2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/version.h>
28 #include "print-tree.h"
29 #include "transaction.h"
32 #include "ref-cache.h"
35 #define PENDING_EXTENT_INSERT 0
36 #define PENDING_EXTENT_DELETE 1
37 #define PENDING_BACKREF_UPDATE 2
39 struct pending_extent_op {
48 struct list_head list;
52 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
53 btrfs_root *extent_root, int all);
54 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
55 btrfs_root *extent_root, int all);
56 static int pin_down_bytes(struct btrfs_trans_handle *trans,
57 struct btrfs_root *root,
58 u64 bytenr, u64 num_bytes, int is_data);
59 static int update_block_group(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root,
61 u64 bytenr, u64 num_bytes, int alloc,
64 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
66 return (cache->flags & bits) == bits;
70 * this adds the block group to the fs_info rb tree for the block group
73 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
74 struct btrfs_block_group_cache *block_group)
77 struct rb_node *parent = NULL;
78 struct btrfs_block_group_cache *cache;
80 spin_lock(&info->block_group_cache_lock);
81 p = &info->block_group_cache_tree.rb_node;
85 cache = rb_entry(parent, struct btrfs_block_group_cache,
87 if (block_group->key.objectid < cache->key.objectid) {
89 } else if (block_group->key.objectid > cache->key.objectid) {
92 spin_unlock(&info->block_group_cache_lock);
97 rb_link_node(&block_group->cache_node, parent, p);
98 rb_insert_color(&block_group->cache_node,
99 &info->block_group_cache_tree);
100 spin_unlock(&info->block_group_cache_lock);
106 * This will return the block group at or after bytenr if contains is 0, else
107 * it will return the block group that contains the bytenr
109 static struct btrfs_block_group_cache *
110 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
113 struct btrfs_block_group_cache *cache, *ret = NULL;
117 spin_lock(&info->block_group_cache_lock);
118 n = info->block_group_cache_tree.rb_node;
121 cache = rb_entry(n, struct btrfs_block_group_cache,
123 end = cache->key.objectid + cache->key.offset - 1;
124 start = cache->key.objectid;
126 if (bytenr < start) {
127 if (!contains && (!ret || start < ret->key.objectid))
130 } else if (bytenr > start) {
131 if (contains && bytenr <= end) {
142 atomic_inc(&ret->count);
143 spin_unlock(&info->block_group_cache_lock);
149 * this is only called by cache_block_group, since we could have freed extents
150 * we need to check the pinned_extents for any extents that can't be used yet
151 * since their free space will be released as soon as the transaction commits.
153 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
154 struct btrfs_fs_info *info, u64 start, u64 end)
156 u64 extent_start, extent_end, size;
159 mutex_lock(&info->pinned_mutex);
160 while (start < end) {
161 ret = find_first_extent_bit(&info->pinned_extents, start,
162 &extent_start, &extent_end,
167 if (extent_start == start) {
168 start = extent_end + 1;
169 } else if (extent_start > start && extent_start < end) {
170 size = extent_start - start;
171 ret = btrfs_add_free_space(block_group, start,
174 start = extent_end + 1;
182 ret = btrfs_add_free_space(block_group, start, size);
185 mutex_unlock(&info->pinned_mutex);
190 static int remove_sb_from_cache(struct btrfs_root *root,
191 struct btrfs_block_group_cache *cache)
198 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
199 bytenr = btrfs_sb_offset(i);
200 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
201 cache->key.objectid, bytenr, 0,
202 &logical, &nr, &stripe_len);
205 btrfs_remove_free_space(cache, logical[nr],
213 static int cache_block_group(struct btrfs_root *root,
214 struct btrfs_block_group_cache *block_group)
216 struct btrfs_path *path;
218 struct btrfs_key key;
219 struct extent_buffer *leaf;
226 root = root->fs_info->extent_root;
228 if (block_group->cached)
231 path = btrfs_alloc_path();
237 * we get into deadlocks with paths held by callers of this function.
238 * since the alloc_mutex is protecting things right now, just
239 * skip the locking here
241 path->skip_locking = 1;
242 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
245 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
246 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
251 leaf = path->nodes[0];
252 slot = path->slots[0];
253 if (slot >= btrfs_header_nritems(leaf)) {
254 ret = btrfs_next_leaf(root, path);
262 btrfs_item_key_to_cpu(leaf, &key, slot);
263 if (key.objectid < block_group->key.objectid)
266 if (key.objectid >= block_group->key.objectid +
267 block_group->key.offset)
270 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
271 add_new_free_space(block_group, root->fs_info, last,
274 last = key.objectid + key.offset;
280 add_new_free_space(block_group, root->fs_info, last,
281 block_group->key.objectid +
282 block_group->key.offset);
284 remove_sb_from_cache(root, block_group);
285 block_group->cached = 1;
288 btrfs_free_path(path);
293 * return the block group that starts at or after bytenr
295 static struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
299 struct btrfs_block_group_cache *cache;
301 cache = block_group_cache_tree_search(info, bytenr, 0);
307 * return the block group that contains teh given bytenr
309 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
313 struct btrfs_block_group_cache *cache;
315 cache = block_group_cache_tree_search(info, bytenr, 1);
320 static inline void put_block_group(struct btrfs_block_group_cache *cache)
322 if (atomic_dec_and_test(&cache->count))
326 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
329 struct list_head *head = &info->space_info;
330 struct list_head *cur;
331 struct btrfs_space_info *found;
332 list_for_each(cur, head) {
333 found = list_entry(cur, struct btrfs_space_info, list);
334 if (found->flags == flags)
340 static u64 div_factor(u64 num, int factor)
349 u64 btrfs_find_block_group(struct btrfs_root *root,
350 u64 search_start, u64 search_hint, int owner)
352 struct btrfs_block_group_cache *cache;
354 u64 last = max(search_hint, search_start);
361 cache = btrfs_lookup_first_block_group(root->fs_info, last);
365 spin_lock(&cache->lock);
366 last = cache->key.objectid + cache->key.offset;
367 used = btrfs_block_group_used(&cache->item);
369 if ((full_search || !cache->ro) &&
370 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
371 if (used + cache->pinned + cache->reserved <
372 div_factor(cache->key.offset, factor)) {
373 group_start = cache->key.objectid;
374 spin_unlock(&cache->lock);
375 put_block_group(cache);
379 spin_unlock(&cache->lock);
380 put_block_group(cache);
388 if (!full_search && factor < 10) {
398 /* simple helper to search for an existing extent at a given offset */
399 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
402 struct btrfs_key key;
403 struct btrfs_path *path;
405 path = btrfs_alloc_path();
407 key.objectid = start;
409 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
410 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
412 btrfs_free_path(path);
417 * Back reference rules. Back refs have three main goals:
419 * 1) differentiate between all holders of references to an extent so that
420 * when a reference is dropped we can make sure it was a valid reference
421 * before freeing the extent.
423 * 2) Provide enough information to quickly find the holders of an extent
424 * if we notice a given block is corrupted or bad.
426 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
427 * maintenance. This is actually the same as #2, but with a slightly
428 * different use case.
430 * File extents can be referenced by:
432 * - multiple snapshots, subvolumes, or different generations in one subvol
433 * - different files inside a single subvolume
434 * - different offsets inside a file (bookend extents in file.c)
436 * The extent ref structure has fields for:
438 * - Objectid of the subvolume root
439 * - Generation number of the tree holding the reference
440 * - objectid of the file holding the reference
441 * - number of references holding by parent node (alway 1 for tree blocks)
443 * Btree leaf may hold multiple references to a file extent. In most cases,
444 * these references are from same file and the corresponding offsets inside
445 * the file are close together.
447 * When a file extent is allocated the fields are filled in:
448 * (root_key.objectid, trans->transid, inode objectid, 1)
450 * When a leaf is cow'd new references are added for every file extent found
451 * in the leaf. It looks similar to the create case, but trans->transid will
452 * be different when the block is cow'd.
454 * (root_key.objectid, trans->transid, inode objectid,
455 * number of references in the leaf)
457 * When a file extent is removed either during snapshot deletion or
458 * file truncation, we find the corresponding back reference and check
459 * the following fields:
461 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
464 * Btree extents can be referenced by:
466 * - Different subvolumes
467 * - Different generations of the same subvolume
469 * When a tree block is created, back references are inserted:
471 * (root->root_key.objectid, trans->transid, level, 1)
473 * When a tree block is cow'd, new back references are added for all the
474 * blocks it points to. If the tree block isn't in reference counted root,
475 * the old back references are removed. These new back references are of
476 * the form (trans->transid will have increased since creation):
478 * (root->root_key.objectid, trans->transid, level, 1)
480 * When a backref is in deleting, the following fields are checked:
482 * if backref was for a tree root:
483 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
485 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
487 * Back Reference Key composing:
489 * The key objectid corresponds to the first byte in the extent, the key
490 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
491 * byte of parent extent. If a extent is tree root, the key offset is set
492 * to the key objectid.
495 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
496 struct btrfs_root *root,
497 struct btrfs_path *path,
498 u64 bytenr, u64 parent,
499 u64 ref_root, u64 ref_generation,
500 u64 owner_objectid, int del)
502 struct btrfs_key key;
503 struct btrfs_extent_ref *ref;
504 struct extent_buffer *leaf;
508 key.objectid = bytenr;
509 key.type = BTRFS_EXTENT_REF_KEY;
512 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
520 leaf = path->nodes[0];
521 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
522 ref_objectid = btrfs_ref_objectid(leaf, ref);
523 if (btrfs_ref_root(leaf, ref) != ref_root ||
524 btrfs_ref_generation(leaf, ref) != ref_generation ||
525 (ref_objectid != owner_objectid &&
526 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
537 * updates all the backrefs that are pending on update_list for the
540 static int noinline update_backrefs(struct btrfs_trans_handle *trans,
541 struct btrfs_root *extent_root,
542 struct btrfs_path *path,
543 struct list_head *update_list)
545 struct btrfs_key key;
546 struct btrfs_extent_ref *ref;
547 struct btrfs_fs_info *info = extent_root->fs_info;
548 struct pending_extent_op *op;
549 struct extent_buffer *leaf;
551 struct list_head *cur = update_list->next;
553 u64 ref_root = extent_root->root_key.objectid;
555 op = list_entry(cur, struct pending_extent_op, list);
558 key.objectid = op->bytenr;
559 key.type = BTRFS_EXTENT_REF_KEY;
560 key.offset = op->orig_parent;
562 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
565 leaf = path->nodes[0];
568 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
570 ref_objectid = btrfs_ref_objectid(leaf, ref);
572 if (btrfs_ref_root(leaf, ref) != ref_root ||
573 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
574 (ref_objectid != op->level &&
575 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
576 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
577 "owner %u\n", op->bytenr, op->orig_parent,
578 ref_root, op->level);
579 btrfs_print_leaf(extent_root, leaf);
583 key.objectid = op->bytenr;
584 key.offset = op->parent;
585 key.type = BTRFS_EXTENT_REF_KEY;
586 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
588 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
589 btrfs_set_ref_generation(leaf, ref, op->generation);
593 list_del_init(&op->list);
594 unlock_extent(&info->extent_ins, op->bytenr,
595 op->bytenr + op->num_bytes - 1, GFP_NOFS);
598 if (cur == update_list) {
599 btrfs_mark_buffer_dirty(path->nodes[0]);
600 btrfs_release_path(extent_root, path);
604 op = list_entry(cur, struct pending_extent_op, list);
607 while (path->slots[0] < btrfs_header_nritems(leaf)) {
608 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
609 if (key.objectid == op->bytenr &&
610 key.type == BTRFS_EXTENT_REF_KEY)
615 btrfs_mark_buffer_dirty(path->nodes[0]);
616 btrfs_release_path(extent_root, path);
623 static int noinline insert_extents(struct btrfs_trans_handle *trans,
624 struct btrfs_root *extent_root,
625 struct btrfs_path *path,
626 struct list_head *insert_list, int nr)
628 struct btrfs_key *keys;
630 struct pending_extent_op *op;
631 struct extent_buffer *leaf;
632 struct list_head *cur = insert_list->next;
633 struct btrfs_fs_info *info = extent_root->fs_info;
634 u64 ref_root = extent_root->root_key.objectid;
635 int i = 0, last = 0, ret;
641 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
645 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
651 list_for_each_entry(op, insert_list, list) {
652 keys[i].objectid = op->bytenr;
653 keys[i].offset = op->num_bytes;
654 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
655 data_size[i] = sizeof(struct btrfs_extent_item);
658 keys[i].objectid = op->bytenr;
659 keys[i].offset = op->parent;
660 keys[i].type = BTRFS_EXTENT_REF_KEY;
661 data_size[i] = sizeof(struct btrfs_extent_ref);
665 op = list_entry(cur, struct pending_extent_op, list);
669 ret = btrfs_insert_some_items(trans, extent_root, path,
670 keys+i, data_size+i, total-i);
676 leaf = path->nodes[0];
677 for (c = 0; c < ret; c++) {
678 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
681 * if the first item we inserted was a backref, then
682 * the EXTENT_ITEM will be the odd c's, else it will
685 if ((ref_first && (c % 2)) ||
686 (!ref_first && !(c % 2))) {
687 struct btrfs_extent_item *itm;
689 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
690 struct btrfs_extent_item);
691 btrfs_set_extent_refs(path->nodes[0], itm, 1);
694 struct btrfs_extent_ref *ref;
696 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
697 struct btrfs_extent_ref);
698 btrfs_set_ref_root(leaf, ref, ref_root);
699 btrfs_set_ref_generation(leaf, ref,
701 btrfs_set_ref_objectid(leaf, ref, op->level);
702 btrfs_set_ref_num_refs(leaf, ref, 1);
707 * using del to see when its ok to free up the
708 * pending_extent_op. In the case where we insert the
709 * last item on the list in order to help do batching
710 * we need to not free the extent op until we actually
711 * insert the extent_item
714 unlock_extent(&info->extent_ins, op->bytenr,
715 op->bytenr + op->num_bytes - 1,
718 list_del_init(&op->list);
720 if (cur != insert_list)
722 struct pending_extent_op,
726 btrfs_mark_buffer_dirty(leaf);
727 btrfs_release_path(extent_root, path);
730 * Ok backref's and items usually go right next to eachother,
731 * but if we could only insert 1 item that means that we
732 * inserted on the end of a leaf, and we have no idea what may
733 * be on the next leaf so we just play it safe. In order to
734 * try and help this case we insert the last thing on our
735 * insert list so hopefully it will end up being the last
736 * thing on the leaf and everything else will be before it,
737 * which will let us insert a whole bunch of items at the same
740 if (ret == 1 && !last && (i + ret < total)) {
742 * last: where we will pick up the next time around
743 * i: our current key to insert, will be total - 1
744 * cur: the current op we are screwing with
749 cur = insert_list->prev;
750 op = list_entry(cur, struct pending_extent_op, list);
753 * ok we successfully inserted the last item on the
754 * list, lets reset everything
756 * i: our current key to insert, so where we left off
758 * last: done with this
759 * cur: the op we are messing with
761 * total: since we inserted the last key, we need to
762 * decrement total so we dont overflow
768 cur = insert_list->next;
769 op = list_entry(cur, struct pending_extent_op,
784 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
785 struct btrfs_root *root,
786 struct btrfs_path *path,
787 u64 bytenr, u64 parent,
788 u64 ref_root, u64 ref_generation,
791 struct btrfs_key key;
792 struct extent_buffer *leaf;
793 struct btrfs_extent_ref *ref;
797 key.objectid = bytenr;
798 key.type = BTRFS_EXTENT_REF_KEY;
801 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
803 leaf = path->nodes[0];
804 ref = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_ref);
806 btrfs_set_ref_root(leaf, ref, ref_root);
807 btrfs_set_ref_generation(leaf, ref, ref_generation);
808 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
809 btrfs_set_ref_num_refs(leaf, ref, 1);
810 } else if (ret == -EEXIST) {
812 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
813 leaf = path->nodes[0];
814 ref = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_extent_ref);
816 if (btrfs_ref_root(leaf, ref) != ref_root ||
817 btrfs_ref_generation(leaf, ref) != ref_generation) {
823 num_refs = btrfs_ref_num_refs(leaf, ref);
824 BUG_ON(num_refs == 0);
825 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
827 existing_owner = btrfs_ref_objectid(leaf, ref);
828 if (existing_owner != owner_objectid &&
829 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
830 btrfs_set_ref_objectid(leaf, ref,
831 BTRFS_MULTIPLE_OBJECTIDS);
837 btrfs_mark_buffer_dirty(path->nodes[0]);
839 btrfs_release_path(root, path);
843 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
844 struct btrfs_root *root,
845 struct btrfs_path *path)
847 struct extent_buffer *leaf;
848 struct btrfs_extent_ref *ref;
852 leaf = path->nodes[0];
853 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
854 num_refs = btrfs_ref_num_refs(leaf, ref);
855 BUG_ON(num_refs == 0);
858 ret = btrfs_del_item(trans, root, path);
860 btrfs_set_ref_num_refs(leaf, ref, num_refs);
861 btrfs_mark_buffer_dirty(leaf);
863 btrfs_release_path(root, path);
867 #ifdef BIO_RW_DISCARD
868 static void btrfs_issue_discard(struct block_device *bdev,
871 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
872 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
874 blkdev_issue_discard(bdev, start >> 9, len >> 9);
879 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
882 #ifdef BIO_RW_DISCARD
884 u64 map_length = num_bytes;
885 struct btrfs_multi_bio *multi = NULL;
887 /* Tell the block device(s) that the sectors can be discarded */
888 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
889 bytenr, &map_length, &multi, 0);
891 struct btrfs_bio_stripe *stripe = multi->stripes;
894 if (map_length > num_bytes)
895 map_length = num_bytes;
897 for (i = 0; i < multi->num_stripes; i++, stripe++) {
898 btrfs_issue_discard(stripe->dev->bdev,
911 static int noinline free_extents(struct btrfs_trans_handle *trans,
912 struct btrfs_root *extent_root,
913 struct list_head *del_list)
915 struct btrfs_fs_info *info = extent_root->fs_info;
916 struct btrfs_path *path;
917 struct btrfs_key key, found_key;
918 struct extent_buffer *leaf;
919 struct list_head *cur;
920 struct pending_extent_op *op;
921 struct btrfs_extent_item *ei;
922 int ret, num_to_del, extent_slot = 0, found_extent = 0;
926 path = btrfs_alloc_path();
932 /* search for the backref for the current ref we want to delete */
933 cur = del_list->next;
934 op = list_entry(cur, struct pending_extent_op, list);
935 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
937 extent_root->root_key.objectid,
938 op->orig_generation, op->level, 1);
940 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
941 "owner %u\n", op->bytenr,
942 extent_root->root_key.objectid, op->orig_generation,
944 btrfs_print_leaf(extent_root, path->nodes[0]);
949 extent_slot = path->slots[0];
954 * if we aren't the first item on the leaf we can move back one and see
955 * if our ref is right next to our extent item
957 if (likely(extent_slot)) {
959 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
961 if (found_key.objectid == op->bytenr &&
962 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
963 found_key.offset == op->num_bytes) {
970 * if we didn't find the extent we need to delete the backref and then
971 * search for the extent item key so we can update its ref count
974 key.objectid = op->bytenr;
975 key.type = BTRFS_EXTENT_ITEM_KEY;
976 key.offset = op->num_bytes;
978 ret = remove_extent_backref(trans, extent_root, path);
980 btrfs_release_path(extent_root, path);
981 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
983 extent_slot = path->slots[0];
986 /* this is where we update the ref count for the extent */
987 leaf = path->nodes[0];
988 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
989 refs = btrfs_extent_refs(leaf, ei);
992 btrfs_set_extent_refs(leaf, ei, refs);
994 btrfs_mark_buffer_dirty(leaf);
997 * This extent needs deleting. The reason cur_slot is extent_slot +
998 * num_to_del is because extent_slot points to the slot where the extent
999 * is, and if the backref was not right next to the extent we will be
1000 * deleting at least 1 item, and will want to start searching at the
1001 * slot directly next to extent_slot. However if we did find the
1002 * backref next to the extent item them we will be deleting at least 2
1003 * items and will want to start searching directly after the ref slot
1006 struct list_head *pos, *n, *end;
1007 int cur_slot = extent_slot+num_to_del;
1011 path->slots[0] = extent_slot;
1012 bytes_freed = op->num_bytes;
1014 mutex_lock(&info->pinned_mutex);
1015 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1016 op->num_bytes, op->level >=
1017 BTRFS_FIRST_FREE_OBJECTID);
1018 mutex_unlock(&info->pinned_mutex);
1023 * we need to see if we can delete multiple things at once, so
1024 * start looping through the list of extents we are wanting to
1025 * delete and see if their extent/backref's are right next to
1026 * eachother and the extents only have 1 ref
1028 for (pos = cur->next; pos != del_list; pos = pos->next) {
1029 struct pending_extent_op *tmp;
1031 tmp = list_entry(pos, struct pending_extent_op, list);
1033 /* we only want to delete extent+ref at this stage */
1034 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1037 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1038 if (found_key.objectid != tmp->bytenr ||
1039 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1040 found_key.offset != tmp->num_bytes)
1043 /* check to make sure this extent only has one ref */
1044 ei = btrfs_item_ptr(leaf, cur_slot,
1045 struct btrfs_extent_item);
1046 if (btrfs_extent_refs(leaf, ei) != 1)
1049 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1050 if (found_key.objectid != tmp->bytenr ||
1051 found_key.type != BTRFS_EXTENT_REF_KEY ||
1052 found_key.offset != tmp->orig_parent)
1056 * the ref is right next to the extent, we can set the
1057 * ref count to 0 since we will delete them both now
1059 btrfs_set_extent_refs(leaf, ei, 0);
1061 /* pin down the bytes for this extent */
1062 mutex_lock(&info->pinned_mutex);
1063 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1064 tmp->num_bytes, tmp->level >=
1065 BTRFS_FIRST_FREE_OBJECTID);
1066 mutex_unlock(&info->pinned_mutex);
1070 * use the del field to tell if we need to go ahead and
1071 * free up the extent when we delete the item or not.
1074 bytes_freed += tmp->num_bytes;
1081 /* update the free space counters */
1082 spin_lock(&info->delalloc_lock);
1083 super_used = btrfs_super_bytes_used(&info->super_copy);
1084 btrfs_set_super_bytes_used(&info->super_copy,
1085 super_used - bytes_freed);
1087 root_used = btrfs_root_used(&extent_root->root_item);
1088 btrfs_set_root_used(&extent_root->root_item,
1089 root_used - bytes_freed);
1090 spin_unlock(&info->delalloc_lock);
1092 /* delete the items */
1093 ret = btrfs_del_items(trans, extent_root, path,
1094 path->slots[0], num_to_del);
1098 * loop through the extents we deleted and do the cleanup work
1101 for (pos = cur, n = pos->next; pos != end;
1102 pos = n, n = pos->next) {
1103 struct pending_extent_op *tmp;
1104 tmp = list_entry(pos, struct pending_extent_op, list);
1107 * remember tmp->del tells us wether or not we pinned
1110 ret = update_block_group(trans, extent_root,
1111 tmp->bytenr, tmp->num_bytes, 0,
1115 list_del_init(&tmp->list);
1116 unlock_extent(&info->extent_ins, tmp->bytenr,
1117 tmp->bytenr + tmp->num_bytes - 1,
1121 } else if (refs && found_extent) {
1123 * the ref and extent were right next to eachother, but the
1124 * extent still has a ref, so just free the backref and keep
1127 ret = remove_extent_backref(trans, extent_root, path);
1130 list_del_init(&op->list);
1131 unlock_extent(&info->extent_ins, op->bytenr,
1132 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1136 * the extent has multiple refs and the backref we were looking
1137 * for was not right next to it, so just unlock and go next,
1140 list_del_init(&op->list);
1141 unlock_extent(&info->extent_ins, op->bytenr,
1142 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1146 btrfs_release_path(extent_root, path);
1147 if (!list_empty(del_list))
1151 btrfs_free_path(path);
1155 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1156 struct btrfs_root *root, u64 bytenr,
1157 u64 orig_parent, u64 parent,
1158 u64 orig_root, u64 ref_root,
1159 u64 orig_generation, u64 ref_generation,
1163 struct btrfs_root *extent_root = root->fs_info->extent_root;
1164 struct btrfs_path *path;
1166 if (root == root->fs_info->extent_root) {
1167 struct pending_extent_op *extent_op;
1170 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1171 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1172 mutex_lock(&root->fs_info->extent_ins_mutex);
1173 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1174 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1176 ret = get_state_private(&root->fs_info->extent_ins,
1179 extent_op = (struct pending_extent_op *)
1180 (unsigned long)priv;
1181 BUG_ON(extent_op->parent != orig_parent);
1182 BUG_ON(extent_op->generation != orig_generation);
1184 extent_op->parent = parent;
1185 extent_op->generation = ref_generation;
1187 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1190 extent_op->type = PENDING_BACKREF_UPDATE;
1191 extent_op->bytenr = bytenr;
1192 extent_op->num_bytes = num_bytes;
1193 extent_op->parent = parent;
1194 extent_op->orig_parent = orig_parent;
1195 extent_op->generation = ref_generation;
1196 extent_op->orig_generation = orig_generation;
1197 extent_op->level = (int)owner_objectid;
1198 INIT_LIST_HEAD(&extent_op->list);
1201 set_extent_bits(&root->fs_info->extent_ins,
1202 bytenr, bytenr + num_bytes - 1,
1203 EXTENT_WRITEBACK, GFP_NOFS);
1204 set_state_private(&root->fs_info->extent_ins,
1205 bytenr, (unsigned long)extent_op);
1207 mutex_unlock(&root->fs_info->extent_ins_mutex);
1211 path = btrfs_alloc_path();
1214 ret = lookup_extent_backref(trans, extent_root, path,
1215 bytenr, orig_parent, orig_root,
1216 orig_generation, owner_objectid, 1);
1219 ret = remove_extent_backref(trans, extent_root, path);
1222 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1223 parent, ref_root, ref_generation,
1226 finish_current_insert(trans, extent_root, 0);
1227 del_pending_extents(trans, extent_root, 0);
1229 btrfs_free_path(path);
1233 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1234 struct btrfs_root *root, u64 bytenr,
1235 u64 orig_parent, u64 parent,
1236 u64 ref_root, u64 ref_generation,
1240 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1241 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1243 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1244 parent, ref_root, ref_root,
1245 ref_generation, ref_generation,
1250 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1251 struct btrfs_root *root, u64 bytenr,
1252 u64 orig_parent, u64 parent,
1253 u64 orig_root, u64 ref_root,
1254 u64 orig_generation, u64 ref_generation,
1257 struct btrfs_path *path;
1259 struct btrfs_key key;
1260 struct extent_buffer *l;
1261 struct btrfs_extent_item *item;
1264 path = btrfs_alloc_path();
1269 key.objectid = bytenr;
1270 key.type = BTRFS_EXTENT_ITEM_KEY;
1271 key.offset = (u64)-1;
1273 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1277 BUG_ON(ret == 0 || path->slots[0] == 0);
1282 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1283 if (key.objectid != bytenr) {
1284 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1285 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1288 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1290 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1291 refs = btrfs_extent_refs(l, item);
1292 btrfs_set_extent_refs(l, item, refs + 1);
1293 btrfs_mark_buffer_dirty(path->nodes[0]);
1295 btrfs_release_path(root->fs_info->extent_root, path);
1298 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1299 path, bytenr, parent,
1300 ref_root, ref_generation,
1303 finish_current_insert(trans, root->fs_info->extent_root, 0);
1304 del_pending_extents(trans, root->fs_info->extent_root, 0);
1306 btrfs_free_path(path);
1310 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1311 struct btrfs_root *root,
1312 u64 bytenr, u64 num_bytes, u64 parent,
1313 u64 ref_root, u64 ref_generation,
1317 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1318 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1320 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1321 0, ref_root, 0, ref_generation,
1326 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1327 struct btrfs_root *root)
1329 finish_current_insert(trans, root->fs_info->extent_root, 1);
1330 del_pending_extents(trans, root->fs_info->extent_root, 1);
1334 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1335 struct btrfs_root *root, u64 bytenr,
1336 u64 num_bytes, u32 *refs)
1338 struct btrfs_path *path;
1340 struct btrfs_key key;
1341 struct extent_buffer *l;
1342 struct btrfs_extent_item *item;
1344 WARN_ON(num_bytes < root->sectorsize);
1345 path = btrfs_alloc_path();
1347 key.objectid = bytenr;
1348 key.offset = num_bytes;
1349 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1350 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1355 btrfs_print_leaf(root, path->nodes[0]);
1356 printk("failed to find block number %Lu\n", bytenr);
1360 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1361 *refs = btrfs_extent_refs(l, item);
1363 btrfs_free_path(path);
1367 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1368 struct btrfs_root *root, u64 objectid, u64 bytenr)
1370 struct btrfs_root *extent_root = root->fs_info->extent_root;
1371 struct btrfs_path *path;
1372 struct extent_buffer *leaf;
1373 struct btrfs_extent_ref *ref_item;
1374 struct btrfs_key key;
1375 struct btrfs_key found_key;
1381 key.objectid = bytenr;
1382 key.offset = (u64)-1;
1383 key.type = BTRFS_EXTENT_ITEM_KEY;
1385 path = btrfs_alloc_path();
1386 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1392 if (path->slots[0] == 0)
1396 leaf = path->nodes[0];
1397 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1399 if (found_key.objectid != bytenr ||
1400 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1403 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1405 leaf = path->nodes[0];
1406 nritems = btrfs_header_nritems(leaf);
1407 if (path->slots[0] >= nritems) {
1408 ret = btrfs_next_leaf(extent_root, path);
1415 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1416 if (found_key.objectid != bytenr)
1419 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1424 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1425 struct btrfs_extent_ref);
1426 ref_root = btrfs_ref_root(leaf, ref_item);
1427 if ((ref_root != root->root_key.objectid &&
1428 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1429 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1433 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1442 btrfs_free_path(path);
1446 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1447 struct extent_buffer *buf, u32 nr_extents)
1449 struct btrfs_key key;
1450 struct btrfs_file_extent_item *fi;
1458 if (!root->ref_cows)
1461 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1463 root_gen = root->root_key.offset;
1466 root_gen = trans->transid - 1;
1469 level = btrfs_header_level(buf);
1470 nritems = btrfs_header_nritems(buf);
1473 struct btrfs_leaf_ref *ref;
1474 struct btrfs_extent_info *info;
1476 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1482 ref->root_gen = root_gen;
1483 ref->bytenr = buf->start;
1484 ref->owner = btrfs_header_owner(buf);
1485 ref->generation = btrfs_header_generation(buf);
1486 ref->nritems = nr_extents;
1487 info = ref->extents;
1489 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1491 btrfs_item_key_to_cpu(buf, &key, i);
1492 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1494 fi = btrfs_item_ptr(buf, i,
1495 struct btrfs_file_extent_item);
1496 if (btrfs_file_extent_type(buf, fi) ==
1497 BTRFS_FILE_EXTENT_INLINE)
1499 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1500 if (disk_bytenr == 0)
1503 info->bytenr = disk_bytenr;
1505 btrfs_file_extent_disk_num_bytes(buf, fi);
1506 info->objectid = key.objectid;
1507 info->offset = key.offset;
1511 ret = btrfs_add_leaf_ref(root, ref, shared);
1512 if (ret == -EEXIST && shared) {
1513 struct btrfs_leaf_ref *old;
1514 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1516 btrfs_remove_leaf_ref(root, old);
1517 btrfs_free_leaf_ref(root, old);
1518 ret = btrfs_add_leaf_ref(root, ref, shared);
1521 btrfs_free_leaf_ref(root, ref);
1527 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1528 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1535 u64 orig_generation;
1537 u32 nr_file_extents = 0;
1538 struct btrfs_key key;
1539 struct btrfs_file_extent_item *fi;
1544 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1545 u64, u64, u64, u64, u64, u64, u64, u64);
1547 ref_root = btrfs_header_owner(buf);
1548 ref_generation = btrfs_header_generation(buf);
1549 orig_root = btrfs_header_owner(orig_buf);
1550 orig_generation = btrfs_header_generation(orig_buf);
1552 nritems = btrfs_header_nritems(buf);
1553 level = btrfs_header_level(buf);
1555 if (root->ref_cows) {
1556 process_func = __btrfs_inc_extent_ref;
1559 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1562 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1564 process_func = __btrfs_update_extent_ref;
1567 for (i = 0; i < nritems; i++) {
1570 btrfs_item_key_to_cpu(buf, &key, i);
1571 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1573 fi = btrfs_item_ptr(buf, i,
1574 struct btrfs_file_extent_item);
1575 if (btrfs_file_extent_type(buf, fi) ==
1576 BTRFS_FILE_EXTENT_INLINE)
1578 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1584 ret = process_func(trans, root, bytenr,
1585 orig_buf->start, buf->start,
1586 orig_root, ref_root,
1587 orig_generation, ref_generation,
1596 bytenr = btrfs_node_blockptr(buf, i);
1597 ret = process_func(trans, root, bytenr,
1598 orig_buf->start, buf->start,
1599 orig_root, ref_root,
1600 orig_generation, ref_generation,
1612 *nr_extents = nr_file_extents;
1614 *nr_extents = nritems;
1622 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1623 struct btrfs_root *root, struct extent_buffer *orig_buf,
1624 struct extent_buffer *buf, int start_slot, int nr)
1631 u64 orig_generation;
1632 struct btrfs_key key;
1633 struct btrfs_file_extent_item *fi;
1639 BUG_ON(start_slot < 0);
1640 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1642 ref_root = btrfs_header_owner(buf);
1643 ref_generation = btrfs_header_generation(buf);
1644 orig_root = btrfs_header_owner(orig_buf);
1645 orig_generation = btrfs_header_generation(orig_buf);
1646 level = btrfs_header_level(buf);
1648 if (!root->ref_cows) {
1650 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1653 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1657 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1660 btrfs_item_key_to_cpu(buf, &key, slot);
1661 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1663 fi = btrfs_item_ptr(buf, slot,
1664 struct btrfs_file_extent_item);
1665 if (btrfs_file_extent_type(buf, fi) ==
1666 BTRFS_FILE_EXTENT_INLINE)
1668 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1671 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1672 orig_buf->start, buf->start,
1673 orig_root, ref_root,
1674 orig_generation, ref_generation,
1679 bytenr = btrfs_node_blockptr(buf, slot);
1680 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1681 orig_buf->start, buf->start,
1682 orig_root, ref_root,
1683 orig_generation, ref_generation,
1695 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1696 struct btrfs_root *root,
1697 struct btrfs_path *path,
1698 struct btrfs_block_group_cache *cache)
1702 struct btrfs_root *extent_root = root->fs_info->extent_root;
1704 struct extent_buffer *leaf;
1706 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1711 leaf = path->nodes[0];
1712 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1713 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1714 btrfs_mark_buffer_dirty(leaf);
1715 btrfs_release_path(extent_root, path);
1717 finish_current_insert(trans, extent_root, 0);
1718 pending_ret = del_pending_extents(trans, extent_root, 0);
1727 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1728 struct btrfs_root *root)
1730 struct btrfs_block_group_cache *cache, *entry;
1734 struct btrfs_path *path;
1737 path = btrfs_alloc_path();
1743 spin_lock(&root->fs_info->block_group_cache_lock);
1744 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1745 n; n = rb_next(n)) {
1746 entry = rb_entry(n, struct btrfs_block_group_cache,
1753 spin_unlock(&root->fs_info->block_group_cache_lock);
1759 last += cache->key.offset;
1761 err = write_one_cache_group(trans, root,
1764 * if we fail to write the cache group, we want
1765 * to keep it marked dirty in hopes that a later
1773 btrfs_free_path(path);
1777 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1779 struct btrfs_block_group_cache *block_group;
1782 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1783 if (!block_group || block_group->ro)
1786 put_block_group(block_group);
1790 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1791 u64 total_bytes, u64 bytes_used,
1792 struct btrfs_space_info **space_info)
1794 struct btrfs_space_info *found;
1796 found = __find_space_info(info, flags);
1798 spin_lock(&found->lock);
1799 found->total_bytes += total_bytes;
1800 found->bytes_used += bytes_used;
1802 spin_unlock(&found->lock);
1803 *space_info = found;
1806 found = kzalloc(sizeof(*found), GFP_NOFS);
1810 list_add(&found->list, &info->space_info);
1811 INIT_LIST_HEAD(&found->block_groups);
1812 init_rwsem(&found->groups_sem);
1813 spin_lock_init(&found->lock);
1814 found->flags = flags;
1815 found->total_bytes = total_bytes;
1816 found->bytes_used = bytes_used;
1817 found->bytes_pinned = 0;
1818 found->bytes_reserved = 0;
1819 found->bytes_readonly = 0;
1821 found->force_alloc = 0;
1822 *space_info = found;
1826 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1828 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1829 BTRFS_BLOCK_GROUP_RAID1 |
1830 BTRFS_BLOCK_GROUP_RAID10 |
1831 BTRFS_BLOCK_GROUP_DUP);
1833 if (flags & BTRFS_BLOCK_GROUP_DATA)
1834 fs_info->avail_data_alloc_bits |= extra_flags;
1835 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1836 fs_info->avail_metadata_alloc_bits |= extra_flags;
1837 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1838 fs_info->avail_system_alloc_bits |= extra_flags;
1842 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1844 spin_lock(&cache->space_info->lock);
1845 spin_lock(&cache->lock);
1847 cache->space_info->bytes_readonly += cache->key.offset -
1848 btrfs_block_group_used(&cache->item);
1851 spin_unlock(&cache->lock);
1852 spin_unlock(&cache->space_info->lock);
1855 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1857 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1859 if (num_devices == 1)
1860 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1861 if (num_devices < 4)
1862 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1864 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1865 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1866 BTRFS_BLOCK_GROUP_RAID10))) {
1867 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1870 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1871 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1872 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1875 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1876 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1877 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1878 (flags & BTRFS_BLOCK_GROUP_DUP)))
1879 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1883 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1884 struct btrfs_root *extent_root, u64 alloc_bytes,
1885 u64 flags, int force)
1887 struct btrfs_space_info *space_info;
1891 mutex_lock(&extent_root->fs_info->chunk_mutex);
1893 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1895 space_info = __find_space_info(extent_root->fs_info, flags);
1897 ret = update_space_info(extent_root->fs_info, flags,
1901 BUG_ON(!space_info);
1903 spin_lock(&space_info->lock);
1904 if (space_info->force_alloc) {
1906 space_info->force_alloc = 0;
1908 if (space_info->full) {
1909 spin_unlock(&space_info->lock);
1913 thresh = space_info->total_bytes - space_info->bytes_readonly;
1914 thresh = div_factor(thresh, 6);
1916 (space_info->bytes_used + space_info->bytes_pinned +
1917 space_info->bytes_reserved + alloc_bytes) < thresh) {
1918 spin_unlock(&space_info->lock);
1921 spin_unlock(&space_info->lock);
1923 ret = btrfs_alloc_chunk(trans, extent_root, flags);
1925 printk("space info full %Lu\n", flags);
1926 space_info->full = 1;
1929 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1933 static int update_block_group(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root,
1935 u64 bytenr, u64 num_bytes, int alloc,
1938 struct btrfs_block_group_cache *cache;
1939 struct btrfs_fs_info *info = root->fs_info;
1940 u64 total = num_bytes;
1945 cache = btrfs_lookup_block_group(info, bytenr);
1948 byte_in_group = bytenr - cache->key.objectid;
1949 WARN_ON(byte_in_group > cache->key.offset);
1951 spin_lock(&cache->space_info->lock);
1952 spin_lock(&cache->lock);
1954 old_val = btrfs_block_group_used(&cache->item);
1955 num_bytes = min(total, cache->key.offset - byte_in_group);
1957 old_val += num_bytes;
1958 cache->space_info->bytes_used += num_bytes;
1960 cache->space_info->bytes_readonly -= num_bytes;
1961 btrfs_set_block_group_used(&cache->item, old_val);
1962 spin_unlock(&cache->lock);
1963 spin_unlock(&cache->space_info->lock);
1965 old_val -= num_bytes;
1966 cache->space_info->bytes_used -= num_bytes;
1968 cache->space_info->bytes_readonly += num_bytes;
1969 btrfs_set_block_group_used(&cache->item, old_val);
1970 spin_unlock(&cache->lock);
1971 spin_unlock(&cache->space_info->lock);
1975 ret = btrfs_discard_extent(root, bytenr,
1979 ret = btrfs_add_free_space(cache, bytenr,
1984 put_block_group(cache);
1986 bytenr += num_bytes;
1991 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1993 struct btrfs_block_group_cache *cache;
1996 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2000 bytenr = cache->key.objectid;
2001 put_block_group(cache);
2006 int btrfs_update_pinned_extents(struct btrfs_root *root,
2007 u64 bytenr, u64 num, int pin)
2010 struct btrfs_block_group_cache *cache;
2011 struct btrfs_fs_info *fs_info = root->fs_info;
2013 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2015 set_extent_dirty(&fs_info->pinned_extents,
2016 bytenr, bytenr + num - 1, GFP_NOFS);
2018 clear_extent_dirty(&fs_info->pinned_extents,
2019 bytenr, bytenr + num - 1, GFP_NOFS);
2022 cache = btrfs_lookup_block_group(fs_info, bytenr);
2024 len = min(num, cache->key.offset -
2025 (bytenr - cache->key.objectid));
2027 spin_lock(&cache->space_info->lock);
2028 spin_lock(&cache->lock);
2029 cache->pinned += len;
2030 cache->space_info->bytes_pinned += len;
2031 spin_unlock(&cache->lock);
2032 spin_unlock(&cache->space_info->lock);
2033 fs_info->total_pinned += len;
2035 spin_lock(&cache->space_info->lock);
2036 spin_lock(&cache->lock);
2037 cache->pinned -= len;
2038 cache->space_info->bytes_pinned -= len;
2039 spin_unlock(&cache->lock);
2040 spin_unlock(&cache->space_info->lock);
2041 fs_info->total_pinned -= len;
2043 btrfs_add_free_space(cache, bytenr, len);
2045 put_block_group(cache);
2052 static int update_reserved_extents(struct btrfs_root *root,
2053 u64 bytenr, u64 num, int reserve)
2056 struct btrfs_block_group_cache *cache;
2057 struct btrfs_fs_info *fs_info = root->fs_info;
2060 cache = btrfs_lookup_block_group(fs_info, bytenr);
2062 len = min(num, cache->key.offset -
2063 (bytenr - cache->key.objectid));
2065 spin_lock(&cache->space_info->lock);
2066 spin_lock(&cache->lock);
2068 cache->reserved += len;
2069 cache->space_info->bytes_reserved += len;
2071 cache->reserved -= len;
2072 cache->space_info->bytes_reserved -= len;
2074 spin_unlock(&cache->lock);
2075 spin_unlock(&cache->space_info->lock);
2076 put_block_group(cache);
2083 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2088 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2091 mutex_lock(&root->fs_info->pinned_mutex);
2093 ret = find_first_extent_bit(pinned_extents, last,
2094 &start, &end, EXTENT_DIRTY);
2097 set_extent_dirty(copy, start, end, GFP_NOFS);
2100 mutex_unlock(&root->fs_info->pinned_mutex);
2104 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root,
2106 struct extent_io_tree *unpin)
2112 mutex_lock(&root->fs_info->pinned_mutex);
2114 ret = find_first_extent_bit(unpin, 0, &start, &end,
2119 ret = btrfs_discard_extent(root, start, end + 1 - start);
2121 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2122 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2124 if (need_resched()) {
2125 mutex_unlock(&root->fs_info->pinned_mutex);
2127 mutex_lock(&root->fs_info->pinned_mutex);
2130 mutex_unlock(&root->fs_info->pinned_mutex);
2134 static int finish_current_insert(struct btrfs_trans_handle *trans,
2135 struct btrfs_root *extent_root, int all)
2142 struct btrfs_fs_info *info = extent_root->fs_info;
2143 struct btrfs_path *path;
2144 struct pending_extent_op *extent_op, *tmp;
2145 struct list_head insert_list, update_list;
2147 int num_inserts = 0, max_inserts;
2149 path = btrfs_alloc_path();
2150 INIT_LIST_HEAD(&insert_list);
2151 INIT_LIST_HEAD(&update_list);
2153 max_inserts = extent_root->leafsize /
2154 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2155 sizeof(struct btrfs_extent_ref) +
2156 sizeof(struct btrfs_extent_item));
2158 mutex_lock(&info->extent_ins_mutex);
2160 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2161 &end, EXTENT_WRITEBACK);
2163 if (skipped && all && !num_inserts) {
2168 mutex_unlock(&info->extent_ins_mutex);
2172 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2176 if (need_resched()) {
2177 mutex_unlock(&info->extent_ins_mutex);
2179 mutex_lock(&info->extent_ins_mutex);
2184 ret = get_state_private(&info->extent_ins, start, &priv);
2186 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2188 if (extent_op->type == PENDING_EXTENT_INSERT) {
2190 list_add_tail(&extent_op->list, &insert_list);
2192 if (num_inserts == max_inserts) {
2193 mutex_unlock(&info->extent_ins_mutex);
2196 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2197 list_add_tail(&extent_op->list, &update_list);
2205 * process the update list, clear the writeback bit for it, and if
2206 * somebody marked this thing for deletion then just unlock it and be
2207 * done, the free_extents will handle it
2209 mutex_lock(&info->extent_ins_mutex);
2210 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2211 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2212 extent_op->bytenr + extent_op->num_bytes - 1,
2213 EXTENT_WRITEBACK, GFP_NOFS);
2214 if (extent_op->del) {
2215 list_del_init(&extent_op->list);
2216 unlock_extent(&info->extent_ins, extent_op->bytenr,
2217 extent_op->bytenr + extent_op->num_bytes
2222 mutex_unlock(&info->extent_ins_mutex);
2225 * still have things left on the update list, go ahead an update
2228 if (!list_empty(&update_list)) {
2229 ret = update_backrefs(trans, extent_root, path, &update_list);
2234 * if no inserts need to be done, but we skipped some extents and we
2235 * need to make sure everything is cleaned then reset everything and
2236 * go back to the beginning
2238 if (!num_inserts && all && skipped) {
2241 INIT_LIST_HEAD(&update_list);
2242 INIT_LIST_HEAD(&insert_list);
2244 } else if (!num_inserts) {
2249 * process the insert extents list. Again if we are deleting this
2250 * extent, then just unlock it, pin down the bytes if need be, and be
2251 * done with it. Saves us from having to actually insert the extent
2252 * into the tree and then subsequently come along and delete it
2254 mutex_lock(&info->extent_ins_mutex);
2255 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2256 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2257 extent_op->bytenr + extent_op->num_bytes - 1,
2258 EXTENT_WRITEBACK, GFP_NOFS);
2259 if (extent_op->del) {
2261 list_del_init(&extent_op->list);
2262 unlock_extent(&info->extent_ins, extent_op->bytenr,
2263 extent_op->bytenr + extent_op->num_bytes
2266 mutex_lock(&extent_root->fs_info->pinned_mutex);
2267 ret = pin_down_bytes(trans, extent_root,
2269 extent_op->num_bytes, 0);
2270 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2272 spin_lock(&info->delalloc_lock);
2273 used = btrfs_super_bytes_used(&info->super_copy);
2274 btrfs_set_super_bytes_used(&info->super_copy,
2275 used - extent_op->num_bytes);
2276 used = btrfs_root_used(&extent_root->root_item);
2277 btrfs_set_root_used(&extent_root->root_item,
2278 used - extent_op->num_bytes);
2279 spin_unlock(&info->delalloc_lock);
2281 ret = update_block_group(trans, extent_root,
2283 extent_op->num_bytes,
2290 mutex_unlock(&info->extent_ins_mutex);
2292 ret = insert_extents(trans, extent_root, path, &insert_list,
2297 * if we broke out of the loop in order to insert stuff because we hit
2298 * the maximum number of inserts at a time we can handle, then loop
2299 * back and pick up where we left off
2301 if (num_inserts == max_inserts) {
2302 INIT_LIST_HEAD(&insert_list);
2303 INIT_LIST_HEAD(&update_list);
2309 * again, if we need to make absolutely sure there are no more pending
2310 * extent operations left and we know that we skipped some, go back to
2311 * the beginning and do it all again
2313 if (all && skipped) {
2314 INIT_LIST_HEAD(&insert_list);
2315 INIT_LIST_HEAD(&update_list);
2322 btrfs_free_path(path);
2326 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2327 struct btrfs_root *root,
2328 u64 bytenr, u64 num_bytes, int is_data)
2331 struct extent_buffer *buf;
2336 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2340 /* we can reuse a block if it hasn't been written
2341 * and it is from this transaction. We can't
2342 * reuse anything from the tree log root because
2343 * it has tiny sub-transactions.
2345 if (btrfs_buffer_uptodate(buf, 0) &&
2346 btrfs_try_tree_lock(buf)) {
2347 u64 header_owner = btrfs_header_owner(buf);
2348 u64 header_transid = btrfs_header_generation(buf);
2349 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2350 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2351 header_transid == trans->transid &&
2352 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2353 clean_tree_block(NULL, root, buf);
2354 btrfs_tree_unlock(buf);
2355 free_extent_buffer(buf);
2358 btrfs_tree_unlock(buf);
2360 free_extent_buffer(buf);
2362 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2369 * remove an extent from the root, returns 0 on success
2371 static int __free_extent(struct btrfs_trans_handle *trans,
2372 struct btrfs_root *root,
2373 u64 bytenr, u64 num_bytes, u64 parent,
2374 u64 root_objectid, u64 ref_generation,
2375 u64 owner_objectid, int pin, int mark_free)
2377 struct btrfs_path *path;
2378 struct btrfs_key key;
2379 struct btrfs_fs_info *info = root->fs_info;
2380 struct btrfs_root *extent_root = info->extent_root;
2381 struct extent_buffer *leaf;
2383 int extent_slot = 0;
2384 int found_extent = 0;
2386 struct btrfs_extent_item *ei;
2389 key.objectid = bytenr;
2390 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2391 key.offset = num_bytes;
2392 path = btrfs_alloc_path();
2397 ret = lookup_extent_backref(trans, extent_root, path,
2398 bytenr, parent, root_objectid,
2399 ref_generation, owner_objectid, 1);
2401 struct btrfs_key found_key;
2402 extent_slot = path->slots[0];
2403 while(extent_slot > 0) {
2405 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2407 if (found_key.objectid != bytenr)
2409 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2410 found_key.offset == num_bytes) {
2414 if (path->slots[0] - extent_slot > 5)
2417 if (!found_extent) {
2418 ret = remove_extent_backref(trans, extent_root, path);
2420 btrfs_release_path(extent_root, path);
2421 ret = btrfs_search_slot(trans, extent_root,
2424 printk(KERN_ERR "umm, got %d back from search"
2425 ", was looking for %Lu\n", ret,
2427 btrfs_print_leaf(extent_root, path->nodes[0]);
2430 extent_slot = path->slots[0];
2433 btrfs_print_leaf(extent_root, path->nodes[0]);
2435 printk("Unable to find ref byte nr %Lu root %Lu "
2436 "gen %Lu owner %Lu\n", bytenr,
2437 root_objectid, ref_generation, owner_objectid);
2440 leaf = path->nodes[0];
2441 ei = btrfs_item_ptr(leaf, extent_slot,
2442 struct btrfs_extent_item);
2443 refs = btrfs_extent_refs(leaf, ei);
2446 btrfs_set_extent_refs(leaf, ei, refs);
2448 btrfs_mark_buffer_dirty(leaf);
2450 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2451 struct btrfs_extent_ref *ref;
2452 ref = btrfs_item_ptr(leaf, path->slots[0],
2453 struct btrfs_extent_ref);
2454 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2455 /* if the back ref and the extent are next to each other
2456 * they get deleted below in one shot
2458 path->slots[0] = extent_slot;
2460 } else if (found_extent) {
2461 /* otherwise delete the extent back ref */
2462 ret = remove_extent_backref(trans, extent_root, path);
2464 /* if refs are 0, we need to setup the path for deletion */
2466 btrfs_release_path(extent_root, path);
2467 ret = btrfs_search_slot(trans, extent_root, &key, path,
2478 mutex_lock(&root->fs_info->pinned_mutex);
2479 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2480 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2481 mutex_unlock(&root->fs_info->pinned_mutex);
2486 /* block accounting for super block */
2487 spin_lock(&info->delalloc_lock);
2488 super_used = btrfs_super_bytes_used(&info->super_copy);
2489 btrfs_set_super_bytes_used(&info->super_copy,
2490 super_used - num_bytes);
2492 /* block accounting for root item */
2493 root_used = btrfs_root_used(&root->root_item);
2494 btrfs_set_root_used(&root->root_item,
2495 root_used - num_bytes);
2496 spin_unlock(&info->delalloc_lock);
2497 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2500 btrfs_release_path(extent_root, path);
2502 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2503 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2507 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2511 btrfs_free_path(path);
2512 finish_current_insert(trans, extent_root, 0);
2517 * find all the blocks marked as pending in the radix tree and remove
2518 * them from the extent map
2520 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2521 btrfs_root *extent_root, int all)
2529 int nr = 0, skipped = 0;
2530 struct extent_io_tree *pending_del;
2531 struct extent_io_tree *extent_ins;
2532 struct pending_extent_op *extent_op;
2533 struct btrfs_fs_info *info = extent_root->fs_info;
2534 struct list_head delete_list;
2536 INIT_LIST_HEAD(&delete_list);
2537 extent_ins = &extent_root->fs_info->extent_ins;
2538 pending_del = &extent_root->fs_info->pending_del;
2541 mutex_lock(&info->extent_ins_mutex);
2543 ret = find_first_extent_bit(pending_del, search, &start, &end,
2546 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 int noinline 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 "we were searching for %Lu bytes, num_bytes %Lu,"
2999 " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
3000 loop, allowed_chunk_alloc);
3004 put_block_group(block_group);
3006 up_read(&space_info->groups_sem);
3010 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3012 struct btrfs_block_group_cache *cache;
3013 struct list_head *l;
3015 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
3016 info->total_bytes - info->bytes_used - info->bytes_pinned -
3017 info->bytes_reserved, (info->full) ? "" : "not ");
3019 down_read(&info->groups_sem);
3020 list_for_each(l, &info->block_groups) {
3021 cache = list_entry(l, struct btrfs_block_group_cache, list);
3022 spin_lock(&cache->lock);
3023 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
3024 "%Lu pinned %Lu reserved\n",
3025 cache->key.objectid, cache->key.offset,
3026 btrfs_block_group_used(&cache->item),
3027 cache->pinned, cache->reserved);
3028 btrfs_dump_free_space(cache, bytes);
3029 spin_unlock(&cache->lock);
3031 up_read(&info->groups_sem);
3034 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3035 struct btrfs_root *root,
3036 u64 num_bytes, u64 min_alloc_size,
3037 u64 empty_size, u64 hint_byte,
3038 u64 search_end, struct btrfs_key *ins,
3042 u64 search_start = 0;
3044 struct btrfs_fs_info *info = root->fs_info;
3047 alloc_profile = info->avail_data_alloc_bits &
3048 info->data_alloc_profile;
3049 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3050 } else if (root == root->fs_info->chunk_root) {
3051 alloc_profile = info->avail_system_alloc_bits &
3052 info->system_alloc_profile;
3053 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3055 alloc_profile = info->avail_metadata_alloc_bits &
3056 info->metadata_alloc_profile;
3057 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3060 data = btrfs_reduce_alloc_profile(root, data);
3062 * the only place that sets empty_size is btrfs_realloc_node, which
3063 * is not called recursively on allocations
3065 if (empty_size || root->ref_cows) {
3066 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3067 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3069 BTRFS_BLOCK_GROUP_METADATA |
3070 (info->metadata_alloc_profile &
3071 info->avail_metadata_alloc_bits), 0);
3073 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3074 num_bytes + 2 * 1024 * 1024, data, 0);
3077 WARN_ON(num_bytes < root->sectorsize);
3078 ret = find_free_extent(trans, root, num_bytes, empty_size,
3079 search_start, search_end, hint_byte, ins,
3080 trans->alloc_exclude_start,
3081 trans->alloc_exclude_nr, data);
3083 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3084 num_bytes = num_bytes >> 1;
3085 num_bytes = num_bytes & ~(root->sectorsize - 1);
3086 num_bytes = max(num_bytes, min_alloc_size);
3087 do_chunk_alloc(trans, root->fs_info->extent_root,
3088 num_bytes, data, 1);
3092 struct btrfs_space_info *sinfo;
3094 sinfo = __find_space_info(root->fs_info, data);
3095 printk("allocation failed flags %Lu, wanted %Lu\n",
3097 dump_space_info(sinfo, num_bytes);
3104 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3106 struct btrfs_block_group_cache *cache;
3109 cache = btrfs_lookup_block_group(root->fs_info, start);
3111 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
3115 ret = btrfs_discard_extent(root, start, len);
3117 btrfs_add_free_space(cache, start, len);
3118 put_block_group(cache);
3119 update_reserved_extents(root, start, len, 0);
3124 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3125 struct btrfs_root *root,
3126 u64 num_bytes, u64 min_alloc_size,
3127 u64 empty_size, u64 hint_byte,
3128 u64 search_end, struct btrfs_key *ins,
3132 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3133 empty_size, hint_byte, search_end, ins,
3135 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3139 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3140 struct btrfs_root *root, u64 parent,
3141 u64 root_objectid, u64 ref_generation,
3142 u64 owner, struct btrfs_key *ins)
3148 u64 num_bytes = ins->offset;
3150 struct btrfs_fs_info *info = root->fs_info;
3151 struct btrfs_root *extent_root = info->extent_root;
3152 struct btrfs_extent_item *extent_item;
3153 struct btrfs_extent_ref *ref;
3154 struct btrfs_path *path;
3155 struct btrfs_key keys[2];
3158 parent = ins->objectid;
3160 /* block accounting for super block */
3161 spin_lock(&info->delalloc_lock);
3162 super_used = btrfs_super_bytes_used(&info->super_copy);
3163 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3165 /* block accounting for root item */
3166 root_used = btrfs_root_used(&root->root_item);
3167 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3168 spin_unlock(&info->delalloc_lock);
3170 if (root == extent_root) {
3171 struct pending_extent_op *extent_op;
3173 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3176 extent_op->type = PENDING_EXTENT_INSERT;
3177 extent_op->bytenr = ins->objectid;
3178 extent_op->num_bytes = ins->offset;
3179 extent_op->parent = parent;
3180 extent_op->orig_parent = 0;
3181 extent_op->generation = ref_generation;
3182 extent_op->orig_generation = 0;
3183 extent_op->level = (int)owner;
3184 INIT_LIST_HEAD(&extent_op->list);
3187 mutex_lock(&root->fs_info->extent_ins_mutex);
3188 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3189 ins->objectid + ins->offset - 1,
3190 EXTENT_WRITEBACK, GFP_NOFS);
3191 set_state_private(&root->fs_info->extent_ins,
3192 ins->objectid, (unsigned long)extent_op);
3193 mutex_unlock(&root->fs_info->extent_ins_mutex);
3197 memcpy(&keys[0], ins, sizeof(*ins));
3198 keys[1].objectid = ins->objectid;
3199 keys[1].type = BTRFS_EXTENT_REF_KEY;
3200 keys[1].offset = parent;
3201 sizes[0] = sizeof(*extent_item);
3202 sizes[1] = sizeof(*ref);
3204 path = btrfs_alloc_path();
3207 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3211 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3212 struct btrfs_extent_item);
3213 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3214 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3215 struct btrfs_extent_ref);
3217 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3218 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3219 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3220 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3222 btrfs_mark_buffer_dirty(path->nodes[0]);
3224 trans->alloc_exclude_start = 0;
3225 trans->alloc_exclude_nr = 0;
3226 btrfs_free_path(path);
3227 finish_current_insert(trans, extent_root, 0);
3228 pending_ret = del_pending_extents(trans, extent_root, 0);
3238 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
3240 printk("update block group failed for %Lu %Lu\n",
3241 ins->objectid, ins->offset);
3248 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3249 struct btrfs_root *root, u64 parent,
3250 u64 root_objectid, u64 ref_generation,
3251 u64 owner, struct btrfs_key *ins)
3255 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3257 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3258 ref_generation, owner, ins);
3259 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3264 * this is used by the tree logging recovery code. It records that
3265 * an extent has been allocated and makes sure to clear the free
3266 * space cache bits as well
3268 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3269 struct btrfs_root *root, u64 parent,
3270 u64 root_objectid, u64 ref_generation,
3271 u64 owner, struct btrfs_key *ins)
3274 struct btrfs_block_group_cache *block_group;
3276 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3277 mutex_lock(&block_group->cache_mutex);
3278 cache_block_group(root, block_group);
3279 mutex_unlock(&block_group->cache_mutex);
3281 ret = btrfs_remove_free_space(block_group, ins->objectid,
3284 put_block_group(block_group);
3285 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3286 ref_generation, owner, ins);
3291 * finds a free extent and does all the dirty work required for allocation
3292 * returns the key for the extent through ins, and a tree buffer for
3293 * the first block of the extent through buf.
3295 * returns 0 if everything worked, non-zero otherwise.
3297 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3298 struct btrfs_root *root,
3299 u64 num_bytes, u64 parent, u64 min_alloc_size,
3300 u64 root_objectid, u64 ref_generation,
3301 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3302 u64 search_end, struct btrfs_key *ins, u64 data)
3306 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3307 min_alloc_size, empty_size, hint_byte,
3308 search_end, ins, data);
3310 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3311 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3312 root_objectid, ref_generation,
3313 owner_objectid, ins);
3317 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3322 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3323 struct btrfs_root *root,
3324 u64 bytenr, u32 blocksize)
3326 struct extent_buffer *buf;
3328 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3330 return ERR_PTR(-ENOMEM);
3331 btrfs_set_header_generation(buf, trans->transid);
3332 btrfs_tree_lock(buf);
3333 clean_tree_block(trans, root, buf);
3334 btrfs_set_buffer_uptodate(buf);
3335 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3336 set_extent_dirty(&root->dirty_log_pages, buf->start,
3337 buf->start + buf->len - 1, GFP_NOFS);
3339 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3340 buf->start + buf->len - 1, GFP_NOFS);
3342 trans->blocks_used++;
3347 * helper function to allocate a block for a given tree
3348 * returns the tree buffer or NULL.
3350 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3351 struct btrfs_root *root,
3352 u32 blocksize, u64 parent,
3359 struct btrfs_key ins;
3361 struct extent_buffer *buf;
3363 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3364 root_objectid, ref_generation, level,
3365 empty_size, hint, (u64)-1, &ins, 0);
3368 return ERR_PTR(ret);
3371 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3375 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3376 struct btrfs_root *root, struct extent_buffer *leaf)
3379 u64 leaf_generation;
3380 struct btrfs_key key;
3381 struct btrfs_file_extent_item *fi;
3386 BUG_ON(!btrfs_is_leaf(leaf));
3387 nritems = btrfs_header_nritems(leaf);
3388 leaf_owner = btrfs_header_owner(leaf);
3389 leaf_generation = btrfs_header_generation(leaf);
3391 for (i = 0; i < nritems; i++) {
3395 btrfs_item_key_to_cpu(leaf, &key, i);
3396 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3398 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3399 if (btrfs_file_extent_type(leaf, fi) ==
3400 BTRFS_FILE_EXTENT_INLINE)
3403 * FIXME make sure to insert a trans record that
3404 * repeats the snapshot del on crash
3406 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3407 if (disk_bytenr == 0)
3410 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3411 btrfs_file_extent_disk_num_bytes(leaf, fi),
3412 leaf->start, leaf_owner, leaf_generation,
3416 atomic_inc(&root->fs_info->throttle_gen);
3417 wake_up(&root->fs_info->transaction_throttle);
3423 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3424 struct btrfs_root *root,
3425 struct btrfs_leaf_ref *ref)
3429 struct btrfs_extent_info *info = ref->extents;
3431 for (i = 0; i < ref->nritems; i++) {
3432 ret = __btrfs_free_extent(trans, root, info->bytenr,
3433 info->num_bytes, ref->bytenr,
3434 ref->owner, ref->generation,
3437 atomic_inc(&root->fs_info->throttle_gen);
3438 wake_up(&root->fs_info->transaction_throttle);
3448 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3453 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3456 #if 0 // some debugging code in case we see problems here
3457 /* if the refs count is one, it won't get increased again. But
3458 * if the ref count is > 1, someone may be decreasing it at
3459 * the same time we are.
3462 struct extent_buffer *eb = NULL;
3463 eb = btrfs_find_create_tree_block(root, start, len);
3465 btrfs_tree_lock(eb);
3467 mutex_lock(&root->fs_info->alloc_mutex);
3468 ret = lookup_extent_ref(NULL, root, start, len, refs);
3470 mutex_unlock(&root->fs_info->alloc_mutex);
3473 btrfs_tree_unlock(eb);
3474 free_extent_buffer(eb);
3477 printk("block %llu went down to one during drop_snap\n",
3478 (unsigned long long)start);
3489 * helper function for drop_snapshot, this walks down the tree dropping ref
3490 * counts as it goes.
3492 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3493 struct btrfs_root *root,
3494 struct btrfs_path *path, int *level)
3500 struct extent_buffer *next;
3501 struct extent_buffer *cur;
3502 struct extent_buffer *parent;
3503 struct btrfs_leaf_ref *ref;
3508 WARN_ON(*level < 0);
3509 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3510 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3511 path->nodes[*level]->len, &refs);
3517 * walk down to the last node level and free all the leaves
3519 while(*level >= 0) {
3520 WARN_ON(*level < 0);
3521 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3522 cur = path->nodes[*level];
3524 if (btrfs_header_level(cur) != *level)
3527 if (path->slots[*level] >=
3528 btrfs_header_nritems(cur))
3531 ret = btrfs_drop_leaf_ref(trans, root, cur);
3535 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3536 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3537 blocksize = btrfs_level_size(root, *level - 1);
3539 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3542 parent = path->nodes[*level];
3543 root_owner = btrfs_header_owner(parent);
3544 root_gen = btrfs_header_generation(parent);
3545 path->slots[*level]++;
3547 ret = __btrfs_free_extent(trans, root, bytenr,
3548 blocksize, parent->start,
3549 root_owner, root_gen,
3553 atomic_inc(&root->fs_info->throttle_gen);
3554 wake_up(&root->fs_info->transaction_throttle);
3560 * at this point, we have a single ref, and since the
3561 * only place referencing this extent is a dead root
3562 * the reference count should never go higher.
3563 * So, we don't need to check it again
3566 ref = btrfs_lookup_leaf_ref(root, bytenr);
3567 if (ref && ref->generation != ptr_gen) {
3568 btrfs_free_leaf_ref(root, ref);
3572 ret = cache_drop_leaf_ref(trans, root, ref);
3574 btrfs_remove_leaf_ref(root, ref);
3575 btrfs_free_leaf_ref(root, ref);
3579 if (printk_ratelimit()) {
3580 printk("leaf ref miss for bytenr %llu\n",
3581 (unsigned long long)bytenr);
3584 next = btrfs_find_tree_block(root, bytenr, blocksize);
3585 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3586 free_extent_buffer(next);
3588 next = read_tree_block(root, bytenr, blocksize,
3593 * this is a debugging check and can go away
3594 * the ref should never go all the way down to 1
3597 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3603 WARN_ON(*level <= 0);
3604 if (path->nodes[*level-1])
3605 free_extent_buffer(path->nodes[*level-1]);
3606 path->nodes[*level-1] = next;
3607 *level = btrfs_header_level(next);
3608 path->slots[*level] = 0;
3612 WARN_ON(*level < 0);
3613 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3615 if (path->nodes[*level] == root->node) {
3616 parent = path->nodes[*level];
3617 bytenr = path->nodes[*level]->start;
3619 parent = path->nodes[*level + 1];
3620 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3623 blocksize = btrfs_level_size(root, *level);
3624 root_owner = btrfs_header_owner(parent);
3625 root_gen = btrfs_header_generation(parent);
3627 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3628 parent->start, root_owner, root_gen,
3630 free_extent_buffer(path->nodes[*level]);
3631 path->nodes[*level] = NULL;
3640 * helper function for drop_subtree, this function is similar to
3641 * walk_down_tree. The main difference is that it checks reference
3642 * counts while tree blocks are locked.
3644 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3645 struct btrfs_root *root,
3646 struct btrfs_path *path, int *level)
3648 struct extent_buffer *next;
3649 struct extent_buffer *cur;
3650 struct extent_buffer *parent;
3657 cur = path->nodes[*level];
3658 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3664 while (*level >= 0) {
3665 cur = path->nodes[*level];
3667 ret = btrfs_drop_leaf_ref(trans, root, cur);
3669 clean_tree_block(trans, root, cur);
3672 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3673 clean_tree_block(trans, root, cur);
3677 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3678 blocksize = btrfs_level_size(root, *level - 1);
3679 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3681 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3682 btrfs_tree_lock(next);
3684 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3688 parent = path->nodes[*level];
3689 ret = btrfs_free_extent(trans, root, bytenr,
3690 blocksize, parent->start,
3691 btrfs_header_owner(parent),
3692 btrfs_header_generation(parent),
3695 path->slots[*level]++;
3696 btrfs_tree_unlock(next);
3697 free_extent_buffer(next);
3701 *level = btrfs_header_level(next);
3702 path->nodes[*level] = next;
3703 path->slots[*level] = 0;
3704 path->locks[*level] = 1;
3708 parent = path->nodes[*level + 1];
3709 bytenr = path->nodes[*level]->start;
3710 blocksize = path->nodes[*level]->len;
3712 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3713 parent->start, btrfs_header_owner(parent),
3714 btrfs_header_generation(parent), *level, 1);
3717 if (path->locks[*level]) {
3718 btrfs_tree_unlock(path->nodes[*level]);
3719 path->locks[*level] = 0;
3721 free_extent_buffer(path->nodes[*level]);
3722 path->nodes[*level] = NULL;
3729 * helper for dropping snapshots. This walks back up the tree in the path
3730 * to find the first node higher up where we haven't yet gone through
3733 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3734 struct btrfs_root *root,
3735 struct btrfs_path *path,
3736 int *level, int max_level)
3740 struct btrfs_root_item *root_item = &root->root_item;
3745 for (i = *level; i < max_level && path->nodes[i]; i++) {
3746 slot = path->slots[i];
3747 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3748 struct extent_buffer *node;
3749 struct btrfs_disk_key disk_key;
3750 node = path->nodes[i];
3753 WARN_ON(*level == 0);
3754 btrfs_node_key(node, &disk_key, path->slots[i]);
3755 memcpy(&root_item->drop_progress,
3756 &disk_key, sizeof(disk_key));
3757 root_item->drop_level = i;
3760 struct extent_buffer *parent;
3761 if (path->nodes[*level] == root->node)
3762 parent = path->nodes[*level];
3764 parent = path->nodes[*level + 1];
3766 root_owner = btrfs_header_owner(parent);
3767 root_gen = btrfs_header_generation(parent);
3769 clean_tree_block(trans, root, path->nodes[*level]);
3770 ret = btrfs_free_extent(trans, root,
3771 path->nodes[*level]->start,
3772 path->nodes[*level]->len,
3773 parent->start, root_owner,
3774 root_gen, *level, 1);
3776 if (path->locks[*level]) {
3777 btrfs_tree_unlock(path->nodes[*level]);
3778 path->locks[*level] = 0;
3780 free_extent_buffer(path->nodes[*level]);
3781 path->nodes[*level] = NULL;
3789 * drop the reference count on the tree rooted at 'snap'. This traverses
3790 * the tree freeing any blocks that have a ref count of zero after being
3793 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3799 struct btrfs_path *path;
3802 struct btrfs_root_item *root_item = &root->root_item;
3804 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3805 path = btrfs_alloc_path();
3808 level = btrfs_header_level(root->node);
3810 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3811 path->nodes[level] = root->node;
3812 extent_buffer_get(root->node);
3813 path->slots[level] = 0;
3815 struct btrfs_key key;
3816 struct btrfs_disk_key found_key;
3817 struct extent_buffer *node;
3819 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3820 level = root_item->drop_level;
3821 path->lowest_level = level;
3822 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3827 node = path->nodes[level];
3828 btrfs_node_key(node, &found_key, path->slots[level]);
3829 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3830 sizeof(found_key)));
3832 * unlock our path, this is safe because only this
3833 * function is allowed to delete this snapshot
3835 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3836 if (path->nodes[i] && path->locks[i]) {
3838 btrfs_tree_unlock(path->nodes[i]);
3843 wret = walk_down_tree(trans, root, path, &level);
3849 wret = walk_up_tree(trans, root, path, &level,
3855 if (trans->transaction->in_commit) {
3859 atomic_inc(&root->fs_info->throttle_gen);
3860 wake_up(&root->fs_info->transaction_throttle);
3862 for (i = 0; i <= orig_level; i++) {
3863 if (path->nodes[i]) {
3864 free_extent_buffer(path->nodes[i]);
3865 path->nodes[i] = NULL;
3869 btrfs_free_path(path);
3873 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3874 struct btrfs_root *root,
3875 struct extent_buffer *node,
3876 struct extent_buffer *parent)
3878 struct btrfs_path *path;
3884 path = btrfs_alloc_path();
3887 BUG_ON(!btrfs_tree_locked(parent));
3888 parent_level = btrfs_header_level(parent);
3889 extent_buffer_get(parent);
3890 path->nodes[parent_level] = parent;
3891 path->slots[parent_level] = btrfs_header_nritems(parent);
3893 BUG_ON(!btrfs_tree_locked(node));
3894 level = btrfs_header_level(node);
3895 extent_buffer_get(node);
3896 path->nodes[level] = node;
3897 path->slots[level] = 0;
3900 wret = walk_down_subtree(trans, root, path, &level);
3906 wret = walk_up_tree(trans, root, path, &level, parent_level);
3913 btrfs_free_path(path);
3917 static unsigned long calc_ra(unsigned long start, unsigned long last,
3920 return min(last, start + nr - 1);
3923 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3928 unsigned long first_index;
3929 unsigned long last_index;
3932 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3933 struct file_ra_state *ra;
3934 struct btrfs_ordered_extent *ordered;
3935 unsigned int total_read = 0;
3936 unsigned int total_dirty = 0;
3939 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3941 mutex_lock(&inode->i_mutex);
3942 first_index = start >> PAGE_CACHE_SHIFT;
3943 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3945 /* make sure the dirty trick played by the caller work */
3946 ret = invalidate_inode_pages2_range(inode->i_mapping,
3947 first_index, last_index);
3951 file_ra_state_init(ra, inode->i_mapping);
3953 for (i = first_index ; i <= last_index; i++) {
3954 if (total_read % ra->ra_pages == 0) {
3955 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3956 calc_ra(i, last_index, ra->ra_pages));
3960 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3962 page = grab_cache_page(inode->i_mapping, i);
3967 if (!PageUptodate(page)) {
3968 btrfs_readpage(NULL, page);
3970 if (!PageUptodate(page)) {
3972 page_cache_release(page);
3977 wait_on_page_writeback(page);
3979 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3980 page_end = page_start + PAGE_CACHE_SIZE - 1;
3981 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3983 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3985 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3987 page_cache_release(page);
3988 btrfs_start_ordered_extent(inode, ordered, 1);
3989 btrfs_put_ordered_extent(ordered);
3992 set_page_extent_mapped(page);
3994 if (i == first_index)
3995 set_extent_bits(io_tree, page_start, page_end,
3996 EXTENT_BOUNDARY, GFP_NOFS);
3997 btrfs_set_extent_delalloc(inode, page_start, page_end);
3999 set_page_dirty(page);
4002 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4004 page_cache_release(page);
4009 mutex_unlock(&inode->i_mutex);
4010 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4014 static int noinline relocate_data_extent(struct inode *reloc_inode,
4015 struct btrfs_key *extent_key,
4018 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4019 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4020 struct extent_map *em;
4021 u64 start = extent_key->objectid - offset;
4022 u64 end = start + extent_key->offset - 1;
4024 em = alloc_extent_map(GFP_NOFS);
4025 BUG_ON(!em || IS_ERR(em));
4028 em->len = extent_key->offset;
4029 em->block_len = extent_key->offset;
4030 em->block_start = extent_key->objectid;
4031 em->bdev = root->fs_info->fs_devices->latest_bdev;
4032 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4034 /* setup extent map to cheat btrfs_readpage */
4035 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4038 spin_lock(&em_tree->lock);
4039 ret = add_extent_mapping(em_tree, em);
4040 spin_unlock(&em_tree->lock);
4041 if (ret != -EEXIST) {
4042 free_extent_map(em);
4045 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4047 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4049 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4052 struct btrfs_ref_path {
4054 u64 nodes[BTRFS_MAX_LEVEL];
4056 u64 root_generation;
4063 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4064 u64 new_nodes[BTRFS_MAX_LEVEL];
4067 struct disk_extent {
4078 static int is_cowonly_root(u64 root_objectid)
4080 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4081 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4082 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4083 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4084 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4085 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4090 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4091 struct btrfs_root *extent_root,
4092 struct btrfs_ref_path *ref_path,
4095 struct extent_buffer *leaf;
4096 struct btrfs_path *path;
4097 struct btrfs_extent_ref *ref;
4098 struct btrfs_key key;
4099 struct btrfs_key found_key;
4105 path = btrfs_alloc_path();
4110 ref_path->lowest_level = -1;
4111 ref_path->current_level = -1;
4112 ref_path->shared_level = -1;
4116 level = ref_path->current_level - 1;
4117 while (level >= -1) {
4119 if (level < ref_path->lowest_level)
4123 bytenr = ref_path->nodes[level];
4125 bytenr = ref_path->extent_start;
4127 BUG_ON(bytenr == 0);
4129 parent = ref_path->nodes[level + 1];
4130 ref_path->nodes[level + 1] = 0;
4131 ref_path->current_level = level;
4132 BUG_ON(parent == 0);
4134 key.objectid = bytenr;
4135 key.offset = parent + 1;
4136 key.type = BTRFS_EXTENT_REF_KEY;
4138 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4143 leaf = path->nodes[0];
4144 nritems = btrfs_header_nritems(leaf);
4145 if (path->slots[0] >= nritems) {
4146 ret = btrfs_next_leaf(extent_root, path);
4151 leaf = path->nodes[0];
4154 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4155 if (found_key.objectid == bytenr &&
4156 found_key.type == BTRFS_EXTENT_REF_KEY) {
4157 if (level < ref_path->shared_level)
4158 ref_path->shared_level = level;
4163 btrfs_release_path(extent_root, path);
4166 /* reached lowest level */
4170 level = ref_path->current_level;
4171 while (level < BTRFS_MAX_LEVEL - 1) {
4174 bytenr = ref_path->nodes[level];
4176 bytenr = ref_path->extent_start;
4178 BUG_ON(bytenr == 0);
4180 key.objectid = bytenr;
4182 key.type = BTRFS_EXTENT_REF_KEY;
4184 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4188 leaf = path->nodes[0];
4189 nritems = btrfs_header_nritems(leaf);
4190 if (path->slots[0] >= nritems) {
4191 ret = btrfs_next_leaf(extent_root, path);
4195 /* the extent was freed by someone */
4196 if (ref_path->lowest_level == level)
4198 btrfs_release_path(extent_root, path);
4201 leaf = path->nodes[0];
4204 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4205 if (found_key.objectid != bytenr ||
4206 found_key.type != BTRFS_EXTENT_REF_KEY) {
4207 /* the extent was freed by someone */
4208 if (ref_path->lowest_level == level) {
4212 btrfs_release_path(extent_root, path);
4216 ref = btrfs_item_ptr(leaf, path->slots[0],
4217 struct btrfs_extent_ref);
4218 ref_objectid = btrfs_ref_objectid(leaf, ref);
4219 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4221 level = (int)ref_objectid;
4222 BUG_ON(level >= BTRFS_MAX_LEVEL);
4223 ref_path->lowest_level = level;
4224 ref_path->current_level = level;
4225 ref_path->nodes[level] = bytenr;
4227 WARN_ON(ref_objectid != level);
4230 WARN_ON(level != -1);
4234 if (ref_path->lowest_level == level) {
4235 ref_path->owner_objectid = ref_objectid;
4236 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4240 * the block is tree root or the block isn't in reference
4243 if (found_key.objectid == found_key.offset ||
4244 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4245 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4246 ref_path->root_generation =
4247 btrfs_ref_generation(leaf, ref);
4249 /* special reference from the tree log */
4250 ref_path->nodes[0] = found_key.offset;
4251 ref_path->current_level = 0;
4258 BUG_ON(ref_path->nodes[level] != 0);
4259 ref_path->nodes[level] = found_key.offset;
4260 ref_path->current_level = level;
4263 * the reference was created in the running transaction,
4264 * no need to continue walking up.
4266 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4267 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4268 ref_path->root_generation =
4269 btrfs_ref_generation(leaf, ref);
4274 btrfs_release_path(extent_root, path);
4277 /* reached max tree level, but no tree root found. */
4280 btrfs_free_path(path);
4284 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4285 struct btrfs_root *extent_root,
4286 struct btrfs_ref_path *ref_path,
4289 memset(ref_path, 0, sizeof(*ref_path));
4290 ref_path->extent_start = extent_start;
4292 return __next_ref_path(trans, extent_root, ref_path, 1);
4295 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4296 struct btrfs_root *extent_root,
4297 struct btrfs_ref_path *ref_path)
4299 return __next_ref_path(trans, extent_root, ref_path, 0);
4302 static int noinline get_new_locations(struct inode *reloc_inode,
4303 struct btrfs_key *extent_key,
4304 u64 offset, int no_fragment,
4305 struct disk_extent **extents,
4308 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4309 struct btrfs_path *path;
4310 struct btrfs_file_extent_item *fi;
4311 struct extent_buffer *leaf;
4312 struct disk_extent *exts = *extents;
4313 struct btrfs_key found_key;
4318 int max = *nr_extents;
4321 WARN_ON(!no_fragment && *extents);
4324 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4329 path = btrfs_alloc_path();
4332 cur_pos = extent_key->objectid - offset;
4333 last_byte = extent_key->objectid + extent_key->offset;
4334 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4344 leaf = path->nodes[0];
4345 nritems = btrfs_header_nritems(leaf);
4346 if (path->slots[0] >= nritems) {
4347 ret = btrfs_next_leaf(root, path);
4352 leaf = path->nodes[0];
4355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4356 if (found_key.offset != cur_pos ||
4357 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4358 found_key.objectid != reloc_inode->i_ino)
4361 fi = btrfs_item_ptr(leaf, path->slots[0],
4362 struct btrfs_file_extent_item);
4363 if (btrfs_file_extent_type(leaf, fi) !=
4364 BTRFS_FILE_EXTENT_REG ||
4365 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4369 struct disk_extent *old = exts;
4371 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4372 memcpy(exts, old, sizeof(*exts) * nr);
4373 if (old != *extents)
4377 exts[nr].disk_bytenr =
4378 btrfs_file_extent_disk_bytenr(leaf, fi);
4379 exts[nr].disk_num_bytes =
4380 btrfs_file_extent_disk_num_bytes(leaf, fi);
4381 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4382 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4383 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4384 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4385 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4386 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4388 BUG_ON(exts[nr].offset > 0);
4389 BUG_ON(exts[nr].compression || exts[nr].encryption);
4390 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4392 cur_pos += exts[nr].num_bytes;
4395 if (cur_pos + offset >= last_byte)
4405 BUG_ON(cur_pos + offset > last_byte);
4406 if (cur_pos + offset < last_byte) {
4412 btrfs_free_path(path);
4414 if (exts != *extents)
4423 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4424 struct btrfs_root *root,
4425 struct btrfs_path *path,
4426 struct btrfs_key *extent_key,
4427 struct btrfs_key *leaf_key,
4428 struct btrfs_ref_path *ref_path,
4429 struct disk_extent *new_extents,
4432 struct extent_buffer *leaf;
4433 struct btrfs_file_extent_item *fi;
4434 struct inode *inode = NULL;
4435 struct btrfs_key key;
4443 int extent_locked = 0;
4447 memcpy(&key, leaf_key, sizeof(key));
4448 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4449 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4450 if (key.objectid < ref_path->owner_objectid ||
4451 (key.objectid == ref_path->owner_objectid &&
4452 key.type < BTRFS_EXTENT_DATA_KEY)) {
4453 key.objectid = ref_path->owner_objectid;
4454 key.type = BTRFS_EXTENT_DATA_KEY;
4460 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4464 leaf = path->nodes[0];
4465 nritems = btrfs_header_nritems(leaf);
4467 if (extent_locked && ret > 0) {
4469 * the file extent item was modified by someone
4470 * before the extent got locked.
4472 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4473 lock_end, GFP_NOFS);
4477 if (path->slots[0] >= nritems) {
4478 if (++nr_scaned > 2)
4481 BUG_ON(extent_locked);
4482 ret = btrfs_next_leaf(root, path);
4487 leaf = path->nodes[0];
4488 nritems = btrfs_header_nritems(leaf);
4491 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4493 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4494 if ((key.objectid > ref_path->owner_objectid) ||
4495 (key.objectid == ref_path->owner_objectid &&
4496 key.type > BTRFS_EXTENT_DATA_KEY) ||
4497 (key.offset >= first_pos + extent_key->offset))
4501 if (inode && key.objectid != inode->i_ino) {
4502 BUG_ON(extent_locked);
4503 btrfs_release_path(root, path);
4504 mutex_unlock(&inode->i_mutex);
4510 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4515 fi = btrfs_item_ptr(leaf, path->slots[0],
4516 struct btrfs_file_extent_item);
4517 extent_type = btrfs_file_extent_type(leaf, fi);
4518 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4519 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4520 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4521 extent_key->objectid)) {
4527 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4528 ext_offset = btrfs_file_extent_offset(leaf, fi);
4530 if (first_pos > key.offset - ext_offset)
4531 first_pos = key.offset - ext_offset;
4533 if (!extent_locked) {
4534 lock_start = key.offset;
4535 lock_end = lock_start + num_bytes - 1;
4537 if (lock_start > key.offset ||
4538 lock_end + 1 < key.offset + num_bytes) {
4539 unlock_extent(&BTRFS_I(inode)->io_tree,
4540 lock_start, lock_end, GFP_NOFS);
4546 btrfs_release_path(root, path);
4548 inode = btrfs_iget_locked(root->fs_info->sb,
4549 key.objectid, root);
4550 if (inode->i_state & I_NEW) {
4551 BTRFS_I(inode)->root = root;
4552 BTRFS_I(inode)->location.objectid =
4554 BTRFS_I(inode)->location.type =
4555 BTRFS_INODE_ITEM_KEY;
4556 BTRFS_I(inode)->location.offset = 0;
4557 btrfs_read_locked_inode(inode);
4558 unlock_new_inode(inode);
4561 * some code call btrfs_commit_transaction while
4562 * holding the i_mutex, so we can't use mutex_lock
4565 if (is_bad_inode(inode) ||
4566 !mutex_trylock(&inode->i_mutex)) {
4569 key.offset = (u64)-1;
4574 if (!extent_locked) {
4575 struct btrfs_ordered_extent *ordered;
4577 btrfs_release_path(root, path);
4579 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4580 lock_end, GFP_NOFS);
4581 ordered = btrfs_lookup_first_ordered_extent(inode,
4584 ordered->file_offset <= lock_end &&
4585 ordered->file_offset + ordered->len > lock_start) {
4586 unlock_extent(&BTRFS_I(inode)->io_tree,
4587 lock_start, lock_end, GFP_NOFS);
4588 btrfs_start_ordered_extent(inode, ordered, 1);
4589 btrfs_put_ordered_extent(ordered);
4590 key.offset += num_bytes;
4594 btrfs_put_ordered_extent(ordered);
4600 if (nr_extents == 1) {
4601 /* update extent pointer in place */
4602 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4603 new_extents[0].disk_bytenr);
4604 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4605 new_extents[0].disk_num_bytes);
4606 btrfs_mark_buffer_dirty(leaf);
4608 btrfs_drop_extent_cache(inode, key.offset,
4609 key.offset + num_bytes - 1, 0);
4611 ret = btrfs_inc_extent_ref(trans, root,
4612 new_extents[0].disk_bytenr,
4613 new_extents[0].disk_num_bytes,
4615 root->root_key.objectid,
4620 ret = btrfs_free_extent(trans, root,
4621 extent_key->objectid,
4624 btrfs_header_owner(leaf),
4625 btrfs_header_generation(leaf),
4629 btrfs_release_path(root, path);
4630 key.offset += num_bytes;
4638 * drop old extent pointer at first, then insert the
4639 * new pointers one bye one
4641 btrfs_release_path(root, path);
4642 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4643 key.offset + num_bytes,
4644 key.offset, &alloc_hint);
4647 for (i = 0; i < nr_extents; i++) {
4648 if (ext_offset >= new_extents[i].num_bytes) {
4649 ext_offset -= new_extents[i].num_bytes;
4652 extent_len = min(new_extents[i].num_bytes -
4653 ext_offset, num_bytes);
4655 ret = btrfs_insert_empty_item(trans, root,
4660 leaf = path->nodes[0];
4661 fi = btrfs_item_ptr(leaf, path->slots[0],
4662 struct btrfs_file_extent_item);
4663 btrfs_set_file_extent_generation(leaf, fi,
4665 btrfs_set_file_extent_type(leaf, fi,
4666 BTRFS_FILE_EXTENT_REG);
4667 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4668 new_extents[i].disk_bytenr);
4669 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4670 new_extents[i].disk_num_bytes);
4671 btrfs_set_file_extent_ram_bytes(leaf, fi,
4672 new_extents[i].ram_bytes);
4674 btrfs_set_file_extent_compression(leaf, fi,
4675 new_extents[i].compression);
4676 btrfs_set_file_extent_encryption(leaf, fi,
4677 new_extents[i].encryption);
4678 btrfs_set_file_extent_other_encoding(leaf, fi,
4679 new_extents[i].other_encoding);
4681 btrfs_set_file_extent_num_bytes(leaf, fi,
4683 ext_offset += new_extents[i].offset;
4684 btrfs_set_file_extent_offset(leaf, fi,
4686 btrfs_mark_buffer_dirty(leaf);
4688 btrfs_drop_extent_cache(inode, key.offset,
4689 key.offset + extent_len - 1, 0);
4691 ret = btrfs_inc_extent_ref(trans, root,
4692 new_extents[i].disk_bytenr,
4693 new_extents[i].disk_num_bytes,
4695 root->root_key.objectid,
4696 trans->transid, key.objectid);
4698 btrfs_release_path(root, path);
4700 inode_add_bytes(inode, extent_len);
4703 num_bytes -= extent_len;
4704 key.offset += extent_len;
4709 BUG_ON(i >= nr_extents);
4713 if (extent_locked) {
4714 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4715 lock_end, GFP_NOFS);
4719 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4720 key.offset >= first_pos + extent_key->offset)
4727 btrfs_release_path(root, path);
4729 mutex_unlock(&inode->i_mutex);
4730 if (extent_locked) {
4731 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4732 lock_end, GFP_NOFS);
4739 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4740 struct btrfs_root *root,
4741 struct extent_buffer *buf, u64 orig_start)
4746 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4747 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4749 level = btrfs_header_level(buf);
4751 struct btrfs_leaf_ref *ref;
4752 struct btrfs_leaf_ref *orig_ref;
4754 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4758 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4760 btrfs_free_leaf_ref(root, orig_ref);
4764 ref->nritems = orig_ref->nritems;
4765 memcpy(ref->extents, orig_ref->extents,
4766 sizeof(ref->extents[0]) * ref->nritems);
4768 btrfs_free_leaf_ref(root, orig_ref);
4770 ref->root_gen = trans->transid;
4771 ref->bytenr = buf->start;
4772 ref->owner = btrfs_header_owner(buf);
4773 ref->generation = btrfs_header_generation(buf);
4774 ret = btrfs_add_leaf_ref(root, ref, 0);
4776 btrfs_free_leaf_ref(root, ref);
4781 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4782 struct extent_buffer *leaf,
4783 struct btrfs_block_group_cache *group,
4784 struct btrfs_root *target_root)
4786 struct btrfs_key key;
4787 struct inode *inode = NULL;
4788 struct btrfs_file_extent_item *fi;
4790 u64 skip_objectid = 0;
4794 nritems = btrfs_header_nritems(leaf);
4795 for (i = 0; i < nritems; i++) {
4796 btrfs_item_key_to_cpu(leaf, &key, i);
4797 if (key.objectid == skip_objectid ||
4798 key.type != BTRFS_EXTENT_DATA_KEY)
4800 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4801 if (btrfs_file_extent_type(leaf, fi) ==
4802 BTRFS_FILE_EXTENT_INLINE)
4804 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4806 if (!inode || inode->i_ino != key.objectid) {
4808 inode = btrfs_ilookup(target_root->fs_info->sb,
4809 key.objectid, target_root, 1);
4812 skip_objectid = key.objectid;
4815 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4817 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4818 key.offset + num_bytes - 1, GFP_NOFS);
4819 btrfs_drop_extent_cache(inode, key.offset,
4820 key.offset + num_bytes - 1, 1);
4821 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4822 key.offset + num_bytes - 1, GFP_NOFS);
4829 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4830 struct btrfs_root *root,
4831 struct extent_buffer *leaf,
4832 struct btrfs_block_group_cache *group,
4833 struct inode *reloc_inode)
4835 struct btrfs_key key;
4836 struct btrfs_key extent_key;
4837 struct btrfs_file_extent_item *fi;
4838 struct btrfs_leaf_ref *ref;
4839 struct disk_extent *new_extent;
4848 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4849 BUG_ON(!new_extent);
4851 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4855 nritems = btrfs_header_nritems(leaf);
4856 for (i = 0; i < nritems; i++) {
4857 btrfs_item_key_to_cpu(leaf, &key, i);
4858 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4860 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4861 if (btrfs_file_extent_type(leaf, fi) ==
4862 BTRFS_FILE_EXTENT_INLINE)
4864 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4865 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4870 if (bytenr >= group->key.objectid + group->key.offset ||
4871 bytenr + num_bytes <= group->key.objectid)
4874 extent_key.objectid = bytenr;
4875 extent_key.offset = num_bytes;
4876 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4878 ret = get_new_locations(reloc_inode, &extent_key,
4879 group->key.objectid, 1,
4880 &new_extent, &nr_extent);
4885 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4886 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4887 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4888 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4890 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4891 new_extent->disk_bytenr);
4892 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4893 new_extent->disk_num_bytes);
4894 btrfs_mark_buffer_dirty(leaf);
4896 ret = btrfs_inc_extent_ref(trans, root,
4897 new_extent->disk_bytenr,
4898 new_extent->disk_num_bytes,
4900 root->root_key.objectid,
4901 trans->transid, key.objectid);
4903 ret = btrfs_free_extent(trans, root,
4904 bytenr, num_bytes, leaf->start,
4905 btrfs_header_owner(leaf),
4906 btrfs_header_generation(leaf),
4912 BUG_ON(ext_index + 1 != ref->nritems);
4913 btrfs_free_leaf_ref(root, ref);
4917 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4918 struct btrfs_root *root)
4920 struct btrfs_root *reloc_root;
4923 if (root->reloc_root) {
4924 reloc_root = root->reloc_root;
4925 root->reloc_root = NULL;
4926 list_add(&reloc_root->dead_list,
4927 &root->fs_info->dead_reloc_roots);
4929 btrfs_set_root_bytenr(&reloc_root->root_item,
4930 reloc_root->node->start);
4931 btrfs_set_root_level(&root->root_item,
4932 btrfs_header_level(reloc_root->node));
4933 memset(&reloc_root->root_item.drop_progress, 0,
4934 sizeof(struct btrfs_disk_key));
4935 reloc_root->root_item.drop_level = 0;
4937 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4938 &reloc_root->root_key,
4939 &reloc_root->root_item);
4945 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4947 struct btrfs_trans_handle *trans;
4948 struct btrfs_root *reloc_root;
4949 struct btrfs_root *prev_root = NULL;
4950 struct list_head dead_roots;
4954 INIT_LIST_HEAD(&dead_roots);
4955 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4957 while (!list_empty(&dead_roots)) {
4958 reloc_root = list_entry(dead_roots.prev,
4959 struct btrfs_root, dead_list);
4960 list_del_init(&reloc_root->dead_list);
4962 BUG_ON(reloc_root->commit_root != NULL);
4964 trans = btrfs_join_transaction(root, 1);
4967 mutex_lock(&root->fs_info->drop_mutex);
4968 ret = btrfs_drop_snapshot(trans, reloc_root);
4971 mutex_unlock(&root->fs_info->drop_mutex);
4973 nr = trans->blocks_used;
4974 ret = btrfs_end_transaction(trans, root);
4976 btrfs_btree_balance_dirty(root, nr);
4979 free_extent_buffer(reloc_root->node);
4981 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4982 &reloc_root->root_key);
4984 mutex_unlock(&root->fs_info->drop_mutex);
4986 nr = trans->blocks_used;
4987 ret = btrfs_end_transaction(trans, root);
4989 btrfs_btree_balance_dirty(root, nr);
4992 prev_root = reloc_root;
4995 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5001 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5003 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5007 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5009 struct btrfs_root *reloc_root;
5010 struct btrfs_trans_handle *trans;
5011 struct btrfs_key location;
5015 mutex_lock(&root->fs_info->tree_reloc_mutex);
5016 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5018 found = !list_empty(&root->fs_info->dead_reloc_roots);
5019 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5022 trans = btrfs_start_transaction(root, 1);
5024 ret = btrfs_commit_transaction(trans, root);
5028 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5029 location.offset = (u64)-1;
5030 location.type = BTRFS_ROOT_ITEM_KEY;
5032 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5033 BUG_ON(!reloc_root);
5034 btrfs_orphan_cleanup(reloc_root);
5038 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5039 struct btrfs_root *root)
5041 struct btrfs_root *reloc_root;
5042 struct extent_buffer *eb;
5043 struct btrfs_root_item *root_item;
5044 struct btrfs_key root_key;
5047 BUG_ON(!root->ref_cows);
5048 if (root->reloc_root)
5051 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5054 ret = btrfs_copy_root(trans, root, root->commit_root,
5055 &eb, BTRFS_TREE_RELOC_OBJECTID);
5058 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5059 root_key.offset = root->root_key.objectid;
5060 root_key.type = BTRFS_ROOT_ITEM_KEY;
5062 memcpy(root_item, &root->root_item, sizeof(root_item));
5063 btrfs_set_root_refs(root_item, 0);
5064 btrfs_set_root_bytenr(root_item, eb->start);
5065 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5066 btrfs_set_root_generation(root_item, trans->transid);
5068 btrfs_tree_unlock(eb);
5069 free_extent_buffer(eb);
5071 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5072 &root_key, root_item);
5076 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5078 BUG_ON(!reloc_root);
5079 reloc_root->last_trans = trans->transid;
5080 reloc_root->commit_root = NULL;
5081 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5083 root->reloc_root = reloc_root;
5088 * Core function of space balance.
5090 * The idea is using reloc trees to relocate tree blocks in reference
5091 * counted roots. There is one reloc tree for each subvol, and all
5092 * reloc trees share same root key objectid. Reloc trees are snapshots
5093 * of the latest committed roots of subvols (root->commit_root).
5095 * To relocate a tree block referenced by a subvol, there are two steps.
5096 * COW the block through subvol's reloc tree, then update block pointer
5097 * in the subvol to point to the new block. Since all reloc trees share
5098 * same root key objectid, doing special handing for tree blocks owned
5099 * by them is easy. Once a tree block has been COWed in one reloc tree,
5100 * we can use the resulting new block directly when the same block is
5101 * required to COW again through other reloc trees. By this way, relocated
5102 * tree blocks are shared between reloc trees, so they are also shared
5105 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5106 struct btrfs_root *root,
5107 struct btrfs_path *path,
5108 struct btrfs_key *first_key,
5109 struct btrfs_ref_path *ref_path,
5110 struct btrfs_block_group_cache *group,
5111 struct inode *reloc_inode)
5113 struct btrfs_root *reloc_root;
5114 struct extent_buffer *eb = NULL;
5115 struct btrfs_key *keys;
5119 int lowest_level = 0;
5122 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5123 lowest_level = ref_path->owner_objectid;
5125 if (!root->ref_cows) {
5126 path->lowest_level = lowest_level;
5127 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5129 path->lowest_level = 0;
5130 btrfs_release_path(root, path);
5134 mutex_lock(&root->fs_info->tree_reloc_mutex);
5135 ret = init_reloc_tree(trans, root);
5137 reloc_root = root->reloc_root;
5139 shared_level = ref_path->shared_level;
5140 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5142 keys = ref_path->node_keys;
5143 nodes = ref_path->new_nodes;
5144 memset(&keys[shared_level + 1], 0,
5145 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5146 memset(&nodes[shared_level + 1], 0,
5147 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5149 if (nodes[lowest_level] == 0) {
5150 path->lowest_level = lowest_level;
5151 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5154 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5155 eb = path->nodes[level];
5156 if (!eb || eb == reloc_root->node)
5158 nodes[level] = eb->start;
5160 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5162 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5165 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5166 eb = path->nodes[0];
5167 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5168 group, reloc_inode);
5171 btrfs_release_path(reloc_root, path);
5173 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5179 * replace tree blocks in the fs tree with tree blocks in
5182 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5185 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5186 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5189 extent_buffer_get(path->nodes[0]);
5190 eb = path->nodes[0];
5191 btrfs_release_path(reloc_root, path);
5192 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5194 free_extent_buffer(eb);
5197 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5198 path->lowest_level = 0;
5202 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5203 struct btrfs_root *root,
5204 struct btrfs_path *path,
5205 struct btrfs_key *first_key,
5206 struct btrfs_ref_path *ref_path)
5210 ret = relocate_one_path(trans, root, path, first_key,
5211 ref_path, NULL, NULL);
5214 if (root == root->fs_info->extent_root)
5215 btrfs_extent_post_op(trans, root);
5220 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5221 struct btrfs_root *extent_root,
5222 struct btrfs_path *path,
5223 struct btrfs_key *extent_key)
5227 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5230 ret = btrfs_del_item(trans, extent_root, path);
5232 btrfs_release_path(extent_root, path);
5236 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5237 struct btrfs_ref_path *ref_path)
5239 struct btrfs_key root_key;
5241 root_key.objectid = ref_path->root_objectid;
5242 root_key.type = BTRFS_ROOT_ITEM_KEY;
5243 if (is_cowonly_root(ref_path->root_objectid))
5244 root_key.offset = 0;
5246 root_key.offset = (u64)-1;
5248 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5251 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5252 struct btrfs_path *path,
5253 struct btrfs_key *extent_key,
5254 struct btrfs_block_group_cache *group,
5255 struct inode *reloc_inode, int pass)
5257 struct btrfs_trans_handle *trans;
5258 struct btrfs_root *found_root;
5259 struct btrfs_ref_path *ref_path = NULL;
5260 struct disk_extent *new_extents = NULL;
5265 struct btrfs_key first_key;
5269 trans = btrfs_start_transaction(extent_root, 1);
5272 if (extent_key->objectid == 0) {
5273 ret = del_extent_zero(trans, extent_root, path, extent_key);
5277 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5283 for (loops = 0; ; loops++) {
5285 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5286 extent_key->objectid);
5288 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5295 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5296 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5299 found_root = read_ref_root(extent_root->fs_info, ref_path);
5300 BUG_ON(!found_root);
5302 * for reference counted tree, only process reference paths
5303 * rooted at the latest committed root.
5305 if (found_root->ref_cows &&
5306 ref_path->root_generation != found_root->root_key.offset)
5309 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5312 * copy data extents to new locations
5314 u64 group_start = group->key.objectid;
5315 ret = relocate_data_extent(reloc_inode,
5324 level = ref_path->owner_objectid;
5327 if (prev_block != ref_path->nodes[level]) {
5328 struct extent_buffer *eb;
5329 u64 block_start = ref_path->nodes[level];
5330 u64 block_size = btrfs_level_size(found_root, level);
5332 eb = read_tree_block(found_root, block_start,
5334 btrfs_tree_lock(eb);
5335 BUG_ON(level != btrfs_header_level(eb));
5338 btrfs_item_key_to_cpu(eb, &first_key, 0);
5340 btrfs_node_key_to_cpu(eb, &first_key, 0);
5342 btrfs_tree_unlock(eb);
5343 free_extent_buffer(eb);
5344 prev_block = block_start;
5347 btrfs_record_root_in_trans(found_root);
5348 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5350 * try to update data extent references while
5351 * keeping metadata shared between snapshots.
5354 ret = relocate_one_path(trans, found_root,
5355 path, &first_key, ref_path,
5356 group, reloc_inode);
5362 * use fallback method to process the remaining
5366 u64 group_start = group->key.objectid;
5367 new_extents = kmalloc(sizeof(*new_extents),
5370 ret = get_new_locations(reloc_inode,
5378 ret = replace_one_extent(trans, found_root,
5380 &first_key, ref_path,
5381 new_extents, nr_extents);
5383 ret = relocate_tree_block(trans, found_root, path,
5384 &first_key, ref_path);
5391 btrfs_end_transaction(trans, extent_root);
5397 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5400 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5401 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5403 num_devices = root->fs_info->fs_devices->rw_devices;
5404 if (num_devices == 1) {
5405 stripped |= BTRFS_BLOCK_GROUP_DUP;
5406 stripped = flags & ~stripped;
5408 /* turn raid0 into single device chunks */
5409 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5412 /* turn mirroring into duplication */
5413 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5414 BTRFS_BLOCK_GROUP_RAID10))
5415 return stripped | BTRFS_BLOCK_GROUP_DUP;
5418 /* they already had raid on here, just return */
5419 if (flags & stripped)
5422 stripped |= BTRFS_BLOCK_GROUP_DUP;
5423 stripped = flags & ~stripped;
5425 /* switch duplicated blocks with raid1 */
5426 if (flags & BTRFS_BLOCK_GROUP_DUP)
5427 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5429 /* turn single device chunks into raid0 */
5430 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5435 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5436 struct btrfs_block_group_cache *shrink_block_group,
5439 struct btrfs_trans_handle *trans;
5440 u64 new_alloc_flags;
5443 spin_lock(&shrink_block_group->lock);
5444 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5445 spin_unlock(&shrink_block_group->lock);
5447 trans = btrfs_start_transaction(root, 1);
5448 spin_lock(&shrink_block_group->lock);
5450 new_alloc_flags = update_block_group_flags(root,
5451 shrink_block_group->flags);
5452 if (new_alloc_flags != shrink_block_group->flags) {
5454 btrfs_block_group_used(&shrink_block_group->item);
5456 calc = shrink_block_group->key.offset;
5458 spin_unlock(&shrink_block_group->lock);
5460 do_chunk_alloc(trans, root->fs_info->extent_root,
5461 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5463 btrfs_end_transaction(trans, root);
5465 spin_unlock(&shrink_block_group->lock);
5469 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5470 struct btrfs_root *root,
5471 u64 objectid, u64 size)
5473 struct btrfs_path *path;
5474 struct btrfs_inode_item *item;
5475 struct extent_buffer *leaf;
5478 path = btrfs_alloc_path();
5482 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5486 leaf = path->nodes[0];
5487 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5488 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5489 btrfs_set_inode_generation(leaf, item, 1);
5490 btrfs_set_inode_size(leaf, item, size);
5491 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5492 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5493 btrfs_mark_buffer_dirty(leaf);
5494 btrfs_release_path(root, path);
5496 btrfs_free_path(path);
5500 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5501 struct btrfs_block_group_cache *group)
5503 struct inode *inode = NULL;
5504 struct btrfs_trans_handle *trans;
5505 struct btrfs_root *root;
5506 struct btrfs_key root_key;
5507 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5510 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5511 root_key.type = BTRFS_ROOT_ITEM_KEY;
5512 root_key.offset = (u64)-1;
5513 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5515 return ERR_CAST(root);
5517 trans = btrfs_start_transaction(root, 1);
5520 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5524 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5527 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5528 group->key.offset, 0, group->key.offset,
5532 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5533 if (inode->i_state & I_NEW) {
5534 BTRFS_I(inode)->root = root;
5535 BTRFS_I(inode)->location.objectid = objectid;
5536 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5537 BTRFS_I(inode)->location.offset = 0;
5538 btrfs_read_locked_inode(inode);
5539 unlock_new_inode(inode);
5540 BUG_ON(is_bad_inode(inode));
5544 BTRFS_I(inode)->index_cnt = group->key.objectid;
5546 err = btrfs_orphan_add(trans, inode);
5548 btrfs_end_transaction(trans, root);
5552 inode = ERR_PTR(err);
5557 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5560 struct btrfs_ordered_sum *sums;
5561 struct btrfs_sector_sum *sector_sum;
5562 struct btrfs_ordered_extent *ordered;
5563 struct btrfs_root *root = BTRFS_I(inode)->root;
5564 struct list_head list;
5569 INIT_LIST_HEAD(&list);
5571 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5572 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5574 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5575 ret = btrfs_lookup_csums_range(root, disk_bytenr,
5576 disk_bytenr + len - 1, &list);
5578 while (!list_empty(&list)) {
5579 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5580 list_del_init(&sums->list);
5582 sector_sum = sums->sums;
5583 sums->bytenr = ordered->start;
5586 while (offset < sums->len) {
5587 sector_sum->bytenr += ordered->start - disk_bytenr;
5589 offset += root->sectorsize;
5592 btrfs_add_ordered_sum(inode, ordered, sums);
5594 btrfs_put_ordered_extent(ordered);
5598 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5600 struct btrfs_trans_handle *trans;
5601 struct btrfs_path *path;
5602 struct btrfs_fs_info *info = root->fs_info;
5603 struct extent_buffer *leaf;
5604 struct inode *reloc_inode;
5605 struct btrfs_block_group_cache *block_group;
5606 struct btrfs_key key;
5615 root = root->fs_info->extent_root;
5617 block_group = btrfs_lookup_block_group(info, group_start);
5618 BUG_ON(!block_group);
5620 printk("btrfs relocating block group %llu flags %llu\n",
5621 (unsigned long long)block_group->key.objectid,
5622 (unsigned long long)block_group->flags);
5624 path = btrfs_alloc_path();
5627 reloc_inode = create_reloc_inode(info, block_group);
5628 BUG_ON(IS_ERR(reloc_inode));
5630 __alloc_chunk_for_shrink(root, block_group, 1);
5631 set_block_group_readonly(block_group);
5633 btrfs_start_delalloc_inodes(info->tree_root);
5634 btrfs_wait_ordered_extents(info->tree_root, 0);
5639 key.objectid = block_group->key.objectid;
5642 cur_byte = key.objectid;
5644 trans = btrfs_start_transaction(info->tree_root, 1);
5645 btrfs_commit_transaction(trans, info->tree_root);
5647 mutex_lock(&root->fs_info->cleaner_mutex);
5648 btrfs_clean_old_snapshots(info->tree_root);
5649 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5650 mutex_unlock(&root->fs_info->cleaner_mutex);
5653 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5657 leaf = path->nodes[0];
5658 nritems = btrfs_header_nritems(leaf);
5659 if (path->slots[0] >= nritems) {
5660 ret = btrfs_next_leaf(root, path);
5667 leaf = path->nodes[0];
5668 nritems = btrfs_header_nritems(leaf);
5671 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5673 if (key.objectid >= block_group->key.objectid +
5674 block_group->key.offset)
5677 if (progress && need_resched()) {
5678 btrfs_release_path(root, path);
5685 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5686 key.objectid + key.offset <= cur_byte) {
5692 cur_byte = key.objectid + key.offset;
5693 btrfs_release_path(root, path);
5695 __alloc_chunk_for_shrink(root, block_group, 0);
5696 ret = relocate_one_extent(root, path, &key, block_group,
5702 key.objectid = cur_byte;
5707 btrfs_release_path(root, path);
5710 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5711 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5714 if (total_found > 0) {
5715 printk("btrfs found %llu extents in pass %d\n",
5716 (unsigned long long)total_found, pass);
5718 if (total_found == skipped && pass > 2) {
5720 reloc_inode = create_reloc_inode(info, block_group);
5726 /* delete reloc_inode */
5729 /* unpin extents in this range */
5730 trans = btrfs_start_transaction(info->tree_root, 1);
5731 btrfs_commit_transaction(trans, info->tree_root);
5733 spin_lock(&block_group->lock);
5734 WARN_ON(block_group->pinned > 0);
5735 WARN_ON(block_group->reserved > 0);
5736 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5737 spin_unlock(&block_group->lock);
5738 put_block_group(block_group);
5741 btrfs_free_path(path);
5745 static int find_first_block_group(struct btrfs_root *root,
5746 struct btrfs_path *path, struct btrfs_key *key)
5749 struct btrfs_key found_key;
5750 struct extent_buffer *leaf;
5753 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5758 slot = path->slots[0];
5759 leaf = path->nodes[0];
5760 if (slot >= btrfs_header_nritems(leaf)) {
5761 ret = btrfs_next_leaf(root, path);
5768 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5770 if (found_key.objectid >= key->objectid &&
5771 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5782 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5784 struct btrfs_block_group_cache *block_group;
5787 spin_lock(&info->block_group_cache_lock);
5788 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5789 block_group = rb_entry(n, struct btrfs_block_group_cache,
5791 rb_erase(&block_group->cache_node,
5792 &info->block_group_cache_tree);
5793 spin_unlock(&info->block_group_cache_lock);
5795 btrfs_remove_free_space_cache(block_group);
5796 down_write(&block_group->space_info->groups_sem);
5797 list_del(&block_group->list);
5798 up_write(&block_group->space_info->groups_sem);
5800 WARN_ON(atomic_read(&block_group->count) != 1);
5803 spin_lock(&info->block_group_cache_lock);
5805 spin_unlock(&info->block_group_cache_lock);
5809 int btrfs_read_block_groups(struct btrfs_root *root)
5811 struct btrfs_path *path;
5813 struct btrfs_block_group_cache *cache;
5814 struct btrfs_fs_info *info = root->fs_info;
5815 struct btrfs_space_info *space_info;
5816 struct btrfs_key key;
5817 struct btrfs_key found_key;
5818 struct extent_buffer *leaf;
5820 root = info->extent_root;
5823 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5824 path = btrfs_alloc_path();
5829 ret = find_first_block_group(root, path, &key);
5837 leaf = path->nodes[0];
5838 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5839 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5845 atomic_set(&cache->count, 1);
5846 spin_lock_init(&cache->lock);
5847 mutex_init(&cache->alloc_mutex);
5848 mutex_init(&cache->cache_mutex);
5849 INIT_LIST_HEAD(&cache->list);
5850 read_extent_buffer(leaf, &cache->item,
5851 btrfs_item_ptr_offset(leaf, path->slots[0]),
5852 sizeof(cache->item));
5853 memcpy(&cache->key, &found_key, sizeof(found_key));
5855 key.objectid = found_key.objectid + found_key.offset;
5856 btrfs_release_path(root, path);
5857 cache->flags = btrfs_block_group_flags(&cache->item);
5859 ret = update_space_info(info, cache->flags, found_key.offset,
5860 btrfs_block_group_used(&cache->item),
5863 cache->space_info = space_info;
5864 down_write(&space_info->groups_sem);
5865 list_add_tail(&cache->list, &space_info->block_groups);
5866 up_write(&space_info->groups_sem);
5868 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5871 set_avail_alloc_bits(root->fs_info, cache->flags);
5872 if (btrfs_chunk_readonly(root, cache->key.objectid))
5873 set_block_group_readonly(cache);
5877 btrfs_free_path(path);
5881 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5882 struct btrfs_root *root, u64 bytes_used,
5883 u64 type, u64 chunk_objectid, u64 chunk_offset,
5887 struct btrfs_root *extent_root;
5888 struct btrfs_block_group_cache *cache;
5890 extent_root = root->fs_info->extent_root;
5892 root->fs_info->last_trans_new_blockgroup = trans->transid;
5894 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5898 cache->key.objectid = chunk_offset;
5899 cache->key.offset = size;
5900 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5901 atomic_set(&cache->count, 1);
5902 spin_lock_init(&cache->lock);
5903 mutex_init(&cache->alloc_mutex);
5904 mutex_init(&cache->cache_mutex);
5905 INIT_LIST_HEAD(&cache->list);
5907 btrfs_set_block_group_used(&cache->item, bytes_used);
5908 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5909 cache->flags = type;
5910 btrfs_set_block_group_flags(&cache->item, type);
5912 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5913 &cache->space_info);
5915 down_write(&cache->space_info->groups_sem);
5916 list_add_tail(&cache->list, &cache->space_info->block_groups);
5917 up_write(&cache->space_info->groups_sem);
5919 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5922 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5923 sizeof(cache->item));
5926 finish_current_insert(trans, extent_root, 0);
5927 ret = del_pending_extents(trans, extent_root, 0);
5929 set_avail_alloc_bits(extent_root->fs_info, type);
5934 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5935 struct btrfs_root *root, u64 group_start)
5937 struct btrfs_path *path;
5938 struct btrfs_block_group_cache *block_group;
5939 struct btrfs_key key;
5942 root = root->fs_info->extent_root;
5944 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5945 BUG_ON(!block_group);
5946 BUG_ON(!block_group->ro);
5948 memcpy(&key, &block_group->key, sizeof(key));
5950 path = btrfs_alloc_path();
5953 btrfs_remove_free_space_cache(block_group);
5954 rb_erase(&block_group->cache_node,
5955 &root->fs_info->block_group_cache_tree);
5956 down_write(&block_group->space_info->groups_sem);
5957 list_del(&block_group->list);
5958 up_write(&block_group->space_info->groups_sem);
5960 spin_lock(&block_group->space_info->lock);
5961 block_group->space_info->total_bytes -= block_group->key.offset;
5962 block_group->space_info->bytes_readonly -= block_group->key.offset;
5963 spin_unlock(&block_group->space_info->lock);
5964 block_group->space_info->full = 0;
5966 put_block_group(block_group);
5967 put_block_group(block_group);
5969 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5975 ret = btrfs_del_item(trans, root, path);
5977 btrfs_free_path(path);