref_path,
5385 new_extents, nr_extents);
5387 ret = relocate_tree_block(trans, found_root, path,
5388 &first_key, ref_path);
5395 btrfs_end_transaction(trans, extent_root);
5401 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5404 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5405 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5407 num_devices = root->fs_info->fs_devices->rw_devices;
5408 if (num_devices == 1) {
5409 stripped |= BTRFS_BLOCK_GROUP_DUP;
5410 stripped = flags & ~stripped;
5412 /* turn raid0 into single device chunks */
5413 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5416 /* turn mirroring into duplication */
5417 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5418 BTRFS_BLOCK_GROUP_RAID10))
5419 return stripped | BTRFS_BLOCK_GROUP_DUP;
5422 /* they already had raid on here, just return */
5423 if (flags & stripped)
5426 stripped |= BTRFS_BLOCK_GROUP_DUP;
5427 stripped = flags & ~stripped;
5429 /* switch duplicated blocks with raid1 */
5430 if (flags & BTRFS_BLOCK_GROUP_DUP)
5431 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5433 /* turn single device chunks into raid0 */
5434 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5439 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5440 struct btrfs_block_group_cache *shrink_block_group,
5443 struct btrfs_trans_handle *trans;
5444 u64 new_alloc_flags;
5447 spin_lock(&shrink_block_group->lock);
5448 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5449 spin_unlock(&shrink_block_group->lock);
5451 trans = btrfs_start_transaction(root, 1);
5452 spin_lock(&shrink_block_group->lock);
5454 new_alloc_flags = update_block_group_flags(root,
5455 shrink_block_group->flags);
5456 if (new_alloc_flags != shrink_block_group->flags) {
5458 btrfs_block_group_used(&shrink_block_group->item);
5460 calc = shrink_block_group->key.offset;
5462 spin_unlock(&shrink_block_group->lock);
5464 do_chunk_alloc(trans, root->fs_info->extent_root,
5465 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5467 btrfs_end_transaction(trans, root);
5469 spin_unlock(&shrink_block_group->lock);
5473 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5474 struct btrfs_root *root,
5475 u64 objectid, u64 size)
5477 struct btrfs_path *path;
5478 struct btrfs_inode_item *item;
5479 struct extent_buffer *leaf;
5482 path = btrfs_alloc_path();
5486 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5490 leaf = path->nodes[0];
5491 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5492 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5493 btrfs_set_inode_generation(leaf, item, 1);
5494 btrfs_set_inode_size(leaf, item, size);
5495 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5496 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5497 btrfs_mark_buffer_dirty(leaf);
5498 btrfs_release_path(root, path);
5500 btrfs_free_path(path);
5504 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5505 struct btrfs_block_group_cache *group)
5507 struct inode *inode = NULL;
5508 struct btrfs_trans_handle *trans;
5509 struct btrfs_root *root;
5510 struct btrfs_key root_key;
5511 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5514 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5515 root_key.type = BTRFS_ROOT_ITEM_KEY;
5516 root_key.offset = (u64)-1;
5517 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5519 return ERR_CAST(root);
5521 trans = btrfs_start_transaction(root, 1);
5524 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5528 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5531 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5532 group->key.offset, 0, group->key.offset,
5536 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5537 if (inode->i_state & I_NEW) {
5538 BTRFS_I(inode)->root = root;
5539 BTRFS_I(inode)->location.objectid = objectid;
5540 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5541 BTRFS_I(inode)->location.offset = 0;
5542 btrfs_read_locked_inode(inode);
5543 unlock_new_inode(inode);
5544 BUG_ON(is_bad_inode(inode));
5548 BTRFS_I(inode)->index_cnt = group->key.objectid;
5550 err = btrfs_orphan_add(trans, inode);
5552 btrfs_end_transaction(trans, root);
5556 inode = ERR_PTR(err);
5561 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5564 struct btrfs_ordered_sum *sums;
5565 struct btrfs_sector_sum *sector_sum;
5566 struct btrfs_ordered_extent *ordered;
5567 struct btrfs_root *root = BTRFS_I(inode)->root;
5568 struct list_head list;
5573 INIT_LIST_HEAD(&list);
5575 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5576 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5578 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5579 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5580 disk_bytenr + len - 1, &list);
5582 while (!list_empty(&list)) {
5583 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5584 list_del_init(&sums->list);
5586 sector_sum = sums->sums;
5587 sums->bytenr = ordered->start;
5590 while (offset < sums->len) {
5591 sector_sum->bytenr += ordered->start - disk_bytenr;
5593 offset += root->sectorsize;
5596 btrfs_add_ordered_sum(inode, ordered, sums);
5598 btrfs_put_ordered_extent(ordered);
5602 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5604 struct btrfs_trans_handle *trans;
5605 struct btrfs_path *path;
5606 struct btrfs_fs_info *info = root->fs_info;
5607 struct extent_buffer *leaf;
5608 struct inode *reloc_inode;
5609 struct btrfs_block_group_cache *block_group;
5610 struct btrfs_key key;
5619 root = root->fs_info->extent_root;
5621 block_group = btrfs_lookup_block_group(info, group_start);
5622 BUG_ON(!block_group);
5624 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5625 (unsigned long long)block_group->key.objectid,
5626 (unsigned long long)block_group->flags);
5628 path = btrfs_alloc_path();
5631 reloc_inode = create_reloc_inode(info, block_group);
5632 BUG_ON(IS_ERR(reloc_inode));
5634 __alloc_chunk_for_shrink(root, block_group, 1);
5635 set_block_group_readonly(block_group);
5637 btrfs_start_delalloc_inodes(info->tree_root);
5638 btrfs_wait_ordered_extents(info->tree_root, 0);
5643 key.objectid = block_group->key.objectid;
5646 cur_byte = key.objectid;
5648 trans = btrfs_start_transaction(info->tree_root, 1);
5649 btrfs_commit_transaction(trans, info->tree_root);
5651 mutex_lock(&root->fs_info->cleaner_mutex);
5652 btrfs_clean_old_snapshots(info->tree_root);
5653 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5654 mutex_unlock(&root->fs_info->cleaner_mutex);
5657 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5661 leaf = path->nodes[0];
5662 nritems = btrfs_header_nritems(leaf);
5663 if (path->slots[0] >= nritems) {
5664 ret = btrfs_next_leaf(root, path);
5671 leaf = path->nodes[0];
5672 nritems = btrfs_header_nritems(leaf);
5675 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5677 if (key.objectid >= block_group->key.objectid +
5678 block_group->key.offset)
5681 if (progress && need_resched()) {
5682 btrfs_release_path(root, path);
5689 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5690 key.objectid + key.offset <= cur_byte) {
5696 cur_byte = key.objectid + key.offset;
5697 btrfs_release_path(root, path);
5699 __alloc_chunk_for_shrink(root, block_group, 0);
5700 ret = relocate_one_extent(root, path, &key, block_group,
5706 key.objectid = cur_byte;
5711 btrfs_release_path(root, path);
5714 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5715 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5718 if (total_found > 0) {
5719 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5720 (unsigned long long)total_found, pass);
5722 if (total_found == skipped && pass > 2) {
5724 reloc_inode = create_reloc_inode(info, block_group);
5730 /* delete reloc_inode */
5733 /* unpin extents in this range */
5734 trans = btrfs_start_transaction(info->tree_root, 1);
5735 btrfs_commit_transaction(trans, info->tree_root);
5737 spin_lock(&block_group->lock);
5738 WARN_ON(block_group->pinned > 0);
5739 WARN_ON(block_group->reserved > 0);
5740 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5741 spin_unlock(&block_group->lock);
5742 put_block_group(block_group);
5745 btrfs_free_path(path);
5749 static int find_first_block_group(struct btrfs_root *root,
5750 struct btrfs_path *path, struct btrfs_key *key)
5753 struct btrfs_key found_key;
5754 struct extent_buffer *leaf;
5757 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5762 slot = path->slots[0];
5763 leaf = path->nodes[0];
5764 if (slot >= btrfs_header_nritems(leaf)) {
5765 ret = btrfs_next_leaf(root, path);
5772 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5774 if (found_key.objectid >= key->objectid &&
5775 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5786 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5788 struct btrfs_block_group_cache *block_group;
5791 spin_lock(&info->block_group_cache_lock);
5792 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5793 block_group = rb_entry(n, struct btrfs_block_group_cache,
5795 rb_erase(&block_group->cache_node,
5796 &info->block_group_cache_tree);
5797 spin_unlock(&info->block_group_cache_lock);
5799 btrfs_remove_free_space_cache(block_group);
5800 down_write(&block_group->space_info->groups_sem);
5801 list_del(&block_group->list);
5802 up_write(&block_group->space_info->groups_sem);
5804 WARN_ON(atomic_read(&block_group->count) != 1);
5807 spin_lock(&info->block_group_cache_lock);
5809 spin_unlock(&info->block_group_cache_lock);
5813 int btrfs_read_block_groups(struct btrfs_root *root)
5815 struct btrfs_path *path;
5817 struct btrfs_block_group_cache *cache;
5818 struct btrfs_fs_info *info = root->fs_info;
5819 struct btrfs_space_info *space_info;
5820 struct btrfs_key key;
5821 struct btrfs_key found_key;
5822 struct extent_buffer *leaf;
5824 root = info->extent_root;
5827 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5828 path = btrfs_alloc_path();
5833 ret = find_first_block_group(root, path, &key);
5841 leaf = path->nodes[0];
5842 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5843 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5849 atomic_set(&cache->count, 1);
5850 spin_lock_init(&cache->lock);
5851 mutex_init(&cache->alloc_mutex);
5852 mutex_init(&cache->cache_mutex);
5853 INIT_LIST_HEAD(&cache->list);
5854 read_extent_buffer(leaf, &cache->item,
5855 btrfs_item_ptr_offset(leaf, path->slots[0]),
5856 sizeof(cache->item));
5857 memcpy(&cache->key, &found_key, sizeof(found_key));
5859 key.objectid = found_key.objectid + found_key.offset;
5860 btrfs_release_path(root, path);
5861 cache->flags = btrfs_block_group_flags(&cache->item);
5863 ret = update_space_info(info, cache->flags, found_key.offset,
5864 btrfs_block_group_used(&cache->item),
5867 cache->space_info = space_info;
5868 down_write(&space_info->groups_sem);
5869 list_add_tail(&cache->list, &space_info->block_groups);
5870 up_write(&space_info->groups_sem);
5872 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5875 set_avail_alloc_bits(root->fs_info, cache->flags);
5876 if (btrfs_chunk_readonly(root, cache->key.objectid))
5877 set_block_group_readonly(cache);
5881 btrfs_free_path(path);
5885 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5886 struct btrfs_root *root, u64 bytes_used,
5887 u64 type, u64 chunk_objectid, u64 chunk_offset,
5891 struct btrfs_root *extent_root;
5892 struct btrfs_block_group_cache *cache;
5894 extent_root = root->fs_info->extent_root;
5896 root->fs_info->last_trans_new_blockgroup = trans->transid;
5898 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5902 cache->key.objectid = chunk_offset;
5903 cache->key.offset = size;
5904 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5905 atomic_set(&cache->count, 1);
5906 spin_lock_init(&cache->lock);
5907 mutex_init(&cache->alloc_mutex);
5908 mutex_init(&cache->cache_mutex);
5909 INIT_LIST_HEAD(&cache->list);
5911 btrfs_set_block_group_used(&cache->item, bytes_used);
5912 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5913 cache->flags = type;
5914 btrfs_set_block_group_flags(&cache->item, type);
5916 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5917 &cache->space_info);
5919 down_write(&cache->space_info->groups_sem);
5920 list_add_tail(&cache->list, &cache->space_info->block_groups);
5921 up_write(&cache->space_info->groups_sem);
5923 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5926 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5927 sizeof(cache->item));
5930 finish_current_insert(trans, extent_root, 0);
5931 ret = del_pending_extents(trans, extent_root, 0);
5933 set_avail_alloc_bits(extent_root->fs_info, type);
5938 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5939 struct btrfs_root *root, u64 group_start)
5941 struct btrfs_path *path;
5942 struct btrfs_block_group_cache *block_group;
5943 struct btrfs_key key;
5946 root = root->fs_info->extent_root;
5948 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5949 BUG_ON(!block_group);
5950 BUG_ON(!block_group->ro);
5952 memcpy(&key, &block_group->key, sizeof(key));
5954 path = btrfs_alloc_path();
5957 spin_lock(&root->fs_info->block_group_cache_lock);
5958 rb_erase(&block_group->cache_node,
5959 &root->fs_info->block_group_cache_tree);
5960 spin_unlock(&root->fs_info->block_group_cache_lock);
5961 btrfs_remove_free_space_cache(block_group);
5962 down_write(&block_group->space_info->groups_sem);
5963 list_del(&block_group->list);
5964 up_write(&block_group->space_info->groups_sem);
5966 spin_lock(&block_group->space_info->lock);
5967 block_group->space_info->total_bytes -= block_group->key.offset;
5968 block_group->space_info->bytes_readonly -= block_group->key.offset;
5969 spin_unlock(&block_group->space_info->lock);
5970 block_group->space_info->full = 0;
5972 put_block_group(block_group);
5973 put_block_group(block_group);
5975 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5981 ret = btrfs_del_item(trans, root, path);
5983 btrfs_free_path(path);