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 {
36 [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
39 int btrfs_super_csum_size(const struct btrfs_super_block *s)
41 u16 t = btrfs_super_csum_type(s);
43 * csum type is validated at mount time
45 return btrfs_csums[t].size;
48 const char *btrfs_super_csum_name(u16 csum_type)
50 /* csum type is validated at mount time */
51 return btrfs_csums[csum_type].name;
54 struct btrfs_path *btrfs_alloc_path(void)
56 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
60 * set all locked nodes in the path to blocking locks. This should
61 * be done before scheduling
63 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
66 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
67 if (!p->nodes[i] || !p->locks[i])
70 * If we currently have a spinning reader or writer lock this
71 * will bump the count of blocking holders and drop the
74 if (p->locks[i] == BTRFS_READ_LOCK) {
75 btrfs_set_lock_blocking_read(p->nodes[i]);
76 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
77 } else if (p->locks[i] == BTRFS_WRITE_LOCK) {
78 btrfs_set_lock_blocking_write(p->nodes[i]);
79 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
84 /* this also releases the path */
85 void btrfs_free_path(struct btrfs_path *p)
89 btrfs_release_path(p);
90 kmem_cache_free(btrfs_path_cachep, p);
94 * path release drops references on the extent buffers in the path
95 * and it drops any locks held by this path
97 * It is safe to call this on paths that no locks or extent buffers held.
99 noinline void btrfs_release_path(struct btrfs_path *p)
103 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
108 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
111 free_extent_buffer(p->nodes[i]);
117 * safely gets a reference on the root node of a tree. A lock
118 * is not taken, so a concurrent writer may put a different node
119 * at the root of the tree. See btrfs_lock_root_node for the
122 * The extent buffer returned by this has a reference taken, so
123 * it won't disappear. It may stop being the root of the tree
124 * at any time because there are no locks held.
126 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
128 struct extent_buffer *eb;
132 eb = rcu_dereference(root->node);
135 * RCU really hurts here, we could free up the root node because
136 * it was COWed but we may not get the new root node yet so do
137 * the inc_not_zero dance and if it doesn't work then
138 * synchronize_rcu and try again.
140 if (atomic_inc_not_zero(&eb->refs)) {
150 /* loop around taking references on and locking the root node of the
151 * tree until you end up with a lock on the root. A locked buffer
152 * is returned, with a reference held.
154 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
156 struct extent_buffer *eb;
159 eb = btrfs_root_node(root);
161 if (eb == root->node)
163 btrfs_tree_unlock(eb);
164 free_extent_buffer(eb);
169 /* loop around taking references on and locking the root node of the
170 * tree until you end up with a lock on the root. A locked buffer
171 * is returned, with a reference held.
173 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
175 struct extent_buffer *eb;
178 eb = btrfs_root_node(root);
179 btrfs_tree_read_lock(eb);
180 if (eb == root->node)
182 btrfs_tree_read_unlock(eb);
183 free_extent_buffer(eb);
188 /* cowonly root (everything not a reference counted cow subvolume), just get
189 * put onto a simple dirty list. transaction.c walks this to make sure they
190 * get properly updated on disk.
192 static void add_root_to_dirty_list(struct btrfs_root *root)
194 struct btrfs_fs_info *fs_info = root->fs_info;
196 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
197 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
200 spin_lock(&fs_info->trans_lock);
201 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
202 /* Want the extent tree to be the last on the list */
203 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
204 list_move_tail(&root->dirty_list,
205 &fs_info->dirty_cowonly_roots);
207 list_move(&root->dirty_list,
208 &fs_info->dirty_cowonly_roots);
210 spin_unlock(&fs_info->trans_lock);
214 * used by snapshot creation to make a copy of a root for a tree with
215 * a given objectid. The buffer with the new root node is returned in
216 * cow_ret, and this func returns zero on success or a negative error code.
218 int btrfs_copy_root(struct btrfs_trans_handle *trans,
219 struct btrfs_root *root,
220 struct extent_buffer *buf,
221 struct extent_buffer **cow_ret, u64 new_root_objectid)
223 struct btrfs_fs_info *fs_info = root->fs_info;
224 struct extent_buffer *cow;
227 struct btrfs_disk_key disk_key;
229 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
230 trans->transid != fs_info->running_transaction->transid);
231 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
232 trans->transid != root->last_trans);
234 level = btrfs_header_level(buf);
236 btrfs_item_key(buf, &disk_key, 0);
238 btrfs_node_key(buf, &disk_key, 0);
240 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
241 &disk_key, level, buf->start, 0);
245 copy_extent_buffer_full(cow, buf);
246 btrfs_set_header_bytenr(cow, cow->start);
247 btrfs_set_header_generation(cow, trans->transid);
248 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
249 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
250 BTRFS_HEADER_FLAG_RELOC);
251 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
252 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
254 btrfs_set_header_owner(cow, new_root_objectid);
256 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
258 WARN_ON(btrfs_header_generation(buf) > trans->transid);
259 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
260 ret = btrfs_inc_ref(trans, root, cow, 1);
262 ret = btrfs_inc_ref(trans, root, cow, 0);
267 btrfs_mark_buffer_dirty(cow);
276 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
277 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
279 MOD_LOG_ROOT_REPLACE,
282 struct tree_mod_root {
287 struct tree_mod_elem {
293 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
296 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
299 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
300 struct btrfs_disk_key key;
303 /* this is used for op == MOD_LOG_MOVE_KEYS */
309 /* this is used for op == MOD_LOG_ROOT_REPLACE */
310 struct tree_mod_root old_root;
314 * Pull a new tree mod seq number for our operation.
316 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
318 return atomic64_inc_return(&fs_info->tree_mod_seq);
322 * This adds a new blocker to the tree mod log's blocker list if the @elem
323 * passed does not already have a sequence number set. So when a caller expects
324 * to record tree modifications, it should ensure to set elem->seq to zero
325 * before calling btrfs_get_tree_mod_seq.
326 * Returns a fresh, unused tree log modification sequence number, even if no new
329 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
330 struct seq_list *elem)
332 write_lock(&fs_info->tree_mod_log_lock);
333 spin_lock(&fs_info->tree_mod_seq_lock);
335 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
336 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
338 spin_unlock(&fs_info->tree_mod_seq_lock);
339 write_unlock(&fs_info->tree_mod_log_lock);
344 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
345 struct seq_list *elem)
347 struct rb_root *tm_root;
348 struct rb_node *node;
349 struct rb_node *next;
350 struct seq_list *cur_elem;
351 struct tree_mod_elem *tm;
352 u64 min_seq = (u64)-1;
353 u64 seq_putting = elem->seq;
358 spin_lock(&fs_info->tree_mod_seq_lock);
359 list_del(&elem->list);
362 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
363 if (cur_elem->seq < min_seq) {
364 if (seq_putting > cur_elem->seq) {
366 * blocker with lower sequence number exists, we
367 * cannot remove anything from the log
369 spin_unlock(&fs_info->tree_mod_seq_lock);
372 min_seq = cur_elem->seq;
375 spin_unlock(&fs_info->tree_mod_seq_lock);
378 * anything that's lower than the lowest existing (read: blocked)
379 * sequence number can be removed from the tree.
381 write_lock(&fs_info->tree_mod_log_lock);
382 tm_root = &fs_info->tree_mod_log;
383 for (node = rb_first(tm_root); node; node = next) {
384 next = rb_next(node);
385 tm = rb_entry(node, struct tree_mod_elem, node);
386 if (tm->seq > min_seq)
388 rb_erase(node, tm_root);
391 write_unlock(&fs_info->tree_mod_log_lock);
395 * key order of the log:
396 * node/leaf start address -> sequence
398 * The 'start address' is the logical address of the *new* root node
399 * for root replace operations, or the logical address of the affected
400 * block for all other operations.
403 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
405 struct rb_root *tm_root;
406 struct rb_node **new;
407 struct rb_node *parent = NULL;
408 struct tree_mod_elem *cur;
410 lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
412 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
414 tm_root = &fs_info->tree_mod_log;
415 new = &tm_root->rb_node;
417 cur = rb_entry(*new, struct tree_mod_elem, node);
419 if (cur->logical < tm->logical)
420 new = &((*new)->rb_left);
421 else if (cur->logical > tm->logical)
422 new = &((*new)->rb_right);
423 else if (cur->seq < tm->seq)
424 new = &((*new)->rb_left);
425 else if (cur->seq > tm->seq)
426 new = &((*new)->rb_right);
431 rb_link_node(&tm->node, parent, new);
432 rb_insert_color(&tm->node, tm_root);
437 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
438 * returns zero with the tree_mod_log_lock acquired. The caller must hold
439 * this until all tree mod log insertions are recorded in the rb tree and then
440 * write unlock fs_info::tree_mod_log_lock.
442 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
443 struct extent_buffer *eb) {
445 if (list_empty(&(fs_info)->tree_mod_seq_list))
447 if (eb && btrfs_header_level(eb) == 0)
450 write_lock(&fs_info->tree_mod_log_lock);
451 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
452 write_unlock(&fs_info->tree_mod_log_lock);
459 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
460 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
461 struct extent_buffer *eb)
464 if (list_empty(&(fs_info)->tree_mod_seq_list))
466 if (eb && btrfs_header_level(eb) == 0)
472 static struct tree_mod_elem *
473 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
474 enum mod_log_op op, gfp_t flags)
476 struct tree_mod_elem *tm;
478 tm = kzalloc(sizeof(*tm), flags);
482 tm->logical = eb->start;
483 if (op != MOD_LOG_KEY_ADD) {
484 btrfs_node_key(eb, &tm->key, slot);
485 tm->blockptr = btrfs_node_blockptr(eb, slot);
489 tm->generation = btrfs_node_ptr_generation(eb, slot);
490 RB_CLEAR_NODE(&tm->node);
495 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
496 enum mod_log_op op, gfp_t flags)
498 struct tree_mod_elem *tm;
501 if (!tree_mod_need_log(eb->fs_info, eb))
504 tm = alloc_tree_mod_elem(eb, slot, op, flags);
508 if (tree_mod_dont_log(eb->fs_info, eb)) {
513 ret = __tree_mod_log_insert(eb->fs_info, tm);
514 write_unlock(&eb->fs_info->tree_mod_log_lock);
521 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
522 int dst_slot, int src_slot, int nr_items)
524 struct tree_mod_elem *tm = NULL;
525 struct tree_mod_elem **tm_list = NULL;
530 if (!tree_mod_need_log(eb->fs_info, eb))
533 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
537 tm = kzalloc(sizeof(*tm), GFP_NOFS);
543 tm->logical = eb->start;
545 tm->move.dst_slot = dst_slot;
546 tm->move.nr_items = nr_items;
547 tm->op = MOD_LOG_MOVE_KEYS;
549 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
550 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
551 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
558 if (tree_mod_dont_log(eb->fs_info, eb))
563 * When we override something during the move, we log these removals.
564 * This can only happen when we move towards the beginning of the
565 * buffer, i.e. dst_slot < src_slot.
567 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
568 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
573 ret = __tree_mod_log_insert(eb->fs_info, tm);
576 write_unlock(&eb->fs_info->tree_mod_log_lock);
581 for (i = 0; i < nr_items; i++) {
582 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
583 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
587 write_unlock(&eb->fs_info->tree_mod_log_lock);
595 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
596 struct tree_mod_elem **tm_list,
602 for (i = nritems - 1; i >= 0; i--) {
603 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
605 for (j = nritems - 1; j > i; j--)
606 rb_erase(&tm_list[j]->node,
607 &fs_info->tree_mod_log);
615 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
616 struct extent_buffer *new_root, int log_removal)
618 struct btrfs_fs_info *fs_info = old_root->fs_info;
619 struct tree_mod_elem *tm = NULL;
620 struct tree_mod_elem **tm_list = NULL;
625 if (!tree_mod_need_log(fs_info, NULL))
628 if (log_removal && btrfs_header_level(old_root) > 0) {
629 nritems = btrfs_header_nritems(old_root);
630 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
636 for (i = 0; i < nritems; i++) {
637 tm_list[i] = alloc_tree_mod_elem(old_root, i,
638 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
646 tm = kzalloc(sizeof(*tm), GFP_NOFS);
652 tm->logical = new_root->start;
653 tm->old_root.logical = old_root->start;
654 tm->old_root.level = btrfs_header_level(old_root);
655 tm->generation = btrfs_header_generation(old_root);
656 tm->op = MOD_LOG_ROOT_REPLACE;
658 if (tree_mod_dont_log(fs_info, NULL))
662 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
664 ret = __tree_mod_log_insert(fs_info, tm);
666 write_unlock(&fs_info->tree_mod_log_lock);
675 for (i = 0; i < nritems; i++)
684 static struct tree_mod_elem *
685 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
688 struct rb_root *tm_root;
689 struct rb_node *node;
690 struct tree_mod_elem *cur = NULL;
691 struct tree_mod_elem *found = NULL;
693 read_lock(&fs_info->tree_mod_log_lock);
694 tm_root = &fs_info->tree_mod_log;
695 node = tm_root->rb_node;
697 cur = rb_entry(node, struct tree_mod_elem, node);
698 if (cur->logical < start) {
699 node = node->rb_left;
700 } else if (cur->logical > start) {
701 node = node->rb_right;
702 } else if (cur->seq < min_seq) {
703 node = node->rb_left;
704 } else if (!smallest) {
705 /* we want the node with the highest seq */
707 BUG_ON(found->seq > cur->seq);
709 node = node->rb_left;
710 } else if (cur->seq > min_seq) {
711 /* we want the node with the smallest seq */
713 BUG_ON(found->seq < cur->seq);
715 node = node->rb_right;
721 read_unlock(&fs_info->tree_mod_log_lock);
727 * this returns the element from the log with the smallest time sequence
728 * value that's in the log (the oldest log item). any element with a time
729 * sequence lower than min_seq will be ignored.
731 static struct tree_mod_elem *
732 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
735 return __tree_mod_log_search(fs_info, start, min_seq, 1);
739 * this returns the element from the log with the largest time sequence
740 * value that's in the log (the most recent log item). any element with
741 * a time sequence lower than min_seq will be ignored.
743 static struct tree_mod_elem *
744 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
746 return __tree_mod_log_search(fs_info, start, min_seq, 0);
749 static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst,
750 struct extent_buffer *src, unsigned long dst_offset,
751 unsigned long src_offset, int nr_items)
753 struct btrfs_fs_info *fs_info = dst->fs_info;
755 struct tree_mod_elem **tm_list = NULL;
756 struct tree_mod_elem **tm_list_add, **tm_list_rem;
760 if (!tree_mod_need_log(fs_info, NULL))
763 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
766 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
771 tm_list_add = tm_list;
772 tm_list_rem = tm_list + nr_items;
773 for (i = 0; i < nr_items; i++) {
774 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
775 MOD_LOG_KEY_REMOVE, GFP_NOFS);
776 if (!tm_list_rem[i]) {
781 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
782 MOD_LOG_KEY_ADD, GFP_NOFS);
783 if (!tm_list_add[i]) {
789 if (tree_mod_dont_log(fs_info, NULL))
793 for (i = 0; i < nr_items; i++) {
794 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
797 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
802 write_unlock(&fs_info->tree_mod_log_lock);
808 for (i = 0; i < nr_items * 2; i++) {
809 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
810 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
814 write_unlock(&fs_info->tree_mod_log_lock);
820 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
822 struct tree_mod_elem **tm_list = NULL;
827 if (btrfs_header_level(eb) == 0)
830 if (!tree_mod_need_log(eb->fs_info, NULL))
833 nritems = btrfs_header_nritems(eb);
834 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
838 for (i = 0; i < nritems; i++) {
839 tm_list[i] = alloc_tree_mod_elem(eb, i,
840 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
847 if (tree_mod_dont_log(eb->fs_info, eb))
850 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
851 write_unlock(&eb->fs_info->tree_mod_log_lock);
859 for (i = 0; i < nritems; i++)
867 * check if the tree block can be shared by multiple trees
869 int btrfs_block_can_be_shared(struct btrfs_root *root,
870 struct extent_buffer *buf)
873 * Tree blocks not in reference counted trees and tree roots
874 * are never shared. If a block was allocated after the last
875 * snapshot and the block was not allocated by tree relocation,
876 * we know the block is not shared.
878 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
879 buf != root->node && buf != root->commit_root &&
880 (btrfs_header_generation(buf) <=
881 btrfs_root_last_snapshot(&root->root_item) ||
882 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
888 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
889 struct btrfs_root *root,
890 struct extent_buffer *buf,
891 struct extent_buffer *cow,
894 struct btrfs_fs_info *fs_info = root->fs_info;
902 * Backrefs update rules:
904 * Always use full backrefs for extent pointers in tree block
905 * allocated by tree relocation.
907 * If a shared tree block is no longer referenced by its owner
908 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
909 * use full backrefs for extent pointers in tree block.
911 * If a tree block is been relocating
912 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
913 * use full backrefs for extent pointers in tree block.
914 * The reason for this is some operations (such as drop tree)
915 * are only allowed for blocks use full backrefs.
918 if (btrfs_block_can_be_shared(root, buf)) {
919 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
920 btrfs_header_level(buf), 1,
926 btrfs_handle_fs_error(fs_info, ret, NULL);
931 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
932 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
933 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
938 owner = btrfs_header_owner(buf);
939 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
940 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
943 if ((owner == root->root_key.objectid ||
944 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
945 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
946 ret = btrfs_inc_ref(trans, root, buf, 1);
950 if (root->root_key.objectid ==
951 BTRFS_TREE_RELOC_OBJECTID) {
952 ret = btrfs_dec_ref(trans, root, buf, 0);
955 ret = btrfs_inc_ref(trans, root, cow, 1);
959 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
962 if (root->root_key.objectid ==
963 BTRFS_TREE_RELOC_OBJECTID)
964 ret = btrfs_inc_ref(trans, root, cow, 1);
966 ret = btrfs_inc_ref(trans, root, cow, 0);
970 if (new_flags != 0) {
971 int level = btrfs_header_level(buf);
973 ret = btrfs_set_disk_extent_flags(trans,
976 new_flags, level, 0);
981 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
982 if (root->root_key.objectid ==
983 BTRFS_TREE_RELOC_OBJECTID)
984 ret = btrfs_inc_ref(trans, root, cow, 1);
986 ret = btrfs_inc_ref(trans, root, cow, 0);
989 ret = btrfs_dec_ref(trans, root, buf, 1);
993 btrfs_clean_tree_block(buf);
999 static struct extent_buffer *alloc_tree_block_no_bg_flush(
1000 struct btrfs_trans_handle *trans,
1001 struct btrfs_root *root,
1003 const struct btrfs_disk_key *disk_key,
1008 struct btrfs_fs_info *fs_info = root->fs_info;
1009 struct extent_buffer *ret;
1012 * If we are COWing a node/leaf from the extent, chunk, device or free
1013 * space trees, make sure that we do not finish block group creation of
1014 * pending block groups. We do this to avoid a deadlock.
1015 * COWing can result in allocation of a new chunk, and flushing pending
1016 * block groups (btrfs_create_pending_block_groups()) can be triggered
1017 * when finishing allocation of a new chunk. Creation of a pending block
1018 * group modifies the extent, chunk, device and free space trees,
1019 * therefore we could deadlock with ourselves since we are holding a
1020 * lock on an extent buffer that btrfs_create_pending_block_groups() may
1022 * For similar reasons, we also need to delay flushing pending block
1023 * groups when splitting a leaf or node, from one of those trees, since
1024 * we are holding a write lock on it and its parent or when inserting a
1025 * new root node for one of those trees.
1027 if (root == fs_info->extent_root ||
1028 root == fs_info->chunk_root ||
1029 root == fs_info->dev_root ||
1030 root == fs_info->free_space_root)
1031 trans->can_flush_pending_bgs = false;
1033 ret = btrfs_alloc_tree_block(trans, root, parent_start,
1034 root->root_key.objectid, disk_key, level,
1036 trans->can_flush_pending_bgs = true;
1042 * does the dirty work in cow of a single block. The parent block (if
1043 * supplied) is updated to point to the new cow copy. The new buffer is marked
1044 * dirty and returned locked. If you modify the block it needs to be marked
1047 * search_start -- an allocation hint for the new block
1049 * empty_size -- a hint that you plan on doing more cow. This is the size in
1050 * bytes the allocator should try to find free next to the block it returns.
1051 * This is just a hint and may be ignored by the allocator.
1053 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1054 struct btrfs_root *root,
1055 struct extent_buffer *buf,
1056 struct extent_buffer *parent, int parent_slot,
1057 struct extent_buffer **cow_ret,
1058 u64 search_start, u64 empty_size)
1060 struct btrfs_fs_info *fs_info = root->fs_info;
1061 struct btrfs_disk_key disk_key;
1062 struct extent_buffer *cow;
1065 int unlock_orig = 0;
1066 u64 parent_start = 0;
1068 if (*cow_ret == buf)
1071 btrfs_assert_tree_locked(buf);
1073 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1074 trans->transid != fs_info->running_transaction->transid);
1075 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1076 trans->transid != root->last_trans);
1078 level = btrfs_header_level(buf);
1081 btrfs_item_key(buf, &disk_key, 0);
1083 btrfs_node_key(buf, &disk_key, 0);
1085 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1086 parent_start = parent->start;
1088 cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1089 level, search_start, empty_size);
1091 return PTR_ERR(cow);
1093 /* cow is set to blocking by btrfs_init_new_buffer */
1095 copy_extent_buffer_full(cow, buf);
1096 btrfs_set_header_bytenr(cow, cow->start);
1097 btrfs_set_header_generation(cow, trans->transid);
1098 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1099 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1100 BTRFS_HEADER_FLAG_RELOC);
1101 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1102 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1104 btrfs_set_header_owner(cow, root->root_key.objectid);
1106 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1108 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1110 btrfs_abort_transaction(trans, ret);
1114 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1115 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1117 btrfs_abort_transaction(trans, ret);
1122 if (buf == root->node) {
1123 WARN_ON(parent && parent != buf);
1124 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1125 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1126 parent_start = buf->start;
1128 extent_buffer_get(cow);
1129 ret = tree_mod_log_insert_root(root->node, cow, 1);
1131 rcu_assign_pointer(root->node, cow);
1133 btrfs_free_tree_block(trans, root, buf, parent_start,
1135 free_extent_buffer(buf);
1136 add_root_to_dirty_list(root);
1138 WARN_ON(trans->transid != btrfs_header_generation(parent));
1139 tree_mod_log_insert_key(parent, parent_slot,
1140 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1141 btrfs_set_node_blockptr(parent, parent_slot,
1143 btrfs_set_node_ptr_generation(parent, parent_slot,
1145 btrfs_mark_buffer_dirty(parent);
1147 ret = tree_mod_log_free_eb(buf);
1149 btrfs_abort_transaction(trans, ret);
1153 btrfs_free_tree_block(trans, root, buf, parent_start,
1157 btrfs_tree_unlock(buf);
1158 free_extent_buffer_stale(buf);
1159 btrfs_mark_buffer_dirty(cow);
1165 * returns the logical address of the oldest predecessor of the given root.
1166 * entries older than time_seq are ignored.
1168 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1169 struct extent_buffer *eb_root, u64 time_seq)
1171 struct tree_mod_elem *tm;
1172 struct tree_mod_elem *found = NULL;
1173 u64 root_logical = eb_root->start;
1180 * the very last operation that's logged for a root is the
1181 * replacement operation (if it is replaced at all). this has
1182 * the logical address of the *new* root, making it the very
1183 * first operation that's logged for this root.
1186 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1191 * if there are no tree operation for the oldest root, we simply
1192 * return it. this should only happen if that (old) root is at
1199 * if there's an operation that's not a root replacement, we
1200 * found the oldest version of our root. normally, we'll find a
1201 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1203 if (tm->op != MOD_LOG_ROOT_REPLACE)
1207 root_logical = tm->old_root.logical;
1211 /* if there's no old root to return, return what we found instead */
1219 * tm is a pointer to the first operation to rewind within eb. then, all
1220 * previous operations will be rewound (until we reach something older than
1224 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1225 u64 time_seq, struct tree_mod_elem *first_tm)
1228 struct rb_node *next;
1229 struct tree_mod_elem *tm = first_tm;
1230 unsigned long o_dst;
1231 unsigned long o_src;
1232 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1234 n = btrfs_header_nritems(eb);
1235 read_lock(&fs_info->tree_mod_log_lock);
1236 while (tm && tm->seq >= time_seq) {
1238 * all the operations are recorded with the operator used for
1239 * the modification. as we're going backwards, we do the
1240 * opposite of each operation here.
1243 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1244 BUG_ON(tm->slot < n);
1246 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1247 case MOD_LOG_KEY_REMOVE:
1248 btrfs_set_node_key(eb, &tm->key, tm->slot);
1249 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1250 btrfs_set_node_ptr_generation(eb, tm->slot,
1254 case MOD_LOG_KEY_REPLACE:
1255 BUG_ON(tm->slot >= n);
1256 btrfs_set_node_key(eb, &tm->key, tm->slot);
1257 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1258 btrfs_set_node_ptr_generation(eb, tm->slot,
1261 case MOD_LOG_KEY_ADD:
1262 /* if a move operation is needed it's in the log */
1265 case MOD_LOG_MOVE_KEYS:
1266 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1267 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1268 memmove_extent_buffer(eb, o_dst, o_src,
1269 tm->move.nr_items * p_size);
1271 case MOD_LOG_ROOT_REPLACE:
1273 * this operation is special. for roots, this must be
1274 * handled explicitly before rewinding.
1275 * for non-roots, this operation may exist if the node
1276 * was a root: root A -> child B; then A gets empty and
1277 * B is promoted to the new root. in the mod log, we'll
1278 * have a root-replace operation for B, a tree block
1279 * that is no root. we simply ignore that operation.
1283 next = rb_next(&tm->node);
1286 tm = rb_entry(next, struct tree_mod_elem, node);
1287 if (tm->logical != first_tm->logical)
1290 read_unlock(&fs_info->tree_mod_log_lock);
1291 btrfs_set_header_nritems(eb, n);
1295 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1296 * is returned. If rewind operations happen, a fresh buffer is returned. The
1297 * returned buffer is always read-locked. If the returned buffer is not the
1298 * input buffer, the lock on the input buffer is released and the input buffer
1299 * is freed (its refcount is decremented).
1301 static struct extent_buffer *
1302 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1303 struct extent_buffer *eb, u64 time_seq)
1305 struct extent_buffer *eb_rewin;
1306 struct tree_mod_elem *tm;
1311 if (btrfs_header_level(eb) == 0)
1314 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1318 btrfs_set_path_blocking(path);
1319 btrfs_set_lock_blocking_read(eb);
1321 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1322 BUG_ON(tm->slot != 0);
1323 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1325 btrfs_tree_read_unlock_blocking(eb);
1326 free_extent_buffer(eb);
1329 btrfs_set_header_bytenr(eb_rewin, eb->start);
1330 btrfs_set_header_backref_rev(eb_rewin,
1331 btrfs_header_backref_rev(eb));
1332 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1333 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1335 eb_rewin = btrfs_clone_extent_buffer(eb);
1337 btrfs_tree_read_unlock_blocking(eb);
1338 free_extent_buffer(eb);
1343 btrfs_tree_read_unlock_blocking(eb);
1344 free_extent_buffer(eb);
1346 btrfs_tree_read_lock(eb_rewin);
1347 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1348 WARN_ON(btrfs_header_nritems(eb_rewin) >
1349 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1355 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1356 * value. If there are no changes, the current root->root_node is returned. If
1357 * anything changed in between, there's a fresh buffer allocated on which the
1358 * rewind operations are done. In any case, the returned buffer is read locked.
1359 * Returns NULL on error (with no locks held).
1361 static inline struct extent_buffer *
1362 get_old_root(struct btrfs_root *root, u64 time_seq)
1364 struct btrfs_fs_info *fs_info = root->fs_info;
1365 struct tree_mod_elem *tm;
1366 struct extent_buffer *eb = NULL;
1367 struct extent_buffer *eb_root;
1368 u64 eb_root_owner = 0;
1369 struct extent_buffer *old;
1370 struct tree_mod_root *old_root = NULL;
1371 u64 old_generation = 0;
1375 eb_root = btrfs_read_lock_root_node(root);
1376 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1380 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1381 old_root = &tm->old_root;
1382 old_generation = tm->generation;
1383 logical = old_root->logical;
1384 level = old_root->level;
1386 logical = eb_root->start;
1387 level = btrfs_header_level(eb_root);
1390 tm = tree_mod_log_search(fs_info, logical, time_seq);
1391 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1392 btrfs_tree_read_unlock(eb_root);
1393 free_extent_buffer(eb_root);
1394 old = read_tree_block(fs_info, logical, 0, level, NULL);
1395 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1397 free_extent_buffer(old);
1399 "failed to read tree block %llu from get_old_root",
1402 eb = btrfs_clone_extent_buffer(old);
1403 free_extent_buffer(old);
1405 } else if (old_root) {
1406 eb_root_owner = btrfs_header_owner(eb_root);
1407 btrfs_tree_read_unlock(eb_root);
1408 free_extent_buffer(eb_root);
1409 eb = alloc_dummy_extent_buffer(fs_info, logical);
1411 btrfs_set_lock_blocking_read(eb_root);
1412 eb = btrfs_clone_extent_buffer(eb_root);
1413 btrfs_tree_read_unlock_blocking(eb_root);
1414 free_extent_buffer(eb_root);
1419 btrfs_tree_read_lock(eb);
1421 btrfs_set_header_bytenr(eb, eb->start);
1422 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1423 btrfs_set_header_owner(eb, eb_root_owner);
1424 btrfs_set_header_level(eb, old_root->level);
1425 btrfs_set_header_generation(eb, old_generation);
1428 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1430 WARN_ON(btrfs_header_level(eb) != 0);
1431 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1436 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1438 struct tree_mod_elem *tm;
1440 struct extent_buffer *eb_root = btrfs_root_node(root);
1442 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1443 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1444 level = tm->old_root.level;
1446 level = btrfs_header_level(eb_root);
1448 free_extent_buffer(eb_root);
1453 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1454 struct btrfs_root *root,
1455 struct extent_buffer *buf)
1457 if (btrfs_is_testing(root->fs_info))
1460 /* Ensure we can see the FORCE_COW bit */
1461 smp_mb__before_atomic();
1464 * We do not need to cow a block if
1465 * 1) this block is not created or changed in this transaction;
1466 * 2) this block does not belong to TREE_RELOC tree;
1467 * 3) the root is not forced COW.
1469 * What is forced COW:
1470 * when we create snapshot during committing the transaction,
1471 * after we've finished copying src root, we must COW the shared
1472 * block to ensure the metadata consistency.
1474 if (btrfs_header_generation(buf) == trans->transid &&
1475 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1476 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1477 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1478 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1484 * cows a single block, see __btrfs_cow_block for the real work.
1485 * This version of it has extra checks so that a block isn't COWed more than
1486 * once per transaction, as long as it hasn't been written yet
1488 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1489 struct btrfs_root *root, struct extent_buffer *buf,
1490 struct extent_buffer *parent, int parent_slot,
1491 struct extent_buffer **cow_ret)
1493 struct btrfs_fs_info *fs_info = root->fs_info;
1497 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1499 "COW'ing blocks on a fs root that's being dropped");
1501 if (trans->transaction != fs_info->running_transaction)
1502 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1504 fs_info->running_transaction->transid);
1506 if (trans->transid != fs_info->generation)
1507 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1508 trans->transid, fs_info->generation);
1510 if (!should_cow_block(trans, root, buf)) {
1511 trans->dirty = true;
1516 search_start = buf->start & ~((u64)SZ_1G - 1);
1519 btrfs_set_lock_blocking_write(parent);
1520 btrfs_set_lock_blocking_write(buf);
1523 * Before CoWing this block for later modification, check if it's
1524 * the subtree root and do the delayed subtree trace if needed.
1526 * Also We don't care about the error, as it's handled internally.
1528 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1529 ret = __btrfs_cow_block(trans, root, buf, parent,
1530 parent_slot, cow_ret, search_start, 0);
1532 trace_btrfs_cow_block(root, buf, *cow_ret);
1538 * helper function for defrag to decide if two blocks pointed to by a
1539 * node are actually close by
1541 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1543 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1545 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1551 * compare two keys in a memcmp fashion
1553 static int comp_keys(const struct btrfs_disk_key *disk,
1554 const struct btrfs_key *k2)
1556 struct btrfs_key k1;
1558 btrfs_disk_key_to_cpu(&k1, disk);
1560 return btrfs_comp_cpu_keys(&k1, k2);
1564 * same as comp_keys only with two btrfs_key's
1566 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1568 if (k1->objectid > k2->objectid)
1570 if (k1->objectid < k2->objectid)
1572 if (k1->type > k2->type)
1574 if (k1->type < k2->type)
1576 if (k1->offset > k2->offset)
1578 if (k1->offset < k2->offset)
1584 * this is used by the defrag code to go through all the
1585 * leaves pointed to by a node and reallocate them so that
1586 * disk order is close to key order
1588 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1589 struct btrfs_root *root, struct extent_buffer *parent,
1590 int start_slot, u64 *last_ret,
1591 struct btrfs_key *progress)
1593 struct btrfs_fs_info *fs_info = root->fs_info;
1594 struct extent_buffer *cur;
1597 u64 search_start = *last_ret;
1607 int progress_passed = 0;
1608 struct btrfs_disk_key disk_key;
1610 parent_level = btrfs_header_level(parent);
1612 WARN_ON(trans->transaction != fs_info->running_transaction);
1613 WARN_ON(trans->transid != fs_info->generation);
1615 parent_nritems = btrfs_header_nritems(parent);
1616 blocksize = fs_info->nodesize;
1617 end_slot = parent_nritems - 1;
1619 if (parent_nritems <= 1)
1622 btrfs_set_lock_blocking_write(parent);
1624 for (i = start_slot; i <= end_slot; i++) {
1625 struct btrfs_key first_key;
1628 btrfs_node_key(parent, &disk_key, i);
1629 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1632 progress_passed = 1;
1633 blocknr = btrfs_node_blockptr(parent, i);
1634 gen = btrfs_node_ptr_generation(parent, i);
1635 btrfs_node_key_to_cpu(parent, &first_key, i);
1636 if (last_block == 0)
1637 last_block = blocknr;
1640 other = btrfs_node_blockptr(parent, i - 1);
1641 close = close_blocks(blocknr, other, blocksize);
1643 if (!close && i < end_slot) {
1644 other = btrfs_node_blockptr(parent, i + 1);
1645 close = close_blocks(blocknr, other, blocksize);
1648 last_block = blocknr;
1652 cur = find_extent_buffer(fs_info, blocknr);
1654 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1657 if (!cur || !uptodate) {
1659 cur = read_tree_block(fs_info, blocknr, gen,
1663 return PTR_ERR(cur);
1664 } else if (!extent_buffer_uptodate(cur)) {
1665 free_extent_buffer(cur);
1668 } else if (!uptodate) {
1669 err = btrfs_read_buffer(cur, gen,
1670 parent_level - 1,&first_key);
1672 free_extent_buffer(cur);
1677 if (search_start == 0)
1678 search_start = last_block;
1680 btrfs_tree_lock(cur);
1681 btrfs_set_lock_blocking_write(cur);
1682 err = __btrfs_cow_block(trans, root, cur, parent, i,
1685 (end_slot - i) * blocksize));
1687 btrfs_tree_unlock(cur);
1688 free_extent_buffer(cur);
1691 search_start = cur->start;
1692 last_block = cur->start;
1693 *last_ret = search_start;
1694 btrfs_tree_unlock(cur);
1695 free_extent_buffer(cur);
1701 * search for key in the extent_buffer. The items start at offset p,
1702 * and they are item_size apart. There are 'max' items in p.
1704 * the slot in the array is returned via slot, and it points to
1705 * the place where you would insert key if it is not found in
1708 * slot may point to max if the key is bigger than all of the keys
1710 static noinline int generic_bin_search(struct extent_buffer *eb,
1711 unsigned long p, int item_size,
1712 const struct btrfs_key *key,
1719 struct btrfs_disk_key *tmp = NULL;
1720 struct btrfs_disk_key unaligned;
1721 unsigned long offset;
1723 unsigned long map_start = 0;
1724 unsigned long map_len = 0;
1728 btrfs_err(eb->fs_info,
1729 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1730 __func__, low, high, eb->start,
1731 btrfs_header_owner(eb), btrfs_header_level(eb));
1735 while (low < high) {
1736 mid = (low + high) / 2;
1737 offset = p + mid * item_size;
1739 if (!kaddr || offset < map_start ||
1740 (offset + sizeof(struct btrfs_disk_key)) >
1741 map_start + map_len) {
1743 err = map_private_extent_buffer(eb, offset,
1744 sizeof(struct btrfs_disk_key),
1745 &kaddr, &map_start, &map_len);
1748 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1750 } else if (err == 1) {
1751 read_extent_buffer(eb, &unaligned,
1752 offset, sizeof(unaligned));
1759 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1762 ret = comp_keys(tmp, key);
1778 * simple bin_search frontend that does the right thing for
1781 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1782 int level, int *slot)
1785 return generic_bin_search(eb,
1786 offsetof(struct btrfs_leaf, items),
1787 sizeof(struct btrfs_item),
1788 key, btrfs_header_nritems(eb),
1791 return generic_bin_search(eb,
1792 offsetof(struct btrfs_node, ptrs),
1793 sizeof(struct btrfs_key_ptr),
1794 key, btrfs_header_nritems(eb),
1798 static void root_add_used(struct btrfs_root *root, u32 size)
1800 spin_lock(&root->accounting_lock);
1801 btrfs_set_root_used(&root->root_item,
1802 btrfs_root_used(&root->root_item) + size);
1803 spin_unlock(&root->accounting_lock);
1806 static void root_sub_used(struct btrfs_root *root, u32 size)
1808 spin_lock(&root->accounting_lock);
1809 btrfs_set_root_used(&root->root_item,
1810 btrfs_root_used(&root->root_item) - size);
1811 spin_unlock(&root->accounting_lock);
1814 /* given a node and slot number, this reads the blocks it points to. The
1815 * extent buffer is returned with a reference taken (but unlocked).
1817 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1820 int level = btrfs_header_level(parent);
1821 struct extent_buffer *eb;
1822 struct btrfs_key first_key;
1824 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1825 return ERR_PTR(-ENOENT);
1829 btrfs_node_key_to_cpu(parent, &first_key, slot);
1830 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1831 btrfs_node_ptr_generation(parent, slot),
1832 level - 1, &first_key);
1833 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1834 free_extent_buffer(eb);
1842 * node level balancing, used to make sure nodes are in proper order for
1843 * item deletion. We balance from the top down, so we have to make sure
1844 * that a deletion won't leave an node completely empty later on.
1846 static noinline int balance_level(struct btrfs_trans_handle *trans,
1847 struct btrfs_root *root,
1848 struct btrfs_path *path, int level)
1850 struct btrfs_fs_info *fs_info = root->fs_info;
1851 struct extent_buffer *right = NULL;
1852 struct extent_buffer *mid;
1853 struct extent_buffer *left = NULL;
1854 struct extent_buffer *parent = NULL;
1858 int orig_slot = path->slots[level];
1863 mid = path->nodes[level];
1865 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1866 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1867 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1869 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1871 if (level < BTRFS_MAX_LEVEL - 1) {
1872 parent = path->nodes[level + 1];
1873 pslot = path->slots[level + 1];
1877 * deal with the case where there is only one pointer in the root
1878 * by promoting the node below to a root
1881 struct extent_buffer *child;
1883 if (btrfs_header_nritems(mid) != 1)
1886 /* promote the child to a root */
1887 child = btrfs_read_node_slot(mid, 0);
1888 if (IS_ERR(child)) {
1889 ret = PTR_ERR(child);
1890 btrfs_handle_fs_error(fs_info, ret, NULL);
1894 btrfs_tree_lock(child);
1895 btrfs_set_lock_blocking_write(child);
1896 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1898 btrfs_tree_unlock(child);
1899 free_extent_buffer(child);
1903 ret = tree_mod_log_insert_root(root->node, child, 1);
1905 rcu_assign_pointer(root->node, child);
1907 add_root_to_dirty_list(root);
1908 btrfs_tree_unlock(child);
1910 path->locks[level] = 0;
1911 path->nodes[level] = NULL;
1912 btrfs_clean_tree_block(mid);
1913 btrfs_tree_unlock(mid);
1914 /* once for the path */
1915 free_extent_buffer(mid);
1917 root_sub_used(root, mid->len);
1918 btrfs_free_tree_block(trans, root, mid, 0, 1);
1919 /* once for the root ptr */
1920 free_extent_buffer_stale(mid);
1923 if (btrfs_header_nritems(mid) >
1924 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1927 left = btrfs_read_node_slot(parent, pslot - 1);
1932 btrfs_tree_lock(left);
1933 btrfs_set_lock_blocking_write(left);
1934 wret = btrfs_cow_block(trans, root, left,
1935 parent, pslot - 1, &left);
1942 right = btrfs_read_node_slot(parent, pslot + 1);
1947 btrfs_tree_lock(right);
1948 btrfs_set_lock_blocking_write(right);
1949 wret = btrfs_cow_block(trans, root, right,
1950 parent, pslot + 1, &right);
1957 /* first, try to make some room in the middle buffer */
1959 orig_slot += btrfs_header_nritems(left);
1960 wret = push_node_left(trans, left, mid, 1);
1966 * then try to empty the right most buffer into the middle
1969 wret = push_node_left(trans, mid, right, 1);
1970 if (wret < 0 && wret != -ENOSPC)
1972 if (btrfs_header_nritems(right) == 0) {
1973 btrfs_clean_tree_block(right);
1974 btrfs_tree_unlock(right);
1975 del_ptr(root, path, level + 1, pslot + 1);
1976 root_sub_used(root, right->len);
1977 btrfs_free_tree_block(trans, root, right, 0, 1);
1978 free_extent_buffer_stale(right);
1981 struct btrfs_disk_key right_key;
1982 btrfs_node_key(right, &right_key, 0);
1983 ret = tree_mod_log_insert_key(parent, pslot + 1,
1984 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1986 btrfs_set_node_key(parent, &right_key, pslot + 1);
1987 btrfs_mark_buffer_dirty(parent);
1990 if (btrfs_header_nritems(mid) == 1) {
1992 * we're not allowed to leave a node with one item in the
1993 * tree during a delete. A deletion from lower in the tree
1994 * could try to delete the only pointer in this node.
1995 * So, pull some keys from the left.
1996 * There has to be a left pointer at this point because
1997 * otherwise we would have pulled some pointers from the
2002 btrfs_handle_fs_error(fs_info, ret, NULL);
2005 wret = balance_node_right(trans, mid, left);
2011 wret = push_node_left(trans, left, mid, 1);
2017 if (btrfs_header_nritems(mid) == 0) {
2018 btrfs_clean_tree_block(mid);
2019 btrfs_tree_unlock(mid);
2020 del_ptr(root, path, level + 1, pslot);
2021 root_sub_used(root, mid->len);
2022 btrfs_free_tree_block(trans, root, mid, 0, 1);
2023 free_extent_buffer_stale(mid);
2026 /* update the parent key to reflect our changes */
2027 struct btrfs_disk_key mid_key;
2028 btrfs_node_key(mid, &mid_key, 0);
2029 ret = tree_mod_log_insert_key(parent, pslot,
2030 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2032 btrfs_set_node_key(parent, &mid_key, pslot);
2033 btrfs_mark_buffer_dirty(parent);
2036 /* update the path */
2038 if (btrfs_header_nritems(left) > orig_slot) {
2039 extent_buffer_get(left);
2040 /* left was locked after cow */
2041 path->nodes[level] = left;
2042 path->slots[level + 1] -= 1;
2043 path->slots[level] = orig_slot;
2045 btrfs_tree_unlock(mid);
2046 free_extent_buffer(mid);
2049 orig_slot -= btrfs_header_nritems(left);
2050 path->slots[level] = orig_slot;
2053 /* double check we haven't messed things up */
2055 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2059 btrfs_tree_unlock(right);
2060 free_extent_buffer(right);
2063 if (path->nodes[level] != left)
2064 btrfs_tree_unlock(left);
2065 free_extent_buffer(left);
2070 /* Node balancing for insertion. Here we only split or push nodes around
2071 * when they are completely full. This is also done top down, so we
2072 * have to be pessimistic.
2074 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2075 struct btrfs_root *root,
2076 struct btrfs_path *path, int level)
2078 struct btrfs_fs_info *fs_info = root->fs_info;
2079 struct extent_buffer *right = NULL;
2080 struct extent_buffer *mid;
2081 struct extent_buffer *left = NULL;
2082 struct extent_buffer *parent = NULL;
2086 int orig_slot = path->slots[level];
2091 mid = path->nodes[level];
2092 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2094 if (level < BTRFS_MAX_LEVEL - 1) {
2095 parent = path->nodes[level + 1];
2096 pslot = path->slots[level + 1];
2102 left = btrfs_read_node_slot(parent, pslot - 1);
2106 /* first, try to make some room in the middle buffer */
2110 btrfs_tree_lock(left);
2111 btrfs_set_lock_blocking_write(left);
2113 left_nr = btrfs_header_nritems(left);
2114 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2117 ret = btrfs_cow_block(trans, root, left, parent,
2122 wret = push_node_left(trans, left, mid, 0);
2128 struct btrfs_disk_key disk_key;
2129 orig_slot += left_nr;
2130 btrfs_node_key(mid, &disk_key, 0);
2131 ret = tree_mod_log_insert_key(parent, pslot,
2132 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2134 btrfs_set_node_key(parent, &disk_key, pslot);
2135 btrfs_mark_buffer_dirty(parent);
2136 if (btrfs_header_nritems(left) > orig_slot) {
2137 path->nodes[level] = left;
2138 path->slots[level + 1] -= 1;
2139 path->slots[level] = orig_slot;
2140 btrfs_tree_unlock(mid);
2141 free_extent_buffer(mid);
2144 btrfs_header_nritems(left);
2145 path->slots[level] = orig_slot;
2146 btrfs_tree_unlock(left);
2147 free_extent_buffer(left);
2151 btrfs_tree_unlock(left);
2152 free_extent_buffer(left);
2154 right = btrfs_read_node_slot(parent, pslot + 1);
2159 * then try to empty the right most buffer into the middle
2164 btrfs_tree_lock(right);
2165 btrfs_set_lock_blocking_write(right);
2167 right_nr = btrfs_header_nritems(right);
2168 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2171 ret = btrfs_cow_block(trans, root, right,
2177 wret = balance_node_right(trans, right, mid);
2183 struct btrfs_disk_key disk_key;
2185 btrfs_node_key(right, &disk_key, 0);
2186 ret = tree_mod_log_insert_key(parent, pslot + 1,
2187 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2189 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2190 btrfs_mark_buffer_dirty(parent);
2192 if (btrfs_header_nritems(mid) <= orig_slot) {
2193 path->nodes[level] = right;
2194 path->slots[level + 1] += 1;
2195 path->slots[level] = orig_slot -
2196 btrfs_header_nritems(mid);
2197 btrfs_tree_unlock(mid);
2198 free_extent_buffer(mid);
2200 btrfs_tree_unlock(right);
2201 free_extent_buffer(right);
2205 btrfs_tree_unlock(right);
2206 free_extent_buffer(right);
2212 * readahead one full node of leaves, finding things that are close
2213 * to the block in 'slot', and triggering ra on them.
2215 static void reada_for_search(struct btrfs_fs_info *fs_info,
2216 struct btrfs_path *path,
2217 int level, int slot, u64 objectid)
2219 struct extent_buffer *node;
2220 struct btrfs_disk_key disk_key;
2225 struct extent_buffer *eb;
2233 if (!path->nodes[level])
2236 node = path->nodes[level];
2238 search = btrfs_node_blockptr(node, slot);
2239 blocksize = fs_info->nodesize;
2240 eb = find_extent_buffer(fs_info, search);
2242 free_extent_buffer(eb);
2248 nritems = btrfs_header_nritems(node);
2252 if (path->reada == READA_BACK) {
2256 } else if (path->reada == READA_FORWARD) {
2261 if (path->reada == READA_BACK && objectid) {
2262 btrfs_node_key(node, &disk_key, nr);
2263 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2266 search = btrfs_node_blockptr(node, nr);
2267 if ((search <= target && target - search <= 65536) ||
2268 (search > target && search - target <= 65536)) {
2269 readahead_tree_block(fs_info, search);
2273 if ((nread > 65536 || nscan > 32))
2278 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2279 struct btrfs_path *path, int level)
2283 struct extent_buffer *parent;
2284 struct extent_buffer *eb;
2289 parent = path->nodes[level + 1];
2293 nritems = btrfs_header_nritems(parent);
2294 slot = path->slots[level + 1];
2297 block1 = btrfs_node_blockptr(parent, slot - 1);
2298 gen = btrfs_node_ptr_generation(parent, slot - 1);
2299 eb = find_extent_buffer(fs_info, block1);
2301 * if we get -eagain from btrfs_buffer_uptodate, we
2302 * don't want to return eagain here. That will loop
2305 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2307 free_extent_buffer(eb);
2309 if (slot + 1 < nritems) {
2310 block2 = btrfs_node_blockptr(parent, slot + 1);
2311 gen = btrfs_node_ptr_generation(parent, slot + 1);
2312 eb = find_extent_buffer(fs_info, block2);
2313 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2315 free_extent_buffer(eb);
2319 readahead_tree_block(fs_info, block1);
2321 readahead_tree_block(fs_info, block2);
2326 * when we walk down the tree, it is usually safe to unlock the higher layers
2327 * in the tree. The exceptions are when our path goes through slot 0, because
2328 * operations on the tree might require changing key pointers higher up in the
2331 * callers might also have set path->keep_locks, which tells this code to keep
2332 * the lock if the path points to the last slot in the block. This is part of
2333 * walking through the tree, and selecting the next slot in the higher block.
2335 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2336 * if lowest_unlock is 1, level 0 won't be unlocked
2338 static noinline void unlock_up(struct btrfs_path *path, int level,
2339 int lowest_unlock, int min_write_lock_level,
2340 int *write_lock_level)
2343 int skip_level = level;
2345 struct extent_buffer *t;
2347 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2348 if (!path->nodes[i])
2350 if (!path->locks[i])
2352 if (!no_skips && path->slots[i] == 0) {
2356 if (!no_skips && path->keep_locks) {
2359 nritems = btrfs_header_nritems(t);
2360 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2365 if (skip_level < i && i >= lowest_unlock)
2369 if (i >= lowest_unlock && i > skip_level) {
2370 btrfs_tree_unlock_rw(t, path->locks[i]);
2372 if (write_lock_level &&
2373 i > min_write_lock_level &&
2374 i <= *write_lock_level) {
2375 *write_lock_level = i - 1;
2382 * This releases any locks held in the path starting at level and
2383 * going all the way up to the root.
2385 * btrfs_search_slot will keep the lock held on higher nodes in a few
2386 * corner cases, such as COW of the block at slot zero in the node. This
2387 * ignores those rules, and it should only be called when there are no
2388 * more updates to be done higher up in the tree.
2390 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2394 if (path->keep_locks)
2397 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2398 if (!path->nodes[i])
2400 if (!path->locks[i])
2402 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2408 * helper function for btrfs_search_slot. The goal is to find a block
2409 * in cache without setting the path to blocking. If we find the block
2410 * we return zero and the path is unchanged.
2412 * If we can't find the block, we set the path blocking and do some
2413 * reada. -EAGAIN is returned and the search must be repeated.
2416 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2417 struct extent_buffer **eb_ret, int level, int slot,
2418 const struct btrfs_key *key)
2420 struct btrfs_fs_info *fs_info = root->fs_info;
2423 struct extent_buffer *b = *eb_ret;
2424 struct extent_buffer *tmp;
2425 struct btrfs_key first_key;
2429 blocknr = btrfs_node_blockptr(b, slot);
2430 gen = btrfs_node_ptr_generation(b, slot);
2431 parent_level = btrfs_header_level(b);
2432 btrfs_node_key_to_cpu(b, &first_key, slot);
2434 tmp = find_extent_buffer(fs_info, blocknr);
2436 /* first we do an atomic uptodate check */
2437 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2439 * Do extra check for first_key, eb can be stale due to
2440 * being cached, read from scrub, or have multiple
2441 * parents (shared tree blocks).
2443 if (btrfs_verify_level_key(tmp,
2444 parent_level - 1, &first_key, gen)) {
2445 free_extent_buffer(tmp);
2452 /* the pages were up to date, but we failed
2453 * the generation number check. Do a full
2454 * read for the generation number that is correct.
2455 * We must do this without dropping locks so
2456 * we can trust our generation number
2458 btrfs_set_path_blocking(p);
2460 /* now we're allowed to do a blocking uptodate check */
2461 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2466 free_extent_buffer(tmp);
2467 btrfs_release_path(p);
2472 * reduce lock contention at high levels
2473 * of the btree by dropping locks before
2474 * we read. Don't release the lock on the current
2475 * level because we need to walk this node to figure
2476 * out which blocks to read.
2478 btrfs_unlock_up_safe(p, level + 1);
2479 btrfs_set_path_blocking(p);
2481 if (p->reada != READA_NONE)
2482 reada_for_search(fs_info, p, level, slot, key->objectid);
2485 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2489 * If the read above didn't mark this buffer up to date,
2490 * it will never end up being up to date. Set ret to EIO now
2491 * and give up so that our caller doesn't loop forever
2494 if (!extent_buffer_uptodate(tmp))
2496 free_extent_buffer(tmp);
2501 btrfs_release_path(p);
2506 * helper function for btrfs_search_slot. This does all of the checks
2507 * for node-level blocks and does any balancing required based on
2510 * If no extra work was required, zero is returned. If we had to
2511 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2515 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2516 struct btrfs_root *root, struct btrfs_path *p,
2517 struct extent_buffer *b, int level, int ins_len,
2518 int *write_lock_level)
2520 struct btrfs_fs_info *fs_info = root->fs_info;
2523 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2524 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2527 if (*write_lock_level < level + 1) {
2528 *write_lock_level = level + 1;
2529 btrfs_release_path(p);
2533 btrfs_set_path_blocking(p);
2534 reada_for_balance(fs_info, p, level);
2535 sret = split_node(trans, root, p, level);
2542 b = p->nodes[level];
2543 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2544 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2547 if (*write_lock_level < level + 1) {
2548 *write_lock_level = level + 1;
2549 btrfs_release_path(p);
2553 btrfs_set_path_blocking(p);
2554 reada_for_balance(fs_info, p, level);
2555 sret = balance_level(trans, root, p, level);
2561 b = p->nodes[level];
2563 btrfs_release_path(p);
2566 BUG_ON(btrfs_header_nritems(b) == 1);
2576 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2577 int level, int *prev_cmp, int *slot)
2579 if (*prev_cmp != 0) {
2580 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2589 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2590 u64 iobjectid, u64 ioff, u8 key_type,
2591 struct btrfs_key *found_key)
2594 struct btrfs_key key;
2595 struct extent_buffer *eb;
2600 key.type = key_type;
2601 key.objectid = iobjectid;
2604 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2608 eb = path->nodes[0];
2609 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2610 ret = btrfs_next_leaf(fs_root, path);
2613 eb = path->nodes[0];
2616 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2617 if (found_key->type != key.type ||
2618 found_key->objectid != key.objectid)
2624 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2625 struct btrfs_path *p,
2626 int write_lock_level)
2628 struct btrfs_fs_info *fs_info = root->fs_info;
2629 struct extent_buffer *b;
2633 /* We try very hard to do read locks on the root */
2634 root_lock = BTRFS_READ_LOCK;
2636 if (p->search_commit_root) {
2638 * The commit roots are read only so we always do read locks,
2639 * and we always must hold the commit_root_sem when doing
2640 * searches on them, the only exception is send where we don't
2641 * want to block transaction commits for a long time, so
2642 * we need to clone the commit root in order to avoid races
2643 * with transaction commits that create a snapshot of one of
2644 * the roots used by a send operation.
2646 if (p->need_commit_sem) {
2647 down_read(&fs_info->commit_root_sem);
2648 b = btrfs_clone_extent_buffer(root->commit_root);
2649 up_read(&fs_info->commit_root_sem);
2651 return ERR_PTR(-ENOMEM);
2654 b = root->commit_root;
2655 extent_buffer_get(b);
2657 level = btrfs_header_level(b);
2659 * Ensure that all callers have set skip_locking when
2660 * p->search_commit_root = 1.
2662 ASSERT(p->skip_locking == 1);
2667 if (p->skip_locking) {
2668 b = btrfs_root_node(root);
2669 level = btrfs_header_level(b);
2674 * If the level is set to maximum, we can skip trying to get the read
2677 if (write_lock_level < BTRFS_MAX_LEVEL) {
2679 * We don't know the level of the root node until we actually
2680 * have it read locked
2682 b = btrfs_read_lock_root_node(root);
2683 level = btrfs_header_level(b);
2684 if (level > write_lock_level)
2687 /* Whoops, must trade for write lock */
2688 btrfs_tree_read_unlock(b);
2689 free_extent_buffer(b);
2692 b = btrfs_lock_root_node(root);
2693 root_lock = BTRFS_WRITE_LOCK;
2695 /* The level might have changed, check again */
2696 level = btrfs_header_level(b);
2699 p->nodes[level] = b;
2700 if (!p->skip_locking)
2701 p->locks[level] = root_lock;
2703 * Callers are responsible for dropping b's references.
2710 * btrfs_search_slot - look for a key in a tree and perform necessary
2711 * modifications to preserve tree invariants.
2713 * @trans: Handle of transaction, used when modifying the tree
2714 * @p: Holds all btree nodes along the search path
2715 * @root: The root node of the tree
2716 * @key: The key we are looking for
2717 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2718 * deletions it's -1. 0 for plain searches
2719 * @cow: boolean should CoW operations be performed. Must always be 1
2720 * when modifying the tree.
2722 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2723 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2725 * If @key is found, 0 is returned and you can find the item in the leaf level
2726 * of the path (level 0)
2728 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2729 * points to the slot where it should be inserted
2731 * If an error is encountered while searching the tree a negative error number
2734 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2735 const struct btrfs_key *key, struct btrfs_path *p,
2736 int ins_len, int cow)
2738 struct extent_buffer *b;
2743 int lowest_unlock = 1;
2744 /* everything at write_lock_level or lower must be write locked */
2745 int write_lock_level = 0;
2746 u8 lowest_level = 0;
2747 int min_write_lock_level;
2750 lowest_level = p->lowest_level;
2751 WARN_ON(lowest_level && ins_len > 0);
2752 WARN_ON(p->nodes[0] != NULL);
2753 BUG_ON(!cow && ins_len);
2758 /* when we are removing items, we might have to go up to level
2759 * two as we update tree pointers Make sure we keep write
2760 * for those levels as well
2762 write_lock_level = 2;
2763 } else if (ins_len > 0) {
2765 * for inserting items, make sure we have a write lock on
2766 * level 1 so we can update keys
2768 write_lock_level = 1;
2772 write_lock_level = -1;
2774 if (cow && (p->keep_locks || p->lowest_level))
2775 write_lock_level = BTRFS_MAX_LEVEL;
2777 min_write_lock_level = write_lock_level;
2781 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2788 level = btrfs_header_level(b);
2791 * setup the path here so we can release it under lock
2792 * contention with the cow code
2795 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2798 * if we don't really need to cow this block
2799 * then we don't want to set the path blocking,
2800 * so we test it here
2802 if (!should_cow_block(trans, root, b)) {
2803 trans->dirty = true;
2808 * must have write locks on this node and the
2811 if (level > write_lock_level ||
2812 (level + 1 > write_lock_level &&
2813 level + 1 < BTRFS_MAX_LEVEL &&
2814 p->nodes[level + 1])) {
2815 write_lock_level = level + 1;
2816 btrfs_release_path(p);
2820 btrfs_set_path_blocking(p);
2822 err = btrfs_cow_block(trans, root, b, NULL, 0,
2825 err = btrfs_cow_block(trans, root, b,
2826 p->nodes[level + 1],
2827 p->slots[level + 1], &b);
2834 p->nodes[level] = b;
2836 * Leave path with blocking locks to avoid massive
2837 * lock context switch, this is made on purpose.
2841 * we have a lock on b and as long as we aren't changing
2842 * the tree, there is no way to for the items in b to change.
2843 * It is safe to drop the lock on our parent before we
2844 * go through the expensive btree search on b.
2846 * If we're inserting or deleting (ins_len != 0), then we might
2847 * be changing slot zero, which may require changing the parent.
2848 * So, we can't drop the lock until after we know which slot
2849 * we're operating on.
2851 if (!ins_len && !p->keep_locks) {
2854 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2855 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2860 ret = key_search(b, key, level, &prev_cmp, &slot);
2866 if (ret && slot > 0) {
2870 p->slots[level] = slot;
2871 err = setup_nodes_for_search(trans, root, p, b, level,
2872 ins_len, &write_lock_level);
2879 b = p->nodes[level];
2880 slot = p->slots[level];
2883 * slot 0 is special, if we change the key
2884 * we have to update the parent pointer
2885 * which means we must have a write lock
2888 if (slot == 0 && ins_len &&
2889 write_lock_level < level + 1) {
2890 write_lock_level = level + 1;
2891 btrfs_release_path(p);
2895 unlock_up(p, level, lowest_unlock,
2896 min_write_lock_level, &write_lock_level);
2898 if (level == lowest_level) {
2904 err = read_block_for_search(root, p, &b, level,
2913 if (!p->skip_locking) {
2914 level = btrfs_header_level(b);
2915 if (level <= write_lock_level) {
2916 if (!btrfs_try_tree_write_lock(b)) {
2917 btrfs_set_path_blocking(p);
2920 p->locks[level] = BTRFS_WRITE_LOCK;
2922 if (!btrfs_tree_read_lock_atomic(b)) {
2923 btrfs_set_path_blocking(p);
2924 btrfs_tree_read_lock(b);
2926 p->locks[level] = BTRFS_READ_LOCK;
2928 p->nodes[level] = b;
2931 p->slots[level] = slot;
2933 btrfs_leaf_free_space(b) < ins_len) {
2934 if (write_lock_level < 1) {
2935 write_lock_level = 1;
2936 btrfs_release_path(p);
2940 btrfs_set_path_blocking(p);
2941 err = split_leaf(trans, root, key,
2942 p, ins_len, ret == 0);
2950 if (!p->search_for_split)
2951 unlock_up(p, level, lowest_unlock,
2952 min_write_lock_level, NULL);
2959 * we don't really know what they plan on doing with the path
2960 * from here on, so for now just mark it as blocking
2962 if (!p->leave_spinning)
2963 btrfs_set_path_blocking(p);
2964 if (ret < 0 && !p->skip_release_on_error)
2965 btrfs_release_path(p);
2970 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2971 * current state of the tree together with the operations recorded in the tree
2972 * modification log to search for the key in a previous version of this tree, as
2973 * denoted by the time_seq parameter.
2975 * Naturally, there is no support for insert, delete or cow operations.
2977 * The resulting path and return value will be set up as if we called
2978 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2980 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2981 struct btrfs_path *p, u64 time_seq)
2983 struct btrfs_fs_info *fs_info = root->fs_info;
2984 struct extent_buffer *b;
2989 int lowest_unlock = 1;
2990 u8 lowest_level = 0;
2993 lowest_level = p->lowest_level;
2994 WARN_ON(p->nodes[0] != NULL);
2996 if (p->search_commit_root) {
2998 return btrfs_search_slot(NULL, root, key, p, 0, 0);
3002 b = get_old_root(root, time_seq);
3007 level = btrfs_header_level(b);
3008 p->locks[level] = BTRFS_READ_LOCK;
3011 level = btrfs_header_level(b);
3012 p->nodes[level] = b;
3015 * we have a lock on b and as long as we aren't changing
3016 * the tree, there is no way to for the items in b to change.
3017 * It is safe to drop the lock on our parent before we
3018 * go through the expensive btree search on b.
3020 btrfs_unlock_up_safe(p, level + 1);
3023 * Since we can unwind ebs we want to do a real search every
3027 ret = key_search(b, key, level, &prev_cmp, &slot);
3033 if (ret && slot > 0) {
3037 p->slots[level] = slot;
3038 unlock_up(p, level, lowest_unlock, 0, NULL);
3040 if (level == lowest_level) {
3046 err = read_block_for_search(root, p, &b, level,
3055 level = btrfs_header_level(b);
3056 if (!btrfs_tree_read_lock_atomic(b)) {
3057 btrfs_set_path_blocking(p);
3058 btrfs_tree_read_lock(b);
3060 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3065 p->locks[level] = BTRFS_READ_LOCK;
3066 p->nodes[level] = b;
3068 p->slots[level] = slot;
3069 unlock_up(p, level, lowest_unlock, 0, NULL);
3075 if (!p->leave_spinning)
3076 btrfs_set_path_blocking(p);
3078 btrfs_release_path(p);
3084 * helper to use instead of search slot if no exact match is needed but
3085 * instead the next or previous item should be returned.
3086 * When find_higher is true, the next higher item is returned, the next lower
3088 * When return_any and find_higher are both true, and no higher item is found,
3089 * return the next lower instead.
3090 * When return_any is true and find_higher is false, and no lower item is found,
3091 * return the next higher instead.
3092 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3095 int btrfs_search_slot_for_read(struct btrfs_root *root,
3096 const struct btrfs_key *key,
3097 struct btrfs_path *p, int find_higher,
3101 struct extent_buffer *leaf;
3104 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3108 * a return value of 1 means the path is at the position where the
3109 * item should be inserted. Normally this is the next bigger item,
3110 * but in case the previous item is the last in a leaf, path points
3111 * to the first free slot in the previous leaf, i.e. at an invalid
3117 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3118 ret = btrfs_next_leaf(root, p);
3124 * no higher item found, return the next
3129 btrfs_release_path(p);
3133 if (p->slots[0] == 0) {
3134 ret = btrfs_prev_leaf(root, p);
3139 if (p->slots[0] == btrfs_header_nritems(leaf))
3146 * no lower item found, return the next
3151 btrfs_release_path(p);
3161 * adjust the pointers going up the tree, starting at level
3162 * making sure the right key of each node is points to 'key'.
3163 * This is used after shifting pointers to the left, so it stops
3164 * fixing up pointers when a given leaf/node is not in slot 0 of the
3168 static void fixup_low_keys(struct btrfs_path *path,
3169 struct btrfs_disk_key *key, int level)
3172 struct extent_buffer *t;
3175 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3176 int tslot = path->slots[i];
3178 if (!path->nodes[i])
3181 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3184 btrfs_set_node_key(t, key, tslot);
3185 btrfs_mark_buffer_dirty(path->nodes[i]);
3194 * This function isn't completely safe. It's the caller's responsibility
3195 * that the new key won't break the order
3197 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3198 struct btrfs_path *path,
3199 const struct btrfs_key *new_key)
3201 struct btrfs_disk_key disk_key;
3202 struct extent_buffer *eb;
3205 eb = path->nodes[0];
3206 slot = path->slots[0];
3208 btrfs_item_key(eb, &disk_key, slot - 1);
3209 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
3211 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3212 slot, btrfs_disk_key_objectid(&disk_key),
3213 btrfs_disk_key_type(&disk_key),
3214 btrfs_disk_key_offset(&disk_key),
3215 new_key->objectid, new_key->type,
3217 btrfs_print_leaf(eb);
3221 if (slot < btrfs_header_nritems(eb) - 1) {
3222 btrfs_item_key(eb, &disk_key, slot + 1);
3223 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
3225 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3226 slot, btrfs_disk_key_objectid(&disk_key),
3227 btrfs_disk_key_type(&disk_key),
3228 btrfs_disk_key_offset(&disk_key),
3229 new_key->objectid, new_key->type,
3231 btrfs_print_leaf(eb);
3236 btrfs_cpu_key_to_disk(&disk_key, new_key);
3237 btrfs_set_item_key(eb, &disk_key, slot);
3238 btrfs_mark_buffer_dirty(eb);
3240 fixup_low_keys(path, &disk_key, 1);
3244 * try to push data from one node into the next node left in the
3247 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3248 * error, and > 0 if there was no room in the left hand block.
3250 static int push_node_left(struct btrfs_trans_handle *trans,
3251 struct extent_buffer *dst,
3252 struct extent_buffer *src, int empty)
3254 struct btrfs_fs_info *fs_info = trans->fs_info;
3260 src_nritems = btrfs_header_nritems(src);
3261 dst_nritems = btrfs_header_nritems(dst);
3262 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3263 WARN_ON(btrfs_header_generation(src) != trans->transid);
3264 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3266 if (!empty && src_nritems <= 8)
3269 if (push_items <= 0)
3273 push_items = min(src_nritems, push_items);
3274 if (push_items < src_nritems) {
3275 /* leave at least 8 pointers in the node if
3276 * we aren't going to empty it
3278 if (src_nritems - push_items < 8) {
3279 if (push_items <= 8)
3285 push_items = min(src_nritems - 8, push_items);
3287 ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3289 btrfs_abort_transaction(trans, ret);
3292 copy_extent_buffer(dst, src,
3293 btrfs_node_key_ptr_offset(dst_nritems),
3294 btrfs_node_key_ptr_offset(0),
3295 push_items * sizeof(struct btrfs_key_ptr));
3297 if (push_items < src_nritems) {
3299 * Don't call tree_mod_log_insert_move here, key removal was
3300 * already fully logged by tree_mod_log_eb_copy above.
3302 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3303 btrfs_node_key_ptr_offset(push_items),
3304 (src_nritems - push_items) *
3305 sizeof(struct btrfs_key_ptr));
3307 btrfs_set_header_nritems(src, src_nritems - push_items);
3308 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3309 btrfs_mark_buffer_dirty(src);
3310 btrfs_mark_buffer_dirty(dst);
3316 * try to push data from one node into the next node right in the
3319 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3320 * error, and > 0 if there was no room in the right hand block.
3322 * this will only push up to 1/2 the contents of the left node over
3324 static int balance_node_right(struct btrfs_trans_handle *trans,
3325 struct extent_buffer *dst,
3326 struct extent_buffer *src)
3328 struct btrfs_fs_info *fs_info = trans->fs_info;
3335 WARN_ON(btrfs_header_generation(src) != trans->transid);
3336 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3338 src_nritems = btrfs_header_nritems(src);
3339 dst_nritems = btrfs_header_nritems(dst);
3340 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3341 if (push_items <= 0)
3344 if (src_nritems < 4)
3347 max_push = src_nritems / 2 + 1;
3348 /* don't try to empty the node */
3349 if (max_push >= src_nritems)
3352 if (max_push < push_items)
3353 push_items = max_push;
3355 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3357 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3358 btrfs_node_key_ptr_offset(0),
3360 sizeof(struct btrfs_key_ptr));
3362 ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3365 btrfs_abort_transaction(trans, ret);
3368 copy_extent_buffer(dst, src,
3369 btrfs_node_key_ptr_offset(0),
3370 btrfs_node_key_ptr_offset(src_nritems - push_items),
3371 push_items * sizeof(struct btrfs_key_ptr));
3373 btrfs_set_header_nritems(src, src_nritems - push_items);
3374 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3376 btrfs_mark_buffer_dirty(src);
3377 btrfs_mark_buffer_dirty(dst);
3383 * helper function to insert a new root level in the tree.
3384 * A new node is allocated, and a single item is inserted to
3385 * point to the existing root
3387 * returns zero on success or < 0 on failure.
3389 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3390 struct btrfs_root *root,
3391 struct btrfs_path *path, int level)
3393 struct btrfs_fs_info *fs_info = root->fs_info;
3395 struct extent_buffer *lower;
3396 struct extent_buffer *c;
3397 struct extent_buffer *old;
3398 struct btrfs_disk_key lower_key;
3401 BUG_ON(path->nodes[level]);
3402 BUG_ON(path->nodes[level-1] != root->node);
3404 lower = path->nodes[level-1];
3406 btrfs_item_key(lower, &lower_key, 0);
3408 btrfs_node_key(lower, &lower_key, 0);
3410 c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3411 root->node->start, 0);
3415 root_add_used(root, fs_info->nodesize);
3417 btrfs_set_header_nritems(c, 1);
3418 btrfs_set_node_key(c, &lower_key, 0);
3419 btrfs_set_node_blockptr(c, 0, lower->start);
3420 lower_gen = btrfs_header_generation(lower);
3421 WARN_ON(lower_gen != trans->transid);
3423 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3425 btrfs_mark_buffer_dirty(c);
3428 ret = tree_mod_log_insert_root(root->node, c, 0);
3430 rcu_assign_pointer(root->node, c);
3432 /* the super has an extra ref to root->node */
3433 free_extent_buffer(old);
3435 add_root_to_dirty_list(root);
3436 extent_buffer_get(c);
3437 path->nodes[level] = c;
3438 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3439 path->slots[level] = 0;
3444 * worker function to insert a single pointer in a node.
3445 * the node should have enough room for the pointer already
3447 * slot and level indicate where you want the key to go, and
3448 * blocknr is the block the key points to.
3450 static void insert_ptr(struct btrfs_trans_handle *trans,
3451 struct btrfs_path *path,
3452 struct btrfs_disk_key *key, u64 bytenr,
3453 int slot, int level)
3455 struct extent_buffer *lower;
3459 BUG_ON(!path->nodes[level]);
3460 btrfs_assert_tree_locked(path->nodes[level]);
3461 lower = path->nodes[level];
3462 nritems = btrfs_header_nritems(lower);
3463 BUG_ON(slot > nritems);
3464 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3465 if (slot != nritems) {
3467 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3471 memmove_extent_buffer(lower,
3472 btrfs_node_key_ptr_offset(slot + 1),
3473 btrfs_node_key_ptr_offset(slot),
3474 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3477 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3481 btrfs_set_node_key(lower, key, slot);
3482 btrfs_set_node_blockptr(lower, slot, bytenr);
3483 WARN_ON(trans->transid == 0);
3484 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3485 btrfs_set_header_nritems(lower, nritems + 1);
3486 btrfs_mark_buffer_dirty(lower);
3490 * split the node at the specified level in path in two.
3491 * The path is corrected to point to the appropriate node after the split
3493 * Before splitting this tries to make some room in the node by pushing
3494 * left and right, if either one works, it returns right away.
3496 * returns 0 on success and < 0 on failure
3498 static noinline int split_node(struct btrfs_trans_handle *trans,
3499 struct btrfs_root *root,
3500 struct btrfs_path *path, int level)
3502 struct btrfs_fs_info *fs_info = root->fs_info;
3503 struct extent_buffer *c;
3504 struct extent_buffer *split;
3505 struct btrfs_disk_key disk_key;
3510 c = path->nodes[level];
3511 WARN_ON(btrfs_header_generation(c) != trans->transid);
3512 if (c == root->node) {
3514 * trying to split the root, lets make a new one
3516 * tree mod log: We don't log_removal old root in
3517 * insert_new_root, because that root buffer will be kept as a
3518 * normal node. We are going to log removal of half of the
3519 * elements below with tree_mod_log_eb_copy. We're holding a
3520 * tree lock on the buffer, which is why we cannot race with
3521 * other tree_mod_log users.
3523 ret = insert_new_root(trans, root, path, level + 1);
3527 ret = push_nodes_for_insert(trans, root, path, level);
3528 c = path->nodes[level];
3529 if (!ret && btrfs_header_nritems(c) <
3530 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3536 c_nritems = btrfs_header_nritems(c);
3537 mid = (c_nritems + 1) / 2;
3538 btrfs_node_key(c, &disk_key, mid);
3540 split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3543 return PTR_ERR(split);
3545 root_add_used(root, fs_info->nodesize);
3546 ASSERT(btrfs_header_level(c) == level);
3548 ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3550 btrfs_abort_transaction(trans, ret);
3553 copy_extent_buffer(split, c,
3554 btrfs_node_key_ptr_offset(0),
3555 btrfs_node_key_ptr_offset(mid),
3556 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3557 btrfs_set_header_nritems(split, c_nritems - mid);
3558 btrfs_set_header_nritems(c, mid);
3561 btrfs_mark_buffer_dirty(c);
3562 btrfs_mark_buffer_dirty(split);
3564 insert_ptr(trans, path, &disk_key, split->start,
3565 path->slots[level + 1] + 1, level + 1);
3567 if (path->slots[level] >= mid) {
3568 path->slots[level] -= mid;
3569 btrfs_tree_unlock(c);
3570 free_extent_buffer(c);
3571 path->nodes[level] = split;
3572 path->slots[level + 1] += 1;
3574 btrfs_tree_unlock(split);
3575 free_extent_buffer(split);
3581 * how many bytes are required to store the items in a leaf. start
3582 * and nr indicate which items in the leaf to check. This totals up the
3583 * space used both by the item structs and the item data
3585 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3587 struct btrfs_item *start_item;
3588 struct btrfs_item *end_item;
3589 struct btrfs_map_token token;
3591 int nritems = btrfs_header_nritems(l);
3592 int end = min(nritems, start + nr) - 1;
3596 btrfs_init_map_token(&token, l);
3597 start_item = btrfs_item_nr(start);
3598 end_item = btrfs_item_nr(end);
3599 data_len = btrfs_token_item_offset(l, start_item, &token) +
3600 btrfs_token_item_size(l, start_item, &token);
3601 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3602 data_len += sizeof(struct btrfs_item) * nr;
3603 WARN_ON(data_len < 0);
3608 * The space between the end of the leaf items and
3609 * the start of the leaf data. IOW, how much room
3610 * the leaf has left for both items and data
3612 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3614 struct btrfs_fs_info *fs_info = leaf->fs_info;
3615 int nritems = btrfs_header_nritems(leaf);
3618 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3621 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3623 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3624 leaf_space_used(leaf, 0, nritems), nritems);
3630 * min slot controls the lowest index we're willing to push to the
3631 * right. We'll push up to and including min_slot, but no lower
3633 static noinline int __push_leaf_right(struct btrfs_path *path,
3634 int data_size, int empty,
3635 struct extent_buffer *right,
3636 int free_space, u32 left_nritems,
3639 struct btrfs_fs_info *fs_info = right->fs_info;
3640 struct extent_buffer *left = path->nodes[0];
3641 struct extent_buffer *upper = path->nodes[1];
3642 struct btrfs_map_token token;
3643 struct btrfs_disk_key disk_key;
3648 struct btrfs_item *item;
3657 nr = max_t(u32, 1, min_slot);
3659 if (path->slots[0] >= left_nritems)
3660 push_space += data_size;
3662 slot = path->slots[1];
3663 i = left_nritems - 1;
3665 item = btrfs_item_nr(i);
3667 if (!empty && push_items > 0) {
3668 if (path->slots[0] > i)
3670 if (path->slots[0] == i) {
3671 int space = btrfs_leaf_free_space(left);
3673 if (space + push_space * 2 > free_space)
3678 if (path->slots[0] == i)
3679 push_space += data_size;
3681 this_item_size = btrfs_item_size(left, item);
3682 if (this_item_size + sizeof(*item) + push_space > free_space)
3686 push_space += this_item_size + sizeof(*item);
3692 if (push_items == 0)
3695 WARN_ON(!empty && push_items == left_nritems);
3697 /* push left to right */
3698 right_nritems = btrfs_header_nritems(right);
3700 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3701 push_space -= leaf_data_end(left);
3703 /* make room in the right data area */
3704 data_end = leaf_data_end(right);
3705 memmove_extent_buffer(right,
3706 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3707 BTRFS_LEAF_DATA_OFFSET + data_end,
3708 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3710 /* copy from the left data area */
3711 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3712 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3713 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3716 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3717 btrfs_item_nr_offset(0),
3718 right_nritems * sizeof(struct btrfs_item));
3720 /* copy the items from left to right */
3721 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3722 btrfs_item_nr_offset(left_nritems - push_items),
3723 push_items * sizeof(struct btrfs_item));
3725 /* update the item pointers */
3726 btrfs_init_map_token(&token, right);
3727 right_nritems += push_items;
3728 btrfs_set_header_nritems(right, right_nritems);
3729 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3730 for (i = 0; i < right_nritems; i++) {
3731 item = btrfs_item_nr(i);
3732 push_space -= btrfs_token_item_size(right, item, &token);
3733 btrfs_set_token_item_offset(right, item, push_space, &token);
3736 left_nritems -= push_items;
3737 btrfs_set_header_nritems(left, left_nritems);
3740 btrfs_mark_buffer_dirty(left);
3742 btrfs_clean_tree_block(left);
3744 btrfs_mark_buffer_dirty(right);
3746 btrfs_item_key(right, &disk_key, 0);
3747 btrfs_set_node_key(upper, &disk_key, slot + 1);
3748 btrfs_mark_buffer_dirty(upper);
3750 /* then fixup the leaf pointer in the path */
3751 if (path->slots[0] >= left_nritems) {
3752 path->slots[0] -= left_nritems;
3753 if (btrfs_header_nritems(path->nodes[0]) == 0)
3754 btrfs_clean_tree_block(path->nodes[0]);
3755 btrfs_tree_unlock(path->nodes[0]);
3756 free_extent_buffer(path->nodes[0]);
3757 path->nodes[0] = right;
3758 path->slots[1] += 1;
3760 btrfs_tree_unlock(right);
3761 free_extent_buffer(right);
3766 btrfs_tree_unlock(right);
3767 free_extent_buffer(right);
3772 * push some data in the path leaf to the right, trying to free up at
3773 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3775 * returns 1 if the push failed because the other node didn't have enough
3776 * room, 0 if everything worked out and < 0 if there were major errors.
3778 * this will push starting from min_slot to the end of the leaf. It won't
3779 * push any slot lower than min_slot
3781 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3782 *root, struct btrfs_path *path,
3783 int min_data_size, int data_size,
3784 int empty, u32 min_slot)
3786 struct extent_buffer *left = path->nodes[0];
3787 struct extent_buffer *right;
3788 struct extent_buffer *upper;
3794 if (!path->nodes[1])
3797 slot = path->slots[1];
3798 upper = path->nodes[1];
3799 if (slot >= btrfs_header_nritems(upper) - 1)
3802 btrfs_assert_tree_locked(path->nodes[1]);
3804 right = btrfs_read_node_slot(upper, slot + 1);
3806 * slot + 1 is not valid or we fail to read the right node,
3807 * no big deal, just return.
3812 btrfs_tree_lock(right);
3813 btrfs_set_lock_blocking_write(right);
3815 free_space = btrfs_leaf_free_space(right);
3816 if (free_space < data_size)
3819 /* cow and double check */
3820 ret = btrfs_cow_block(trans, root, right, upper,
3825 free_space = btrfs_leaf_free_space(right);
3826 if (free_space < data_size)
3829 left_nritems = btrfs_header_nritems(left);
3830 if (left_nritems == 0)
3833 if (path->slots[0] == left_nritems && !empty) {
3834 /* Key greater than all keys in the leaf, right neighbor has
3835 * enough room for it and we're not emptying our leaf to delete
3836 * it, therefore use right neighbor to insert the new item and
3837 * no need to touch/dirty our left leaf. */
3838 btrfs_tree_unlock(left);
3839 free_extent_buffer(left);
3840 path->nodes[0] = right;
3846 return __push_leaf_right(path, min_data_size, empty,
3847 right, free_space, left_nritems, min_slot);
3849 btrfs_tree_unlock(right);
3850 free_extent_buffer(right);
3855 * push some data in the path leaf to the left, trying to free up at
3856 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3858 * max_slot can put a limit on how far into the leaf we'll push items. The
3859 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3862 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3863 int empty, struct extent_buffer *left,
3864 int free_space, u32 right_nritems,
3867 struct btrfs_fs_info *fs_info = left->fs_info;
3868 struct btrfs_disk_key disk_key;
3869 struct extent_buffer *right = path->nodes[0];
3873 struct btrfs_item *item;
3874 u32 old_left_nritems;
3878 u32 old_left_item_size;
3879 struct btrfs_map_token token;
3882 nr = min(right_nritems, max_slot);
3884 nr = min(right_nritems - 1, max_slot);
3886 for (i = 0; i < nr; i++) {
3887 item = btrfs_item_nr(i);
3889 if (!empty && push_items > 0) {
3890 if (path->slots[0] < i)
3892 if (path->slots[0] == i) {
3893 int space = btrfs_leaf_free_space(right);
3895 if (space + push_space * 2 > free_space)
3900 if (path->slots[0] == i)
3901 push_space += data_size;
3903 this_item_size = btrfs_item_size(right, item);
3904 if (this_item_size + sizeof(*item) + push_space > free_space)
3908 push_space += this_item_size + sizeof(*item);
3911 if (push_items == 0) {
3915 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3917 /* push data from right to left */
3918 copy_extent_buffer(left, right,
3919 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3920 btrfs_item_nr_offset(0),
3921 push_items * sizeof(struct btrfs_item));
3923 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3924 btrfs_item_offset_nr(right, push_items - 1);
3926 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3927 leaf_data_end(left) - push_space,
3928 BTRFS_LEAF_DATA_OFFSET +
3929 btrfs_item_offset_nr(right, push_items - 1),
3931 old_left_nritems = btrfs_header_nritems(left);
3932 BUG_ON(old_left_nritems <= 0);
3934 btrfs_init_map_token(&token, left);
3935 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3936 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3939 item = btrfs_item_nr(i);
3941 ioff = btrfs_token_item_offset(left, item, &token);
3942 btrfs_set_token_item_offset(left, item,
3943 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3946 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3948 /* fixup right node */
3949 if (push_items > right_nritems)
3950 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3953 if (push_items < right_nritems) {
3954 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3955 leaf_data_end(right);
3956 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3957 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3958 BTRFS_LEAF_DATA_OFFSET +
3959 leaf_data_end(right), push_space);
3961 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3962 btrfs_item_nr_offset(push_items),
3963 (btrfs_header_nritems(right) - push_items) *
3964 sizeof(struct btrfs_item));
3967 btrfs_init_map_token(&token, right);
3968 right_nritems -= push_items;
3969 btrfs_set_header_nritems(right, right_nritems);
3970 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3971 for (i = 0; i < right_nritems; i++) {
3972 item = btrfs_item_nr(i);
3974 push_space = push_space - btrfs_token_item_size(right,
3976 btrfs_set_token_item_offset(right, item, push_space, &token);
3979 btrfs_mark_buffer_dirty(left);
3981 btrfs_mark_buffer_dirty(right);
3983 btrfs_clean_tree_block(right);
3985 btrfs_item_key(right, &disk_key, 0);
3986 fixup_low_keys(path, &disk_key, 1);
3988 /* then fixup the leaf pointer in the path */
3989 if (path->slots[0] < push_items) {
3990 path->slots[0] += old_left_nritems;
3991 btrfs_tree_unlock(path->nodes[0]);
3992 free_extent_buffer(path->nodes[0]);
3993 path->nodes[0] = left;
3994 path->slots[1] -= 1;
3996 btrfs_tree_unlock(left);
3997 free_extent_buffer(left);
3998 path->slots[0] -= push_items;
4000 BUG_ON(path->slots[0] < 0);
4003 btrfs_tree_unlock(left);
4004 free_extent_buffer(left);
4009 * push some data in the path leaf to the left, trying to free up at
4010 * least data_size bytes. returns zero if the push worked, nonzero otherwise
4012 * max_slot can put a limit on how far into the leaf we'll push items. The
4013 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
4016 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
4017 *root, struct btrfs_path *path, int min_data_size,
4018 int data_size, int empty, u32 max_slot)
4020 struct extent_buffer *right = path->nodes[0];
4021 struct extent_buffer *left;
4027 slot = path->slots[1];
4030 if (!path->nodes[1])
4033 right_nritems = btrfs_header_nritems(right);
4034 if (right_nritems == 0)
4037 btrfs_assert_tree_locked(path->nodes[1]);
4039 left = btrfs_read_node_slot(path->nodes[1], slot - 1);
4041 * slot - 1 is not valid or we fail to read the left node,
4042 * no big deal, just return.
4047 btrfs_tree_lock(left);
4048 btrfs_set_lock_blocking_write(left);
4050 free_space = btrfs_leaf_free_space(left);
4051 if (free_space < data_size) {
4056 /* cow and double check */
4057 ret = btrfs_cow_block(trans, root, left,
4058 path->nodes[1], slot - 1, &left);
4060 /* we hit -ENOSPC, but it isn't fatal here */
4066 free_space = btrfs_leaf_free_space(left);
4067 if (free_space < data_size) {
4072 return __push_leaf_left(path, min_data_size,
4073 empty, left, free_space, right_nritems,
4076 btrfs_tree_unlock(left);
4077 free_extent_buffer(left);
4082 * split the path's leaf in two, making sure there is at least data_size
4083 * available for the resulting leaf level of the path.
4085 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4086 struct btrfs_path *path,
4087 struct extent_buffer *l,
4088 struct extent_buffer *right,
4089 int slot, int mid, int nritems)
4091 struct btrfs_fs_info *fs_info = trans->fs_info;
4095 struct btrfs_disk_key disk_key;
4096 struct btrfs_map_token token;
4098 nritems = nritems - mid;
4099 btrfs_set_header_nritems(right, nritems);
4100 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4102 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4103 btrfs_item_nr_offset(mid),
4104 nritems * sizeof(struct btrfs_item));
4106 copy_extent_buffer(right, l,
4107 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4108 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4109 leaf_data_end(l), data_copy_size);
4111 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4113 btrfs_init_map_token(&token, right);
4114 for (i = 0; i < nritems; i++) {
4115 struct btrfs_item *item = btrfs_item_nr(i);
4118 ioff = btrfs_token_item_offset(right, item, &token);
4119 btrfs_set_token_item_offset(right, item,
4120 ioff + rt_data_off, &token);
4123 btrfs_set_header_nritems(l, mid);
4124 btrfs_item_key(right, &disk_key, 0);
4125 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4127 btrfs_mark_buffer_dirty(right);
4128 btrfs_mark_buffer_dirty(l);
4129 BUG_ON(path->slots[0] != slot);
4132 btrfs_tree_unlock(path->nodes[0]);
4133 free_extent_buffer(path->nodes[0]);
4134 path->nodes[0] = right;
4135 path->slots[0] -= mid;
4136 path->slots[1] += 1;
4138 btrfs_tree_unlock(right);
4139 free_extent_buffer(right);
4142 BUG_ON(path->slots[0] < 0);
4146 * double splits happen when we need to insert a big item in the middle
4147 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4148 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4151 * We avoid this by trying to push the items on either side of our target
4152 * into the adjacent leaves. If all goes well we can avoid the double split
4155 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4156 struct btrfs_root *root,
4157 struct btrfs_path *path,
4164 int space_needed = data_size;
4166 slot = path->slots[0];
4167 if (slot < btrfs_header_nritems(path->nodes[0]))
4168 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4171 * try to push all the items after our slot into the
4174 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4181 nritems = btrfs_header_nritems(path->nodes[0]);
4183 * our goal is to get our slot at the start or end of a leaf. If
4184 * we've done so we're done
4186 if (path->slots[0] == 0 || path->slots[0] == nritems)
4189 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4192 /* try to push all the items before our slot into the next leaf */
4193 slot = path->slots[0];
4194 space_needed = data_size;
4196 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4197 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4210 * split the path's leaf in two, making sure there is at least data_size
4211 * available for the resulting leaf level of the path.
4213 * returns 0 if all went well and < 0 on failure.
4215 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4216 struct btrfs_root *root,
4217 const struct btrfs_key *ins_key,
4218 struct btrfs_path *path, int data_size,
4221 struct btrfs_disk_key disk_key;
4222 struct extent_buffer *l;
4226 struct extent_buffer *right;
4227 struct btrfs_fs_info *fs_info = root->fs_info;
4231 int num_doubles = 0;
4232 int tried_avoid_double = 0;
4235 slot = path->slots[0];
4236 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4237 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4240 /* first try to make some room by pushing left and right */
4241 if (data_size && path->nodes[1]) {
4242 int space_needed = data_size;
4244 if (slot < btrfs_header_nritems(l))
4245 space_needed -= btrfs_leaf_free_space(l);
4247 wret = push_leaf_right(trans, root, path, space_needed,
4248 space_needed, 0, 0);
4252 space_needed = data_size;
4254 space_needed -= btrfs_leaf_free_space(l);
4255 wret = push_leaf_left(trans, root, path, space_needed,
4256 space_needed, 0, (u32)-1);
4262 /* did the pushes work? */
4263 if (btrfs_leaf_free_space(l) >= data_size)
4267 if (!path->nodes[1]) {
4268 ret = insert_new_root(trans, root, path, 1);
4275 slot = path->slots[0];
4276 nritems = btrfs_header_nritems(l);
4277 mid = (nritems + 1) / 2;
4281 leaf_space_used(l, mid, nritems - mid) + data_size >
4282 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4283 if (slot >= nritems) {
4287 if (mid != nritems &&
4288 leaf_space_used(l, mid, nritems - mid) +
4289 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4290 if (data_size && !tried_avoid_double)
4291 goto push_for_double;
4297 if (leaf_space_used(l, 0, mid) + data_size >
4298 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4299 if (!extend && data_size && slot == 0) {
4301 } else if ((extend || !data_size) && slot == 0) {
4305 if (mid != nritems &&
4306 leaf_space_used(l, mid, nritems - mid) +
4307 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4308 if (data_size && !tried_avoid_double)
4309 goto push_for_double;
4317 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4319 btrfs_item_key(l, &disk_key, mid);
4321 right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4324 return PTR_ERR(right);
4326 root_add_used(root, fs_info->nodesize);
4330 btrfs_set_header_nritems(right, 0);
4331 insert_ptr(trans, path, &disk_key,
4332 right->start, path->slots[1] + 1, 1);
4333 btrfs_tree_unlock(path->nodes[0]);
4334 free_extent_buffer(path->nodes[0]);
4335 path->nodes[0] = right;
4337 path->slots[1] += 1;
4339 btrfs_set_header_nritems(right, 0);
4340 insert_ptr(trans, path, &disk_key,
4341 right->start, path->slots[1], 1);
4342 btrfs_tree_unlock(path->nodes[0]);
4343 free_extent_buffer(path->nodes[0]);
4344 path->nodes[0] = right;
4346 if (path->slots[1] == 0)
4347 fixup_low_keys(path, &disk_key, 1);
4350 * We create a new leaf 'right' for the required ins_len and
4351 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4352 * the content of ins_len to 'right'.
4357 copy_for_split(trans, path, l, right, slot, mid, nritems);
4360 BUG_ON(num_doubles != 0);
4368 push_for_double_split(trans, root, path, data_size);
4369 tried_avoid_double = 1;
4370 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4375 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root,
4377 struct btrfs_path *path, int ins_len)
4379 struct btrfs_key key;
4380 struct extent_buffer *leaf;
4381 struct btrfs_file_extent_item *fi;
4386 leaf = path->nodes[0];
4387 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4389 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4390 key.type != BTRFS_EXTENT_CSUM_KEY);
4392 if (btrfs_leaf_free_space(leaf) >= ins_len)
4395 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4396 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4397 fi = btrfs_item_ptr(leaf, path->slots[0],
4398 struct btrfs_file_extent_item);
4399 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4401 btrfs_release_path(path);
4403 path->keep_locks = 1;
4404 path->search_for_split = 1;
4405 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4406 path->search_for_split = 0;
4413 leaf = path->nodes[0];
4414 /* if our item isn't there, return now */
4415 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4418 /* the leaf has changed, it now has room. return now */
4419 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4422 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4423 fi = btrfs_item_ptr(leaf, path->slots[0],
4424 struct btrfs_file_extent_item);
4425 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4429 btrfs_set_path_blocking(path);
4430 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4434 path->keep_locks = 0;
4435 btrfs_unlock_up_safe(path, 1);
4438 path->keep_locks = 0;
4442 static noinline int split_item(struct btrfs_path *path,
4443 const struct btrfs_key *new_key,
4444 unsigned long split_offset)
4446 struct extent_buffer *leaf;
4447 struct btrfs_item *item;
4448 struct btrfs_item *new_item;
4454 struct btrfs_disk_key disk_key;
4456 leaf = path->nodes[0];
4457 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4459 btrfs_set_path_blocking(path);
4461 item = btrfs_item_nr(path->slots[0]);
4462 orig_offset = btrfs_item_offset(leaf, item);
4463 item_size = btrfs_item_size(leaf, item);
4465 buf = kmalloc(item_size, GFP_NOFS);
4469 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4470 path->slots[0]), item_size);
4472 slot = path->slots[0] + 1;
4473 nritems = btrfs_header_nritems(leaf);
4474 if (slot != nritems) {
4475 /* shift the items */
4476 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4477 btrfs_item_nr_offset(slot),
4478 (nritems - slot) * sizeof(struct btrfs_item));
4481 btrfs_cpu_key_to_disk(&disk_key, new_key);
4482 btrfs_set_item_key(leaf, &disk_key, slot);
4484 new_item = btrfs_item_nr(slot);
4486 btrfs_set_item_offset(leaf, new_item, orig_offset);
4487 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4489 btrfs_set_item_offset(leaf, item,
4490 orig_offset + item_size - split_offset);
4491 btrfs_set_item_size(leaf, item, split_offset);
4493 btrfs_set_header_nritems(leaf, nritems + 1);
4495 /* write the data for the start of the original item */
4496 write_extent_buffer(leaf, buf,
4497 btrfs_item_ptr_offset(leaf, path->slots[0]),
4500 /* write the data for the new item */
4501 write_extent_buffer(leaf, buf + split_offset,
4502 btrfs_item_ptr_offset(leaf, slot),
4503 item_size - split_offset);
4504 btrfs_mark_buffer_dirty(leaf);
4506 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4512 * This function splits a single item into two items,
4513 * giving 'new_key' to the new item and splitting the
4514 * old one at split_offset (from the start of the item).
4516 * The path may be released by this operation. After
4517 * the split, the path is pointing to the old item. The
4518 * new item is going to be in the same node as the old one.
4520 * Note, the item being split must be smaller enough to live alone on
4521 * a tree block with room for one extra struct btrfs_item
4523 * This allows us to split the item in place, keeping a lock on the
4524 * leaf the entire time.
4526 int btrfs_split_item(struct btrfs_trans_handle *trans,
4527 struct btrfs_root *root,
4528 struct btrfs_path *path,
4529 const struct btrfs_key *new_key,
4530 unsigned long split_offset)
4533 ret = setup_leaf_for_split(trans, root, path,
4534 sizeof(struct btrfs_item));
4538 ret = split_item(path, new_key, split_offset);
4543 * This function duplicate a item, giving 'new_key' to the new item.
4544 * It guarantees both items live in the same tree leaf and the new item
4545 * is contiguous with the original item.
4547 * This allows us to split file extent in place, keeping a lock on the
4548 * leaf the entire time.
4550 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4551 struct btrfs_root *root,
4552 struct btrfs_path *path,
4553 const struct btrfs_key *new_key)
4555 struct extent_buffer *leaf;
4559 leaf = path->nodes[0];
4560 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4561 ret = setup_leaf_for_split(trans, root, path,
4562 item_size + sizeof(struct btrfs_item));
4567 setup_items_for_insert(root, path, new_key, &item_size,
4568 item_size, item_size +
4569 sizeof(struct btrfs_item), 1);
4570 leaf = path->nodes[0];
4571 memcpy_extent_buffer(leaf,
4572 btrfs_item_ptr_offset(leaf, path->slots[0]),
4573 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4579 * make the item pointed to by the path smaller. new_size indicates
4580 * how small to make it, and from_end tells us if we just chop bytes
4581 * off the end of the item or if we shift the item to chop bytes off
4584 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4587 struct extent_buffer *leaf;
4588 struct btrfs_item *item;
4590 unsigned int data_end;
4591 unsigned int old_data_start;
4592 unsigned int old_size;
4593 unsigned int size_diff;
4595 struct btrfs_map_token token;
4597 leaf = path->nodes[0];
4598 slot = path->slots[0];
4600 old_size = btrfs_item_size_nr(leaf, slot);
4601 if (old_size == new_size)
4604 nritems = btrfs_header_nritems(leaf);
4605 data_end = leaf_data_end(leaf);
4607 old_data_start = btrfs_item_offset_nr(leaf, slot);
4609 size_diff = old_size - new_size;
4612 BUG_ON(slot >= nritems);
4615 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4617 /* first correct the data pointers */
4618 btrfs_init_map_token(&token, leaf);
4619 for (i = slot; i < nritems; i++) {
4621 item = btrfs_item_nr(i);
4623 ioff = btrfs_token_item_offset(leaf, item, &token);
4624 btrfs_set_token_item_offset(leaf, item,
4625 ioff + size_diff, &token);
4628 /* shift the data */
4630 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4631 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4632 data_end, old_data_start + new_size - data_end);
4634 struct btrfs_disk_key disk_key;
4637 btrfs_item_key(leaf, &disk_key, slot);
4639 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4641 struct btrfs_file_extent_item *fi;
4643 fi = btrfs_item_ptr(leaf, slot,
4644 struct btrfs_file_extent_item);
4645 fi = (struct btrfs_file_extent_item *)(
4646 (unsigned long)fi - size_diff);
4648 if (btrfs_file_extent_type(leaf, fi) ==
4649 BTRFS_FILE_EXTENT_INLINE) {
4650 ptr = btrfs_item_ptr_offset(leaf, slot);
4651 memmove_extent_buffer(leaf, ptr,
4653 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4657 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4658 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4659 data_end, old_data_start - data_end);
4661 offset = btrfs_disk_key_offset(&disk_key);
4662 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4663 btrfs_set_item_key(leaf, &disk_key, slot);
4665 fixup_low_keys(path, &disk_key, 1);
4668 item = btrfs_item_nr(slot);
4669 btrfs_set_item_size(leaf, item, new_size);
4670 btrfs_mark_buffer_dirty(leaf);
4672 if (btrfs_leaf_free_space(leaf) < 0) {
4673 btrfs_print_leaf(leaf);
4679 * make the item pointed to by the path bigger, data_size is the added size.
4681 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4684 struct extent_buffer *leaf;
4685 struct btrfs_item *item;
4687 unsigned int data_end;
4688 unsigned int old_data;
4689 unsigned int old_size;
4691 struct btrfs_map_token token;
4693 leaf = path->nodes[0];
4695 nritems = btrfs_header_nritems(leaf);
4696 data_end = leaf_data_end(leaf);
4698 if (btrfs_leaf_free_space(leaf) < data_size) {
4699 btrfs_print_leaf(leaf);
4702 slot = path->slots[0];
4703 old_data = btrfs_item_end_nr(leaf, slot);
4706 if (slot >= nritems) {
4707 btrfs_print_leaf(leaf);
4708 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4714 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4716 /* first correct the data pointers */
4717 btrfs_init_map_token(&token, leaf);
4718 for (i = slot; i < nritems; i++) {
4720 item = btrfs_item_nr(i);
4722 ioff = btrfs_token_item_offset(leaf, item, &token);
4723 btrfs_set_token_item_offset(leaf, item,
4724 ioff - data_size, &token);
4727 /* shift the data */
4728 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4729 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4730 data_end, old_data - data_end);
4732 data_end = old_data;
4733 old_size = btrfs_item_size_nr(leaf, slot);
4734 item = btrfs_item_nr(slot);
4735 btrfs_set_item_size(leaf, item, old_size + data_size);
4736 btrfs_mark_buffer_dirty(leaf);
4738 if (btrfs_leaf_free_space(leaf) < 0) {
4739 btrfs_print_leaf(leaf);
4745 * this is a helper for btrfs_insert_empty_items, the main goal here is
4746 * to save stack depth by doing the bulk of the work in a function
4747 * that doesn't call btrfs_search_slot
4749 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4750 const struct btrfs_key *cpu_key, u32 *data_size,
4751 u32 total_data, u32 total_size, int nr)
4753 struct btrfs_fs_info *fs_info = root->fs_info;
4754 struct btrfs_item *item;
4757 unsigned int data_end;
4758 struct btrfs_disk_key disk_key;
4759 struct extent_buffer *leaf;
4761 struct btrfs_map_token token;
4763 if (path->slots[0] == 0) {
4764 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4765 fixup_low_keys(path, &disk_key, 1);
4767 btrfs_unlock_up_safe(path, 1);
4769 leaf = path->nodes[0];
4770 slot = path->slots[0];
4772 nritems = btrfs_header_nritems(leaf);
4773 data_end = leaf_data_end(leaf);
4775 if (btrfs_leaf_free_space(leaf) < total_size) {
4776 btrfs_print_leaf(leaf);
4777 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4778 total_size, btrfs_leaf_free_space(leaf));
4782 btrfs_init_map_token(&token, leaf);
4783 if (slot != nritems) {
4784 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4786 if (old_data < data_end) {
4787 btrfs_print_leaf(leaf);
4788 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4789 slot, old_data, data_end);
4793 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4795 /* first correct the data pointers */
4796 for (i = slot; i < nritems; i++) {
4799 item = btrfs_item_nr(i);
4800 ioff = btrfs_token_item_offset(leaf, item, &token);
4801 btrfs_set_token_item_offset(leaf, item,
4802 ioff - total_data, &token);
4804 /* shift the items */
4805 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4806 btrfs_item_nr_offset(slot),
4807 (nritems - slot) * sizeof(struct btrfs_item));
4809 /* shift the data */
4810 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4811 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4812 data_end, old_data - data_end);
4813 data_end = old_data;
4816 /* setup the item for the new data */
4817 for (i = 0; i < nr; i++) {
4818 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4819 btrfs_set_item_key(leaf, &disk_key, slot + i);
4820 item = btrfs_item_nr(slot + i);
4821 btrfs_set_token_item_offset(leaf, item,
4822 data_end - data_size[i], &token);
4823 data_end -= data_size[i];
4824 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4827 btrfs_set_header_nritems(leaf, nritems + nr);
4828 btrfs_mark_buffer_dirty(leaf);
4830 if (btrfs_leaf_free_space(leaf) < 0) {
4831 btrfs_print_leaf(leaf);
4837 * Given a key and some data, insert items into the tree.
4838 * This does all the path init required, making room in the tree if needed.
4840 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4841 struct btrfs_root *root,
4842 struct btrfs_path *path,
4843 const struct btrfs_key *cpu_key, u32 *data_size,
4852 for (i = 0; i < nr; i++)
4853 total_data += data_size[i];
4855 total_size = total_data + (nr * sizeof(struct btrfs_item));
4856 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4862 slot = path->slots[0];
4865 setup_items_for_insert(root, path, cpu_key, data_size,
4866 total_data, total_size, nr);
4871 * Given a key and some data, insert an item into the tree.
4872 * This does all the path init required, making room in the tree if needed.
4874 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4875 const struct btrfs_key *cpu_key, void *data,
4879 struct btrfs_path *path;
4880 struct extent_buffer *leaf;
4883 path = btrfs_alloc_path();
4886 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4888 leaf = path->nodes[0];
4889 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4890 write_extent_buffer(leaf, data, ptr, data_size);
4891 btrfs_mark_buffer_dirty(leaf);
4893 btrfs_free_path(path);
4898 * delete the pointer from a given node.
4900 * the tree should have been previously balanced so the deletion does not
4903 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4904 int level, int slot)
4906 struct extent_buffer *parent = path->nodes[level];
4910 nritems = btrfs_header_nritems(parent);
4911 if (slot != nritems - 1) {
4913 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4914 nritems - slot - 1);
4917 memmove_extent_buffer(parent,
4918 btrfs_node_key_ptr_offset(slot),
4919 btrfs_node_key_ptr_offset(slot + 1),
4920 sizeof(struct btrfs_key_ptr) *
4921 (nritems - slot - 1));
4923 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4929 btrfs_set_header_nritems(parent, nritems);
4930 if (nritems == 0 && parent == root->node) {
4931 BUG_ON(btrfs_header_level(root->node) != 1);
4932 /* just turn the root into a leaf and break */
4933 btrfs_set_header_level(root->node, 0);
4934 } else if (slot == 0) {
4935 struct btrfs_disk_key disk_key;
4937 btrfs_node_key(parent, &disk_key, 0);
4938 fixup_low_keys(path, &disk_key, level + 1);
4940 btrfs_mark_buffer_dirty(parent);
4944 * a helper function to delete the leaf pointed to by path->slots[1] and
4947 * This deletes the pointer in path->nodes[1] and frees the leaf
4948 * block extent. zero is returned if it all worked out, < 0 otherwise.
4950 * The path must have already been setup for deleting the leaf, including
4951 * all the proper balancing. path->nodes[1] must be locked.
4953 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4954 struct btrfs_root *root,
4955 struct btrfs_path *path,
4956 struct extent_buffer *leaf)
4958 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4959 del_ptr(root, path, 1, path->slots[1]);
4962 * btrfs_free_extent is expensive, we want to make sure we
4963 * aren't holding any locks when we call it
4965 btrfs_unlock_up_safe(path, 0);
4967 root_sub_used(root, leaf->len);
4969 extent_buffer_get(leaf);
4970 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4971 free_extent_buffer_stale(leaf);
4974 * delete the item at the leaf level in path. If that empties
4975 * the leaf, remove it from the tree
4977 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4978 struct btrfs_path *path, int slot, int nr)
4980 struct btrfs_fs_info *fs_info = root->fs_info;
4981 struct extent_buffer *leaf;
4982 struct btrfs_item *item;
4990 leaf = path->nodes[0];
4991 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4993 for (i = 0; i < nr; i++)
4994 dsize += btrfs_item_size_nr(leaf, slot + i);
4996 nritems = btrfs_header_nritems(leaf);
4998 if (slot + nr != nritems) {
4999 int data_end = leaf_data_end(leaf);
5000 struct btrfs_map_token token;
5002 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
5004 BTRFS_LEAF_DATA_OFFSET + data_end,
5005 last_off - data_end);
5007 btrfs_init_map_token(&token, leaf);
5008 for (i = slot + nr; i < nritems; i++) {
5011 item = btrfs_item_nr(i);
5012 ioff = btrfs_token_item_offset(leaf, item, &token);
5013 btrfs_set_token_item_offset(leaf, item,
5014 ioff + dsize, &token);
5017 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
5018 btrfs_item_nr_offset(slot + nr),
5019 sizeof(struct btrfs_item) *
5020 (nritems - slot - nr));
5022 btrfs_set_header_nritems(leaf, nritems - nr);
5025 /* delete the leaf if we've emptied it */
5027 if (leaf == root->node) {
5028 btrfs_set_header_level(leaf, 0);
5030 btrfs_set_path_blocking(path);
5031 btrfs_clean_tree_block(leaf);
5032 btrfs_del_leaf(trans, root, path, leaf);
5035 int used = leaf_space_used(leaf, 0, nritems);
5037 struct btrfs_disk_key disk_key;
5039 btrfs_item_key(leaf, &disk_key, 0);
5040 fixup_low_keys(path, &disk_key, 1);
5043 /* delete the leaf if it is mostly empty */
5044 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5045 /* push_leaf_left fixes the path.
5046 * make sure the path still points to our leaf
5047 * for possible call to del_ptr below
5049 slot = path->slots[1];
5050 extent_buffer_get(leaf);
5052 btrfs_set_path_blocking(path);
5053 wret = push_leaf_left(trans, root, path, 1, 1,
5055 if (wret < 0 && wret != -ENOSPC)
5058 if (path->nodes[0] == leaf &&
5059 btrfs_header_nritems(leaf)) {
5060 wret = push_leaf_right(trans, root, path, 1,
5062 if (wret < 0 && wret != -ENOSPC)
5066 if (btrfs_header_nritems(leaf) == 0) {
5067 path->slots[1] = slot;
5068 btrfs_del_leaf(trans, root, path, leaf);
5069 free_extent_buffer(leaf);
5072 /* if we're still in the path, make sure
5073 * we're dirty. Otherwise, one of the
5074 * push_leaf functions must have already
5075 * dirtied this buffer
5077 if (path->nodes[0] == leaf)
5078 btrfs_mark_buffer_dirty(leaf);
5079 free_extent_buffer(leaf);
5082 btrfs_mark_buffer_dirty(leaf);
5089 * search the tree again to find a leaf with lesser keys
5090 * returns 0 if it found something or 1 if there are no lesser leaves.
5091 * returns < 0 on io errors.
5093 * This may release the path, and so you may lose any locks held at the
5096 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5098 struct btrfs_key key;
5099 struct btrfs_disk_key found_key;
5102 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5104 if (key.offset > 0) {
5106 } else if (key.type > 0) {
5108 key.offset = (u64)-1;
5109 } else if (key.objectid > 0) {
5112 key.offset = (u64)-1;
5117 btrfs_release_path(path);
5118 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5121 btrfs_item_key(path->nodes[0], &found_key, 0);
5122 ret = comp_keys(&found_key, &key);
5124 * We might have had an item with the previous key in the tree right
5125 * before we released our path. And after we released our path, that
5126 * item might have been pushed to the first slot (0) of the leaf we
5127 * were holding due to a tree balance. Alternatively, an item with the
5128 * previous key can exist as the only element of a leaf (big fat item).
5129 * Therefore account for these 2 cases, so that our callers (like
5130 * btrfs_previous_item) don't miss an existing item with a key matching
5131 * the previous key we computed above.
5139 * A helper function to walk down the tree starting at min_key, and looking
5140 * for nodes or leaves that are have a minimum transaction id.
5141 * This is used by the btree defrag code, and tree logging
5143 * This does not cow, but it does stuff the starting key it finds back
5144 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5145 * key and get a writable path.
5147 * This honors path->lowest_level to prevent descent past a given level
5150 * min_trans indicates the oldest transaction that you are interested
5151 * in walking through. Any nodes or leaves older than min_trans are
5152 * skipped over (without reading them).
5154 * returns zero if something useful was found, < 0 on error and 1 if there
5155 * was nothing in the tree that matched the search criteria.
5157 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5158 struct btrfs_path *path,
5161 struct extent_buffer *cur;
5162 struct btrfs_key found_key;
5168 int keep_locks = path->keep_locks;
5170 path->keep_locks = 1;
5172 cur = btrfs_read_lock_root_node(root);
5173 level = btrfs_header_level(cur);
5174 WARN_ON(path->nodes[level]);
5175 path->nodes[level] = cur;
5176 path->locks[level] = BTRFS_READ_LOCK;
5178 if (btrfs_header_generation(cur) < min_trans) {
5183 nritems = btrfs_header_nritems(cur);
5184 level = btrfs_header_level(cur);
5185 sret = btrfs_bin_search(cur, min_key, level, &slot);
5191 /* at the lowest level, we're done, setup the path and exit */
5192 if (level == path->lowest_level) {
5193 if (slot >= nritems)
5196 path->slots[level] = slot;
5197 btrfs_item_key_to_cpu(cur, &found_key, slot);
5200 if (sret && slot > 0)
5203 * check this node pointer against the min_trans parameters.
5204 * If it is too old, old, skip to the next one.
5206 while (slot < nritems) {
5209 gen = btrfs_node_ptr_generation(cur, slot);
5210 if (gen < min_trans) {
5218 * we didn't find a candidate key in this node, walk forward
5219 * and find another one
5221 if (slot >= nritems) {
5222 path->slots[level] = slot;
5223 btrfs_set_path_blocking(path);
5224 sret = btrfs_find_next_key(root, path, min_key, level,
5227 btrfs_release_path(path);
5233 /* save our key for returning back */
5234 btrfs_node_key_to_cpu(cur, &found_key, slot);
5235 path->slots[level] = slot;
5236 if (level == path->lowest_level) {
5240 btrfs_set_path_blocking(path);
5241 cur = btrfs_read_node_slot(cur, slot);
5247 btrfs_tree_read_lock(cur);
5249 path->locks[level - 1] = BTRFS_READ_LOCK;
5250 path->nodes[level - 1] = cur;
5251 unlock_up(path, level, 1, 0, NULL);
5254 path->keep_locks = keep_locks;
5256 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5257 btrfs_set_path_blocking(path);
5258 memcpy(min_key, &found_key, sizeof(found_key));
5264 * this is similar to btrfs_next_leaf, but does not try to preserve
5265 * and fixup the path. It looks for and returns the next key in the
5266 * tree based on the current path and the min_trans parameters.
5268 * 0 is returned if another key is found, < 0 if there are any errors
5269 * and 1 is returned if there are no higher keys in the tree
5271 * path->keep_locks should be set to 1 on the search made before
5272 * calling this function.
5274 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5275 struct btrfs_key *key, int level, u64 min_trans)
5278 struct extent_buffer *c;
5280 WARN_ON(!path->keep_locks && !path->skip_locking);
5281 while (level < BTRFS_MAX_LEVEL) {
5282 if (!path->nodes[level])
5285 slot = path->slots[level] + 1;
5286 c = path->nodes[level];
5288 if (slot >= btrfs_header_nritems(c)) {
5291 struct btrfs_key cur_key;
5292 if (level + 1 >= BTRFS_MAX_LEVEL ||
5293 !path->nodes[level + 1])
5296 if (path->locks[level + 1] || path->skip_locking) {
5301 slot = btrfs_header_nritems(c) - 1;
5303 btrfs_item_key_to_cpu(c, &cur_key, slot);
5305 btrfs_node_key_to_cpu(c, &cur_key, slot);
5307 orig_lowest = path->lowest_level;
5308 btrfs_release_path(path);
5309 path->lowest_level = level;
5310 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5312 path->lowest_level = orig_lowest;
5316 c = path->nodes[level];
5317 slot = path->slots[level];
5324 btrfs_item_key_to_cpu(c, key, slot);
5326 u64 gen = btrfs_node_ptr_generation(c, slot);
5328 if (gen < min_trans) {
5332 btrfs_node_key_to_cpu(c, key, slot);
5340 * search the tree again to find a leaf with greater keys
5341 * returns 0 if it found something or 1 if there are no greater leaves.
5342 * returns < 0 on io errors.
5344 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5346 return btrfs_next_old_leaf(root, path, 0);
5349 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5354 struct extent_buffer *c;
5355 struct extent_buffer *next;
5356 struct btrfs_key key;
5359 int old_spinning = path->leave_spinning;
5360 int next_rw_lock = 0;
5362 nritems = btrfs_header_nritems(path->nodes[0]);
5366 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5371 btrfs_release_path(path);
5373 path->keep_locks = 1;
5374 path->leave_spinning = 1;
5377 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5379 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5380 path->keep_locks = 0;
5385 nritems = btrfs_header_nritems(path->nodes[0]);
5387 * by releasing the path above we dropped all our locks. A balance
5388 * could have added more items next to the key that used to be
5389 * at the very end of the block. So, check again here and
5390 * advance the path if there are now more items available.
5392 if (nritems > 0 && path->slots[0] < nritems - 1) {
5399 * So the above check misses one case:
5400 * - after releasing the path above, someone has removed the item that
5401 * used to be at the very end of the block, and balance between leafs
5402 * gets another one with bigger key.offset to replace it.
5404 * This one should be returned as well, or we can get leaf corruption
5405 * later(esp. in __btrfs_drop_extents()).
5407 * And a bit more explanation about this check,
5408 * with ret > 0, the key isn't found, the path points to the slot
5409 * where it should be inserted, so the path->slots[0] item must be the
5412 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5417 while (level < BTRFS_MAX_LEVEL) {
5418 if (!path->nodes[level]) {
5423 slot = path->slots[level] + 1;
5424 c = path->nodes[level];
5425 if (slot >= btrfs_header_nritems(c)) {
5427 if (level == BTRFS_MAX_LEVEL) {
5435 btrfs_tree_unlock_rw(next, next_rw_lock);
5436 free_extent_buffer(next);
5440 next_rw_lock = path->locks[level];
5441 ret = read_block_for_search(root, path, &next, level,
5447 btrfs_release_path(path);
5451 if (!path->skip_locking) {
5452 ret = btrfs_try_tree_read_lock(next);
5453 if (!ret && time_seq) {
5455 * If we don't get the lock, we may be racing
5456 * with push_leaf_left, holding that lock while
5457 * itself waiting for the leaf we've currently
5458 * locked. To solve this situation, we give up
5459 * on our lock and cycle.
5461 free_extent_buffer(next);
5462 btrfs_release_path(path);
5467 btrfs_set_path_blocking(path);
5468 btrfs_tree_read_lock(next);
5470 next_rw_lock = BTRFS_READ_LOCK;
5474 path->slots[level] = slot;
5477 c = path->nodes[level];
5478 if (path->locks[level])
5479 btrfs_tree_unlock_rw(c, path->locks[level]);
5481 free_extent_buffer(c);
5482 path->nodes[level] = next;
5483 path->slots[level] = 0;
5484 if (!path->skip_locking)
5485 path->locks[level] = next_rw_lock;
5489 ret = read_block_for_search(root, path, &next, level,
5495 btrfs_release_path(path);
5499 if (!path->skip_locking) {
5500 ret = btrfs_try_tree_read_lock(next);
5502 btrfs_set_path_blocking(path);
5503 btrfs_tree_read_lock(next);
5505 next_rw_lock = BTRFS_READ_LOCK;
5510 unlock_up(path, 0, 1, 0, NULL);
5511 path->leave_spinning = old_spinning;
5513 btrfs_set_path_blocking(path);
5519 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5520 * searching until it gets past min_objectid or finds an item of 'type'
5522 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5524 int btrfs_previous_item(struct btrfs_root *root,
5525 struct btrfs_path *path, u64 min_objectid,
5528 struct btrfs_key found_key;
5529 struct extent_buffer *leaf;
5534 if (path->slots[0] == 0) {
5535 btrfs_set_path_blocking(path);
5536 ret = btrfs_prev_leaf(root, path);
5542 leaf = path->nodes[0];
5543 nritems = btrfs_header_nritems(leaf);
5546 if (path->slots[0] == nritems)
5549 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5550 if (found_key.objectid < min_objectid)
5552 if (found_key.type == type)
5554 if (found_key.objectid == min_objectid &&
5555 found_key.type < type)
5562 * search in extent tree to find a previous Metadata/Data extent item with
5565 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5567 int btrfs_previous_extent_item(struct btrfs_root *root,
5568 struct btrfs_path *path, u64 min_objectid)
5570 struct btrfs_key found_key;
5571 struct extent_buffer *leaf;
5576 if (path->slots[0] == 0) {
5577 btrfs_set_path_blocking(path);
5578 ret = btrfs_prev_leaf(root, path);
5584 leaf = path->nodes[0];
5585 nritems = btrfs_header_nritems(leaf);
5588 if (path->slots[0] == nritems)
5591 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5592 if (found_key.objectid < min_objectid)
5594 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5595 found_key.type == BTRFS_METADATA_ITEM_KEY)
5597 if (found_key.objectid == min_objectid &&
5598 found_key.type < BTRFS_EXTENT_ITEM_KEY)