1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007,2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/rbtree.h>
12 #include "transaction.h"
13 #include "print-tree.h"
18 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
19 *root, struct btrfs_path *path, int level);
20 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
21 const struct btrfs_key *ins_key, struct btrfs_path *path,
22 int data_size, int extend);
23 static int push_node_left(struct btrfs_trans_handle *trans,
24 struct extent_buffer *dst,
25 struct extent_buffer *src, int empty);
26 static int balance_node_right(struct btrfs_trans_handle *trans,
27 struct extent_buffer *dst_buf,
28 struct extent_buffer *src_buf);
29 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
32 static const struct btrfs_csums {
35 const char driver[12];
37 [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
38 [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
39 [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
40 [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
41 .driver = "blake2b-256" },
44 int btrfs_super_csum_size(const struct btrfs_super_block *s)
46 u16 t = btrfs_super_csum_type(s);
48 * csum type is validated at mount time
50 return btrfs_csums[t].size;
53 const char *btrfs_super_csum_name(u16 csum_type)
55 /* csum type is validated at mount time */
56 return btrfs_csums[csum_type].name;
60 * Return driver name if defined, otherwise the name that's also a valid driver
63 const char *btrfs_super_csum_driver(u16 csum_type)
65 /* csum type is validated at mount time */
66 return btrfs_csums[csum_type].driver[0] ?
67 btrfs_csums[csum_type].driver :
68 btrfs_csums[csum_type].name;
71 size_t __const btrfs_get_num_csums(void)
73 return ARRAY_SIZE(btrfs_csums);
76 struct btrfs_path *btrfs_alloc_path(void)
78 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
81 /* this also releases the path */
82 void btrfs_free_path(struct btrfs_path *p)
86 btrfs_release_path(p);
87 kmem_cache_free(btrfs_path_cachep, p);
91 * path release drops references on the extent buffers in the path
92 * and it drops any locks held by this path
94 * It is safe to call this on paths that no locks or extent buffers held.
96 noinline void btrfs_release_path(struct btrfs_path *p)
100 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
105 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
108 free_extent_buffer(p->nodes[i]);
114 * safely gets a reference on the root node of a tree. A lock
115 * is not taken, so a concurrent writer may put a different node
116 * at the root of the tree. See btrfs_lock_root_node for the
119 * The extent buffer returned by this has a reference taken, so
120 * it won't disappear. It may stop being the root of the tree
121 * at any time because there are no locks held.
123 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
125 struct extent_buffer *eb;
129 eb = rcu_dereference(root->node);
132 * RCU really hurts here, we could free up the root node because
133 * it was COWed but we may not get the new root node yet so do
134 * the inc_not_zero dance and if it doesn't work then
135 * synchronize_rcu and try again.
137 if (atomic_inc_not_zero(&eb->refs)) {
148 * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
149 * just get put onto a simple dirty list. Transaction walks this list to make
150 * sure they get properly updated on disk.
152 static void add_root_to_dirty_list(struct btrfs_root *root)
154 struct btrfs_fs_info *fs_info = root->fs_info;
156 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
157 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
160 spin_lock(&fs_info->trans_lock);
161 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
162 /* Want the extent tree to be the last on the list */
163 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
164 list_move_tail(&root->dirty_list,
165 &fs_info->dirty_cowonly_roots);
167 list_move(&root->dirty_list,
168 &fs_info->dirty_cowonly_roots);
170 spin_unlock(&fs_info->trans_lock);
174 * used by snapshot creation to make a copy of a root for a tree with
175 * a given objectid. The buffer with the new root node is returned in
176 * cow_ret, and this func returns zero on success or a negative error code.
178 int btrfs_copy_root(struct btrfs_trans_handle *trans,
179 struct btrfs_root *root,
180 struct extent_buffer *buf,
181 struct extent_buffer **cow_ret, u64 new_root_objectid)
183 struct btrfs_fs_info *fs_info = root->fs_info;
184 struct extent_buffer *cow;
187 struct btrfs_disk_key disk_key;
189 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
190 trans->transid != fs_info->running_transaction->transid);
191 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
192 trans->transid != root->last_trans);
194 level = btrfs_header_level(buf);
196 btrfs_item_key(buf, &disk_key, 0);
198 btrfs_node_key(buf, &disk_key, 0);
200 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
201 &disk_key, level, buf->start, 0);
205 copy_extent_buffer_full(cow, buf);
206 btrfs_set_header_bytenr(cow, cow->start);
207 btrfs_set_header_generation(cow, trans->transid);
208 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
209 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
210 BTRFS_HEADER_FLAG_RELOC);
211 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
212 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
214 btrfs_set_header_owner(cow, new_root_objectid);
216 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
218 WARN_ON(btrfs_header_generation(buf) > trans->transid);
219 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
220 ret = btrfs_inc_ref(trans, root, cow, 1);
222 ret = btrfs_inc_ref(trans, root, cow, 0);
227 btrfs_mark_buffer_dirty(cow);
236 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
237 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
239 MOD_LOG_ROOT_REPLACE,
242 struct tree_mod_root {
247 struct tree_mod_elem {
253 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
256 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
259 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
260 struct btrfs_disk_key key;
263 /* this is used for op == MOD_LOG_MOVE_KEYS */
269 /* this is used for op == MOD_LOG_ROOT_REPLACE */
270 struct tree_mod_root old_root;
274 * Pull a new tree mod seq number for our operation.
276 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
278 return atomic64_inc_return(&fs_info->tree_mod_seq);
282 * This adds a new blocker to the tree mod log's blocker list if the @elem
283 * passed does not already have a sequence number set. So when a caller expects
284 * to record tree modifications, it should ensure to set elem->seq to zero
285 * before calling btrfs_get_tree_mod_seq.
286 * Returns a fresh, unused tree log modification sequence number, even if no new
289 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
290 struct seq_list *elem)
292 write_lock(&fs_info->tree_mod_log_lock);
294 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
295 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
297 write_unlock(&fs_info->tree_mod_log_lock);
302 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
303 struct seq_list *elem)
305 struct rb_root *tm_root;
306 struct rb_node *node;
307 struct rb_node *next;
308 struct tree_mod_elem *tm;
309 u64 min_seq = (u64)-1;
310 u64 seq_putting = elem->seq;
315 write_lock(&fs_info->tree_mod_log_lock);
316 list_del(&elem->list);
319 if (!list_empty(&fs_info->tree_mod_seq_list)) {
320 struct seq_list *first;
322 first = list_first_entry(&fs_info->tree_mod_seq_list,
323 struct seq_list, list);
324 if (seq_putting > first->seq) {
326 * Blocker with lower sequence number exists, we
327 * cannot remove anything from the log.
329 write_unlock(&fs_info->tree_mod_log_lock);
332 min_seq = first->seq;
336 * anything that's lower than the lowest existing (read: blocked)
337 * sequence number can be removed from the tree.
339 tm_root = &fs_info->tree_mod_log;
340 for (node = rb_first(tm_root); node; node = next) {
341 next = rb_next(node);
342 tm = rb_entry(node, struct tree_mod_elem, node);
343 if (tm->seq >= min_seq)
345 rb_erase(node, tm_root);
348 write_unlock(&fs_info->tree_mod_log_lock);
352 * key order of the log:
353 * node/leaf start address -> sequence
355 * The 'start address' is the logical address of the *new* root node
356 * for root replace operations, or the logical address of the affected
357 * block for all other operations.
360 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
362 struct rb_root *tm_root;
363 struct rb_node **new;
364 struct rb_node *parent = NULL;
365 struct tree_mod_elem *cur;
367 lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
369 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
371 tm_root = &fs_info->tree_mod_log;
372 new = &tm_root->rb_node;
374 cur = rb_entry(*new, struct tree_mod_elem, node);
376 if (cur->logical < tm->logical)
377 new = &((*new)->rb_left);
378 else if (cur->logical > tm->logical)
379 new = &((*new)->rb_right);
380 else if (cur->seq < tm->seq)
381 new = &((*new)->rb_left);
382 else if (cur->seq > tm->seq)
383 new = &((*new)->rb_right);
388 rb_link_node(&tm->node, parent, new);
389 rb_insert_color(&tm->node, tm_root);
394 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
395 * returns zero with the tree_mod_log_lock acquired. The caller must hold
396 * this until all tree mod log insertions are recorded in the rb tree and then
397 * write unlock fs_info::tree_mod_log_lock.
399 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
400 struct extent_buffer *eb) {
402 if (list_empty(&(fs_info)->tree_mod_seq_list))
404 if (eb && btrfs_header_level(eb) == 0)
407 write_lock(&fs_info->tree_mod_log_lock);
408 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
409 write_unlock(&fs_info->tree_mod_log_lock);
416 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
417 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
418 struct extent_buffer *eb)
421 if (list_empty(&(fs_info)->tree_mod_seq_list))
423 if (eb && btrfs_header_level(eb) == 0)
429 static struct tree_mod_elem *
430 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
431 enum mod_log_op op, gfp_t flags)
433 struct tree_mod_elem *tm;
435 tm = kzalloc(sizeof(*tm), flags);
439 tm->logical = eb->start;
440 if (op != MOD_LOG_KEY_ADD) {
441 btrfs_node_key(eb, &tm->key, slot);
442 tm->blockptr = btrfs_node_blockptr(eb, slot);
446 tm->generation = btrfs_node_ptr_generation(eb, slot);
447 RB_CLEAR_NODE(&tm->node);
452 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
453 enum mod_log_op op, gfp_t flags)
455 struct tree_mod_elem *tm;
458 if (!tree_mod_need_log(eb->fs_info, eb))
461 tm = alloc_tree_mod_elem(eb, slot, op, flags);
465 if (tree_mod_dont_log(eb->fs_info, eb)) {
470 ret = __tree_mod_log_insert(eb->fs_info, tm);
471 write_unlock(&eb->fs_info->tree_mod_log_lock);
478 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
479 int dst_slot, int src_slot, int nr_items)
481 struct tree_mod_elem *tm = NULL;
482 struct tree_mod_elem **tm_list = NULL;
487 if (!tree_mod_need_log(eb->fs_info, eb))
490 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
494 tm = kzalloc(sizeof(*tm), GFP_NOFS);
500 tm->logical = eb->start;
502 tm->move.dst_slot = dst_slot;
503 tm->move.nr_items = nr_items;
504 tm->op = MOD_LOG_MOVE_KEYS;
506 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
507 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
508 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
515 if (tree_mod_dont_log(eb->fs_info, eb))
520 * When we override something during the move, we log these removals.
521 * This can only happen when we move towards the beginning of the
522 * buffer, i.e. dst_slot < src_slot.
524 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
525 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
530 ret = __tree_mod_log_insert(eb->fs_info, tm);
533 write_unlock(&eb->fs_info->tree_mod_log_lock);
538 for (i = 0; i < nr_items; i++) {
539 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
540 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
544 write_unlock(&eb->fs_info->tree_mod_log_lock);
552 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
553 struct tree_mod_elem **tm_list,
559 for (i = nritems - 1; i >= 0; i--) {
560 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
562 for (j = nritems - 1; j > i; j--)
563 rb_erase(&tm_list[j]->node,
564 &fs_info->tree_mod_log);
572 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
573 struct extent_buffer *new_root, int log_removal)
575 struct btrfs_fs_info *fs_info = old_root->fs_info;
576 struct tree_mod_elem *tm = NULL;
577 struct tree_mod_elem **tm_list = NULL;
582 if (!tree_mod_need_log(fs_info, NULL))
585 if (log_removal && btrfs_header_level(old_root) > 0) {
586 nritems = btrfs_header_nritems(old_root);
587 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
593 for (i = 0; i < nritems; i++) {
594 tm_list[i] = alloc_tree_mod_elem(old_root, i,
595 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
603 tm = kzalloc(sizeof(*tm), GFP_NOFS);
609 tm->logical = new_root->start;
610 tm->old_root.logical = old_root->start;
611 tm->old_root.level = btrfs_header_level(old_root);
612 tm->generation = btrfs_header_generation(old_root);
613 tm->op = MOD_LOG_ROOT_REPLACE;
615 if (tree_mod_dont_log(fs_info, NULL))
619 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
621 ret = __tree_mod_log_insert(fs_info, tm);
623 write_unlock(&fs_info->tree_mod_log_lock);
632 for (i = 0; i < nritems; i++)
641 static struct tree_mod_elem *
642 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
645 struct rb_root *tm_root;
646 struct rb_node *node;
647 struct tree_mod_elem *cur = NULL;
648 struct tree_mod_elem *found = NULL;
650 read_lock(&fs_info->tree_mod_log_lock);
651 tm_root = &fs_info->tree_mod_log;
652 node = tm_root->rb_node;
654 cur = rb_entry(node, struct tree_mod_elem, node);
655 if (cur->logical < start) {
656 node = node->rb_left;
657 } else if (cur->logical > start) {
658 node = node->rb_right;
659 } else if (cur->seq < min_seq) {
660 node = node->rb_left;
661 } else if (!smallest) {
662 /* we want the node with the highest seq */
664 BUG_ON(found->seq > cur->seq);
666 node = node->rb_left;
667 } else if (cur->seq > min_seq) {
668 /* we want the node with the smallest seq */
670 BUG_ON(found->seq < cur->seq);
672 node = node->rb_right;
678 read_unlock(&fs_info->tree_mod_log_lock);
684 * this returns the element from the log with the smallest time sequence
685 * value that's in the log (the oldest log item). any element with a time
686 * sequence lower than min_seq will be ignored.
688 static struct tree_mod_elem *
689 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
692 return __tree_mod_log_search(fs_info, start, min_seq, 1);
696 * this returns the element from the log with the largest time sequence
697 * value that's in the log (the most recent log item). any element with
698 * a time sequence lower than min_seq will be ignored.
700 static struct tree_mod_elem *
701 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
703 return __tree_mod_log_search(fs_info, start, min_seq, 0);
706 static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst,
707 struct extent_buffer *src, unsigned long dst_offset,
708 unsigned long src_offset, int nr_items)
710 struct btrfs_fs_info *fs_info = dst->fs_info;
712 struct tree_mod_elem **tm_list = NULL;
713 struct tree_mod_elem **tm_list_add, **tm_list_rem;
717 if (!tree_mod_need_log(fs_info, NULL))
720 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
723 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
728 tm_list_add = tm_list;
729 tm_list_rem = tm_list + nr_items;
730 for (i = 0; i < nr_items; i++) {
731 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
732 MOD_LOG_KEY_REMOVE, GFP_NOFS);
733 if (!tm_list_rem[i]) {
738 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
739 MOD_LOG_KEY_ADD, GFP_NOFS);
740 if (!tm_list_add[i]) {
746 if (tree_mod_dont_log(fs_info, NULL))
750 for (i = 0; i < nr_items; i++) {
751 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
754 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
759 write_unlock(&fs_info->tree_mod_log_lock);
765 for (i = 0; i < nr_items * 2; i++) {
766 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
767 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
771 write_unlock(&fs_info->tree_mod_log_lock);
777 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
779 struct tree_mod_elem **tm_list = NULL;
784 if (btrfs_header_level(eb) == 0)
787 if (!tree_mod_need_log(eb->fs_info, NULL))
790 nritems = btrfs_header_nritems(eb);
791 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
795 for (i = 0; i < nritems; i++) {
796 tm_list[i] = alloc_tree_mod_elem(eb, i,
797 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
804 if (tree_mod_dont_log(eb->fs_info, eb))
807 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
808 write_unlock(&eb->fs_info->tree_mod_log_lock);
816 for (i = 0; i < nritems; i++)
824 * check if the tree block can be shared by multiple trees
826 int btrfs_block_can_be_shared(struct btrfs_root *root,
827 struct extent_buffer *buf)
830 * Tree blocks not in shareable trees and tree roots are never shared.
831 * If a block was allocated after the last snapshot and the block was
832 * not allocated by tree relocation, we know the block is not shared.
834 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
835 buf != root->node && buf != root->commit_root &&
836 (btrfs_header_generation(buf) <=
837 btrfs_root_last_snapshot(&root->root_item) ||
838 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
844 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
845 struct btrfs_root *root,
846 struct extent_buffer *buf,
847 struct extent_buffer *cow,
850 struct btrfs_fs_info *fs_info = root->fs_info;
858 * Backrefs update rules:
860 * Always use full backrefs for extent pointers in tree block
861 * allocated by tree relocation.
863 * If a shared tree block is no longer referenced by its owner
864 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
865 * use full backrefs for extent pointers in tree block.
867 * If a tree block is been relocating
868 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
869 * use full backrefs for extent pointers in tree block.
870 * The reason for this is some operations (such as drop tree)
871 * are only allowed for blocks use full backrefs.
874 if (btrfs_block_can_be_shared(root, buf)) {
875 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
876 btrfs_header_level(buf), 1,
882 btrfs_handle_fs_error(fs_info, ret, NULL);
887 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
888 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
889 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
894 owner = btrfs_header_owner(buf);
895 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
896 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
899 if ((owner == root->root_key.objectid ||
900 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
901 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
902 ret = btrfs_inc_ref(trans, root, buf, 1);
906 if (root->root_key.objectid ==
907 BTRFS_TREE_RELOC_OBJECTID) {
908 ret = btrfs_dec_ref(trans, root, buf, 0);
911 ret = btrfs_inc_ref(trans, root, cow, 1);
915 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
918 if (root->root_key.objectid ==
919 BTRFS_TREE_RELOC_OBJECTID)
920 ret = btrfs_inc_ref(trans, root, cow, 1);
922 ret = btrfs_inc_ref(trans, root, cow, 0);
926 if (new_flags != 0) {
927 int level = btrfs_header_level(buf);
929 ret = btrfs_set_disk_extent_flags(trans, buf,
930 new_flags, level, 0);
935 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
936 if (root->root_key.objectid ==
937 BTRFS_TREE_RELOC_OBJECTID)
938 ret = btrfs_inc_ref(trans, root, cow, 1);
940 ret = btrfs_inc_ref(trans, root, cow, 0);
943 ret = btrfs_dec_ref(trans, root, buf, 1);
947 btrfs_clean_tree_block(buf);
953 static struct extent_buffer *alloc_tree_block_no_bg_flush(
954 struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
957 const struct btrfs_disk_key *disk_key,
962 struct btrfs_fs_info *fs_info = root->fs_info;
963 struct extent_buffer *ret;
966 * If we are COWing a node/leaf from the extent, chunk, device or free
967 * space trees, make sure that we do not finish block group creation of
968 * pending block groups. We do this to avoid a deadlock.
969 * COWing can result in allocation of a new chunk, and flushing pending
970 * block groups (btrfs_create_pending_block_groups()) can be triggered
971 * when finishing allocation of a new chunk. Creation of a pending block
972 * group modifies the extent, chunk, device and free space trees,
973 * therefore we could deadlock with ourselves since we are holding a
974 * lock on an extent buffer that btrfs_create_pending_block_groups() may
976 * For similar reasons, we also need to delay flushing pending block
977 * groups when splitting a leaf or node, from one of those trees, since
978 * we are holding a write lock on it and its parent or when inserting a
979 * new root node for one of those trees.
981 if (root == fs_info->extent_root ||
982 root == fs_info->chunk_root ||
983 root == fs_info->dev_root ||
984 root == fs_info->free_space_root)
985 trans->can_flush_pending_bgs = false;
987 ret = btrfs_alloc_tree_block(trans, root, parent_start,
988 root->root_key.objectid, disk_key, level,
990 trans->can_flush_pending_bgs = true;
996 * does the dirty work in cow of a single block. The parent block (if
997 * supplied) is updated to point to the new cow copy. The new buffer is marked
998 * dirty and returned locked. If you modify the block it needs to be marked
1001 * search_start -- an allocation hint for the new block
1003 * empty_size -- a hint that you plan on doing more cow. This is the size in
1004 * bytes the allocator should try to find free next to the block it returns.
1005 * This is just a hint and may be ignored by the allocator.
1007 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1008 struct btrfs_root *root,
1009 struct extent_buffer *buf,
1010 struct extent_buffer *parent, int parent_slot,
1011 struct extent_buffer **cow_ret,
1012 u64 search_start, u64 empty_size)
1014 struct btrfs_fs_info *fs_info = root->fs_info;
1015 struct btrfs_disk_key disk_key;
1016 struct extent_buffer *cow;
1019 int unlock_orig = 0;
1020 u64 parent_start = 0;
1022 if (*cow_ret == buf)
1025 btrfs_assert_tree_locked(buf);
1027 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
1028 trans->transid != fs_info->running_transaction->transid);
1029 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
1030 trans->transid != root->last_trans);
1032 level = btrfs_header_level(buf);
1035 btrfs_item_key(buf, &disk_key, 0);
1037 btrfs_node_key(buf, &disk_key, 0);
1039 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1040 parent_start = parent->start;
1042 cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1043 level, search_start, empty_size);
1045 return PTR_ERR(cow);
1047 /* cow is set to blocking by btrfs_init_new_buffer */
1049 copy_extent_buffer_full(cow, buf);
1050 btrfs_set_header_bytenr(cow, cow->start);
1051 btrfs_set_header_generation(cow, trans->transid);
1052 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1053 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1054 BTRFS_HEADER_FLAG_RELOC);
1055 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1056 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1058 btrfs_set_header_owner(cow, root->root_key.objectid);
1060 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1062 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1064 btrfs_abort_transaction(trans, ret);
1068 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
1069 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1071 btrfs_abort_transaction(trans, ret);
1076 if (buf == root->node) {
1077 WARN_ON(parent && parent != buf);
1078 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1079 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1080 parent_start = buf->start;
1082 atomic_inc(&cow->refs);
1083 ret = tree_mod_log_insert_root(root->node, cow, 1);
1085 rcu_assign_pointer(root->node, cow);
1087 btrfs_free_tree_block(trans, root, buf, parent_start,
1089 free_extent_buffer(buf);
1090 add_root_to_dirty_list(root);
1092 WARN_ON(trans->transid != btrfs_header_generation(parent));
1093 tree_mod_log_insert_key(parent, parent_slot,
1094 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1095 btrfs_set_node_blockptr(parent, parent_slot,
1097 btrfs_set_node_ptr_generation(parent, parent_slot,
1099 btrfs_mark_buffer_dirty(parent);
1101 ret = tree_mod_log_free_eb(buf);
1103 btrfs_abort_transaction(trans, ret);
1107 btrfs_free_tree_block(trans, root, buf, parent_start,
1111 btrfs_tree_unlock(buf);
1112 free_extent_buffer_stale(buf);
1113 btrfs_mark_buffer_dirty(cow);
1119 * returns the logical address of the oldest predecessor of the given root.
1120 * entries older than time_seq are ignored.
1122 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1123 struct extent_buffer *eb_root, u64 time_seq)
1125 struct tree_mod_elem *tm;
1126 struct tree_mod_elem *found = NULL;
1127 u64 root_logical = eb_root->start;
1134 * the very last operation that's logged for a root is the
1135 * replacement operation (if it is replaced at all). this has
1136 * the logical address of the *new* root, making it the very
1137 * first operation that's logged for this root.
1140 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1145 * if there are no tree operation for the oldest root, we simply
1146 * return it. this should only happen if that (old) root is at
1153 * if there's an operation that's not a root replacement, we
1154 * found the oldest version of our root. normally, we'll find a
1155 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1157 if (tm->op != MOD_LOG_ROOT_REPLACE)
1161 root_logical = tm->old_root.logical;
1165 /* if there's no old root to return, return what we found instead */
1173 * tm is a pointer to the first operation to rewind within eb. then, all
1174 * previous operations will be rewound (until we reach something older than
1178 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1179 u64 time_seq, struct tree_mod_elem *first_tm)
1182 struct rb_node *next;
1183 struct tree_mod_elem *tm = first_tm;
1184 unsigned long o_dst;
1185 unsigned long o_src;
1186 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1188 n = btrfs_header_nritems(eb);
1189 read_lock(&fs_info->tree_mod_log_lock);
1190 while (tm && tm->seq >= time_seq) {
1192 * all the operations are recorded with the operator used for
1193 * the modification. as we're going backwards, we do the
1194 * opposite of each operation here.
1197 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1198 BUG_ON(tm->slot < n);
1200 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1201 case MOD_LOG_KEY_REMOVE:
1202 btrfs_set_node_key(eb, &tm->key, tm->slot);
1203 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1204 btrfs_set_node_ptr_generation(eb, tm->slot,
1208 case MOD_LOG_KEY_REPLACE:
1209 BUG_ON(tm->slot >= n);
1210 btrfs_set_node_key(eb, &tm->key, tm->slot);
1211 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1212 btrfs_set_node_ptr_generation(eb, tm->slot,
1215 case MOD_LOG_KEY_ADD:
1216 /* if a move operation is needed it's in the log */
1219 case MOD_LOG_MOVE_KEYS:
1220 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1221 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1222 memmove_extent_buffer(eb, o_dst, o_src,
1223 tm->move.nr_items * p_size);
1225 case MOD_LOG_ROOT_REPLACE:
1227 * this operation is special. for roots, this must be
1228 * handled explicitly before rewinding.
1229 * for non-roots, this operation may exist if the node
1230 * was a root: root A -> child B; then A gets empty and
1231 * B is promoted to the new root. in the mod log, we'll
1232 * have a root-replace operation for B, a tree block
1233 * that is no root. we simply ignore that operation.
1237 next = rb_next(&tm->node);
1240 tm = rb_entry(next, struct tree_mod_elem, node);
1241 if (tm->logical != first_tm->logical)
1244 read_unlock(&fs_info->tree_mod_log_lock);
1245 btrfs_set_header_nritems(eb, n);
1249 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1250 * is returned. If rewind operations happen, a fresh buffer is returned. The
1251 * returned buffer is always read-locked. If the returned buffer is not the
1252 * input buffer, the lock on the input buffer is released and the input buffer
1253 * is freed (its refcount is decremented).
1255 static struct extent_buffer *
1256 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1257 struct extent_buffer *eb, u64 time_seq)
1259 struct extent_buffer *eb_rewin;
1260 struct tree_mod_elem *tm;
1265 if (btrfs_header_level(eb) == 0)
1268 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1272 btrfs_set_path_blocking(path);
1273 btrfs_set_lock_blocking_read(eb);
1275 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1276 BUG_ON(tm->slot != 0);
1277 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1279 btrfs_tree_read_unlock_blocking(eb);
1280 free_extent_buffer(eb);
1283 btrfs_set_header_bytenr(eb_rewin, eb->start);
1284 btrfs_set_header_backref_rev(eb_rewin,
1285 btrfs_header_backref_rev(eb));
1286 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1287 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1289 eb_rewin = btrfs_clone_extent_buffer(eb);
1291 btrfs_tree_read_unlock_blocking(eb);
1292 free_extent_buffer(eb);
1297 btrfs_tree_read_unlock_blocking(eb);
1298 free_extent_buffer(eb);
1300 btrfs_tree_read_lock(eb_rewin);
1301 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1302 WARN_ON(btrfs_header_nritems(eb_rewin) >
1303 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1309 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1310 * value. If there are no changes, the current root->root_node is returned. If
1311 * anything changed in between, there's a fresh buffer allocated on which the
1312 * rewind operations are done. In any case, the returned buffer is read locked.
1313 * Returns NULL on error (with no locks held).
1315 static inline struct extent_buffer *
1316 get_old_root(struct btrfs_root *root, u64 time_seq)
1318 struct btrfs_fs_info *fs_info = root->fs_info;
1319 struct tree_mod_elem *tm;
1320 struct extent_buffer *eb = NULL;
1321 struct extent_buffer *eb_root;
1322 u64 eb_root_owner = 0;
1323 struct extent_buffer *old;
1324 struct tree_mod_root *old_root = NULL;
1325 u64 old_generation = 0;
1329 eb_root = btrfs_read_lock_root_node(root);
1330 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1334 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1335 old_root = &tm->old_root;
1336 old_generation = tm->generation;
1337 logical = old_root->logical;
1338 level = old_root->level;
1340 logical = eb_root->start;
1341 level = btrfs_header_level(eb_root);
1344 tm = tree_mod_log_search(fs_info, logical, time_seq);
1345 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1346 btrfs_tree_read_unlock(eb_root);
1347 free_extent_buffer(eb_root);
1348 old = read_tree_block(fs_info, logical, 0, level, NULL);
1349 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1351 free_extent_buffer(old);
1353 "failed to read tree block %llu from get_old_root",
1356 eb = btrfs_clone_extent_buffer(old);
1357 free_extent_buffer(old);
1359 } else if (old_root) {
1360 eb_root_owner = btrfs_header_owner(eb_root);
1361 btrfs_tree_read_unlock(eb_root);
1362 free_extent_buffer(eb_root);
1363 eb = alloc_dummy_extent_buffer(fs_info, logical);
1365 btrfs_set_lock_blocking_read(eb_root);
1366 eb = btrfs_clone_extent_buffer(eb_root);
1367 btrfs_tree_read_unlock_blocking(eb_root);
1368 free_extent_buffer(eb_root);
1373 btrfs_tree_read_lock(eb);
1375 btrfs_set_header_bytenr(eb, eb->start);
1376 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1377 btrfs_set_header_owner(eb, eb_root_owner);
1378 btrfs_set_header_level(eb, old_root->level);
1379 btrfs_set_header_generation(eb, old_generation);
1382 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1384 WARN_ON(btrfs_header_level(eb) != 0);
1385 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1390 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1392 struct tree_mod_elem *tm;
1394 struct extent_buffer *eb_root = btrfs_root_node(root);
1396 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1397 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1398 level = tm->old_root.level;
1400 level = btrfs_header_level(eb_root);
1402 free_extent_buffer(eb_root);
1407 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root,
1409 struct extent_buffer *buf)
1411 if (btrfs_is_testing(root->fs_info))
1414 /* Ensure we can see the FORCE_COW bit */
1415 smp_mb__before_atomic();
1418 * We do not need to cow a block if
1419 * 1) this block is not created or changed in this transaction;
1420 * 2) this block does not belong to TREE_RELOC tree;
1421 * 3) the root is not forced COW.
1423 * What is forced COW:
1424 * when we create snapshot during committing the transaction,
1425 * after we've finished copying src root, we must COW the shared
1426 * block to ensure the metadata consistency.
1428 if (btrfs_header_generation(buf) == trans->transid &&
1429 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1430 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1431 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1432 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1438 * cows a single block, see __btrfs_cow_block for the real work.
1439 * This version of it has extra checks so that a block isn't COWed more than
1440 * once per transaction, as long as it hasn't been written yet
1442 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1443 struct btrfs_root *root, struct extent_buffer *buf,
1444 struct extent_buffer *parent, int parent_slot,
1445 struct extent_buffer **cow_ret)
1447 struct btrfs_fs_info *fs_info = root->fs_info;
1451 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1453 "COW'ing blocks on a fs root that's being dropped");
1455 if (trans->transaction != fs_info->running_transaction)
1456 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1458 fs_info->running_transaction->transid);
1460 if (trans->transid != fs_info->generation)
1461 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1462 trans->transid, fs_info->generation);
1464 if (!should_cow_block(trans, root, buf)) {
1465 trans->dirty = true;
1470 search_start = buf->start & ~((u64)SZ_1G - 1);
1473 btrfs_set_lock_blocking_write(parent);
1474 btrfs_set_lock_blocking_write(buf);
1477 * Before CoWing this block for later modification, check if it's
1478 * the subtree root and do the delayed subtree trace if needed.
1480 * Also We don't care about the error, as it's handled internally.
1482 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1483 ret = __btrfs_cow_block(trans, root, buf, parent,
1484 parent_slot, cow_ret, search_start, 0);
1486 trace_btrfs_cow_block(root, buf, *cow_ret);
1492 * helper function for defrag to decide if two blocks pointed to by a
1493 * node are actually close by
1495 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1497 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1499 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1505 * compare two keys in a memcmp fashion
1507 static int comp_keys(const struct btrfs_disk_key *disk,
1508 const struct btrfs_key *k2)
1510 struct btrfs_key k1;
1512 btrfs_disk_key_to_cpu(&k1, disk);
1514 return btrfs_comp_cpu_keys(&k1, k2);
1518 * same as comp_keys only with two btrfs_key's
1520 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1522 if (k1->objectid > k2->objectid)
1524 if (k1->objectid < k2->objectid)
1526 if (k1->type > k2->type)
1528 if (k1->type < k2->type)
1530 if (k1->offset > k2->offset)
1532 if (k1->offset < k2->offset)
1538 * this is used by the defrag code to go through all the
1539 * leaves pointed to by a node and reallocate them so that
1540 * disk order is close to key order
1542 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1543 struct btrfs_root *root, struct extent_buffer *parent,
1544 int start_slot, u64 *last_ret,
1545 struct btrfs_key *progress)
1547 struct btrfs_fs_info *fs_info = root->fs_info;
1548 struct extent_buffer *cur;
1551 u64 search_start = *last_ret;
1561 int progress_passed = 0;
1562 struct btrfs_disk_key disk_key;
1564 parent_level = btrfs_header_level(parent);
1566 WARN_ON(trans->transaction != fs_info->running_transaction);
1567 WARN_ON(trans->transid != fs_info->generation);
1569 parent_nritems = btrfs_header_nritems(parent);
1570 blocksize = fs_info->nodesize;
1571 end_slot = parent_nritems - 1;
1573 if (parent_nritems <= 1)
1576 btrfs_set_lock_blocking_write(parent);
1578 for (i = start_slot; i <= end_slot; i++) {
1579 struct btrfs_key first_key;
1582 btrfs_node_key(parent, &disk_key, i);
1583 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1586 progress_passed = 1;
1587 blocknr = btrfs_node_blockptr(parent, i);
1588 gen = btrfs_node_ptr_generation(parent, i);
1589 btrfs_node_key_to_cpu(parent, &first_key, i);
1590 if (last_block == 0)
1591 last_block = blocknr;
1594 other = btrfs_node_blockptr(parent, i - 1);
1595 close = close_blocks(blocknr, other, blocksize);
1597 if (!close && i < end_slot) {
1598 other = btrfs_node_blockptr(parent, i + 1);
1599 close = close_blocks(blocknr, other, blocksize);
1602 last_block = blocknr;
1606 cur = find_extent_buffer(fs_info, blocknr);
1608 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1611 if (!cur || !uptodate) {
1613 cur = read_tree_block(fs_info, blocknr, gen,
1617 return PTR_ERR(cur);
1618 } else if (!extent_buffer_uptodate(cur)) {
1619 free_extent_buffer(cur);
1622 } else if (!uptodate) {
1623 err = btrfs_read_buffer(cur, gen,
1624 parent_level - 1,&first_key);
1626 free_extent_buffer(cur);
1631 if (search_start == 0)
1632 search_start = last_block;
1634 btrfs_tree_lock(cur);
1635 btrfs_set_lock_blocking_write(cur);
1636 err = __btrfs_cow_block(trans, root, cur, parent, i,
1639 (end_slot - i) * blocksize));
1641 btrfs_tree_unlock(cur);
1642 free_extent_buffer(cur);
1645 search_start = cur->start;
1646 last_block = cur->start;
1647 *last_ret = search_start;
1648 btrfs_tree_unlock(cur);
1649 free_extent_buffer(cur);
1655 * search for key in the extent_buffer. The items start at offset p,
1656 * and they are item_size apart. There are 'max' items in p.
1658 * the slot in the array is returned via slot, and it points to
1659 * the place where you would insert key if it is not found in
1662 * slot may point to max if the key is bigger than all of the keys
1664 static noinline int generic_bin_search(struct extent_buffer *eb,
1665 unsigned long p, int item_size,
1666 const struct btrfs_key *key,
1672 const int key_size = sizeof(struct btrfs_disk_key);
1675 btrfs_err(eb->fs_info,
1676 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1677 __func__, low, high, eb->start,
1678 btrfs_header_owner(eb), btrfs_header_level(eb));
1682 while (low < high) {
1684 unsigned long offset;
1685 struct btrfs_disk_key *tmp;
1686 struct btrfs_disk_key unaligned;
1689 mid = (low + high) / 2;
1690 offset = p + mid * item_size;
1691 oip = offset_in_page(offset);
1693 if (oip + key_size <= PAGE_SIZE) {
1694 const unsigned long idx = offset >> PAGE_SHIFT;
1695 char *kaddr = page_address(eb->pages[idx]);
1697 tmp = (struct btrfs_disk_key *)(kaddr + oip);
1699 read_extent_buffer(eb, &unaligned, offset, key_size);
1703 ret = comp_keys(tmp, key);
1719 * simple bin_search frontend that does the right thing for
1722 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1725 if (btrfs_header_level(eb) == 0)
1726 return generic_bin_search(eb,
1727 offsetof(struct btrfs_leaf, items),
1728 sizeof(struct btrfs_item),
1729 key, btrfs_header_nritems(eb),
1732 return generic_bin_search(eb,
1733 offsetof(struct btrfs_node, ptrs),
1734 sizeof(struct btrfs_key_ptr),
1735 key, btrfs_header_nritems(eb),
1739 static void root_add_used(struct btrfs_root *root, u32 size)
1741 spin_lock(&root->accounting_lock);
1742 btrfs_set_root_used(&root->root_item,
1743 btrfs_root_used(&root->root_item) + size);
1744 spin_unlock(&root->accounting_lock);
1747 static void root_sub_used(struct btrfs_root *root, u32 size)
1749 spin_lock(&root->accounting_lock);
1750 btrfs_set_root_used(&root->root_item,
1751 btrfs_root_used(&root->root_item) - size);
1752 spin_unlock(&root->accounting_lock);
1755 /* given a node and slot number, this reads the blocks it points to. The
1756 * extent buffer is returned with a reference taken (but unlocked).
1758 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1761 int level = btrfs_header_level(parent);
1762 struct extent_buffer *eb;
1763 struct btrfs_key first_key;
1765 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1766 return ERR_PTR(-ENOENT);
1770 btrfs_node_key_to_cpu(parent, &first_key, slot);
1771 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1772 btrfs_node_ptr_generation(parent, slot),
1773 level - 1, &first_key);
1774 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1775 free_extent_buffer(eb);
1783 * node level balancing, used to make sure nodes are in proper order for
1784 * item deletion. We balance from the top down, so we have to make sure
1785 * that a deletion won't leave an node completely empty later on.
1787 static noinline int balance_level(struct btrfs_trans_handle *trans,
1788 struct btrfs_root *root,
1789 struct btrfs_path *path, int level)
1791 struct btrfs_fs_info *fs_info = root->fs_info;
1792 struct extent_buffer *right = NULL;
1793 struct extent_buffer *mid;
1794 struct extent_buffer *left = NULL;
1795 struct extent_buffer *parent = NULL;
1799 int orig_slot = path->slots[level];
1804 mid = path->nodes[level];
1806 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1807 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1808 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1810 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1812 if (level < BTRFS_MAX_LEVEL - 1) {
1813 parent = path->nodes[level + 1];
1814 pslot = path->slots[level + 1];
1818 * deal with the case where there is only one pointer in the root
1819 * by promoting the node below to a root
1822 struct extent_buffer *child;
1824 if (btrfs_header_nritems(mid) != 1)
1827 /* promote the child to a root */
1828 child = btrfs_read_node_slot(mid, 0);
1829 if (IS_ERR(child)) {
1830 ret = PTR_ERR(child);
1831 btrfs_handle_fs_error(fs_info, ret, NULL);
1835 btrfs_tree_lock(child);
1836 btrfs_set_lock_blocking_write(child);
1837 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1839 btrfs_tree_unlock(child);
1840 free_extent_buffer(child);
1844 ret = tree_mod_log_insert_root(root->node, child, 1);
1846 rcu_assign_pointer(root->node, child);
1848 add_root_to_dirty_list(root);
1849 btrfs_tree_unlock(child);
1851 path->locks[level] = 0;
1852 path->nodes[level] = NULL;
1853 btrfs_clean_tree_block(mid);
1854 btrfs_tree_unlock(mid);
1855 /* once for the path */
1856 free_extent_buffer(mid);
1858 root_sub_used(root, mid->len);
1859 btrfs_free_tree_block(trans, root, mid, 0, 1);
1860 /* once for the root ptr */
1861 free_extent_buffer_stale(mid);
1864 if (btrfs_header_nritems(mid) >
1865 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1868 left = btrfs_read_node_slot(parent, pslot - 1);
1873 btrfs_tree_lock(left);
1874 btrfs_set_lock_blocking_write(left);
1875 wret = btrfs_cow_block(trans, root, left,
1876 parent, pslot - 1, &left);
1883 right = btrfs_read_node_slot(parent, pslot + 1);
1888 btrfs_tree_lock(right);
1889 btrfs_set_lock_blocking_write(right);
1890 wret = btrfs_cow_block(trans, root, right,
1891 parent, pslot + 1, &right);
1898 /* first, try to make some room in the middle buffer */
1900 orig_slot += btrfs_header_nritems(left);
1901 wret = push_node_left(trans, left, mid, 1);
1907 * then try to empty the right most buffer into the middle
1910 wret = push_node_left(trans, mid, right, 1);
1911 if (wret < 0 && wret != -ENOSPC)
1913 if (btrfs_header_nritems(right) == 0) {
1914 btrfs_clean_tree_block(right);
1915 btrfs_tree_unlock(right);
1916 del_ptr(root, path, level + 1, pslot + 1);
1917 root_sub_used(root, right->len);
1918 btrfs_free_tree_block(trans, root, right, 0, 1);
1919 free_extent_buffer_stale(right);
1922 struct btrfs_disk_key right_key;
1923 btrfs_node_key(right, &right_key, 0);
1924 ret = tree_mod_log_insert_key(parent, pslot + 1,
1925 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1927 btrfs_set_node_key(parent, &right_key, pslot + 1);
1928 btrfs_mark_buffer_dirty(parent);
1931 if (btrfs_header_nritems(mid) == 1) {
1933 * we're not allowed to leave a node with one item in the
1934 * tree during a delete. A deletion from lower in the tree
1935 * could try to delete the only pointer in this node.
1936 * So, pull some keys from the left.
1937 * There has to be a left pointer at this point because
1938 * otherwise we would have pulled some pointers from the
1943 btrfs_handle_fs_error(fs_info, ret, NULL);
1946 wret = balance_node_right(trans, mid, left);
1952 wret = push_node_left(trans, left, mid, 1);
1958 if (btrfs_header_nritems(mid) == 0) {
1959 btrfs_clean_tree_block(mid);
1960 btrfs_tree_unlock(mid);
1961 del_ptr(root, path, level + 1, pslot);
1962 root_sub_used(root, mid->len);
1963 btrfs_free_tree_block(trans, root, mid, 0, 1);
1964 free_extent_buffer_stale(mid);
1967 /* update the parent key to reflect our changes */
1968 struct btrfs_disk_key mid_key;
1969 btrfs_node_key(mid, &mid_key, 0);
1970 ret = tree_mod_log_insert_key(parent, pslot,
1971 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1973 btrfs_set_node_key(parent, &mid_key, pslot);
1974 btrfs_mark_buffer_dirty(parent);
1977 /* update the path */
1979 if (btrfs_header_nritems(left) > orig_slot) {
1980 atomic_inc(&left->refs);
1981 /* left was locked after cow */
1982 path->nodes[level] = left;
1983 path->slots[level + 1] -= 1;
1984 path->slots[level] = orig_slot;
1986 btrfs_tree_unlock(mid);
1987 free_extent_buffer(mid);
1990 orig_slot -= btrfs_header_nritems(left);
1991 path->slots[level] = orig_slot;
1994 /* double check we haven't messed things up */
1996 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2000 btrfs_tree_unlock(right);
2001 free_extent_buffer(right);
2004 if (path->nodes[level] != left)
2005 btrfs_tree_unlock(left);
2006 free_extent_buffer(left);
2011 /* Node balancing for insertion. Here we only split or push nodes around
2012 * when they are completely full. This is also done top down, so we
2013 * have to be pessimistic.
2015 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2016 struct btrfs_root *root,
2017 struct btrfs_path *path, int level)
2019 struct btrfs_fs_info *fs_info = root->fs_info;
2020 struct extent_buffer *right = NULL;
2021 struct extent_buffer *mid;
2022 struct extent_buffer *left = NULL;
2023 struct extent_buffer *parent = NULL;
2027 int orig_slot = path->slots[level];
2032 mid = path->nodes[level];
2033 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2035 if (level < BTRFS_MAX_LEVEL - 1) {
2036 parent = path->nodes[level + 1];
2037 pslot = path->slots[level + 1];
2043 left = btrfs_read_node_slot(parent, pslot - 1);
2047 /* first, try to make some room in the middle buffer */
2051 btrfs_tree_lock(left);
2052 btrfs_set_lock_blocking_write(left);
2054 left_nr = btrfs_header_nritems(left);
2055 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2058 ret = btrfs_cow_block(trans, root, left, parent,
2063 wret = push_node_left(trans, left, mid, 0);
2069 struct btrfs_disk_key disk_key;
2070 orig_slot += left_nr;
2071 btrfs_node_key(mid, &disk_key, 0);
2072 ret = tree_mod_log_insert_key(parent, pslot,
2073 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2075 btrfs_set_node_key(parent, &disk_key, pslot);
2076 btrfs_mark_buffer_dirty(parent);
2077 if (btrfs_header_nritems(left) > orig_slot) {
2078 path->nodes[level] = left;
2079 path->slots[level + 1] -= 1;
2080 path->slots[level] = orig_slot;
2081 btrfs_tree_unlock(mid);
2082 free_extent_buffer(mid);
2085 btrfs_header_nritems(left);
2086 path->slots[level] = orig_slot;
2087 btrfs_tree_unlock(left);
2088 free_extent_buffer(left);
2092 btrfs_tree_unlock(left);
2093 free_extent_buffer(left);
2095 right = btrfs_read_node_slot(parent, pslot + 1);
2100 * then try to empty the right most buffer into the middle
2105 btrfs_tree_lock(right);
2106 btrfs_set_lock_blocking_write(right);
2108 right_nr = btrfs_header_nritems(right);
2109 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2112 ret = btrfs_cow_block(trans, root, right,
2118 wret = balance_node_right(trans, right, mid);
2124 struct btrfs_disk_key disk_key;
2126 btrfs_node_key(right, &disk_key, 0);
2127 ret = tree_mod_log_insert_key(parent, pslot + 1,
2128 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2130 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2131 btrfs_mark_buffer_dirty(parent);
2133 if (btrfs_header_nritems(mid) <= orig_slot) {
2134 path->nodes[level] = right;
2135 path->slots[level + 1] += 1;
2136 path->slots[level] = orig_slot -
2137 btrfs_header_nritems(mid);
2138 btrfs_tree_unlock(mid);
2139 free_extent_buffer(mid);
2141 btrfs_tree_unlock(right);
2142 free_extent_buffer(right);
2146 btrfs_tree_unlock(right);
2147 free_extent_buffer(right);
2153 * readahead one full node of leaves, finding things that are close
2154 * to the block in 'slot', and triggering ra on them.
2156 static void reada_for_search(struct btrfs_fs_info *fs_info,
2157 struct btrfs_path *path,
2158 int level, int slot, u64 objectid)
2160 struct extent_buffer *node;
2161 struct btrfs_disk_key disk_key;
2166 struct extent_buffer *eb;
2174 if (!path->nodes[level])
2177 node = path->nodes[level];
2179 search = btrfs_node_blockptr(node, slot);
2180 blocksize = fs_info->nodesize;
2181 eb = find_extent_buffer(fs_info, search);
2183 free_extent_buffer(eb);
2189 nritems = btrfs_header_nritems(node);
2193 if (path->reada == READA_BACK) {
2197 } else if (path->reada == READA_FORWARD) {
2202 if (path->reada == READA_BACK && objectid) {
2203 btrfs_node_key(node, &disk_key, nr);
2204 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2207 search = btrfs_node_blockptr(node, nr);
2208 if ((search <= target && target - search <= 65536) ||
2209 (search > target && search - target <= 65536)) {
2210 readahead_tree_block(fs_info, search);
2214 if ((nread > 65536 || nscan > 32))
2219 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2220 struct btrfs_path *path, int level)
2224 struct extent_buffer *parent;
2225 struct extent_buffer *eb;
2230 parent = path->nodes[level + 1];
2234 nritems = btrfs_header_nritems(parent);
2235 slot = path->slots[level + 1];
2238 block1 = btrfs_node_blockptr(parent, slot - 1);
2239 gen = btrfs_node_ptr_generation(parent, slot - 1);
2240 eb = find_extent_buffer(fs_info, block1);
2242 * if we get -eagain from btrfs_buffer_uptodate, we
2243 * don't want to return eagain here. That will loop
2246 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2248 free_extent_buffer(eb);
2250 if (slot + 1 < nritems) {
2251 block2 = btrfs_node_blockptr(parent, slot + 1);
2252 gen = btrfs_node_ptr_generation(parent, slot + 1);
2253 eb = find_extent_buffer(fs_info, block2);
2254 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2256 free_extent_buffer(eb);
2260 readahead_tree_block(fs_info, block1);
2262 readahead_tree_block(fs_info, block2);
2267 * when we walk down the tree, it is usually safe to unlock the higher layers
2268 * in the tree. The exceptions are when our path goes through slot 0, because
2269 * operations on the tree might require changing key pointers higher up in the
2272 * callers might also have set path->keep_locks, which tells this code to keep
2273 * the lock if the path points to the last slot in the block. This is part of
2274 * walking through the tree, and selecting the next slot in the higher block.
2276 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2277 * if lowest_unlock is 1, level 0 won't be unlocked
2279 static noinline void unlock_up(struct btrfs_path *path, int level,
2280 int lowest_unlock, int min_write_lock_level,
2281 int *write_lock_level)
2284 int skip_level = level;
2286 struct extent_buffer *t;
2288 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2289 if (!path->nodes[i])
2291 if (!path->locks[i])
2293 if (!no_skips && path->slots[i] == 0) {
2297 if (!no_skips && path->keep_locks) {
2300 nritems = btrfs_header_nritems(t);
2301 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2306 if (skip_level < i && i >= lowest_unlock)
2310 if (i >= lowest_unlock && i > skip_level) {
2311 btrfs_tree_unlock_rw(t, path->locks[i]);
2313 if (write_lock_level &&
2314 i > min_write_lock_level &&
2315 i <= *write_lock_level) {
2316 *write_lock_level = i - 1;
2323 * helper function for btrfs_search_slot. The goal is to find a block
2324 * in cache without setting the path to blocking. If we find the block
2325 * we return zero and the path is unchanged.
2327 * If we can't find the block, we set the path blocking and do some
2328 * reada. -EAGAIN is returned and the search must be repeated.
2331 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2332 struct extent_buffer **eb_ret, int level, int slot,
2333 const struct btrfs_key *key)
2335 struct btrfs_fs_info *fs_info = root->fs_info;
2338 struct extent_buffer *tmp;
2339 struct btrfs_key first_key;
2343 blocknr = btrfs_node_blockptr(*eb_ret, slot);
2344 gen = btrfs_node_ptr_generation(*eb_ret, slot);
2345 parent_level = btrfs_header_level(*eb_ret);
2346 btrfs_node_key_to_cpu(*eb_ret, &first_key, slot);
2348 tmp = find_extent_buffer(fs_info, blocknr);
2350 /* first we do an atomic uptodate check */
2351 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2353 * Do extra check for first_key, eb can be stale due to
2354 * being cached, read from scrub, or have multiple
2355 * parents (shared tree blocks).
2357 if (btrfs_verify_level_key(tmp,
2358 parent_level - 1, &first_key, gen)) {
2359 free_extent_buffer(tmp);
2366 /* the pages were up to date, but we failed
2367 * the generation number check. Do a full
2368 * read for the generation number that is correct.
2369 * We must do this without dropping locks so
2370 * we can trust our generation number
2372 btrfs_set_path_blocking(p);
2374 /* now we're allowed to do a blocking uptodate check */
2375 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2380 free_extent_buffer(tmp);
2381 btrfs_release_path(p);
2386 * reduce lock contention at high levels
2387 * of the btree by dropping locks before
2388 * we read. Don't release the lock on the current
2389 * level because we need to walk this node to figure
2390 * out which blocks to read.
2392 btrfs_unlock_up_safe(p, level + 1);
2393 btrfs_set_path_blocking(p);
2395 if (p->reada != READA_NONE)
2396 reada_for_search(fs_info, p, level, slot, key->objectid);
2399 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2403 * If the read above didn't mark this buffer up to date,
2404 * it will never end up being up to date. Set ret to EIO now
2405 * and give up so that our caller doesn't loop forever
2408 if (!extent_buffer_uptodate(tmp))
2410 free_extent_buffer(tmp);
2415 btrfs_release_path(p);
2420 * helper function for btrfs_search_slot. This does all of the checks
2421 * for node-level blocks and does any balancing required based on
2424 * If no extra work was required, zero is returned. If we had to
2425 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2429 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2430 struct btrfs_root *root, struct btrfs_path *p,
2431 struct extent_buffer *b, int level, int ins_len,
2432 int *write_lock_level)
2434 struct btrfs_fs_info *fs_info = root->fs_info;
2437 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2438 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2441 if (*write_lock_level < level + 1) {
2442 *write_lock_level = level + 1;
2443 btrfs_release_path(p);
2447 btrfs_set_path_blocking(p);
2448 reada_for_balance(fs_info, p, level);
2449 sret = split_node(trans, root, p, level);
2456 b = p->nodes[level];
2457 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2458 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2461 if (*write_lock_level < level + 1) {
2462 *write_lock_level = level + 1;
2463 btrfs_release_path(p);
2467 btrfs_set_path_blocking(p);
2468 reada_for_balance(fs_info, p, level);
2469 sret = balance_level(trans, root, p, level);
2475 b = p->nodes[level];
2477 btrfs_release_path(p);
2480 BUG_ON(btrfs_header_nritems(b) == 1);
2490 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2491 u64 iobjectid, u64 ioff, u8 key_type,
2492 struct btrfs_key *found_key)
2495 struct btrfs_key key;
2496 struct extent_buffer *eb;
2501 key.type = key_type;
2502 key.objectid = iobjectid;
2505 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2509 eb = path->nodes[0];
2510 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2511 ret = btrfs_next_leaf(fs_root, path);
2514 eb = path->nodes[0];
2517 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2518 if (found_key->type != key.type ||
2519 found_key->objectid != key.objectid)
2525 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2526 struct btrfs_path *p,
2527 int write_lock_level)
2529 struct btrfs_fs_info *fs_info = root->fs_info;
2530 struct extent_buffer *b;
2534 /* We try very hard to do read locks on the root */
2535 root_lock = BTRFS_READ_LOCK;
2537 if (p->search_commit_root) {
2539 * The commit roots are read only so we always do read locks,
2540 * and we always must hold the commit_root_sem when doing
2541 * searches on them, the only exception is send where we don't
2542 * want to block transaction commits for a long time, so
2543 * we need to clone the commit root in order to avoid races
2544 * with transaction commits that create a snapshot of one of
2545 * the roots used by a send operation.
2547 if (p->need_commit_sem) {
2548 down_read(&fs_info->commit_root_sem);
2549 b = btrfs_clone_extent_buffer(root->commit_root);
2550 up_read(&fs_info->commit_root_sem);
2552 return ERR_PTR(-ENOMEM);
2555 b = root->commit_root;
2556 atomic_inc(&b->refs);
2558 level = btrfs_header_level(b);
2560 * Ensure that all callers have set skip_locking when
2561 * p->search_commit_root = 1.
2563 ASSERT(p->skip_locking == 1);
2568 if (p->skip_locking) {
2569 b = btrfs_root_node(root);
2570 level = btrfs_header_level(b);
2575 * If the level is set to maximum, we can skip trying to get the read
2578 if (write_lock_level < BTRFS_MAX_LEVEL) {
2580 * We don't know the level of the root node until we actually
2581 * have it read locked
2583 b = btrfs_read_lock_root_node(root);
2584 level = btrfs_header_level(b);
2585 if (level > write_lock_level)
2588 /* Whoops, must trade for write lock */
2589 btrfs_tree_read_unlock(b);
2590 free_extent_buffer(b);
2593 b = btrfs_lock_root_node(root);
2594 root_lock = BTRFS_WRITE_LOCK;
2596 /* The level might have changed, check again */
2597 level = btrfs_header_level(b);
2600 p->nodes[level] = b;
2601 if (!p->skip_locking)
2602 p->locks[level] = root_lock;
2604 * Callers are responsible for dropping b's references.
2611 * btrfs_search_slot - look for a key in a tree and perform necessary
2612 * modifications to preserve tree invariants.
2614 * @trans: Handle of transaction, used when modifying the tree
2615 * @p: Holds all btree nodes along the search path
2616 * @root: The root node of the tree
2617 * @key: The key we are looking for
2618 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2619 * deletions it's -1. 0 for plain searches
2620 * @cow: boolean should CoW operations be performed. Must always be 1
2621 * when modifying the tree.
2623 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2624 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2626 * If @key is found, 0 is returned and you can find the item in the leaf level
2627 * of the path (level 0)
2629 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2630 * points to the slot where it should be inserted
2632 * If an error is encountered while searching the tree a negative error number
2635 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2636 const struct btrfs_key *key, struct btrfs_path *p,
2637 int ins_len, int cow)
2639 struct extent_buffer *b;
2644 int lowest_unlock = 1;
2645 /* everything at write_lock_level or lower must be write locked */
2646 int write_lock_level = 0;
2647 u8 lowest_level = 0;
2648 int min_write_lock_level;
2651 lowest_level = p->lowest_level;
2652 WARN_ON(lowest_level && ins_len > 0);
2653 WARN_ON(p->nodes[0] != NULL);
2654 BUG_ON(!cow && ins_len);
2659 /* when we are removing items, we might have to go up to level
2660 * two as we update tree pointers Make sure we keep write
2661 * for those levels as well
2663 write_lock_level = 2;
2664 } else if (ins_len > 0) {
2666 * for inserting items, make sure we have a write lock on
2667 * level 1 so we can update keys
2669 write_lock_level = 1;
2673 write_lock_level = -1;
2675 if (cow && (p->keep_locks || p->lowest_level))
2676 write_lock_level = BTRFS_MAX_LEVEL;
2678 min_write_lock_level = write_lock_level;
2682 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2691 level = btrfs_header_level(b);
2694 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2697 * if we don't really need to cow this block
2698 * then we don't want to set the path blocking,
2699 * so we test it here
2701 if (!should_cow_block(trans, root, b)) {
2702 trans->dirty = true;
2707 * must have write locks on this node and the
2710 if (level > write_lock_level ||
2711 (level + 1 > write_lock_level &&
2712 level + 1 < BTRFS_MAX_LEVEL &&
2713 p->nodes[level + 1])) {
2714 write_lock_level = level + 1;
2715 btrfs_release_path(p);
2719 btrfs_set_path_blocking(p);
2721 err = btrfs_cow_block(trans, root, b, NULL, 0,
2724 err = btrfs_cow_block(trans, root, b,
2725 p->nodes[level + 1],
2726 p->slots[level + 1], &b);
2733 p->nodes[level] = b;
2735 * Leave path with blocking locks to avoid massive
2736 * lock context switch, this is made on purpose.
2740 * we have a lock on b and as long as we aren't changing
2741 * the tree, there is no way to for the items in b to change.
2742 * It is safe to drop the lock on our parent before we
2743 * go through the expensive btree search on b.
2745 * If we're inserting or deleting (ins_len != 0), then we might
2746 * be changing slot zero, which may require changing the parent.
2747 * So, we can't drop the lock until after we know which slot
2748 * we're operating on.
2750 if (!ins_len && !p->keep_locks) {
2753 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2754 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2760 * If btrfs_bin_search returns an exact match (prev_cmp == 0)
2761 * we can safely assume the target key will always be in slot 0
2762 * on lower levels due to the invariants BTRFS' btree provides,
2763 * namely that a btrfs_key_ptr entry always points to the
2764 * lowest key in the child node, thus we can skip searching
2767 if (prev_cmp == 0) {
2771 ret = btrfs_bin_search(b, key, &slot);
2778 p->slots[level] = slot;
2780 btrfs_leaf_free_space(b) < ins_len) {
2781 if (write_lock_level < 1) {
2782 write_lock_level = 1;
2783 btrfs_release_path(p);
2787 btrfs_set_path_blocking(p);
2788 err = split_leaf(trans, root, key,
2789 p, ins_len, ret == 0);
2797 if (!p->search_for_split)
2798 unlock_up(p, level, lowest_unlock,
2799 min_write_lock_level, NULL);
2802 if (ret && slot > 0) {
2806 p->slots[level] = slot;
2807 err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
2815 b = p->nodes[level];
2816 slot = p->slots[level];
2819 * Slot 0 is special, if we change the key we have to update
2820 * the parent pointer which means we must have a write lock on
2823 if (slot == 0 && ins_len && write_lock_level < level + 1) {
2824 write_lock_level = level + 1;
2825 btrfs_release_path(p);
2829 unlock_up(p, level, lowest_unlock, min_write_lock_level,
2832 if (level == lowest_level) {
2838 err = read_block_for_search(root, p, &b, level, slot, key);
2846 if (!p->skip_locking) {
2847 level = btrfs_header_level(b);
2848 if (level <= write_lock_level) {
2849 if (!btrfs_try_tree_write_lock(b)) {
2850 btrfs_set_path_blocking(p);
2853 p->locks[level] = BTRFS_WRITE_LOCK;
2855 if (!btrfs_tree_read_lock_atomic(b)) {
2856 btrfs_set_path_blocking(p);
2857 btrfs_tree_read_lock(b);
2859 p->locks[level] = BTRFS_READ_LOCK;
2861 p->nodes[level] = b;
2867 * we don't really know what they plan on doing with the path
2868 * from here on, so for now just mark it as blocking
2870 if (!p->leave_spinning)
2871 btrfs_set_path_blocking(p);
2872 if (ret < 0 && !p->skip_release_on_error)
2873 btrfs_release_path(p);
2878 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2879 * current state of the tree together with the operations recorded in the tree
2880 * modification log to search for the key in a previous version of this tree, as
2881 * denoted by the time_seq parameter.
2883 * Naturally, there is no support for insert, delete or cow operations.
2885 * The resulting path and return value will be set up as if we called
2886 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2888 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2889 struct btrfs_path *p, u64 time_seq)
2891 struct btrfs_fs_info *fs_info = root->fs_info;
2892 struct extent_buffer *b;
2897 int lowest_unlock = 1;
2898 u8 lowest_level = 0;
2900 lowest_level = p->lowest_level;
2901 WARN_ON(p->nodes[0] != NULL);
2903 if (p->search_commit_root) {
2905 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2909 b = get_old_root(root, time_seq);
2914 level = btrfs_header_level(b);
2915 p->locks[level] = BTRFS_READ_LOCK;
2920 level = btrfs_header_level(b);
2921 p->nodes[level] = b;
2924 * we have a lock on b and as long as we aren't changing
2925 * the tree, there is no way to for the items in b to change.
2926 * It is safe to drop the lock on our parent before we
2927 * go through the expensive btree search on b.
2929 btrfs_unlock_up_safe(p, level + 1);
2931 ret = btrfs_bin_search(b, key, &slot);
2936 p->slots[level] = slot;
2937 unlock_up(p, level, lowest_unlock, 0, NULL);
2941 if (ret && slot > 0) {
2945 p->slots[level] = slot;
2946 unlock_up(p, level, lowest_unlock, 0, NULL);
2948 if (level == lowest_level) {
2954 err = read_block_for_search(root, p, &b, level, slot, key);
2962 level = btrfs_header_level(b);
2963 if (!btrfs_tree_read_lock_atomic(b)) {
2964 btrfs_set_path_blocking(p);
2965 btrfs_tree_read_lock(b);
2967 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
2972 p->locks[level] = BTRFS_READ_LOCK;
2973 p->nodes[level] = b;
2977 if (!p->leave_spinning)
2978 btrfs_set_path_blocking(p);
2980 btrfs_release_path(p);
2986 * helper to use instead of search slot if no exact match is needed but
2987 * instead the next or previous item should be returned.
2988 * When find_higher is true, the next higher item is returned, the next lower
2990 * When return_any and find_higher are both true, and no higher item is found,
2991 * return the next lower instead.
2992 * When return_any is true and find_higher is false, and no lower item is found,
2993 * return the next higher instead.
2994 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2997 int btrfs_search_slot_for_read(struct btrfs_root *root,
2998 const struct btrfs_key *key,
2999 struct btrfs_path *p, int find_higher,
3003 struct extent_buffer *leaf;
3006 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3010 * a return value of 1 means the path is at the position where the
3011 * item should be inserted. Normally this is the next bigger item,
3012 * but in case the previous item is the last in a leaf, path points
3013 * to the first free slot in the previous leaf, i.e. at an invalid
3019 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3020 ret = btrfs_next_leaf(root, p);
3026 * no higher item found, return the next
3031 btrfs_release_path(p);
3035 if (p->slots[0] == 0) {
3036 ret = btrfs_prev_leaf(root, p);
3041 if (p->slots[0] == btrfs_header_nritems(leaf))
3048 * no lower item found, return the next
3053 btrfs_release_path(p);
3063 * adjust the pointers going up the tree, starting at level
3064 * making sure the right key of each node is points to 'key'.
3065 * This is used after shifting pointers to the left, so it stops
3066 * fixing up pointers when a given leaf/node is not in slot 0 of the
3070 static void fixup_low_keys(struct btrfs_path *path,
3071 struct btrfs_disk_key *key, int level)
3074 struct extent_buffer *t;
3077 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3078 int tslot = path->slots[i];
3080 if (!path->nodes[i])
3083 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3086 btrfs_set_node_key(t, key, tslot);
3087 btrfs_mark_buffer_dirty(path->nodes[i]);
3096 * This function isn't completely safe. It's the caller's responsibility
3097 * that the new key won't break the order
3099 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3100 struct btrfs_path *path,
3101 const struct btrfs_key *new_key)
3103 struct btrfs_disk_key disk_key;
3104 struct extent_buffer *eb;
3107 eb = path->nodes[0];
3108 slot = path->slots[0];
3110 btrfs_item_key(eb, &disk_key, slot - 1);
3111 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
3113 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3114 slot, btrfs_disk_key_objectid(&disk_key),
3115 btrfs_disk_key_type(&disk_key),
3116 btrfs_disk_key_offset(&disk_key),
3117 new_key->objectid, new_key->type,
3119 btrfs_print_leaf(eb);
3123 if (slot < btrfs_header_nritems(eb) - 1) {
3124 btrfs_item_key(eb, &disk_key, slot + 1);
3125 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
3127 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3128 slot, btrfs_disk_key_objectid(&disk_key),
3129 btrfs_disk_key_type(&disk_key),
3130 btrfs_disk_key_offset(&disk_key),
3131 new_key->objectid, new_key->type,
3133 btrfs_print_leaf(eb);
3138 btrfs_cpu_key_to_disk(&disk_key, new_key);
3139 btrfs_set_item_key(eb, &disk_key, slot);
3140 btrfs_mark_buffer_dirty(eb);
3142 fixup_low_keys(path, &disk_key, 1);
3146 * try to push data from one node into the next node left in the
3149 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3150 * error, and > 0 if there was no room in the left hand block.
3152 static int push_node_left(struct btrfs_trans_handle *trans,
3153 struct extent_buffer *dst,
3154 struct extent_buffer *src, int empty)
3156 struct btrfs_fs_info *fs_info = trans->fs_info;
3162 src_nritems = btrfs_header_nritems(src);
3163 dst_nritems = btrfs_header_nritems(dst);
3164 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3165 WARN_ON(btrfs_header_generation(src) != trans->transid);
3166 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3168 if (!empty && src_nritems <= 8)
3171 if (push_items <= 0)
3175 push_items = min(src_nritems, push_items);
3176 if (push_items < src_nritems) {
3177 /* leave at least 8 pointers in the node if
3178 * we aren't going to empty it
3180 if (src_nritems - push_items < 8) {
3181 if (push_items <= 8)
3187 push_items = min(src_nritems - 8, push_items);
3189 ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3191 btrfs_abort_transaction(trans, ret);
3194 copy_extent_buffer(dst, src,
3195 btrfs_node_key_ptr_offset(dst_nritems),
3196 btrfs_node_key_ptr_offset(0),
3197 push_items * sizeof(struct btrfs_key_ptr));
3199 if (push_items < src_nritems) {
3201 * Don't call tree_mod_log_insert_move here, key removal was
3202 * already fully logged by tree_mod_log_eb_copy above.
3204 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3205 btrfs_node_key_ptr_offset(push_items),
3206 (src_nritems - push_items) *
3207 sizeof(struct btrfs_key_ptr));
3209 btrfs_set_header_nritems(src, src_nritems - push_items);
3210 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3211 btrfs_mark_buffer_dirty(src);
3212 btrfs_mark_buffer_dirty(dst);
3218 * try to push data from one node into the next node right in the
3221 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3222 * error, and > 0 if there was no room in the right hand block.
3224 * this will only push up to 1/2 the contents of the left node over
3226 static int balance_node_right(struct btrfs_trans_handle *trans,
3227 struct extent_buffer *dst,
3228 struct extent_buffer *src)
3230 struct btrfs_fs_info *fs_info = trans->fs_info;
3237 WARN_ON(btrfs_header_generation(src) != trans->transid);
3238 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3240 src_nritems = btrfs_header_nritems(src);
3241 dst_nritems = btrfs_header_nritems(dst);
3242 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3243 if (push_items <= 0)
3246 if (src_nritems < 4)
3249 max_push = src_nritems / 2 + 1;
3250 /* don't try to empty the node */
3251 if (max_push >= src_nritems)
3254 if (max_push < push_items)
3255 push_items = max_push;
3257 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3259 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3260 btrfs_node_key_ptr_offset(0),
3262 sizeof(struct btrfs_key_ptr));
3264 ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3267 btrfs_abort_transaction(trans, ret);
3270 copy_extent_buffer(dst, src,
3271 btrfs_node_key_ptr_offset(0),
3272 btrfs_node_key_ptr_offset(src_nritems - push_items),
3273 push_items * sizeof(struct btrfs_key_ptr));
3275 btrfs_set_header_nritems(src, src_nritems - push_items);
3276 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3278 btrfs_mark_buffer_dirty(src);
3279 btrfs_mark_buffer_dirty(dst);
3285 * helper function to insert a new root level in the tree.
3286 * A new node is allocated, and a single item is inserted to
3287 * point to the existing root
3289 * returns zero on success or < 0 on failure.
3291 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3292 struct btrfs_root *root,
3293 struct btrfs_path *path, int level)
3295 struct btrfs_fs_info *fs_info = root->fs_info;
3297 struct extent_buffer *lower;
3298 struct extent_buffer *c;
3299 struct extent_buffer *old;
3300 struct btrfs_disk_key lower_key;
3303 BUG_ON(path->nodes[level]);
3304 BUG_ON(path->nodes[level-1] != root->node);
3306 lower = path->nodes[level-1];
3308 btrfs_item_key(lower, &lower_key, 0);
3310 btrfs_node_key(lower, &lower_key, 0);
3312 c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3313 root->node->start, 0);
3317 root_add_used(root, fs_info->nodesize);
3319 btrfs_set_header_nritems(c, 1);
3320 btrfs_set_node_key(c, &lower_key, 0);
3321 btrfs_set_node_blockptr(c, 0, lower->start);
3322 lower_gen = btrfs_header_generation(lower);
3323 WARN_ON(lower_gen != trans->transid);
3325 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3327 btrfs_mark_buffer_dirty(c);
3330 ret = tree_mod_log_insert_root(root->node, c, 0);
3332 rcu_assign_pointer(root->node, c);
3334 /* the super has an extra ref to root->node */
3335 free_extent_buffer(old);
3337 add_root_to_dirty_list(root);
3338 atomic_inc(&c->refs);
3339 path->nodes[level] = c;
3340 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3341 path->slots[level] = 0;
3346 * worker function to insert a single pointer in a node.
3347 * the node should have enough room for the pointer already
3349 * slot and level indicate where you want the key to go, and
3350 * blocknr is the block the key points to.
3352 static void insert_ptr(struct btrfs_trans_handle *trans,
3353 struct btrfs_path *path,
3354 struct btrfs_disk_key *key, u64 bytenr,
3355 int slot, int level)
3357 struct extent_buffer *lower;
3361 BUG_ON(!path->nodes[level]);
3362 btrfs_assert_tree_locked(path->nodes[level]);
3363 lower = path->nodes[level];
3364 nritems = btrfs_header_nritems(lower);
3365 BUG_ON(slot > nritems);
3366 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3367 if (slot != nritems) {
3369 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3373 memmove_extent_buffer(lower,
3374 btrfs_node_key_ptr_offset(slot + 1),
3375 btrfs_node_key_ptr_offset(slot),
3376 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3379 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3383 btrfs_set_node_key(lower, key, slot);
3384 btrfs_set_node_blockptr(lower, slot, bytenr);
3385 WARN_ON(trans->transid == 0);
3386 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3387 btrfs_set_header_nritems(lower, nritems + 1);
3388 btrfs_mark_buffer_dirty(lower);
3392 * split the node at the specified level in path in two.
3393 * The path is corrected to point to the appropriate node after the split
3395 * Before splitting this tries to make some room in the node by pushing
3396 * left and right, if either one works, it returns right away.
3398 * returns 0 on success and < 0 on failure
3400 static noinline int split_node(struct btrfs_trans_handle *trans,
3401 struct btrfs_root *root,
3402 struct btrfs_path *path, int level)
3404 struct btrfs_fs_info *fs_info = root->fs_info;
3405 struct extent_buffer *c;
3406 struct extent_buffer *split;
3407 struct btrfs_disk_key disk_key;
3412 c = path->nodes[level];
3413 WARN_ON(btrfs_header_generation(c) != trans->transid);
3414 if (c == root->node) {
3416 * trying to split the root, lets make a new one
3418 * tree mod log: We don't log_removal old root in
3419 * insert_new_root, because that root buffer will be kept as a
3420 * normal node. We are going to log removal of half of the
3421 * elements below with tree_mod_log_eb_copy. We're holding a
3422 * tree lock on the buffer, which is why we cannot race with
3423 * other tree_mod_log users.
3425 ret = insert_new_root(trans, root, path, level + 1);
3429 ret = push_nodes_for_insert(trans, root, path, level);
3430 c = path->nodes[level];
3431 if (!ret && btrfs_header_nritems(c) <
3432 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3438 c_nritems = btrfs_header_nritems(c);
3439 mid = (c_nritems + 1) / 2;
3440 btrfs_node_key(c, &disk_key, mid);
3442 split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3445 return PTR_ERR(split);
3447 root_add_used(root, fs_info->nodesize);
3448 ASSERT(btrfs_header_level(c) == level);
3450 ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3452 btrfs_abort_transaction(trans, ret);
3455 copy_extent_buffer(split, c,
3456 btrfs_node_key_ptr_offset(0),
3457 btrfs_node_key_ptr_offset(mid),
3458 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3459 btrfs_set_header_nritems(split, c_nritems - mid);
3460 btrfs_set_header_nritems(c, mid);
3463 btrfs_mark_buffer_dirty(c);
3464 btrfs_mark_buffer_dirty(split);
3466 insert_ptr(trans, path, &disk_key, split->start,
3467 path->slots[level + 1] + 1, level + 1);
3469 if (path->slots[level] >= mid) {
3470 path->slots[level] -= mid;
3471 btrfs_tree_unlock(c);
3472 free_extent_buffer(c);
3473 path->nodes[level] = split;
3474 path->slots[level + 1] += 1;
3476 btrfs_tree_unlock(split);
3477 free_extent_buffer(split);
3483 * how many bytes are required to store the items in a leaf. start
3484 * and nr indicate which items in the leaf to check. This totals up the
3485 * space used both by the item structs and the item data
3487 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3489 struct btrfs_item *start_item;
3490 struct btrfs_item *end_item;
3492 int nritems = btrfs_header_nritems(l);
3493 int end = min(nritems, start + nr) - 1;
3497 start_item = btrfs_item_nr(start);
3498 end_item = btrfs_item_nr(end);
3499 data_len = btrfs_item_offset(l, start_item) +
3500 btrfs_item_size(l, start_item);
3501 data_len = data_len - btrfs_item_offset(l, end_item);
3502 data_len += sizeof(struct btrfs_item) * nr;
3503 WARN_ON(data_len < 0);
3508 * The space between the end of the leaf items and
3509 * the start of the leaf data. IOW, how much room
3510 * the leaf has left for both items and data
3512 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3514 struct btrfs_fs_info *fs_info = leaf->fs_info;
3515 int nritems = btrfs_header_nritems(leaf);
3518 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3521 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3523 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3524 leaf_space_used(leaf, 0, nritems), nritems);
3530 * min slot controls the lowest index we're willing to push to the
3531 * right. We'll push up to and including min_slot, but no lower
3533 static noinline int __push_leaf_right(struct btrfs_path *path,
3534 int data_size, int empty,
3535 struct extent_buffer *right,
3536 int free_space, u32 left_nritems,
3539 struct btrfs_fs_info *fs_info = right->fs_info;
3540 struct extent_buffer *left = path->nodes[0];
3541 struct extent_buffer *upper = path->nodes[1];
3542 struct btrfs_map_token token;
3543 struct btrfs_disk_key disk_key;
3548 struct btrfs_item *item;
3557 nr = max_t(u32, 1, min_slot);
3559 if (path->slots[0] >= left_nritems)
3560 push_space += data_size;
3562 slot = path->slots[1];
3563 i = left_nritems - 1;
3565 item = btrfs_item_nr(i);
3567 if (!empty && push_items > 0) {
3568 if (path->slots[0] > i)
3570 if (path->slots[0] == i) {
3571 int space = btrfs_leaf_free_space(left);
3573 if (space + push_space * 2 > free_space)
3578 if (path->slots[0] == i)
3579 push_space += data_size;
3581 this_item_size = btrfs_item_size(left, item);
3582 if (this_item_size + sizeof(*item) + push_space > free_space)
3586 push_space += this_item_size + sizeof(*item);
3592 if (push_items == 0)
3595 WARN_ON(!empty && push_items == left_nritems);
3597 /* push left to right */
3598 right_nritems = btrfs_header_nritems(right);
3600 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3601 push_space -= leaf_data_end(left);
3603 /* make room in the right data area */
3604 data_end = leaf_data_end(right);
3605 memmove_extent_buffer(right,
3606 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3607 BTRFS_LEAF_DATA_OFFSET + data_end,
3608 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3610 /* copy from the left data area */
3611 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3612 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3613 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3616 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3617 btrfs_item_nr_offset(0),
3618 right_nritems * sizeof(struct btrfs_item));
3620 /* copy the items from left to right */
3621 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3622 btrfs_item_nr_offset(left_nritems - push_items),
3623 push_items * sizeof(struct btrfs_item));
3625 /* update the item pointers */
3626 btrfs_init_map_token(&token, right);
3627 right_nritems += push_items;
3628 btrfs_set_header_nritems(right, right_nritems);
3629 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3630 for (i = 0; i < right_nritems; i++) {
3631 item = btrfs_item_nr(i);
3632 push_space -= btrfs_token_item_size(&token, item);
3633 btrfs_set_token_item_offset(&token, item, push_space);
3636 left_nritems -= push_items;
3637 btrfs_set_header_nritems(left, left_nritems);
3640 btrfs_mark_buffer_dirty(left);
3642 btrfs_clean_tree_block(left);
3644 btrfs_mark_buffer_dirty(right);
3646 btrfs_item_key(right, &disk_key, 0);
3647 btrfs_set_node_key(upper, &disk_key, slot + 1);
3648 btrfs_mark_buffer_dirty(upper);
3650 /* then fixup the leaf pointer in the path */
3651 if (path->slots[0] >= left_nritems) {
3652 path->slots[0] -= left_nritems;
3653 if (btrfs_header_nritems(path->nodes[0]) == 0)
3654 btrfs_clean_tree_block(path->nodes[0]);
3655 btrfs_tree_unlock(path->nodes[0]);
3656 free_extent_buffer(path->nodes[0]);
3657 path->nodes[0] = right;
3658 path->slots[1] += 1;
3660 btrfs_tree_unlock(right);
3661 free_extent_buffer(right);
3666 btrfs_tree_unlock(right);
3667 free_extent_buffer(right);
3672 * push some data in the path leaf to the right, trying to free up at
3673 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3675 * returns 1 if the push failed because the other node didn't have enough
3676 * room, 0 if everything worked out and < 0 if there were major errors.
3678 * this will push starting from min_slot to the end of the leaf. It won't
3679 * push any slot lower than min_slot
3681 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3682 *root, struct btrfs_path *path,
3683 int min_data_size, int data_size,
3684 int empty, u32 min_slot)
3686 struct extent_buffer *left = path->nodes[0];
3687 struct extent_buffer *right;
3688 struct extent_buffer *upper;
3694 if (!path->nodes[1])
3697 slot = path->slots[1];
3698 upper = path->nodes[1];
3699 if (slot >= btrfs_header_nritems(upper) - 1)
3702 btrfs_assert_tree_locked(path->nodes[1]);
3704 right = btrfs_read_node_slot(upper, slot + 1);
3706 * slot + 1 is not valid or we fail to read the right node,
3707 * no big deal, just return.
3712 btrfs_tree_lock(right);
3713 btrfs_set_lock_blocking_write(right);
3715 free_space = btrfs_leaf_free_space(right);
3716 if (free_space < data_size)
3719 /* cow and double check */
3720 ret = btrfs_cow_block(trans, root, right, upper,
3725 free_space = btrfs_leaf_free_space(right);
3726 if (free_space < data_size)
3729 left_nritems = btrfs_header_nritems(left);
3730 if (left_nritems == 0)
3733 if (path->slots[0] == left_nritems && !empty) {
3734 /* Key greater than all keys in the leaf, right neighbor has
3735 * enough room for it and we're not emptying our leaf to delete
3736 * it, therefore use right neighbor to insert the new item and
3737 * no need to touch/dirty our left leaf. */
3738 btrfs_tree_unlock(left);
3739 free_extent_buffer(left);
3740 path->nodes[0] = right;
3746 return __push_leaf_right(path, min_data_size, empty,
3747 right, free_space, left_nritems, min_slot);
3749 btrfs_tree_unlock(right);
3750 free_extent_buffer(right);
3755 * push some data in the path leaf to the left, trying to free up at
3756 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3758 * max_slot can put a limit on how far into the leaf we'll push items. The
3759 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3762 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3763 int empty, struct extent_buffer *left,
3764 int free_space, u32 right_nritems,
3767 struct btrfs_fs_info *fs_info = left->fs_info;
3768 struct btrfs_disk_key disk_key;
3769 struct extent_buffer *right = path->nodes[0];
3773 struct btrfs_item *item;
3774 u32 old_left_nritems;
3778 u32 old_left_item_size;
3779 struct btrfs_map_token token;
3782 nr = min(right_nritems, max_slot);
3784 nr = min(right_nritems - 1, max_slot);
3786 for (i = 0; i < nr; i++) {
3787 item = btrfs_item_nr(i);
3789 if (!empty && push_items > 0) {
3790 if (path->slots[0] < i)
3792 if (path->slots[0] == i) {
3793 int space = btrfs_leaf_free_space(right);
3795 if (space + push_space * 2 > free_space)
3800 if (path->slots[0] == i)
3801 push_space += data_size;
3803 this_item_size = btrfs_item_size(right, item);
3804 if (this_item_size + sizeof(*item) + push_space > free_space)
3808 push_space += this_item_size + sizeof(*item);
3811 if (push_items == 0) {
3815 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3817 /* push data from right to left */
3818 copy_extent_buffer(left, right,
3819 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3820 btrfs_item_nr_offset(0),
3821 push_items * sizeof(struct btrfs_item));
3823 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3824 btrfs_item_offset_nr(right, push_items - 1);
3826 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3827 leaf_data_end(left) - push_space,
3828 BTRFS_LEAF_DATA_OFFSET +
3829 btrfs_item_offset_nr(right, push_items - 1),
3831 old_left_nritems = btrfs_header_nritems(left);
3832 BUG_ON(old_left_nritems <= 0);
3834 btrfs_init_map_token(&token, left);
3835 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3836 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3839 item = btrfs_item_nr(i);
3841 ioff = btrfs_token_item_offset(&token, item);
3842 btrfs_set_token_item_offset(&token, item,
3843 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size));
3845 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3847 /* fixup right node */
3848 if (push_items > right_nritems)
3849 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3852 if (push_items < right_nritems) {
3853 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3854 leaf_data_end(right);
3855 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3856 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3857 BTRFS_LEAF_DATA_OFFSET +
3858 leaf_data_end(right), push_space);
3860 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3861 btrfs_item_nr_offset(push_items),
3862 (btrfs_header_nritems(right) - push_items) *
3863 sizeof(struct btrfs_item));
3866 btrfs_init_map_token(&token, right);
3867 right_nritems -= push_items;
3868 btrfs_set_header_nritems(right, right_nritems);
3869 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3870 for (i = 0; i < right_nritems; i++) {
3871 item = btrfs_item_nr(i);
3873 push_space = push_space - btrfs_token_item_size(&token, item);
3874 btrfs_set_token_item_offset(&token, item, push_space);
3877 btrfs_mark_buffer_dirty(left);
3879 btrfs_mark_buffer_dirty(right);
3881 btrfs_clean_tree_block(right);
3883 btrfs_item_key(right, &disk_key, 0);
3884 fixup_low_keys(path, &disk_key, 1);
3886 /* then fixup the leaf pointer in the path */
3887 if (path->slots[0] < push_items) {
3888 path->slots[0] += old_left_nritems;
3889 btrfs_tree_unlock(path->nodes[0]);
3890 free_extent_buffer(path->nodes[0]);
3891 path->nodes[0] = left;
3892 path->slots[1] -= 1;
3894 btrfs_tree_unlock(left);
3895 free_extent_buffer(left);
3896 path->slots[0] -= push_items;
3898 BUG_ON(path->slots[0] < 0);
3901 btrfs_tree_unlock(left);
3902 free_extent_buffer(left);
3907 * push some data in the path leaf to the left, trying to free up at
3908 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3910 * max_slot can put a limit on how far into the leaf we'll push items. The
3911 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3914 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3915 *root, struct btrfs_path *path, int min_data_size,
3916 int data_size, int empty, u32 max_slot)
3918 struct extent_buffer *right = path->nodes[0];
3919 struct extent_buffer *left;
3925 slot = path->slots[1];
3928 if (!path->nodes[1])
3931 right_nritems = btrfs_header_nritems(right);
3932 if (right_nritems == 0)
3935 btrfs_assert_tree_locked(path->nodes[1]);
3937 left = btrfs_read_node_slot(path->nodes[1], slot - 1);
3939 * slot - 1 is not valid or we fail to read the left node,
3940 * no big deal, just return.
3945 btrfs_tree_lock(left);
3946 btrfs_set_lock_blocking_write(left);
3948 free_space = btrfs_leaf_free_space(left);
3949 if (free_space < data_size) {
3954 /* cow and double check */
3955 ret = btrfs_cow_block(trans, root, left,
3956 path->nodes[1], slot - 1, &left);
3958 /* we hit -ENOSPC, but it isn't fatal here */
3964 free_space = btrfs_leaf_free_space(left);
3965 if (free_space < data_size) {
3970 return __push_leaf_left(path, min_data_size,
3971 empty, left, free_space, right_nritems,
3974 btrfs_tree_unlock(left);
3975 free_extent_buffer(left);
3980 * split the path's leaf in two, making sure there is at least data_size
3981 * available for the resulting leaf level of the path.
3983 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3984 struct btrfs_path *path,
3985 struct extent_buffer *l,
3986 struct extent_buffer *right,
3987 int slot, int mid, int nritems)
3989 struct btrfs_fs_info *fs_info = trans->fs_info;
3993 struct btrfs_disk_key disk_key;
3994 struct btrfs_map_token token;
3996 nritems = nritems - mid;
3997 btrfs_set_header_nritems(right, nritems);
3998 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4000 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4001 btrfs_item_nr_offset(mid),
4002 nritems * sizeof(struct btrfs_item));
4004 copy_extent_buffer(right, l,
4005 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4006 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4007 leaf_data_end(l), data_copy_size);
4009 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4011 btrfs_init_map_token(&token, right);
4012 for (i = 0; i < nritems; i++) {
4013 struct btrfs_item *item = btrfs_item_nr(i);
4016 ioff = btrfs_token_item_offset(&token, item);
4017 btrfs_set_token_item_offset(&token, item, ioff + rt_data_off);
4020 btrfs_set_header_nritems(l, mid);
4021 btrfs_item_key(right, &disk_key, 0);
4022 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4024 btrfs_mark_buffer_dirty(right);
4025 btrfs_mark_buffer_dirty(l);
4026 BUG_ON(path->slots[0] != slot);
4029 btrfs_tree_unlock(path->nodes[0]);
4030 free_extent_buffer(path->nodes[0]);
4031 path->nodes[0] = right;
4032 path->slots[0] -= mid;
4033 path->slots[1] += 1;
4035 btrfs_tree_unlock(right);
4036 free_extent_buffer(right);
4039 BUG_ON(path->slots[0] < 0);
4043 * double splits happen when we need to insert a big item in the middle
4044 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4045 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4048 * We avoid this by trying to push the items on either side of our target
4049 * into the adjacent leaves. If all goes well we can avoid the double split
4052 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4053 struct btrfs_root *root,
4054 struct btrfs_path *path,
4061 int space_needed = data_size;
4063 slot = path->slots[0];
4064 if (slot < btrfs_header_nritems(path->nodes[0]))
4065 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4068 * try to push all the items after our slot into the
4071 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4078 nritems = btrfs_header_nritems(path->nodes[0]);
4080 * our goal is to get our slot at the start or end of a leaf. If
4081 * we've done so we're done
4083 if (path->slots[0] == 0 || path->slots[0] == nritems)
4086 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4089 /* try to push all the items before our slot into the next leaf */
4090 slot = path->slots[0];
4091 space_needed = data_size;
4093 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4094 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4107 * split the path's leaf in two, making sure there is at least data_size
4108 * available for the resulting leaf level of the path.
4110 * returns 0 if all went well and < 0 on failure.
4112 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4113 struct btrfs_root *root,
4114 const struct btrfs_key *ins_key,
4115 struct btrfs_path *path, int data_size,
4118 struct btrfs_disk_key disk_key;
4119 struct extent_buffer *l;
4123 struct extent_buffer *right;
4124 struct btrfs_fs_info *fs_info = root->fs_info;
4128 int num_doubles = 0;
4129 int tried_avoid_double = 0;
4132 slot = path->slots[0];
4133 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4134 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4137 /* first try to make some room by pushing left and right */
4138 if (data_size && path->nodes[1]) {
4139 int space_needed = data_size;
4141 if (slot < btrfs_header_nritems(l))
4142 space_needed -= btrfs_leaf_free_space(l);
4144 wret = push_leaf_right(trans, root, path, space_needed,
4145 space_needed, 0, 0);
4149 space_needed = data_size;
4151 space_needed -= btrfs_leaf_free_space(l);
4152 wret = push_leaf_left(trans, root, path, space_needed,
4153 space_needed, 0, (u32)-1);
4159 /* did the pushes work? */
4160 if (btrfs_leaf_free_space(l) >= data_size)
4164 if (!path->nodes[1]) {
4165 ret = insert_new_root(trans, root, path, 1);
4172 slot = path->slots[0];
4173 nritems = btrfs_header_nritems(l);
4174 mid = (nritems + 1) / 2;
4178 leaf_space_used(l, mid, nritems - mid) + data_size >
4179 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4180 if (slot >= nritems) {
4184 if (mid != nritems &&
4185 leaf_space_used(l, mid, nritems - mid) +
4186 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4187 if (data_size && !tried_avoid_double)
4188 goto push_for_double;
4194 if (leaf_space_used(l, 0, mid) + data_size >
4195 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4196 if (!extend && data_size && slot == 0) {
4198 } else if ((extend || !data_size) && slot == 0) {
4202 if (mid != nritems &&
4203 leaf_space_used(l, mid, nritems - mid) +
4204 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4205 if (data_size && !tried_avoid_double)
4206 goto push_for_double;
4214 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4216 btrfs_item_key(l, &disk_key, mid);
4218 right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4221 return PTR_ERR(right);
4223 root_add_used(root, fs_info->nodesize);
4227 btrfs_set_header_nritems(right, 0);
4228 insert_ptr(trans, path, &disk_key,
4229 right->start, path->slots[1] + 1, 1);
4230 btrfs_tree_unlock(path->nodes[0]);
4231 free_extent_buffer(path->nodes[0]);
4232 path->nodes[0] = right;
4234 path->slots[1] += 1;
4236 btrfs_set_header_nritems(right, 0);
4237 insert_ptr(trans, path, &disk_key,
4238 right->start, path->slots[1], 1);
4239 btrfs_tree_unlock(path->nodes[0]);
4240 free_extent_buffer(path->nodes[0]);
4241 path->nodes[0] = right;
4243 if (path->slots[1] == 0)
4244 fixup_low_keys(path, &disk_key, 1);
4247 * We create a new leaf 'right' for the required ins_len and
4248 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4249 * the content of ins_len to 'right'.
4254 copy_for_split(trans, path, l, right, slot, mid, nritems);
4257 BUG_ON(num_doubles != 0);
4265 push_for_double_split(trans, root, path, data_size);
4266 tried_avoid_double = 1;
4267 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4272 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4273 struct btrfs_root *root,
4274 struct btrfs_path *path, int ins_len)
4276 struct btrfs_key key;
4277 struct extent_buffer *leaf;
4278 struct btrfs_file_extent_item *fi;
4283 leaf = path->nodes[0];
4284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4286 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4287 key.type != BTRFS_EXTENT_CSUM_KEY);
4289 if (btrfs_leaf_free_space(leaf) >= ins_len)
4292 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4293 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4294 fi = btrfs_item_ptr(leaf, path->slots[0],
4295 struct btrfs_file_extent_item);
4296 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4298 btrfs_release_path(path);
4300 path->keep_locks = 1;
4301 path->search_for_split = 1;
4302 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4303 path->search_for_split = 0;
4310 leaf = path->nodes[0];
4311 /* if our item isn't there, return now */
4312 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4315 /* the leaf has changed, it now has room. return now */
4316 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4319 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4320 fi = btrfs_item_ptr(leaf, path->slots[0],
4321 struct btrfs_file_extent_item);
4322 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4326 btrfs_set_path_blocking(path);
4327 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4331 path->keep_locks = 0;
4332 btrfs_unlock_up_safe(path, 1);
4335 path->keep_locks = 0;
4339 static noinline int split_item(struct btrfs_path *path,
4340 const struct btrfs_key *new_key,
4341 unsigned long split_offset)
4343 struct extent_buffer *leaf;
4344 struct btrfs_item *item;
4345 struct btrfs_item *new_item;
4351 struct btrfs_disk_key disk_key;
4353 leaf = path->nodes[0];
4354 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4356 btrfs_set_path_blocking(path);
4358 item = btrfs_item_nr(path->slots[0]);
4359 orig_offset = btrfs_item_offset(leaf, item);
4360 item_size = btrfs_item_size(leaf, item);
4362 buf = kmalloc(item_size, GFP_NOFS);
4366 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4367 path->slots[0]), item_size);
4369 slot = path->slots[0] + 1;
4370 nritems = btrfs_header_nritems(leaf);
4371 if (slot != nritems) {
4372 /* shift the items */
4373 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4374 btrfs_item_nr_offset(slot),
4375 (nritems - slot) * sizeof(struct btrfs_item));
4378 btrfs_cpu_key_to_disk(&disk_key, new_key);
4379 btrfs_set_item_key(leaf, &disk_key, slot);
4381 new_item = btrfs_item_nr(slot);
4383 btrfs_set_item_offset(leaf, new_item, orig_offset);
4384 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4386 btrfs_set_item_offset(leaf, item,
4387 orig_offset + item_size - split_offset);
4388 btrfs_set_item_size(leaf, item, split_offset);
4390 btrfs_set_header_nritems(leaf, nritems + 1);
4392 /* write the data for the start of the original item */
4393 write_extent_buffer(leaf, buf,
4394 btrfs_item_ptr_offset(leaf, path->slots[0]),
4397 /* write the data for the new item */
4398 write_extent_buffer(leaf, buf + split_offset,
4399 btrfs_item_ptr_offset(leaf, slot),
4400 item_size - split_offset);
4401 btrfs_mark_buffer_dirty(leaf);
4403 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4409 * This function splits a single item into two items,
4410 * giving 'new_key' to the new item and splitting the
4411 * old one at split_offset (from the start of the item).
4413 * The path may be released by this operation. After
4414 * the split, the path is pointing to the old item. The
4415 * new item is going to be in the same node as the old one.
4417 * Note, the item being split must be smaller enough to live alone on
4418 * a tree block with room for one extra struct btrfs_item
4420 * This allows us to split the item in place, keeping a lock on the
4421 * leaf the entire time.
4423 int btrfs_split_item(struct btrfs_trans_handle *trans,
4424 struct btrfs_root *root,
4425 struct btrfs_path *path,
4426 const struct btrfs_key *new_key,
4427 unsigned long split_offset)
4430 ret = setup_leaf_for_split(trans, root, path,
4431 sizeof(struct btrfs_item));
4435 ret = split_item(path, new_key, split_offset);
4440 * This function duplicate a item, giving 'new_key' to the new item.
4441 * It guarantees both items live in the same tree leaf and the new item
4442 * is contiguous with the original item.
4444 * This allows us to split file extent in place, keeping a lock on the
4445 * leaf the entire time.
4447 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4448 struct btrfs_root *root,
4449 struct btrfs_path *path,
4450 const struct btrfs_key *new_key)
4452 struct extent_buffer *leaf;
4456 leaf = path->nodes[0];
4457 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4458 ret = setup_leaf_for_split(trans, root, path,
4459 item_size + sizeof(struct btrfs_item));
4464 setup_items_for_insert(root, path, new_key, &item_size,
4465 item_size, item_size +
4466 sizeof(struct btrfs_item), 1);
4467 leaf = path->nodes[0];
4468 memcpy_extent_buffer(leaf,
4469 btrfs_item_ptr_offset(leaf, path->slots[0]),
4470 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4476 * make the item pointed to by the path smaller. new_size indicates
4477 * how small to make it, and from_end tells us if we just chop bytes
4478 * off the end of the item or if we shift the item to chop bytes off
4481 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4484 struct extent_buffer *leaf;
4485 struct btrfs_item *item;
4487 unsigned int data_end;
4488 unsigned int old_data_start;
4489 unsigned int old_size;
4490 unsigned int size_diff;
4492 struct btrfs_map_token token;
4494 leaf = path->nodes[0];
4495 slot = path->slots[0];
4497 old_size = btrfs_item_size_nr(leaf, slot);
4498 if (old_size == new_size)
4501 nritems = btrfs_header_nritems(leaf);
4502 data_end = leaf_data_end(leaf);
4504 old_data_start = btrfs_item_offset_nr(leaf, slot);
4506 size_diff = old_size - new_size;
4509 BUG_ON(slot >= nritems);
4512 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4514 /* first correct the data pointers */
4515 btrfs_init_map_token(&token, leaf);
4516 for (i = slot; i < nritems; i++) {
4518 item = btrfs_item_nr(i);
4520 ioff = btrfs_token_item_offset(&token, item);
4521 btrfs_set_token_item_offset(&token, item, ioff + size_diff);
4524 /* shift the data */
4526 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4527 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4528 data_end, old_data_start + new_size - data_end);
4530 struct btrfs_disk_key disk_key;
4533 btrfs_item_key(leaf, &disk_key, slot);
4535 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4537 struct btrfs_file_extent_item *fi;
4539 fi = btrfs_item_ptr(leaf, slot,
4540 struct btrfs_file_extent_item);
4541 fi = (struct btrfs_file_extent_item *)(
4542 (unsigned long)fi - size_diff);
4544 if (btrfs_file_extent_type(leaf, fi) ==
4545 BTRFS_FILE_EXTENT_INLINE) {
4546 ptr = btrfs_item_ptr_offset(leaf, slot);
4547 memmove_extent_buffer(leaf, ptr,
4549 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4553 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4554 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4555 data_end, old_data_start - data_end);
4557 offset = btrfs_disk_key_offset(&disk_key);
4558 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4559 btrfs_set_item_key(leaf, &disk_key, slot);
4561 fixup_low_keys(path, &disk_key, 1);
4564 item = btrfs_item_nr(slot);
4565 btrfs_set_item_size(leaf, item, new_size);
4566 btrfs_mark_buffer_dirty(leaf);
4568 if (btrfs_leaf_free_space(leaf) < 0) {
4569 btrfs_print_leaf(leaf);
4575 * make the item pointed to by the path bigger, data_size is the added size.
4577 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4580 struct extent_buffer *leaf;
4581 struct btrfs_item *item;
4583 unsigned int data_end;
4584 unsigned int old_data;
4585 unsigned int old_size;
4587 struct btrfs_map_token token;
4589 leaf = path->nodes[0];
4591 nritems = btrfs_header_nritems(leaf);
4592 data_end = leaf_data_end(leaf);
4594 if (btrfs_leaf_free_space(leaf) < data_size) {
4595 btrfs_print_leaf(leaf);
4598 slot = path->slots[0];
4599 old_data = btrfs_item_end_nr(leaf, slot);
4602 if (slot >= nritems) {
4603 btrfs_print_leaf(leaf);
4604 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4610 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4612 /* first correct the data pointers */
4613 btrfs_init_map_token(&token, leaf);
4614 for (i = slot; i < nritems; i++) {
4616 item = btrfs_item_nr(i);
4618 ioff = btrfs_token_item_offset(&token, item);
4619 btrfs_set_token_item_offset(&token, item, ioff - data_size);
4622 /* shift the data */
4623 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4624 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4625 data_end, old_data - data_end);
4627 data_end = old_data;
4628 old_size = btrfs_item_size_nr(leaf, slot);
4629 item = btrfs_item_nr(slot);
4630 btrfs_set_item_size(leaf, item, old_size + data_size);
4631 btrfs_mark_buffer_dirty(leaf);
4633 if (btrfs_leaf_free_space(leaf) < 0) {
4634 btrfs_print_leaf(leaf);
4640 * this is a helper for btrfs_insert_empty_items, the main goal here is
4641 * to save stack depth by doing the bulk of the work in a function
4642 * that doesn't call btrfs_search_slot
4644 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4645 const struct btrfs_key *cpu_key, u32 *data_size,
4646 u32 total_data, u32 total_size, int nr)
4648 struct btrfs_fs_info *fs_info = root->fs_info;
4649 struct btrfs_item *item;
4652 unsigned int data_end;
4653 struct btrfs_disk_key disk_key;
4654 struct extent_buffer *leaf;
4656 struct btrfs_map_token token;
4658 if (path->slots[0] == 0) {
4659 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4660 fixup_low_keys(path, &disk_key, 1);
4662 btrfs_unlock_up_safe(path, 1);
4664 leaf = path->nodes[0];
4665 slot = path->slots[0];
4667 nritems = btrfs_header_nritems(leaf);
4668 data_end = leaf_data_end(leaf);
4670 if (btrfs_leaf_free_space(leaf) < total_size) {
4671 btrfs_print_leaf(leaf);
4672 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4673 total_size, btrfs_leaf_free_space(leaf));
4677 btrfs_init_map_token(&token, leaf);
4678 if (slot != nritems) {
4679 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4681 if (old_data < data_end) {
4682 btrfs_print_leaf(leaf);
4683 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4684 slot, old_data, data_end);
4688 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4690 /* first correct the data pointers */
4691 for (i = slot; i < nritems; i++) {
4694 item = btrfs_item_nr(i);
4695 ioff = btrfs_token_item_offset(&token, item);
4696 btrfs_set_token_item_offset(&token, item,
4699 /* shift the items */
4700 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4701 btrfs_item_nr_offset(slot),
4702 (nritems - slot) * sizeof(struct btrfs_item));
4704 /* shift the data */
4705 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4706 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4707 data_end, old_data - data_end);
4708 data_end = old_data;
4711 /* setup the item for the new data */
4712 for (i = 0; i < nr; i++) {
4713 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4714 btrfs_set_item_key(leaf, &disk_key, slot + i);
4715 item = btrfs_item_nr(slot + i);
4716 btrfs_set_token_item_offset(&token, item, data_end - data_size[i]);
4717 data_end -= data_size[i];
4718 btrfs_set_token_item_size(&token, item, data_size[i]);
4721 btrfs_set_header_nritems(leaf, nritems + nr);
4722 btrfs_mark_buffer_dirty(leaf);
4724 if (btrfs_leaf_free_space(leaf) < 0) {
4725 btrfs_print_leaf(leaf);
4731 * Given a key and some data, insert items into the tree.
4732 * This does all the path init required, making room in the tree if needed.
4734 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4735 struct btrfs_root *root,
4736 struct btrfs_path *path,
4737 const struct btrfs_key *cpu_key, u32 *data_size,
4746 for (i = 0; i < nr; i++)
4747 total_data += data_size[i];
4749 total_size = total_data + (nr * sizeof(struct btrfs_item));
4750 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4756 slot = path->slots[0];
4759 setup_items_for_insert(root, path, cpu_key, data_size,
4760 total_data, total_size, nr);
4765 * Given a key and some data, insert an item into the tree.
4766 * This does all the path init required, making room in the tree if needed.
4768 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4769 const struct btrfs_key *cpu_key, void *data,
4773 struct btrfs_path *path;
4774 struct extent_buffer *leaf;
4777 path = btrfs_alloc_path();
4780 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4782 leaf = path->nodes[0];
4783 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4784 write_extent_buffer(leaf, data, ptr, data_size);
4785 btrfs_mark_buffer_dirty(leaf);
4787 btrfs_free_path(path);
4792 * delete the pointer from a given node.
4794 * the tree should have been previously balanced so the deletion does not
4797 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4798 int level, int slot)
4800 struct extent_buffer *parent = path->nodes[level];
4804 nritems = btrfs_header_nritems(parent);
4805 if (slot != nritems - 1) {
4807 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4808 nritems - slot - 1);
4811 memmove_extent_buffer(parent,
4812 btrfs_node_key_ptr_offset(slot),
4813 btrfs_node_key_ptr_offset(slot + 1),
4814 sizeof(struct btrfs_key_ptr) *
4815 (nritems - slot - 1));
4817 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4823 btrfs_set_header_nritems(parent, nritems);
4824 if (nritems == 0 && parent == root->node) {
4825 BUG_ON(btrfs_header_level(root->node) != 1);
4826 /* just turn the root into a leaf and break */
4827 btrfs_set_header_level(root->node, 0);
4828 } else if (slot == 0) {
4829 struct btrfs_disk_key disk_key;
4831 btrfs_node_key(parent, &disk_key, 0);
4832 fixup_low_keys(path, &disk_key, level + 1);
4834 btrfs_mark_buffer_dirty(parent);
4838 * a helper function to delete the leaf pointed to by path->slots[1] and
4841 * This deletes the pointer in path->nodes[1] and frees the leaf
4842 * block extent. zero is returned if it all worked out, < 0 otherwise.
4844 * The path must have already been setup for deleting the leaf, including
4845 * all the proper balancing. path->nodes[1] must be locked.
4847 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4848 struct btrfs_root *root,
4849 struct btrfs_path *path,
4850 struct extent_buffer *leaf)
4852 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4853 del_ptr(root, path, 1, path->slots[1]);
4856 * btrfs_free_extent is expensive, we want to make sure we
4857 * aren't holding any locks when we call it
4859 btrfs_unlock_up_safe(path, 0);
4861 root_sub_used(root, leaf->len);
4863 atomic_inc(&leaf->refs);
4864 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4865 free_extent_buffer_stale(leaf);
4868 * delete the item at the leaf level in path. If that empties
4869 * the leaf, remove it from the tree
4871 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4872 struct btrfs_path *path, int slot, int nr)
4874 struct btrfs_fs_info *fs_info = root->fs_info;
4875 struct extent_buffer *leaf;
4876 struct btrfs_item *item;
4884 leaf = path->nodes[0];
4885 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4887 for (i = 0; i < nr; i++)
4888 dsize += btrfs_item_size_nr(leaf, slot + i);
4890 nritems = btrfs_header_nritems(leaf);
4892 if (slot + nr != nritems) {
4893 int data_end = leaf_data_end(leaf);
4894 struct btrfs_map_token token;
4896 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4898 BTRFS_LEAF_DATA_OFFSET + data_end,
4899 last_off - data_end);
4901 btrfs_init_map_token(&token, leaf);
4902 for (i = slot + nr; i < nritems; i++) {
4905 item = btrfs_item_nr(i);
4906 ioff = btrfs_token_item_offset(&token, item);
4907 btrfs_set_token_item_offset(&token, item, ioff + dsize);
4910 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4911 btrfs_item_nr_offset(slot + nr),
4912 sizeof(struct btrfs_item) *
4913 (nritems - slot - nr));
4915 btrfs_set_header_nritems(leaf, nritems - nr);
4918 /* delete the leaf if we've emptied it */
4920 if (leaf == root->node) {
4921 btrfs_set_header_level(leaf, 0);
4923 btrfs_set_path_blocking(path);
4924 btrfs_clean_tree_block(leaf);
4925 btrfs_del_leaf(trans, root, path, leaf);
4928 int used = leaf_space_used(leaf, 0, nritems);
4930 struct btrfs_disk_key disk_key;
4932 btrfs_item_key(leaf, &disk_key, 0);
4933 fixup_low_keys(path, &disk_key, 1);
4936 /* delete the leaf if it is mostly empty */
4937 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4938 /* push_leaf_left fixes the path.
4939 * make sure the path still points to our leaf
4940 * for possible call to del_ptr below
4942 slot = path->slots[1];
4943 atomic_inc(&leaf->refs);
4945 btrfs_set_path_blocking(path);
4946 wret = push_leaf_left(trans, root, path, 1, 1,
4948 if (wret < 0 && wret != -ENOSPC)
4951 if (path->nodes[0] == leaf &&
4952 btrfs_header_nritems(leaf)) {
4953 wret = push_leaf_right(trans, root, path, 1,
4955 if (wret < 0 && wret != -ENOSPC)
4959 if (btrfs_header_nritems(leaf) == 0) {
4960 path->slots[1] = slot;
4961 btrfs_del_leaf(trans, root, path, leaf);
4962 free_extent_buffer(leaf);
4965 /* if we're still in the path, make sure
4966 * we're dirty. Otherwise, one of the
4967 * push_leaf functions must have already
4968 * dirtied this buffer
4970 if (path->nodes[0] == leaf)
4971 btrfs_mark_buffer_dirty(leaf);
4972 free_extent_buffer(leaf);
4975 btrfs_mark_buffer_dirty(leaf);
4982 * search the tree again to find a leaf with lesser keys
4983 * returns 0 if it found something or 1 if there are no lesser leaves.
4984 * returns < 0 on io errors.
4986 * This may release the path, and so you may lose any locks held at the
4989 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4991 struct btrfs_key key;
4992 struct btrfs_disk_key found_key;
4995 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4997 if (key.offset > 0) {
4999 } else if (key.type > 0) {
5001 key.offset = (u64)-1;
5002 } else if (key.objectid > 0) {
5005 key.offset = (u64)-1;
5010 btrfs_release_path(path);
5011 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5014 btrfs_item_key(path->nodes[0], &found_key, 0);
5015 ret = comp_keys(&found_key, &key);
5017 * We might have had an item with the previous key in the tree right
5018 * before we released our path. And after we released our path, that
5019 * item might have been pushed to the first slot (0) of the leaf we
5020 * were holding due to a tree balance. Alternatively, an item with the
5021 * previous key can exist as the only element of a leaf (big fat item).
5022 * Therefore account for these 2 cases, so that our callers (like
5023 * btrfs_previous_item) don't miss an existing item with a key matching
5024 * the previous key we computed above.
5032 * A helper function to walk down the tree starting at min_key, and looking
5033 * for nodes or leaves that are have a minimum transaction id.
5034 * This is used by the btree defrag code, and tree logging
5036 * This does not cow, but it does stuff the starting key it finds back
5037 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5038 * key and get a writable path.
5040 * This honors path->lowest_level to prevent descent past a given level
5043 * min_trans indicates the oldest transaction that you are interested
5044 * in walking through. Any nodes or leaves older than min_trans are
5045 * skipped over (without reading them).
5047 * returns zero if something useful was found, < 0 on error and 1 if there
5048 * was nothing in the tree that matched the search criteria.
5050 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5051 struct btrfs_path *path,
5054 struct extent_buffer *cur;
5055 struct btrfs_key found_key;
5061 int keep_locks = path->keep_locks;
5063 path->keep_locks = 1;
5065 cur = btrfs_read_lock_root_node(root);
5066 level = btrfs_header_level(cur);
5067 WARN_ON(path->nodes[level]);
5068 path->nodes[level] = cur;
5069 path->locks[level] = BTRFS_READ_LOCK;
5071 if (btrfs_header_generation(cur) < min_trans) {
5076 nritems = btrfs_header_nritems(cur);
5077 level = btrfs_header_level(cur);
5078 sret = btrfs_bin_search(cur, min_key, &slot);
5084 /* at the lowest level, we're done, setup the path and exit */
5085 if (level == path->lowest_level) {
5086 if (slot >= nritems)
5089 path->slots[level] = slot;
5090 btrfs_item_key_to_cpu(cur, &found_key, slot);
5093 if (sret && slot > 0)
5096 * check this node pointer against the min_trans parameters.
5097 * If it is too old, old, skip to the next one.
5099 while (slot < nritems) {
5102 gen = btrfs_node_ptr_generation(cur, slot);
5103 if (gen < min_trans) {
5111 * we didn't find a candidate key in this node, walk forward
5112 * and find another one
5114 if (slot >= nritems) {
5115 path->slots[level] = slot;
5116 btrfs_set_path_blocking(path);
5117 sret = btrfs_find_next_key(root, path, min_key, level,
5120 btrfs_release_path(path);
5126 /* save our key for returning back */
5127 btrfs_node_key_to_cpu(cur, &found_key, slot);
5128 path->slots[level] = slot;
5129 if (level == path->lowest_level) {
5133 btrfs_set_path_blocking(path);
5134 cur = btrfs_read_node_slot(cur, slot);
5140 btrfs_tree_read_lock(cur);
5142 path->locks[level - 1] = BTRFS_READ_LOCK;
5143 path->nodes[level - 1] = cur;
5144 unlock_up(path, level, 1, 0, NULL);
5147 path->keep_locks = keep_locks;
5149 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5150 btrfs_set_path_blocking(path);
5151 memcpy(min_key, &found_key, sizeof(found_key));
5157 * this is similar to btrfs_next_leaf, but does not try to preserve
5158 * and fixup the path. It looks for and returns the next key in the
5159 * tree based on the current path and the min_trans parameters.
5161 * 0 is returned if another key is found, < 0 if there are any errors
5162 * and 1 is returned if there are no higher keys in the tree
5164 * path->keep_locks should be set to 1 on the search made before
5165 * calling this function.
5167 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5168 struct btrfs_key *key, int level, u64 min_trans)
5171 struct extent_buffer *c;
5173 WARN_ON(!path->keep_locks && !path->skip_locking);
5174 while (level < BTRFS_MAX_LEVEL) {
5175 if (!path->nodes[level])
5178 slot = path->slots[level] + 1;
5179 c = path->nodes[level];
5181 if (slot >= btrfs_header_nritems(c)) {
5184 struct btrfs_key cur_key;
5185 if (level + 1 >= BTRFS_MAX_LEVEL ||
5186 !path->nodes[level + 1])
5189 if (path->locks[level + 1] || path->skip_locking) {
5194 slot = btrfs_header_nritems(c) - 1;
5196 btrfs_item_key_to_cpu(c, &cur_key, slot);
5198 btrfs_node_key_to_cpu(c, &cur_key, slot);
5200 orig_lowest = path->lowest_level;
5201 btrfs_release_path(path);
5202 path->lowest_level = level;
5203 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5205 path->lowest_level = orig_lowest;
5209 c = path->nodes[level];
5210 slot = path->slots[level];
5217 btrfs_item_key_to_cpu(c, key, slot);
5219 u64 gen = btrfs_node_ptr_generation(c, slot);
5221 if (gen < min_trans) {
5225 btrfs_node_key_to_cpu(c, key, slot);
5233 * search the tree again to find a leaf with greater keys
5234 * returns 0 if it found something or 1 if there are no greater leaves.
5235 * returns < 0 on io errors.
5237 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5239 return btrfs_next_old_leaf(root, path, 0);
5242 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5247 struct extent_buffer *c;
5248 struct extent_buffer *next;
5249 struct btrfs_key key;
5252 int old_spinning = path->leave_spinning;
5253 int next_rw_lock = 0;
5255 nritems = btrfs_header_nritems(path->nodes[0]);
5259 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5264 btrfs_release_path(path);
5266 path->keep_locks = 1;
5267 path->leave_spinning = 1;
5270 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5272 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5273 path->keep_locks = 0;
5278 nritems = btrfs_header_nritems(path->nodes[0]);
5280 * by releasing the path above we dropped all our locks. A balance
5281 * could have added more items next to the key that used to be
5282 * at the very end of the block. So, check again here and
5283 * advance the path if there are now more items available.
5285 if (nritems > 0 && path->slots[0] < nritems - 1) {
5292 * So the above check misses one case:
5293 * - after releasing the path above, someone has removed the item that
5294 * used to be at the very end of the block, and balance between leafs
5295 * gets another one with bigger key.offset to replace it.
5297 * This one should be returned as well, or we can get leaf corruption
5298 * later(esp. in __btrfs_drop_extents()).
5300 * And a bit more explanation about this check,
5301 * with ret > 0, the key isn't found, the path points to the slot
5302 * where it should be inserted, so the path->slots[0] item must be the
5305 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5310 while (level < BTRFS_MAX_LEVEL) {
5311 if (!path->nodes[level]) {
5316 slot = path->slots[level] + 1;
5317 c = path->nodes[level];
5318 if (slot >= btrfs_header_nritems(c)) {
5320 if (level == BTRFS_MAX_LEVEL) {
5328 btrfs_tree_unlock_rw(next, next_rw_lock);
5329 free_extent_buffer(next);
5333 next_rw_lock = path->locks[level];
5334 ret = read_block_for_search(root, path, &next, level,
5340 btrfs_release_path(path);
5344 if (!path->skip_locking) {
5345 ret = btrfs_try_tree_read_lock(next);
5346 if (!ret && time_seq) {
5348 * If we don't get the lock, we may be racing
5349 * with push_leaf_left, holding that lock while
5350 * itself waiting for the leaf we've currently
5351 * locked. To solve this situation, we give up
5352 * on our lock and cycle.
5354 free_extent_buffer(next);
5355 btrfs_release_path(path);
5360 btrfs_set_path_blocking(path);
5361 btrfs_tree_read_lock(next);
5363 next_rw_lock = BTRFS_READ_LOCK;
5367 path->slots[level] = slot;
5370 c = path->nodes[level];
5371 if (path->locks[level])
5372 btrfs_tree_unlock_rw(c, path->locks[level]);
5374 free_extent_buffer(c);
5375 path->nodes[level] = next;
5376 path->slots[level] = 0;
5377 if (!path->skip_locking)
5378 path->locks[level] = next_rw_lock;
5382 ret = read_block_for_search(root, path, &next, level,
5388 btrfs_release_path(path);
5392 if (!path->skip_locking) {
5393 ret = btrfs_try_tree_read_lock(next);
5395 btrfs_set_path_blocking(path);
5396 btrfs_tree_read_lock(next);
5398 next_rw_lock = BTRFS_READ_LOCK;
5403 unlock_up(path, 0, 1, 0, NULL);
5404 path->leave_spinning = old_spinning;
5406 btrfs_set_path_blocking(path);
5412 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5413 * searching until it gets past min_objectid or finds an item of 'type'
5415 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5417 int btrfs_previous_item(struct btrfs_root *root,
5418 struct btrfs_path *path, u64 min_objectid,
5421 struct btrfs_key found_key;
5422 struct extent_buffer *leaf;
5427 if (path->slots[0] == 0) {
5428 btrfs_set_path_blocking(path);
5429 ret = btrfs_prev_leaf(root, path);
5435 leaf = path->nodes[0];
5436 nritems = btrfs_header_nritems(leaf);
5439 if (path->slots[0] == nritems)
5442 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5443 if (found_key.objectid < min_objectid)
5445 if (found_key.type == type)
5447 if (found_key.objectid == min_objectid &&
5448 found_key.type < type)
5455 * search in extent tree to find a previous Metadata/Data extent item with
5458 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5460 int btrfs_previous_extent_item(struct btrfs_root *root,
5461 struct btrfs_path *path, u64 min_objectid)
5463 struct btrfs_key found_key;
5464 struct extent_buffer *leaf;
5469 if (path->slots[0] == 0) {
5470 btrfs_set_path_blocking(path);
5471 ret = btrfs_prev_leaf(root, path);
5477 leaf = path->nodes[0];
5478 nritems = btrfs_header_nritems(leaf);
5481 if (path->slots[0] == nritems)
5484 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5485 if (found_key.objectid < min_objectid)
5487 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5488 found_key.type == BTRFS_METADATA_ITEM_KEY)
5490 if (found_key.objectid == min_objectid &&
5491 found_key.type < BTRFS_EXTENT_ITEM_KEY)